Datasets:

Hendrik-a

/

arxiv

like 0

we have combined the large snia database of the canada - france - hawaii telescope supernova legacy survey and catalogs of galaxies with photometric redshifts , vla 1.4 ghz radio sources , and spitzer infrared sources . we present eight sneia in early - type host galaxies which have counterparts in the radio and infrared source catalogs . we find the snia rate in subsets of radio and infrared early - type galaxies is times the rate in all early - type galaxies , and that any enhancement is always . rates in these subsets are consistent with predictions of the two component `` a+b '' snia rate model . since infrared properties of radio snia hosts indicate dust obscured star formation , we incorporate infrared star formation rates into the `` a+b '' model . we also show the properties of sneia in radio and infrared galaxies suggest the hosts contain dust and support a continuum of delay time distributions for sneia , although other delay time distributions can not be ruled out based on our data .

0912.1646

type ia supernovae ( sneia ) are currently the best cosmological tool to investigate the dark energy driving the accelerated expansion of the universe @xcite . though generally accepted as thermonuclear explosions of carbon - oxygen white dwarfs which have accreted sufficient mass from their companion , two competitive models exist for the companion scenario : the single - degenerate ( main sequence or red giant companion ) and double - degenerate ( another white dwarf ) . since the time delay between star formation and sneia explosion is scenario - dependent , rates in hosts of varying star formation rate ( sfr ) help constrain this . in fact , snia rates are known to be correlated with host morphology , b - k color , and by inference also with star formation @xcite . as such , a galaxy s snia rate is commonly expressed as the sum of a `` delayed '' component from old stellar populations and a `` prompt '' component from young stellar populations . these components are parametrized as `` a '' and `` b '' , proportional to a galaxy s mass and sfr respectively @xcite . this `` a+b '' two - component model can be matched with progenitor populations that have distributions of delay times as described by mannucci et al . ( 2006 ) , and pritchet et al . ( 2008 ) . the c99 catalog matched with very large array ( vla ) 1.4ghz radio data revealed the specific snia rate may be enhanced by 27 times in radio - loud ( @xmath3 ) over radio - quiet ( @xmath4 ) early - type galaxies ( della valle et al . 2005 , hereafter dv05 ) . although winds from active galactic nuclei ( agn ) could increase accretion rates of the interstellar medium ( ism ) onto white dwarfs , a process thought to trigger classical nova eruptions @xcite , dv05 reject this explanation for the enhanced specific snia rate in radio - loud galaxies . they find galaxy interactions are most likely to cause the radio emission and supply the necessary snia progenitors via stellar capture during dwarf accretion or star formation induced by major mergers @xcite . the existence of bright infrared counterparts for radio galaxies has been well documented and often attributed to dust obscured star formation coeval with the agn @xcite . by matching a hypothetical galaxy interaction and agn activity timeline with a recurring star formation model of ten @xmath5 year long episodes each separated by @xmath6 years , mannucci et al . ( 2006 ) find the enhanced specific snia rate in radio - loud early - type galaxies is best fit by a bimodal delay - time distribution ( dtd ) in which the `` prompt '' ( b ) delay time is constrained to just @xmath7 years . they suggest this implies two physical populations of snia progenitors , but note a broad single - population dtd could not be ruled out . as this controversial implication relies on a rate enhancement found with 21 photometrically identified sneia from dv05 , we look to confirm this in the large database of spectroscopically typed sneia from the canada - france hawaii telescope ( cfht ) supernova legacy survey ( snls ) . we use photometric redshift galaxy catalogs @xcite and 1.4 ghz radio catalogs to calculate the snia rate in radio - loud early - type galaxies , and also infrared source catalogs to look for obscured star formation in the radio - loud snia host galaxies . dust extinction in starbursts hinders sn detection : only one snia has been detected , and the snia rate remains unconstrained in starburst galaxies @xcite . we use the infrared catalogs to calculate the snia rate in bright and luminous infrared galaxies , which are known to experience bursts of star formation up to @xmath8 . we also statistically compare the snia rates in radio and infrared galaxies to predictions of the two - component `` a+b '' model , and perform an identical analysis with the low redshift c99 catalog . in [ s2obs ] we describe the snls and c99 snia databases and galaxy catalogs , the overlapping radio and infrared source catalogs , and present the radio and infrared properties of snls and c99 snia early - type host galaxies . we derive the snia rate in radio and infrared galaxies in [ srates ] , compare to expectations of the two - component `` a+b '' model in [ sabp ] , and reanalyze with relaxed data constraints in [ srelax ] . in [ ssnprop ] we juxtapose the properties of the sneia in radio and infrared host galaxies with known correlations between snia properties and stellar populations . finally , in [ sdisc ] we discuss the implications of our results regarding current snia science and future surveys , and conclude in [ sconc ] . here we describe the combination and processing of private and public catalogs of supernovae , galaxies , and radio and infrared sources . from 2003 to 2008 , cfhtls deep component imaged four 1 deg@xmath9 fields ( d1d4 ) every three to four nights ( during dark / grey time , when visible ) in four megacam filters ( @xmath10 ) to a depth @xmath11 ; this was accompanied by a strong spectroscopic campaign to follow up on as many potential snia candidates as possible @xcite . after final reductions of all data , the snls will have discovered and identified hundreds of sneia , provide the best direct constraints on the dark energy equation of state parameter @xmath12 @xcite and be a useful compliment to other cosmological techniques @xcite . we use the 91 and 64 sneia identified prior to 2006 dec 31 in cfhtls deep fields 1 and 2 respectively ( @xmath13 of the final four - field total ) , because suitable vla radio surveys cover d1 and d2 only , and up to this date the snls spectroscopic analysis and typing is complete . the ilbert et al . ( 2006 ) galaxy catalog incorporates vimos vlt deep survey spectroscopic redshifts to calibrate their spectral energy distribution ( sed ) fitting routine , resulting in accurate ( @xmath14 ) photometric redshifts for galaxies in the four snls deep fields . in optimizing the photo - z calculation , the accuracies of the seds are compromised ( ilbert , private communication ) and the distribution of sed types is discontinuous . to solve this we fit 51 seds , interpolated from coleman et al . ( 1980 ) and kinney et al . ( 1996 ) templates made by stephen gwyn at the canadian astronomical data centre , to the catalog galaxies , and estimate galaxy stellar masses and star formation rates using fits of this library of seds to the models of buzzoni ( 2005 ) . we correct these for systematic offsets ( of about a factor of 2 ) to agree with the pegase models @xcite . our sed types include e / s0 ( ellipticals or early - types ) , sbc and scd ( spiral galaxies ) , irr ( irregular galaxies ) , and sb ( starbursts , @xmath15 ) . to ensure catalog purity we restrict the catalog as recommended by ilbert et al . ( 2006 ) , and limit to @xmath16 . we identify snia hosts as the nearest ilbert catalog galaxy unless there are two within 2 arcseconds , in which case redshift is used to discriminate . in d1 and d2 , 22 and 15 sneia have no catalog galaxy within 5 arcseconds ( maximum host offset ) and can not be used ; this mainly includes sneia in hosts which do not meet our catalog restrictions ( 17 and 12 ) , but also a few snia whose hosts are not in the original galaxy catalgs ( 5 and 3 ; many are near field edges , or close to foreground stars ) . for sneia with ilbert catalog hosts , iterative outlier rejection is applied to the residual dispersion between host photometric and snia spectroscopic redshifts for each deep field , resulting in photometric redshift uncertainties of @xmath17 and @xmath18 , rejecting 2 and 6 sneia hosts as outliers in d1 and d2 . we use snia with spectroscopic redshifts @xmath19 , the limit to which the snls snia sample is nearly complete @xcite . these restrictions yield 33 and 29 usable sneia in d1 and d2 respectively ; possible effects of the redshift restriction are considered in [ srelax ] . for every galaxy we calculate its snia observed rate per year , @xmath20 , from the two - component `` a+b '' model applying the time dilation correction to our observed frame of reference , as shown in equation [ eab ] : @xmath21 where @xmath22 is stellar mass ( @xmath23 ) , @xmath24 is star formation rate ( @xmath23 / yr ) , and the a and b values are from sullivan et al . ( 2006 ) : @xmath25 and @xmath26 . the rate @xmath20 is the number of sneia expected in a galaxy per year of observing . we use it to determine an effective control time , @xmath27 , to account for snls survey efficiencies such as detection and spectroscopic completeness , and the lengths of snls observing seasons ( approximately 6 months ) , for fields d1 and d2 separately ( i.e. to have @xmath28 and @xmath29 ) as follows : @xmath30 where @xmath31 is the total number of snls sneia observed in the field prior to 2006 dec 31 with @xmath32 , and @xmath33 is the total number of sneia expected in all @xmath32 galaxies per year of observing the field . we find @xmath34 and @xmath35 where @xmath27 has units of years . the vla - virmos 1.4 ghz deep survey covers d1 to a @xmath36 flux of @xmath37 , with a mean rms noise @xmath38 , and 75% completeness at fluxes @xmath3980180@xmath40 @xcite . of the 1054 radio sources , 503 optical counterparts are identified as the closest galaxy within 2 arcseconds ( maximum vla positional error ) . away from the galaxy catalog s masked regions around foreground stars our recovery rate is @xmath41 , similar to ciliegi et al . the vla - cosmos 1.4 ghz large project covers d2 to @xmath42 with a mean rms noise of @xmath43 @xcite , for which we match @xmath44 of radio sources with a galaxy . radio luminosities are derived from photometric redshifts of galaxy counterparts , include a @xmath45 bandpass correction @xcite , and are plotted in figure [ fs2rlumvsz ] . a galaxy is radio - loud if @xmath3 , the faint - end limit of the radio luminosity function , although exact limits for radio - loud , -faint , and -quiet galaxies are not universal @xcite . ledlow & owen ( 1996 ) find @xmath46 ( @xmath47 ) of elliptical galaxies with absolute r magnitude @xmath48 are radio - loud , and amongst c99 elliptical galaxies dv05 find @xmath49 ( @xmath50 ) are radio - loud . in our sample of sed type e / s0 galaxies with @xmath32 we find 48% of galaxies with @xmath51 are radio - loud , suggesting we can identify half of the radio - loud population amongst optically bright galaxies due to our higher minimum radio flux limit . the apparent bimodality of vla - cosmos ( d2 ) radio luminosities in figure [ fs2rlumvsz ] is a result of combining the integrated fluxes of resolved and unresolved sources . the flux of the bimodality valley corresponds to the lower limit of integrated fluxes for resolved sources , @xmath52 , as shown in the right - hand plot of figure 17 from schinnerer et al . coincidentally , it also corresponds to the vla - virmos ( d1 ) lower limit . this may indicate vla - virmos and vla - cosmos sample slightly different radio source populations , with more faint , unresolved sources in d2 . we consider any influence of this on our results in [ srelax ] . the spitzer wide - area infrared extragalactic ( swire ) survey covers d1 to a flux of @xmath53 ( where subscript on s denotes wavelength ) , and to @xmath54 @xcite . swire objects with @xmath55 and stellarity @xmath56 are most likely stars or qsos and rejected from the catalog . the spitzer cosmic evolution survey ( s - cosmos ) covers d2 to @xmath57 and @xmath58 after required aperture corrections @xcite . objects flagged as likely compromised by nearby bright sources are rejected . both catalogs are available via the nasa infrared space archive . we reject foreground objects and qsos , identify optical counterparts as in [ ss2radio ] , and limit to @xmath19 as in [ ss2radio ] . the fraction of galaxies infrared counterparts in all four irac bands _ and _ mips is @xmath59 ; these are plotted on infrared color - color diagrams in figure [ fs2ircc ] which also show agn boundaries derived from spitzer first look survey data of comparable flux - density limits ( @xmath60 and @xmath61 ) and redshifts ( @xmath62 ) @xcite . sajina et al . ( 2005 ) show the region on the irac color - color plot ( left ) is actually appropriate for agn of redshifts @xmath63 . it should be noted that not all galaxies in these regions are agn - dominated at ir wavelengths . we convert @xmath64 to k - band stellar mass via balogh et al . ( 2007 ) , and @xmath65 to bolometric infrared luminosity @xmath66 and infrared star formation rate @xmath67 with publicly available templates and codes @xcite . plots of @xmath66 versus redshift in figure [ fs2irlumvsz ] show our sample of luminous infrared galaxies ( lirg , @xmath68 ) is complete to @xmath69 . the conversion of @xmath66 to @xmath67 is appropriate for starburst - dominated infrared emission , but ir color - color diagrams in figure [ fs2ircc ] show some ir sources may be agn - dominated ; uncertainties introduced by applying this conversion to potentially agn - dominated sources are discussed in [ ssabpir ] . figure [ fs2optircomp ] shows the optical and infrared masses agree very well , but @xmath70 is 310 times lower than @xmath67 . having sampled the brightest infrared galaxies which are most likely to have obscured star formation , this factor is and should be higher than the usual @xmath71 @xcite . we find eight snls snia host galaxies with radio and/or infrared counterparts and @xmath19 , the properties of which are given in table [ ts2rlessn ] and plotted in figures [ fs2rlumvsz ] to [ fs2optircomp ] . image stamps centred on snia coordinates are shown in figure [ fs2images ] . beyond the redshift limit of @xmath72 , there are no radio snia hosts , but one with an infrared counterpart : the sbc type host of snia 04d2ca at @xmath73 is classified as a lirg . in [ srelax ] we discuss the effects of extending our redshift range to @xmath74 . of these eight snia hosts , five have sed type e / s0 and will be included in the analysis : four are radio - loud ( @xmath3 ) , three are bright infrared galaxies ( birg , @xmath75 ) , and two are luminous infrared galaxies ( lirg , @xmath68 ) . galaxies which are bright at mid - infrared wavelengths ( mir , for example @xmath76 ) are described as being starburst or agn dominated , meaning they contain dust which absorbs high - energy photons from either young stars or agn , and re - emits them in the mir . it may seem the two radio - loud snia host galaxies with sed type e / s0 ( normally quiescent ) and infrared counterparts must be agn dominated , but there are several reasons why this is not the case . first , the radio - loud e / s0 hosts do not lie along the `` agn plume '' of the irac color - color plot ( although 04d1jg and 05d1hn hosts are in its `` agn zone '' ) , and are outside the `` agn zone '' of the irac+mips color - color plot as shown in figure [ fs2ircc ] . second , agn - dominated infrared emission originates from a hot dusty torus tens of parsecs across @xcite and would be unresolved , but the infrared sources associated with these two galaxies have are flagged as extended . third , a visual inspection of radio and infrared snia hosts in @xmath77 and galex ( d2 only ) images show all but the late - type host of 06d2ff are quite faint to invisible in the ultraviolet , which can be indicative of dust obscured star formation . fourth , not all starbursts appear morphologically irregular ( e.g. figure 4.68 of binney & merrifield ( 1998 ) , in which a profoundly disturbed galaxy appears as an elliptical in a shallow exposure ) . we therefore accept the conversion of @xmath66 into @xmath67 for these radio - loud e / s0 snia host galaxies , as discussed in [ ss2ir ] . the cappellaro et al . ( 1999 ) catalog is a combination of visual and photographic searches from cappellaro et al . ( 1993 ) and evans et al . ( 1989 ) , provided to us by e. cappellaro . sn data includes name , type ( classified photometrically ) , and host name ; galaxy catalog includes hubble type , recession velocity , b - band luminosity and control time . for radio counterparts we use the same methods , matching criteria , flux limits , and radio source catalogs as dv05 : the nrao vla sky survey @xcite , the parkes - mit - nrao survey @xcite , and the gb6 survey @xcite . we convert radio flux to luminosity as in [ ss2radio ] . table [ tdve ] shows we do not recover the same radio luminosities but do classify the same hosts as radio - loud , except for that of sn1983j which dv05 classify as `` borderline '' radio - loud and thus add 0.5 snia to the number of radio - loud hosts . we find a lower radio power and would not do this . galaxy classifications from the nasa extragalactic database ( ned ) for the early - type snia host sample of dv05 ( their table 2 ) are reproduced in table [ tdve ] , showing some are peculiar ( possibly merging ) and/or have emission lines . we derive mass and star formation rate from morphology and optical magnitude as in [ ss2ilbcat ] , and k - band stellar masses from the 2 micron all - sky survey @xcite as in [ ss2ir ] . we compile infrared data from two infrared science archive catalogs , the iras faint source catalog @xcite and the iras cataloged galaxies and quasars @xcite , and convert to infrared luminosity as described by sanders & mirabel ( 1996 ) . we also calculate the expected number of snia for each galaxy from the two - component `` a+b '' model as described in [ ss2add ] , and use the given given galaxy control times ( the time during which an exploding sn would be detectable ) which also account for sne detection biases @xcite . optical properties have an uncertainty of @xmath78@xmath79 from conversions between galaxy type conventions , magnitude bands , and mass and sfr models . the properties of early - type c99 snia hosts presented in table [ tdve ] suggest they may not all be simply quiescent ellipticals , perhaps containing more snia progenitors ( of the ` prompt ' or ` b ' component variety ) than previously expected . for example , ngc 1275 is a lirg and likely contains dust - obscured star formation and a young stellar component despite being morphologically early - type . in this section we present snls and c99 snia rates in a variety of early - type host galaxy subsets : radio - loud ( @xmath3 ) , radio - loudest ( @xmath80 ) , bright infrared galaxies or birg ( @xmath75 ) , and luminous infrared galaxies or lirg ( @xmath68 ) . the number of snls snia observed in d12 prior to 2006 dec 21 for each subset , @xmath31 , is shown in table [ tnsn ] . also shown ( in brackets ) is @xmath81 , where c is the effective control time from equation [ ec ] and accounts for snls detection efficiencies and observing seasons . thus , @xmath81 is the number of snia which explode per year of snls observations . also quoted are poisson uncertainties on @xmath81 at the 0.84 ( @xmath82 ) confidence level from gehrels ( 1986 ) . the amount of mass in galaxies in each subset is given in table [ tmgal ] . we calculate the specific snia rate in each galaxy subset , @xmath83 , via : @xmath84}}.\ ] ] the @xmath45 in the denominator is a time dilation factor for each galaxy , which converts the units for @xmath85 from observed to rest - frame time . specific snia rates are given in table [ tsnrates ] in the commonly used `` supernovae unit '' snum , which is equal to one supernova per century per @xmath86 . uncertainties are propagated from those in table [ tnsn ] but _ do not include _ the uncertainty in galaxy masses . ratios of snia rates in radio and ir subsets over all e / s0 galaxies are given in table [ tsnratios ] . these tables also contain equivalent values for the c99 snia and galaxy catalogs . we use radio luminosities derived in [ ss2c99 ] to classify field galaxies , but initially use radio luminosities from dv05 to classify snia hosts in order to make a direct comparison with their results . we revise our results with our own luminosities and discuss in [ ssratesradio ] and [ ssabpradio ] . with snls data , we find the specific snia rate in radio - loud early - type galaxies to be @xmath87 snum , within 1@xmath88 of the c99 result in table [ tsnrates ] . whereas dv05 quote a @xmath89 times snia rate enhancement in radio - loud over radio - quiet early - type galaxies , vla radio catalog incompleteness means we can identify only half of the radio - loud population ( [ ss2radio ] ) and can not isolate a radio - quiet set for the snls ( see figure [ fs2rlumvsz ] ) . in lieu of comparing with a radio - quiet set we consider the ratio between radio - loud and _ all _ early - types as a lower limit , and find it consistent with no enhancement at the 1@xmath88 and 2@xmath88 levels for the snls and c99 catalogs respectively , as shown in table [ tsnratios ] . can we correct our radio incompleteness , and does it affect our results ? while we could use radio - loud galaxy mass functions to estimate the mass missing from our radio - loud sample , we could not know how many snia this mass hosted , so can not correct the snia rates for radio incompleteness . instead , we simply use the limited mass in radio - loud galaxies we _ do _ find , and the limited sample of snia hosted by it , as a representative sample . any snia rate enhancement in radio - loud galaxies would also appear in this limited sample , assuming it is not proportional to @xmath90 ( i.e. agn power ) . the snia rate is not expected to be proportional to @xmath90 because agn activity is not suspected of directly influencing the probability of an snia explosion , as discussed in [ s1intro ] ; any evolution in the physical properties of agn is not expected to influence our results either . our radio incompleteness is thus unlikely to affect our results - they are a suitable comparison sample to c99 . as an extra test , we present the specific snia rates amongst the `` radio - loudest '' ( @xmath80 ) early - type galaxies in table [ tsnrates ] , and their ratio over all early - types in table [ tsnratios ] . unlike the radio - loud , the radio - loudest sample from snls catalogs is complete to @xmath72 . while the snls snia rate in the radio - loudest early - type galaxies shows no increase over the rate in all early - types , the c99 catalog shows a @xmath91 enhancement which is twice that for the radio - loud sample . however , this is within 3@xmath88 of a null result , and additional factors may artificially inflate this rate as discussed below . although we do not recover precisely the same radio luminosities for c99 early - type hosts as dv05 , we have initially used their classifications in order to directly compare with their results . as shown in table [ tdve ] , we would place ngc 1316 in the radio - loud subset but not the radio - loudest , and would not classify ngc 3106 as `` borderline '' radio - loud . also , sn1968a in ngc 1275 is listed as type - i , not type - ia . although rare , core - collapse supernovae have been observed in early - type galaxies ; this is often due to misclassification of late - types as early - type , especially for galaxies with hi and radio emission @xcite . ngc 1275 is a known star forming early - type galaxy , has the highest @xmath66 of all c99 early - type hosts ( table [ tdve ] ) , and is precisely the sort to potentially host a ccsn . considering these factors , we have instead 5 , 8 , and 20 snia hosts in the radio - loudest , radio - loud , and all early - type samples , lowering their rates to @xmath92 , @xmath93 , and @xmath94 snum respectively . ratios for the radio - loudest and radio - loud categories drop to @xmath95 and @xmath96 , bringing the null result within 2@xmath88 and 1@xmath88 respectively . with snls data , we find the specific snia rate in early - type birg and lirg to be @xmath97 and @xmath98 snum respectively . as presented in table [ tsnrates ] the c99 data results agree within 1@xmath88 . again we consider the ratios between b / lirg and _ all _ early - types as a lower limit on any possible enhancement , as given in table [ tsnratios ] . it appears the snia rate is a factor of 35 times higher in b / lirg , but this is not more significant than 1@xmath88 ( 2@xmath88 for c99 birg ) . while we could have isolated all early - type non - lirg for a ratio , the lirg fraction is small enough that this would not be significantly different from the all early - types sample . as discussed in [ ssratesradio ] , sn1968a in early - type lirg ngc 1275 may not have been a type ia . also , sn1983j and sn1991q do not have a definitive classification . the scenario that these three are not type ia drops the rate in c99 birg early - type galaxies to @xmath99 snum , consistent with the snia rate in all early - type galaxies . we also note that although one snia host is optically classified as starburst ( see [ ss2ilbcat ] ) , it does not have an ir counterpart and can not be included here . specific snia rates in subsets of radio and infrared early - type host galaxies are @xmath015 times the rate in all early - type galaxies , and all enhancement have low significance ( 2@xmath88 at most ) . in this section we test whether these rates are also consistent with the two - component `` a+b '' model , and whether the potential enhancements might simply be due to star formation providing additional snia progenitors . to do this , we statistically compare the observed number of snia in radio and infrared early - type galaxies to that predicted by the two - component `` a+b '' model . for the snls catalog , the total number of snia predicted by `` a+b '' over the survey s duration is @xmath100 for an individual galaxy , where @xmath20 and @xmath27 are explained in [ ss2add ] . for the c99 catalog , @xmath20 is converted to @xmath101 using given observational control times . where @xmath31 is the number observed , the poisson probability of observing @xmath102 given an expected number @xmath103 is expressed by @xcite : @xmath104 when @xmath105 , the _ summed _ poisson probability of observing @xmath102 _ or more _ is the integral of this from @xmath102 to @xmath106 ; when @xmath107 the probability of observing @xmath102 _ or less _ is the integral from @xmath108 to @xmath102 . it assesses whether the observed number of snia is consistent with the `` a+b '' model in any given galaxy subset , as used for early - type cluster galaxies in graham et al . summed probabilities @xmath109 are considered statistically significant results . we calculate summed poisson probabilities for the same snls and c99 radio and infrared galaxy subsets described in [ srates ] . optical _ masses and star formation rates are used in the `` a+b '' model , we refer to the total expected number of snia as @xmath110 and the associated summed probability as @xmath111 . we can also derive @xmath112 and @xmath113 by using infrared sfr if a galaxy has a mips counterpart ( using optical sfr if not ) , an account for contributions from dust - obscured star formation to the `` prompt '' component . when substituting @xmath67 into the `` a+b '' model in equation [ eab ] , we must alter the `` b '' value of sullivan et al . ( 2006 ) because it was derived from _ optical _ star formation rates , and typically @xmath114 @xcite . although figure [ fs2optircomp ] shows @xmath115@xmath116 , this is for the small fraction of brightest infrared galaxies in which this factor is , in most cases , due to dust - obscured star formation . had @xmath67 been used for all regular galaxies , the derived b value would be @xmath117 . summed poisson probabilities for snls radio - loud and radio - loudest early - type galaxy samples are @xmath118 as in table [ tpp ] , indicating no significant deviation between observations and `` a+b '' model predictions . as discussed in [ ssratesradio ] , it is unlikely the snls results are affected by our radio incompleteness . however , results for the c99 samples suggest significant to _ very _ significant deviations in the samples of radio - loud and radio - loudest early - type galaxies . these results are unchanged with the consideration of infrared sfr . could this be due to a @xmath119 variation between derived b values from different surveys @xcite ? for the c99 radio - loud sample , using @xmath120b yields @xmath121 and @xmath122 , but @xmath123b is necessary to raise @xmath124 and @xmath125 for the c99 radio - loudest early - type subset , so a change in b value can not be the whole story . as discussed in [ ssratesradio ] , the actual number of snia in the radio - loud and -loudest subsets of c99 may be 8 and 5 respectively . this would increase @xmath111 and @xmath113 to @xmath126 for both subsets , which we do not consider significant . summed poisson probabilities in table [ tpp ] for snls birg and lirg early - type samples _ hints _ at a possible excess of sneia over optical `` a+b '' model predictions , and the inclusion of @xmath67 does raise @xmath112 and @xmath113 , but these results are not statistically significant . the same can be said for c99 birg , though as discussed in [ ssratesradio ] the number of c99 snia in birgs may be just 1 , in good agreement with @xmath110 . in general we find the number of snia in birg and lirg is consistent with predictions of the `` a+b '' model . we note in [ ss2ir ] that infrared emission of potentially agn - dominated galaxies has been converted to @xmath67 with a starburst - dominated template , thereby artificially inflating @xmath127 . identifying galaxies which lay along the `` agn plume '' ( left plot , plume between @xmath128 ) * and * in the `` agn zone '' ( right plot ) of figure [ fs2ircc ] as the most likely to be agn , we find this population contributes 2040% _ of the additional _ snia predicted when @xmath67 is included , as in 2040% of ( @xmath112)-(@xmath110 ) . to consider whether altering the constraints on snls data effects our results we try , in turn : extending our redshift range to @xmath129 , including galaxies with sed type sbc , and different combination methods for our radio catalogs . sixteen snls sneia early - type host galaxies have @xmath130 , but due to the flux limits of our radio catalogs , none are associated with a radio or infrared counterpart ( only the sbc type host of snls sneia 04d2ca at @xmath73 is detected as a lirg ) . we determine the effective control time @xmath27 for d1 and d2 in redshift ranges 0.60.8 ( @xmath131 and @xmath132 ) and 0.81.0 ( @xmath133 and @xmath134 ) to account for varying snls detection efficiencies . the specific snia rate in all early - type galaxies to @xmath129 is @xmath135 snum , not significantly different from that for @xmath32 . extending to @xmath129 adds mass - but not snia - to the radio and infrared early - type subsets , reducing rates and ratios to insignificant levels and increasing summed poisson probabilities to @xmath136 . no new information is gained from extending to @xmath129 as radio and ir catalogs are not deep enough , and doing so only dilutes our sample . there are 34 galaxies with sed type e / s0 to sbc and @xmath32 , 6 radio - loud and 5 birg as in table [ ts2rlessn ] . the specific snia rate amongst all e / s0sbc galaxies is @xmath137 snum , ( higher than for all e / s0 ) ; in the radio - loud set it is @xmath138 snum ( similar ) and in the birg set is @xmath139 snum ( lower ) . no statistically significant snia rate enhancement in radio or infrared over all e / s0sbc galaxies is observed , and in these sets the summed poisson probability is @xmath140 . including sbc type galaxies adds to our total snia in radio and infrared hosts , but their naturally larger b component makes it harder to tell whether any contribution to the b component is being made from star formation associated with radio and infrared emission . the deeper flux limit of the vla - cosmos in d2 may result in slightly different radio source populations between d1 and d2 . to investigate whether this effects our results , we first restrict vla - cosmos to the same flux limit as vla - virmos , @xmath141 . only 2 snia in radio - loud early - type galaxies are observed , which decreases the specific snia rate to @xmath142 snum , increases its ratio over all early - types to @xmath143 , and raises the summed poisson probability to @xmath144 . secondly , we limit our analysis to the deeper radio catalog in d2 and are left with one radio - loud e / s0 snia host , which yields a rate of @xmath145 snum , a ratio over all e / s0 @xmath146 , and @xmath147 . in general we conclude that any differences in the radio source populations are not major contributors to the results of this experiment . it is well established that snia light curve shape and peak luminosity are correlated : more slowly - declining light curves reach a brighter peak luminosity . light curves are parametrized by @xmath148 , the decrease in magnitude over the first 15 days after maximum light @xcite , or by stretch , @xmath149 , the amount that a template ( average ) light curve must be `` stretched '' to fit the observations where @xmath150 for brighter , slower declining sneia @xcite . snia peak luminosity depends on the mass of @xmath151ni synthesized during the explosion @xcite ; whether this mass is influenced more by the metallicity or age of the white dwarf progenitor star is under much debate @xcite . snia properties are also correlated with qualities of the host galaxies , where late - type star - forming galaxies host mostly bright , slowly declining sneia , and early - type elliptical galaxies with little to no star - formation host faint , rapidly declining sneia @xcite . gallagher et al . ( 2005 ) compiled a database of 57 local sneia and found the mean @xmath148 in late - type ( sa to peculiar ) host galaxies was @xmath152 , but in early - type ( e / s0 ) hosts @xmath153 . sullivan et al . ( 2006 ) show the median stretch of snls sneia in late - type star - forming galaxies is greater than in early - types . this is supported by howell et al . ( 2007 ) who use light curve fitting routine sifto @xcite to constrain the stretch distribution of `` b '' ( prompt ) component sneia ( associated with young stellar populations ) to be centred at @xmath154 with @xmath155 , and of `` a '' ( delayed ) component sneia ( associated with old stellar populations ) to be centred at @xmath156 with @xmath157 . if sneia in radio - loud early - type hosts are associated with young stars , their stretch values should be consistent with the `` b '' component stretch distribution . sifto stretch values for snls sneia in radio galaxies are given in table [ ts2rlessn ] . ( snia 06d2je was not included in the snls third - year results , but preliminary fits show its stretch is similar to 05d1hn ) . stretches for snia 04d1jg , 05d1hn , and 06d2ck are consistent with the `` b '' component distribution of howell et al . ( 2007 ) . individually , 05d1hn and 06d2ck are more consistent with originating from the `` a '' component at the 1@xmath88 level , but collectively these three have a mean stretch of @xmath158 with @xmath1590.057 , which overlaps the 1@xmath88 uncertainty of the `` a '' component stretch distribution . amongst c99 sneia in radio - loud early - type hosts , five of eight have a @xmath148 given in table 2 of dv05 , @xmath160 . their mean is @xmath161 with @xmath162 which is consistent with @xmath153 for sneia in early - type hosts @xcite . instead of a numerical comparison , dv05 find that the distribution of decline rates for sneia in radio - loud early - type hosts is intermediate between that found for late - type and early - type hosts . they suggest this supports a _ continuum _ of snia and host galaxies properties , from young progenitors in very active galaxies , intermediate age progenitors in active galaxies , and old progenitors in passive galaxies . stretch values for the three snls sneia in radio - loud early - type galaxies also support this . two snls sneia found in radio - loud _ late_-type hosts , 05d1by and 06d2ff , have stretch values consistent with the `` b '' and `` a '' component distributions respectively which is normal since late - types have an underlying old stellar population . two snls sneia in non - radio birg hosts , 04d2bt ( early - type ) and 06d2ca ( late - type ) , have stretch values consistent with the `` a '' and `` b '' component distributions which also agree with their host type . in general we find the properties of radio and infrared snia hosts , and their environments , to resemble dusty starburst galaxies ( [ ss2hostradioir ] ) , and expect their snia light curves to be reddened and extinguished . snls analysis finds 05d1by , 05d1hn , 06d2ff , and 04d2bt are particularly red , and the first three have large hubble residuals , indicating this is an unusually red , faint sample of sneia . it is interesting that 04d1jg and 06d2ca are not particularly red or faint , because these two snia have the _ brightest _ infrared hosts . perhaps they originated on the near side of the galaxy and escaped dust extinction ( neither have large host offsets , 2 and 0.5 arcseconds respectively ) . also , the host of 06d2ca is sed type scd and likely has unobscured star forming regions . as a final note , snia 05d1by is also spectroscopically peculiar in that the @xmath163 silicon line is very broad , indicating fast ejecta velocities . it is similar to sn 2001ay @xcite , which had an atypically slow light curve decline given its unremarkable peak magnitude , a very luminous near - ir magnitude at @xmath164 days post maximum , and an early - type host@xcite . such peculiar spectra comprise @xmath165 of all sneia , but in this small sample it is impossible to comment on whether they are more or less common in active galaxies . with the c99 and snls catalogs we recover the snia rate in radio - loud early - type galaxies of dv05 , and for the first time we calculate snia rates in the radio - loudest , birg , and lirg subsets of early - type galaxies . we find these rates are @xmath015 times the rate in all early - type galaxies , at that any potential enhancement is @xmath166 . for the first time , we have incorporated infrared star formation rates into the two - component `` a+b '' model . while it does slightly increase the number of snia predicted , more so in b / lirg than in radio - loud(est ) samples , ultimately _ all _ snls results are consistent with the two - component `` a+b '' model . the same can not initially be said for all c99 results , but the observed number of c99 sneia may be an upper limit due to snia and/or host misclassifications . the benefit of considering infrared host galaxy properties to account for possible dust obscured star formation is best shown in figure [ frates ] , which plots the specific snia rate as a function of mean specific star formation rate ( ssfr ) for each early - type galaxy subset . blue symbols indicate where ssfr was derived from optical data only , and red symbols where @xmath67 was substituted for all galaxies with counterparts in the spitzer ir catalogs . curved lines show the two - component `` a+b '' model from sullivan et al . although error bars are large and all rates are within @xmath2 of the optical `` a+b '' model , when plotted with @xmath167 the snls snia rates in b / lirg and radio - loud(est ) subsets do align more closely with `` a+b '' . the same can be said for c99 b / lirg , and the apparent discrepancy for the c99 radio - loud(est ) subset may be due to misclassifications discussed in [ ssratesradio ] . these results are not significant , but are included to demonstrate the utility of such an analysis for future snia rates . dv05 suggested the snia rate enhancement in radio - loud early - type galaxies may be due to galaxy mergers triggering radio activity and providing extra snia progenitors by induced star formation . some aspects of the snls data are consistent with this : the evidence for dust - obscured star formation in snia radio hosts , the agreement between observations and `` a+b '' model predictions in snia radio hosts , and the properties of snia in radio hosts being similar to those associated with intermediate age stellar populations . however , the discrepancy between observations and `` a+b '' model predictions for c99 radio galaxies remains . in [ ssratesradio ] and [ ssabpradio ] we discuss possible snia and/or host misclassifications , but might it be the two - component model is simply inappropriate for radio galaxies ? it is a linear approximation to a relation that is not necessarily linear , and uses current sfr as a measure of the sfr when the white dwarf progenitor star was born - the parameter actually related to the number of snia progenitors . in regular spiral galaxies where the sfr remains constant for @xmath168 to @xmath169 gyr this is essentially true , but not necessarily for galaxies experiencing interactions and/or mergers , and perhaps episodic star formation ( discussed below ) , on timescales of @xmath170 years ( dv05 ) . this is not a failure of the `` a+b '' model , merely a limitation of its application to individual galaxies . in large samples this effect should approximately `` average out '' . with such a hypothetical episodic sfr , mannucci et al . ( 2006 ) show the radio - loud snia rate enhancement of dv05 is best fit with a _ bimodal _ delay - time distribution ( dtd ) for snia in which @xmath171 of the prompt component explodes in @xmath172 years , and suggest two physically distinct populations of snia progenitors . however , we find the snls snia rate in radio galaxies agrees with the `` a+b '' model , which is consistent with _ continuous _ dtd s @xcite , and that the properties of sneia in radio hosts do not constrain them to very young stellar populations ( as does dv05 ) . as a final note , the suggestion by dv05 that galaxy interactions and/or mergers cause both the radio activity and enhanced snia rate is consistent with the well documented correlation between clustered environments and radio galaxies @xcite . figure [ fs2images ] shows image stamps of radio snia host environments ; galaxies within @xmath173 ( i.e. @xmath174 ) are marked . this figure demonstrates that most radio loud snia hosts have nearby neighbours , as expected for elliptical galaxies . as reported in graham et al . ( 2008 ) , the host of 05d1by is located in the outskirts of a galaxy cluster . we also find the galaxy density within @xmath175 of this host shows clustering over the background field galaxy distribution , as determined by a `` significance parameter '' @xcite ; the small - scale environments of 04d1jg and 06d2ff hosts also show clustering , but not more so than the average snia environment . in general we find the properties of radio snia hosts and their environments are consistent with dusty starburst galaxies in non - isolated regions . a full analysis of snia in small groups and pairs using spectroscopic ( i.e. higher accuracy ) galaxy redshifts will be presented in an upcoming paper . based on the snls catalog , snia rates in radio and infrared early - type galaxies are @xmath015 times the rate in all early - type galaxies , but any enhancement is @xmath166 . rates in these subsets are consistent with predictions of the two component `` a+b '' snia rate model . the infrared properties of snia radio - loud early - type host galaxies suggest the presence of dust obscured star formation , and we have for the first time incorporated @xmath67 in the `` a+b '' model . in the c99 catalog , radio - loudest snia hosts are consistent with the `` a+b '' model only if some sneia and/or host galaxies have been misclassified . in general we find the properties of snia in radio and infrared galaxies support a _ continuum _ of snia and host galaxies properties , from young progenitors in very active galaxies to old progenitors in passive galaxies , as did dv05 . also these snia are fainter and redder than other snia , consistent with a dusty environment , and one was spectroscopically peculiar . to continue the investigation of snia rates in different stellar populations , observations in the cfht deep fields at near- to far - infrared and sub - millimeter would be very useful . radio coverage of the remaining two cfhtls deep fields could double the number of high - redshift snia in radio - loud early - type hosts available for analysis , but requires @xmath176 hours per field to achieve the depth of vla - virmos ( d1 ) and @xmath177 hours per field for vla - cosmos depth ( d2 ) . future work including spectroscopic galaxy redshifts will allow an analysis of the small scale environments of snia hosts , and snia rates and properties in small groups and interacting pairs . we gratefully acknowledge the cfht queued service observations team , all snls collaboration members , olivier ilbert and henry mccracken for early access to and correspondence regarding the photometric redshift galaxy catalog , and enrico cappellaro for access to the c99 galaxy and supernova samples . mlg acknowledges colin borys , dave patton , and stephane arnouts for their advice . this work is based in part on observations obtained with megaprime / megacam , a joint project of cfht and cea / dapnia , at the cfht which is operated by the national research council ( nrc ) of canada , the institut national des science de lunivers of the centre national de la recherche scientifique ( cnrs ) of france , and the university of hawaii . this work is also based in part of data products produced at the canadian astronomy data centre as part of the cfht legacy survey , a collaborative project of nrc and cnrs . this work is also based in part on on observations obtained at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the 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 ministerio de ciencia , tecnologa e innovacin productiva ( argentina ) gemini identification numbers of the programs under which these observations were taken are : gs-2003b - q-8 , gn-2003b - q-9 , gs-2004a - q-11 , gn-2004a - q-19 , gs-2004b - q-31 , gn-2004b - q-16 , gs-2005a - q-11 , gn-2005a - q-11 , gs-2005b - q-6 , gn-2005b - q-7 , gn-2006a - q-7 and gn-2006b - q-10 . this research has made use of the nasa/ ipac infrared science archive , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . this publication has also made 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 work has been supported by nserc and the university of victoria . mlg gratefully acknowledges the financial support of the province of british columbia through the ministry of advanced education . ms acknowledges support from the royal society . s. fabbro acknowledges support from fundao para a cincia e tecnologia , portugal , under grant sfrh / bpd/31817/006 and project pocti / cte - ast/57664/2004 . , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] , 30x30 arcseconds , centred on sneia for host galaxies detected in radio and/or infrared source catalogs . right to left , top to bottom : 04d1jg , 05d1by , 05d1hn , 06d2ck , 06d2ff , 06d2je , 04d2bt , and 06d2ca . white crosses mark sneia , and white plus signs mark nearby galaxies within @xmath173 ( i.e. @xmath174 ) . [ fs2images],title="fig:",width=151,height=151 ] lccccccccccc snia & snia & snia & snia & host & host & host & host & host radio & host ir & host ir + snls i d & @xmath178 & color & stretch & sed & @xmath179 & mass & ssfr & luminosity & luminosity & ssfr + & & & & & & [ @xmath180 & [ @xmath181 & [ @xmath182 & [ @xmath183 & [ @xmath181 + 04d1jg & 0.584 & -0.092 & @xmath184 & e / s0 & -22.99 & 43.5 & 0.15 & 14.5 & 46.47 & 1.56 + 05d1by & 0.299 & 0.736 & @xmath185 & sbc & -21.17 & 5.43 & 0.82 & 2.85 & 8.972 & 2.66 + 05d1hn & 0.149 & 0.334 & @xmath186 & e / s0 & -22.08 & 20.2 & 0.08 & 2.82 & 12.95 & 1.09 + 06d2ck & 0.552 & -0.020 & @xmath187 & e / s0 & -22.50 & 32.1 & 0.03 & 5.30 & @xmath188 & @xmath188 + 06d2ff & 0.345 & 0.299 & @xmath189 & sbc & -21.94 & 9.19 & 1.17 & 4.16 & 7.871 & 1.84 + 06d2je & 0.418 & @xmath188 & @xmath188 & e / s0 & -19.79 & 2.16 & 0.25 & 1.73 & @xmath188 & @xmath188 + 04d2bt & 0.220 & 0.222 & @xmath190 & e / s0 & -21.40 & 10.1 & 0.08 & 0.00 & 2.193 & 0.35 + 06d2ca & 0.531 & 0.047 & @xmath191 & scd & -21.03 & 0.81 & 6.83 & 0.00 & 33.56 & 24.6 + llcclccc sn ia name & galaxy & hubble & radio & ned & optical & infrared & infrared + ( and type ) & name & type & luminosity & galaxy type & ssfr & luminosity & ssfr + & & & [ @xmath182 & & [ @xmath181 & [ @xmath183 & [ @xmath192 + 1961h ( ia ) & ngc 4564 & -4.9 & @xmath1930.008 ( @xmath1930.010 ) & e6 & 0.00 & @xmath188 & @xmath188 + 1968a ( i ) & ngc 1275 & -1.6 & 720.0 ( 1462 ) & cd ; pec ; nlrg & 0.26 & 17.5 & 0.63 + 1970j ( ia ) & ngc 7619 & -5.0 & 0.540 ( 0.590 ) & e & 0.00 & @xmath188 & @xmath188 + 1972j ( i ) & ngc 7634 & -2.0 & @xmath1930.040 ( @xmath1930.050 ) & sb0 & 0.27 & @xmath188 & @xmath188 + 1980i ( ia ) & ngc 4374 & -4.7 & 20.00 ( 31.18 ) & e1 ; liner ; sy2 & 0.00 & 0.2 & 0.02 + 1980n ( ia ) & ngc 1316 & -1.9 & 1000 . ( 1.752 ) & sab ; liner & 0.25 & 0.7 & 0.03 + 1981d ( ia ) & ngc 1316 & -1.9 & 1000 . ( 1.752 ) & sab ; liner & 0.25 & 0.7 & 0.03 + 1981 g ( ia ) & ngc 4874 & -4.0 & 24.00 ( 27.98 ) & cd & 0.03 & @xmath188 & @xmath188 + 1983 g ( ia ) & ngc 4753 & -2.2 & 0.530 ( @xmath1930.015 ) & i0 & 0.26 & 0.6 & 0.08 + 1983j ( ? ) & ngc 3106 & -1.9 & 0.920 ( 0.692 ) & s0 & 0.26 & 1.7 & 0.13 + 1986 g ( ia ) & ngc 5128 & -2.2 & 18.00 ( 52.39 ) & s0 ; pec ; sy2 & 0.25 & 1.6 & 0.20 + 1990 m ( ia ) & ngc 5493 & -2.1 & @xmath1930.040 ( @xmath1930.051 ) & s0 ; pec & 0.25 & @xmath188 & @xmath188 + 1991q ( ? ) & ngc 4926a & -1.6 & 0.310 ( 0.325 ) & s0 ; pec ? ; sbrst & 0.24 & 6.2 & 2.86 + 1991bg ( ia ) & ngc 4374 & -4.7 & 20.00 ( 31.18 ) & e1 ; liner ; sy2 & 0.00 & 0.2 & 0.02 + 1991bi ( ia ) & ngc 5127 & -4.9 & 8.700 ( 9.843 ) & e ; pec & 0.00 & @xmath188 & @xmath188 + 1992a ( ia ) & ngc 1380 & -1.9 & @xmath1930.020 ( @xmath1930.018 ) & sa0 & 0.25 & 0.4 & 0.06 + 1992bo ( ia ) & e352-g57 & -1.5 & @xmath1930.160 ( @xmath1930.175 ) & sb ; pec & 0.27 & @xmath188 & @xmath188 + 1993c ( ia ) & ngc 2954 & -4.9 & @xmath1930.080 ( @xmath1930.105 ) & e & 0.00 & @xmath188 & @xmath188 + 1993ah ( ia ) & e471-g27 & -2.0 & 41.00 ( 44.77 ) & sbb & 0.26 & @xmath188 & @xmath188 + 1994d ( ia ) & ngc 4526 & -2.0 & 0.042 ( 0.047 ) & sab & 0.26 & 0.7 & 0.15 + 1996x ( ia ) & ngc 5061 & -5.0 & @xmath1930.030 ( @xmath1930.033 ) & e0 & 0.00 & @xmath188 & @xmath188 + lcc sample & snls & c99 + all & 38440.4 ( 4337 ) & 23315.0 ( 2079 ) + radio - loud & 3879.3 ( 165 ) & 7724.5 ( 314 ) + radio - loudest & 1638.8 ( 46 ) & 2825.5 ( 92 ) + birg & 1650.3 ( 179 ) & 1805.7 ( 166 ) + lirg & 747.0 ( 73 ) & 469.2 ( 22 ) + lcc sample & snls & c99 + & [ snum ] & [ snum ] + all & @xmath204 & @xmath205 + radio - loud & @xmath87 & @xmath206 + radio - loudest & @xmath207 & @xmath208 + birg & @xmath209 & @xmath210 + lirg & @xmath98 & @xmath211 + llccccc & @xmath31 & @xmath110 & @xmath111 & @xmath112 & @xmath113 + radio - loud & snls & 4 & 2.16 & 0.17 & 2.38 & 0.22 + & c99 & 9.5 & 4.84 & 0.04 & 5.10 & 0.05 + radio - loudest & snls & 1 & 0.90 & 0.59 & 1.00 & 0.73 + & c99 & 8 & 2.35 & 0.003 & 2.39 & 0.003 + birg & snls & 3 & 1.03 & 0.09 & 1.52 & 0.20 + & c99 & 4 & 1.39 & 0.05 & 1.77 & 0.10 + lirg & snls & 2 & 0.44 & 0.07 & 0.83 & 0.20 + & c99 & 1.0 & 0.26 & 0.23 & 0.43 & 0.35 +

here we present a model to study the micro - plastic regime of a stress - strain curve . in this model an explicit dislocation population represents the mobile dislocation content and an internal shear - stress field represents a mean - field description of the immobile dislocation content . the mobile dislocations are constrained to a simple dipolar mat geometry and modelled via a dislocation dynamics algorithm , whilst the shear - stress field is chosen to be a sinusoidal function of distance along the mat direction . the sinusoidal function , defined by a periodic length and a shear - stress amplitude , is interpreted to represent a pre - existing micro - structure . these model parameters , along with the mobile dislocation density , are found to admit a diversity of micro - plastic behaviour involving intermittent plasticity in the form of a scale - free avalanche phenomenon , with an exponent and scaling - collapse for the strain burst magnitude distribution that is in agreement with mean - field theory and similar to that seen in experiment and more complex dislocation dynamics simulations .

1205.1486

in 1964 , using state - of - the - art torsion experiments , tinder and co - workers were able to achieve a strain resolution of @xmath0 for sub mm - sized samples to study the micro - plastic regime of highly pure poly - crystal cu samples @xcite , followed by tests on zn single crystals some years later @xcite . their plastic - strain versus stress curves contained plateaus of stress which were attributed to the occurrence of discrete dislocation glide activity over a length scale comparable to the dislocation spacing of an assumed three - dimensional network . indeed the authors write in ref . @xcite , `` the results suggest that an important fraction of the total strain , in the initial stages of deformation , involved motion of a few favourably situated dislocation segments through distances large enough to form new interactions with other elements of the three dimensional network . if this were so , then most elements of the network must have been relatively immobile , making little or no contribution to the strain . '' that most dislocations remain immobile remains a contemporary viewpoint @xcite . another more recently pursued route is to probe discrete dislocation activity via the stress - strain curve of micron sized focused ion beam ( fib ) milled single crystals . here , advantage is taken of a nano - indentation platform equipped with a flat punch tip to compress micro-/nano - crystals @xcite . in addition to the sub - nanometer displacement resolution of the system , which notably has a lower strain resolution than the above mentioned torsion experiments , the sample size is decreased to the micron range and below . as a result , the strain associated with the discrete dislocation activity is increased to an easily detectable magnitude . whilst this more contemporary work has been primarily motivated by the `` smaller - is - stronger '' size - effect paradigm @xcite , an extensive analysis of the statistics of the discrete dislocation activity has revealed power - law behavior in the distribution of strain - burst magnitudes giving an exponent of @xmath1 @xcite . similar exponents can also be found in bulk samples via detailed analysis of load displacement signals , where structural evolution is also seen to occur @xcite . the very recent work of dahmen and co - workers @xcite suggest that the variation in literature exponent values could be due to the different stress intervals used to bin the strain burst magnitude data , where a total integration over the stress interval to material value should , within mean field theory , give an exponent equal to precisely two . such exponents are indicative of crackling @xcite or barkhausen noise , and more generally of avalanche phenomena , indicating that dislocation mediated plastic deformation belongs to a universality class that encompasses many natural phenomenon over a variety of different length and timescales . indeed , similar power - law exponents can also be found for metallic glasses @xcite in which the underlying plastic deformation is fundamentally different to crystalline metals . another class of experiments revealing the intermittent nature of dislocation dynamics is the acoustic emission monitoring of ice @xcite . such experiments measure the acoustic energy released by intermittent dislocation activity during constant stress deformation ( in the tertiary creep regime ) . indeed , via single sensor acoustic emission signals , such dislocation activity can be well characterised in time revealing power - law behaviour with exponents of 1.6 to 1.8 , which is very similar to that seen via the micro - compression stress - strain curves @xcite . furthermore , via multiple sensor monitoring , time and space clustering of dislocation avalanches could be observed @xcite . it was found that avalanche epi - centres were correlated in space according to a non - integer power - law exponent indicating scale - free clustering and at short enough times such clusters were correlated in time indicating collective activity . that such scale - invariant dislocation activity occurs , is a signature of an underlying complex dislocation based micro - structure . an entity whose properties and evolution under an applied stress play a central role in the more general subjects of material strength and strain hardening @xcite . due to the complex dynamics and evolution of the dislocation structure , computer simulation based approaches have helped greatly over the past decades to clarify the underlying dislocation based mechanisms responsible for such structural evolution . one such method is the so - called dislocation dynamics ( dd ) approach . early work involved two dimensional arrays of straight edge dislocations interacting via elasticity using single- and multi - slip geometries @xcite . these works demonstrate that under an external stress , dislocation patterning emerges analogous to what is seen in static transmission electron microscopy experiments . other and more recent works have developed these numerical techniques in terms of efficiency @xcite , strain boundary - conditions and obstacle / composite geometries @xcite to study more complex patterning such as the emergence of granular cell structures and the study of grain boundary network evolution @xcite , strain hardening , and material strength in both bulk and confined volume systems @xcite . analogous methodologies and investigations have also been carried out in three dimensions @xcite . intermittent dislocation activity , in the form of dislocation avalanches , has also been studied using the dd simulation technique @xcite . such simulations have produced power - law exponents of the distribution of strain burst magnitudes similar to that seen in experiments , indicating some degree of scale - free behaviour , where dislocations are arranged in meta - stable cell and/or wall structures , with only a minor fraction of the dislocation population moving intermittently , thereby creating discrete strain jumps . the so - called self organised criticality ( soc ) view of the dislocation network state offers one theoretical platform for the understanding of the observed universality @xcite , in which the material system organises itself into a configuration that is critical , resulting in scale - free behaviour upon transiting to a new realisation of the critical state . originally developed to describe sand - pile dynamics @xcite , the approach is somewhat at odds with the historical viewpoint that dislocation structure evolution is primarily driven by equilibrium driving forces such as that embodied in low energy structures ( les ) theory @xcite and in a wide range of strain hardening theories . the finding that the occurrence of avalanche phenomenon is insensitive to the nature of the forming immobile dislocation network , the slip geometry , the deformation mode , and the details of the dislocation dynamics and spatial dimension is a central hallmark of soc , which is robust against the details of the underlying physical model . this motivates investigating plastic flow with simpler models that explicitly do not take into account the fine details of individually interacting sessile and mobile dislocations , naturally shifting the focus of plastic flow from complex dislocation structural evolution to the interaction of a minute mobile dislocation population with a simplified description of the sessile dislocation population . such an approach is analogous to the study of dislocations in the presence of pinning potentials @xcite and more generally to coarse grained models of plasticity that study the depinning transition ( see ref . @xcite and references therein , and also refs . dislocation field theories are also well able to study intermittency , however unlike dislocation dynamics based methods which are able to simulate only small plastic strains , these models are able to incorporate dislocation transport over non - neglible distances and therefore able to simulate significant structural evolution @xcite . in this work , a simple model is therefore proposed in which a dipolar dislocation mat @xcite is embedded into an internal static sinusoidal stress field defined by a wave - length and a shear - stress amplitude . the explicit dislocation population is modelled via a standard dd algorithm . it is found that , similar to very complex and detailed 3d - dd simulations , the resulting stress - strain curves evolve in a discrete manner that reflects an underlying intermittent plasticity originating from irreversible changes in dislocation configuration , and that the distribution of the corresponding strain - burst magnitudes reveal both extremal value statistics and scale - free avalanche behaviour . thus , although the complex details of microscopic dislocation mechanisms and structure are omitted , the simple model is still able to capture the fundamental properties of intermittent micro - plastic flow . in the following section , the model is formally introduced and the dd technique described , and in sec . [ sec_loading ] various loading modes to produce a stress - strain curve are presented . [ sec_results ] presents the dd simulations for a wide range of model parameters to investigate their effect on deformation behaviour . the statistical analysis of the stress at which intermittent activity occurs , as well as the distribution of related strain - bursts , are presented in sec . [ sec_statistics ] . finally , sec . [ sec_discussion ] discusses the context for the model , where it is argued that its applicable range is restricted to the micro - plastic region of the stress - strain curve a regime where significant structural evolution and work hardening are , to a large extent , absent @xcite . the proposed model consists of two parallel slip planes , in the @xmath2 plane , separated by a distance @xmath3 along the @xmath4 direction . each slip plane is populated by infinitely long straight edge dislocations where one slip plane contains @xmath5 dislocations with burgers vector @xmath6 and the other @xmath7 dislocations with burgers vector @xmath8 . equal numbers of each type of dislocation are considered to ensure no net burgers vector content : @xmath9 . such a structure is traditionally known as a simple dipolar mat @xcite . to this , a time independent internal sinusoidal shear - stress field is added of the form @xmath10 which is parametrised by a shear - stress amplitude @xmath11 and a periodic length scale @xmath12 . fig . [ schematic_fig ] displays a schematic of the model system in which the line direction of each straight dislocation is perpendicular to the plane of the figure . the dislocation density is defined by the total number of dislocations divided by the area of the system . the area of the system is defined by its length along @xmath13 , defined as @xmath14 , and its spatial extent along @xmath4 , defined as @xmath15 . detailed discussion on the origin of the internal sinusoidal stress field is deferred to sec . [ sec_discussion ] , however , its existence should be viewed as a simplified representation of a pre - existing ( and unchanging ) immobile dislocation population . the explicit dislocations , schematically shown in fig . [ schematic_fig ] , are therefore to be viewed as the mobile dislocation population of the system , and are quantified by their density @xmath16 . in addition to the force arising from the internal shear - stress field , each dislocation will experience a force of elastic origin from all other dislocations within the system . since it is assumed that the dislocations are unable to leave their slip plane , the only relevant force per unit dislocation length component will be along the @xmath13 direction and , via isotropic elasticity , may be calculated from the well known expression between two edge dislocations ( labelled @xmath17 and @xmath18 ) on parallel slip planes @xcite @xmath19 here @xmath20 is the isotropic shear modulus and @xmath21 possion s ratio of the isotropic elastic medium , @xmath22 ( @xmath23 ) is the burgers vector in the x - direction of the @xmath17th ( @xmath18th ) dislocation , and @xmath24 is the two dimensional vector defining the dislocations spatial separation . presently a model isotropic cu system is implemented , in which the shear modulus is taken as @xmath25 gpa , the possion s ratio as @xmath26 , and the burgers vector magnitude as @xmath27 . for the present work , periodic boundary conditions along @xmath14 , the dipolar mat direction , and open boundary conditions along @xmath3 , are assumed . due to the long range nature of eqn . [ eqddint ] , the correct treatment of periodicity invovles the summation of all dislocation image contributions to the force per unit dislocation length on a given dislocation . for the considered one - dimensional periodicity , an exact solution to such a summation is tractable , and is given by @xmath28}{d^{2}\left(\cos\left(\frac{2\pi\delta x}{d}\right)-\cosh\left(\frac{2\pi\delta y}{d}\right)\right)^{2}}. \label{eqddintimage}\end{aligned}\ ] ] a simple derivation of this equation is detailed in appendix [ appendix_a ] . the temporal evolution of a particular dislocation configuration is characterised by the choice of an empirical mobility law . due to the actual discreteness of the lattice at the atomic scale , a dislocation segment must overcome an energy barrier associated with the local shearing of atoms in order to move an atomic distance . this so - called peierls energy barrier and the associated peierls stress @xcite , the stress at which the dislocation can begin to move ( defined at a given temperature ) , results in the dislocation moving quasi - statically from atomic lattice site to atomic lattice site . at the meso - scopic scale this results in over - damped motion where the dislocation s velocity is proportional to the force acting on the dislocation the present mobility law . the material specific proportionality constant is referred to as a damping parameter and is dependent on dislocation type , geometry , and on temperature . the equation of motion along the @xmath13 direction for the @xmath17th dislocation is then given by @xmath29 where @xmath30 is the damping coefficient , which for cu is @xmath31 pa s @xcite . in eqn . [ eqofmotion ] , @xmath32 , is the total force per unit dislocation length acting on the dislocation , @xmath33b_{x , i}+\sum_{j\ne i}f_{x , ij}.\label{eqtotalforce}\ ] ] here @xmath34 is an externally applied homogeneous shear - stress field and @xmath35 is the static sinusoidal shear - stress field defined in eqn . [ eqinternal ] . the numerical solution of eqn . [ eqofmotion ] constitutes the dislocation dynamics ( dd ) algorithm presently used in which an appropriate finite time - step , @xmath36 , is used to integrate the equations of motion . the corresponding shear - strain response @xmath37 within this @xmath36 is calculated via @xmath38 it is again emphasised that over the time - scale of @xmath36 all atomic scale aspects are averaged over and inertial effects are ignored . since the edge dislocations are infinitely long and straight such dynamics falls into the class of two dimensional dd modelling . the dislocation density is given by @xmath39 . for the present work , the periodic length @xmath14 is chosen to define the distance @xmath3 between the two populated slip systems via @xmath40 . this sets the mean distance between dislocations along the @xmath41 and @xmath42direction to be the same , and defines the dislocation density as @xmath43 . thus , @xmath44 , and choosing a value of the dislocation density will fix the scale of interaction between the two parallel slip planes of the dipolar mat geometry see appendix a. the motivation for such a restriction is to give the dislocation density a greater bulk - like relevance , where in the bulk limit it represents an isotropic density of dislocations . other definitions of @xmath3 are , of course , possible . to simulate a loading experiment and thus a stress - strain curve , a sample must be produced and a loading mode chosen . presently , sample preparation involves the chosen number of dislocations being initially placed at random positions within the dipolar mat geometry , and the structure relaxed via eqn . [ eqofmotion ] to minimise the force on each dislocation to within a chosen tolerance . this is performed at @xmath45 . in the present work , no attempt is therefore made to determine a low ( or lowest ) energy dislocation structure , an approach that is compatible with the fact that the explicit dislocations within the model are only those of the mobile variety . this aspect is found to be crucial to the properties of the model and will be discussed in more detail in sec . [ sec_discussion ] . there exist a number of ways in which a deformation simulation can be done . the first such loading mode is referred to as `` stress - relaxed '' and involves incrementing the external shear - stress by a value @xmath46 ( @xmath47 ) , and relaxing the structure until the the sum of the dislocation force magnitudes , or equivalently , the dislocation velocity magnitudes , varies by less than the fraction @xmath0 . during the relaxation , the associated plastic shear - strain increment may be calculated as the sum of eqn . [ shear_strain_eqn ] over all @xmath36 time steps of the relaxation . once convergence is obtained this shear - strain increment is added to the total plastic shear - strain and the cycle is repeated until the desired stress - strain curve is obtained . [ fig_stress_relaxed_ss_example]a displays a typical shear - stress versus plastic shear - strain curve during the initial stages of loading . discrete dislocation activity , via strain bursts , is clearly evident and is separated by continuous an - elastic regions . by adding the elastic shear - strain , @xmath48 , to the plastic shear - strain , the total shear - strain is obtained . [ fig_stress_relaxed_ss_example]b displays the corresponding shear - stress versus total shear - strain curve for a small range of stresses and an approximately linear stress - strain curve with intermittent stress plateaus is seen . when displaying a similar curve for the full range of stresses , as seen in fig . [ fig_stress_relaxed_ss_example]a , only a straight line is resolvable indicating that such strain bursts are well within the micro - plastic regime of deformation . at larger stresses a plastic flow regime is entered which will be investigated in more detail in subsequent sections . experimentally , two distinct deformation modes can be used : displacement controlled and load controlled . here we consider an inherently force controlled testing device . in such a case , displacement controlled testing is done by adjusting the applied load via a feed - back loop such that the displacement rate is held at a fixed value throughout the loading , whereas for load controlled experiments , the applied load simply increases at a chosen rate . for the present model , these deformation modes correspond to a constant shear - strain rate and constant shear - stress rate loading condition . to obtain a stress - strain curve with a constant shear - stress rate , a numerical value for the applied stress rate , @xmath49 , is chosen . this then defines a stress increment @xmath50 , where @xmath36 is the time - step used to evolve the dislocation network according to eqn . [ eqofmotion ] . thus at every simulation iteration , the stress is increased by @xmath46 and the configuration evolves in time by an amount @xmath36 . to implement a constant shear - strain rate loading mode , @xmath51 , a numerical value is chosen and the appropriate @xmath46 stress increment , to achieve such a strain rate , is performed every simulation step . the actual value of @xmath46 is determined by assuming that the total strain rate decomposes additively into an elastic and plastic component : @xmath52 giving @xmath53 or @xmath54 where @xmath55 is the strain increment per simulation iteration , calculated via eqn . [ shear_strain_eqn ] . thus the stress increment @xmath46 is determined by the correction needed to achieve the required constant strain rate for the next simulation iteration . in the above , due to the simplified geometry of the model ( fig . [ schematic_fig ] ) , a pure shear modulus , @xmath20 , rather than a young s modulus is used . to investigate how such loading modes affect the discrete dislocation activity seen in fig . [ fig_stress_relaxed_ss_example ] , the appropriate stress - rate and strain - rate must be chosen . this is done by first choosing @xmath51 , from which @xmath49 is obtained via @xmath56 to ensure that in the elastic / an - elastic regime both deformation modes have the same total strain rate . [ fig_burst_modes ] displays a single strain burst in all three considered loading modes . for the constant strain rate and stress rate modes two strain rates are considered : 0.1 s@xmath57 and 1.0 s@xmath57 . for the `` stress - relaxed '' loading mode a sharp plateau is evident in fig . [ fig_burst_modes]a with an identically zero gradient . in this region , the constant stress rate mode also exhibits a plateau but with a ( non - zero ) positive gradient since , during the evolution of the strain burst , the stress is rising at the chosen rate . also , the stress at which the strain burst initiates is somewhat higher ( and increasing with increasing strain rate ) than that in the stress relaxed mode indicating a strain - rate effect . in this regard , the `` stress - relaxed '' deformation mode can be considered as the zero stress - rate limit of the constant stress rate loading mode in which the dislocation configuration always has time to relax before the next stress increment . for the constant strain - rate loading mode , the onset of the strain burst occurs at similar stresses to that of the constant stress rate , however as the strain - burst evolves , the stress decreases to maintain the chosen strain - rate . [ fig_burst_modes]b displays the same strain burst with the stress now as a function of the total strain . the greatest effect is seen in the constant strain - rate mode where due to the drop in stress during the strain - burst , there is a rapid drop in elastic strain . [ fig_burst_modes_strain_rate]a displays the plastic strain rate as a function of plastic strain for the burst shown in fig . [ fig_burst_modes ] ( for a constant strain rate of 1.0 s@xmath57 ) . the elastic strain rate response as dictated by eqn . [ constsr ] is also shown and correspondingly reduces to compensate for the rise in the plastic strain rate . data for a constant strain rate of 0.1 s@xmath57 differs little from the 1.0 s@xmath57 data , a result ( along with the high strain - rates of fig . [ fig_burst_modes_strain_rate]a ) due to the driven zero temperature nature of the simulation . it is noted that for the constant strain rate loading mode , the loading system responds instantaneously ( to within @xmath36 ) to any discrete plastic event ( hence leading to the backwards curvature seen in fig . [ fig_burst_modes]b ) . to model an instrumentally realistic device , a more complex differential equation than that of eqn . [ constsr ] would need to be developed , which takes into account the finite delay time and resolution of the loading apparatus . . b ) average dislocation velocity during the same dislocation burst in `` stress relaxed '' deformation mode.,scaledwidth=95.0% ] fig . [ fig_burst_modes_strain_rate]b now shows the average dislocation velocity during the `` stress relaxed '' deformation mode for four consecutive applied shear stresses in which the first constant stress relaxation undergoes the irreversible plastic strain rate seen in fig . [ fig_burst_modes ] . for this stress value , an initial relaxation of the dislocation velocities is seen , followed by a rapid acceleration of the dislocations corresponding to the emerging instability and the eventual irreversible plastic event of fig . [ fig_burst_modes ] . this continues until a maximum velocity is reached at which the dislocations then begin to decelerate until the convergence criterion is met . three further constant stress increments result in an immediate relaxation of the dislocation velocities corresponding to a reversible ( anelastic ) relaxation of the dislocation configuration . in this section the `` stress - relaxed '' loading mode is used to investigate the influence of the model parameters on the stress - strain curve . the influence of the sinusoidal shear - stress field amplitude is first investigated . [ different_tau_fig]a displays the resulting shear - stress versus plastic shear - strain curves for three different choices of @xmath11 ; namely 100mpa , 10mpa and 1mpa for a system with @xmath58 @xmath59 m and @xmath60 @xmath59 m , containing @xmath61 dislocations . this gives a mobile dislocation density of @xmath62 . the figure demonstrates that the choice of @xmath11 strongly controls the stress at which macroscopic plastic flow occurs . in fig . [ different_tau_fig]a , the vertical axis is plotted as a logarithmic scale to reveal the fine structure of the micro - plastic region . in all cases discrete strain bursts are evident . [ different_tau_fig]b displays the corresponding shear - stress versus total shear - strain curves . the small plastic shear - strain values evident in figure [ different_tau_fig]a and the sharp yield transition in [ different_tau_fig]b emphasise that the shear - stress versus plastic shear - strain data prior to flow is clearly in the micro - plastic regime of the stress - strain curve . values , for a mobile dislocation density @xmath63 , and b ) the corresponding stress versus total strain.,scaledwidth=95.0% ] in figs . [ different_rho_fig]a - b , @xmath16 is now varied from @xmath63 to @xmath64 by changing the number of dislocations for a fixed @xmath65 @xmath59 m and @xmath60 @xmath59 m . here @xmath11 is 50 mpa [ different_rho_fig]a - b demonstrate that with increasing mobile dislocation density the flow stress decreases from @xmath11 . thus at low enough dislocation densities ( as in fig . [ different_tau_fig ] ) , @xmath11 defines the flow stress of the system , a natural result since a well isolated dislocation will be primarily affected by the sinusoidal shear - stress field . [ different_rho_fig]a - b also demonstrate that the micro - plastic regime broadens with increasing dislocation density and that the first discrete dislocation event occurs at a decreasing applied homogeneous shear - stress ( see arrowed regions in fig . [ different_rho_fig]b ) . figs . [ different_rho_fig]c - d show similar data for @xmath66 mpa , for the same values of dislocation density . although a similar trend is seen , the deviation away from the maximum flow stress of @xmath66 is a proportionally stronger function of the increasing dislocation density . mpa and ( lower panels ) @xmath66 mpa . the right panels display the same data with strain plotted using a logarithmic scale to emphasise the initial micro - plastic strain region where the arrows ( in b ) indicate the first occurrence of a strain response.,scaledwidth=95.0% ] the remaining micro - structural variable that can be varied is the characteristic length scale of the sinusoidal stress field , @xmath12 . [ different_lambda_fig ] displays stress versus plastic strain curves as @xmath12 increases from @xmath67 m to @xmath68 m , with @xmath66 mpa and 50 mpa , @xmath69 m and @xmath62 . here a larger @xmath14 was used to obtain a better statistical sample of the dislocation environment at the length scale of @xmath12 . the chosen dislocation density is similar to that of fig . [ different_tau_fig ] in which @xmath11 largely determined the flow stress . inspection of fig . [ different_lambda_fig ] reveals that with increasing @xmath12 the micro - plastic regime is broadened due to increasing strain burst magnitudes with increasing @xmath12 . for small strain ( figs . [ different_lambda_fig]a and [ different_lambda_fig]c ) , it also appears that the flow stress regime occurs at a reduced stress ( for increasing @xmath12 ) , however at larger strains ( figs . [ different_lambda_fig]b and [ different_lambda_fig]d ) flow stresses approximately equal to @xmath11 are eventually reached . it is also noted that with increasing @xmath12 the strain burst magnitude increases . with decreasing @xmath11 , the deviation of the curves away from @xmath11 is a stronger function of increasing @xmath12 resulting in a somewhat broader micro - plastic region . choices , where for each simulation @xmath70 mpa ( upper panels ) and @xmath66 mpa ( lower panels ) , @xmath69 m and a mobile dislocation density @xmath63 . the left panels show the initial strain regime and the right panels show a larger strain range.,scaledwidth=95.0% ] the central properties of the present model are reflected in figures [ different_tau_fig ] , [ different_rho_fig ] and [ different_lambda_fig ] . if the dislocation density is low enough or @xmath11 sufficiently high , then @xmath11 will primarily determine the shear - stress at which the extended plastic flow regime begins . in this limit , the internal shear - stress that each dislocation experiences is due to the sinusoidal stress field and when @xmath71 approaches @xmath11 , the plastic flow regime is immediately encountered with a negligible micro - plastic region . by increasing the mobile dislocation density or decreasing @xmath11 the micro - plastic regime broadens and also the stress at which extended plastic flow occurs decreases due to the increasing influence of the elastic interaction between the dislocations . this latter feature is also reflected in the decreasing stress at which the first strain burst is seen . by increasing @xmath12 , the strain burst magnitude increases which broadens the micro - plastic regime . the stress versus plastic strain behaviour for three different values of @xmath14 , using @xmath66 mpa and @xmath16 equal to @xmath63 is presently investigated . for all three simulations @xmath60 @xmath59 m . [ different_d_fig ] displays the resulting stress - strain curves demonstrating that there is no strong overall @xmath14 dependence . indeed , the stress at which the first strain burst occurs differs little for the three samples : @xmath72 mpa for @xmath73 , @xmath74 mpa for @xmath75 and @xmath76 mpa for @xmath77 . it is seen , however , that with increasing @xmath14 the scale of the discrete plastic strain bursts becomes finer , leading to small - scale differences in the curves . inspection of the minimum magnitude of the plastic strain bursts observed in this figure , and in figures [ different_tau_fig ] and [ different_rho_fig ] , reveals it to be approximately @xmath78 , indicating that the strain bursts correspond to dislocation motion over a distance equal to at least @xmath12 . it is worth noting that the existence of such _ minimum _ discrete strain bursts is an artifact of the periodicity used to approximate a system of infinite extent . indeed , in the limit @xmath79 the minimum strain bursts would approach zero . with @xmath66 mpa and a mobile dislocation density @xmath63 . b ) a close up of the stress - stain curves indicating that with increasing periodicity the minimum discrete strain bursts reduce in strain magnitude.,scaledwidth=95.0% ] to better understand the relationship between the mobile dislocation density and @xmath12 , a series of simulations is performed to investigate the statistics of the shear - stress at which the first strain burst occurs . for computational efficiency these simulations were done using the constant stress - rate loading mode with @xmath80 @xmath59 m and @xmath70 mpa . [ first_burst_fig]a displays such shear - stress values as a function of @xmath12 for three different mobile dislocation densities . for each @xmath12 and dislocation density value , the results of many different simulations are shown to indicate the degree of statistical variation . inspection reveals that by increasing the dislocation density , the first - burst stress value becomes increasingly dependent on the value of @xmath12 , decreasing with decreasing @xmath12 . for all three considered dislocation densities , the degree of scatter increases with decreasing @xmath12 . for three dislocation densities for @xmath70 mpa and b ) @xmath66 mpa at a dislocation density of @xmath81.,scaledwidth=95.0% ] the state of the dislocation configuration prior to loading provides an understanding of the results of fig . [ first_burst_fig]a : @xmath82 dislocations are introduced into the dipolar mat by randomly placing them along the two possible slip planes and the system is then relaxed to a local equilibrium configuration . for large enough @xmath12 this relaxation process will result in similar numbers of dislocations for each @xmath12 unit , producing a dislocation configuration that is increasingly ordered . however , as @xmath12 decreases , there will exist an increasingly varying amount of dislocations for each @xmath12 unit , and the original disorder associated with the random initial positions of the dislocations is increasingly preserved . these two trends constitute the origin of the increased scatter and general decrease seen in fig . [ first_burst_fig]a with respect to decreasing @xmath12 , since the first - burst stress value is directly related to the lowest critical shear - stress required for an irreversible configurational change to occur . that is , the first - burst stress probes the extremal values of the dislocation environment and a greater variation in dislocation environment will naturally lead to a greater variation in the first - burst stress , which also increases scatter and results in an on - average reduced stress magnitude . dislocation interaction enhances this effect by generally increasing the degree of disorder in the low - lambda limit . the above result is , in fact , dependent on the system size ( the periodicity length , @xmath14 ) being finite . as the system size grows at a fixed value of @xmath12 and dislocation density , the chances for an extremal dislocation configuration to occur increases , resulting in a increased scatter and reduction of the first - burst shear - stress . it is therefore expected that as the system size increases , the saturation in first burst stresses as a function of increasing @xmath12 will be shifted to larger values of @xmath12 . such a size effect must also be reflected in the choice of @xmath11 which , along with the dislocation interaction , determines the degree of configurational disorder in the initial state . for a given value of @xmath12 , a smaller value of @xmath11 should increase the disorder due to the increasing influence of the dislocation interaction on the initial configuration . [ first_burst_fig]b displays a similar figure of first burst stresses as a function @xmath12 using a value of @xmath66 mpa and , indeed , demonstrates a shift to the right of the saturation region . although these results are an artifact of the system size , they reveal that through proper choice of model parameters , the intermittent plasticity seen in the past sections is driven by the extreme value statistics associated with the initial dislocation configuration a behaviour that is to be equally expected in real crystals . the statistical properties of the strain burst magnitudes are now investigated for a system with @xmath66 mpa , @xmath60 @xmath59 m , and @xmath16 equal to @xmath63 . three systems with @xmath83 , @xmath84 and @xmath85 @xmath59 m are first considered . the stress - strain curves for the @xmath73 and @xmath75 systems span up to the flow stress regime at approximately @xmath66 mpa , whereas for the @xmath77 data is only available up to approximately 9.4 mpa . [ strain_burst_fig ] plots the magnitude of the strain burst as a function of the shear - stress at which it occurred , for all three samples . in the first instance , the plastic strain magnitudes for each stress step ( as calculated via eqn . 6 ) were analysed and it was found that irreversible plasticity via discrete strain bursts begins at approximately 3 mpa with magnitudes equal to @xmath78 ( indicated by the corresponding coloured arrows ) . with increasing stress , larger strain bursts occur with magnitudes equal to multiples of the basic strain burst unit for each sample . [ strain_burst_fig]b displays the entire range of strain bursts using a vertical log - scale , and it is seen that the spread in strain bursts increases rapidly as the shear - stress approaches the flow regime at approximately @xmath66 mpa . . the arrows indicate the ( minimum ) fundamental strain bursts for each sample , indicating that larger strain bursts occur as multiples of these values . b ) vertical log plot for the entire applied stress range demonstrating that as the flow stress regime is approached individual strain bursts can become large.,scaledwidth=95.0% ] plotting the strain - burst magnitude distribution derived from the data of fig . [ strain_burst_fig ] , reveals a power law behaviour which is similar for each of the three considered systems . however from these graphs ( not shown ) , a reliable exponent could not be obtained and to investigate this aspect further improved statistics is required . [ strain_burst_dist1_fig]a shows strain - burst magnitude distributions from a much larger number of events , plotted using a log - log scale , for three different choices of parameters : 1 ) @xmath60 @xmath59 m and @xmath66 mpa ( same as that of fig . [ strain_burst_fig ] ) , 2 ) @xmath86 @xmath59 m and @xmath66 mpa , and 3 ) @xmath60 @xmath59 m and @xmath70 mpa . for all simulations @xmath80 @xmath59 m and @xmath87 m@xmath88 . for 1 ) , eleven deformation simulations were performed , and for 2 ) and 3 ) sixteen independent simulations were performed for each system . all distributions were obtained by logarithmic binning of identified irreversible plastic strain events . inspection of this figure reveals good power law behaviour over a number of decades . this is particularly the case for the parameter set 1 ) , in which more statistics could be obtained from each individual curve . indeed , for this parameter - set , fig . [ first_burst_fig ] shows that the first - burst stresses are low and exhibit strong scatter indicating significant intermittent plasticity . for this parameter set , a power law fit with an exponent of @xmath89 was found . visual inspection of the remaining two parameter sets reveals an approximately similar exponent . for parameter set 2 ) , finite size effects for the largest strain burst magnitudes are apparent . this is to be expected since at @xmath86 @xmath59 m , there exists just 80 @xmath12 lengths within the @xmath80 @xmath59 m system , whereas for parameter sets 1 ) and 3 ) , where @xmath60 , there exist 400 @xmath12 lengths . the above mentioned data of fig . [ strain_burst_dist1_fig]a was obtained by logarithmic binning of all discrete irreversible plastic events in the micro - plastic regime . in the recent mean - field theory ( mft ) work of dahmen and co - workers @xcite this is referred to as the stress integrated distribution , @xmath90 in which @xmath91 is the strain burst magnitude , whose integrand is given by the stress dependent distribution function , @xmath92 where @xmath93 is the critical stress at which extended plastic flow occurs , and the exponents @xmath94 and @xmath95 , and the scaling function @xmath96 are universal . from mft theory @xmath97 and @xmath98 . performing the stress integral gives @xmath99 where the last step uses the mft exponents . the data of fig . [ strain_burst_dist1_fig]a is therefore in excellent agreement with that of mft . further insight can be given via the stress - binned complementary cumulative distribution , @xmath100 whose stress integral gives a power law @xmath101 . [ strain_burst_dist1_fig]b , plots this quantity for the case of @xmath60 @xmath102 and @xmath66 mpa and gives an exponent equal to @xmath103 . fig . [ strain_burst_dist1_fig]b also plots @xmath104 for a selection of stress intervals revealing strong cut - off effects due to the tunable criticality of the mft model . following ref . @xcite , these curves are found to collapse when appropriately scaled via the function @xmath105 onto the universal mean field result , as shown in fig . [ strain_burst_dist1_fig]c , giving quantitative evidence that scale free avalanche behavoir does indeed occur for the current model . here @xmath93 is taken as @xmath11 . similar behaviour is seen for the case of @xmath86 @xmath59 m and @xmath66 but with @xmath106 in which the @xmath107 is an adjustment parameter to correct for finite size effects @xcite . for the case of @xmath60 @xmath59 m and @xmath70 the statistics was not of a sufficient quality to perform the procedure . when compared to more realistic two and three dimensional simulations that include dislocation reactions and much larger dislocation numbers , it might seem remarkable that such a simple model is able to exhibit scale - free behaviour . that this is the case , is a central feature of the soc phenomenon in which the universal features of the critical dislocation configuration are robust against the details of the underlying physical model . moreover , due to this simplicity , extremely good power law behaviour is obtained over many decades for a minimal computational effort . @xmath59 m and @xmath66 mpa and c ) via an appropriate scaling procedure where @xmath108 all stress dependent complementary cumulative distributions collapse on the universal curve given by mean field theory see ref . the simulations of sec . [ sec_results ] demonstrate that two dominant factors control the characteristic stress scale of the stress - strain curve . these are the choice of @xmath11 and the mobile dislocation density . figs . [ different_tau_fig ] and [ different_rho_fig ] demonstrate that @xmath11 sets an upper stress limit for extended plastic flow to occur . this upper limit is approached when the mobile dislocation density is sufficiently low that the sinusoidal stress field is the dominant contribution to the stress field each dislocation feels . by increasing the mobile dislocation content , the characteristic shear - stress scale reduces due to the increasing role of the internal stress fields arising from the elastic interaction between dislocations . at the same time a broadening of the micro - plastic regime is observed , which very much is a general property of micro - yielding , where a greater number of mobile edge dislocations correlates with larger measurable plastic strain @xcite . figs . [ different_lambda_fig ] to [ first_burst_fig ] demonstrate the effect of @xmath12 ( and @xmath14 ) on the stress - strain curve is somewhat subtler . the choice of @xmath12 influences the initial configuration of the mobile dislocation population , thereby affecting the statistics of the stress at which the first strain - burst occurs and also the way in which the extended plastic flow regime is reached . [ strain_burst_dist1_fig ] shows , however , that the statistics of the strain burst magnitude is insensitive to all three discussed parameters , reflecting the universality of soc with respect to micro structural details . how shall one quantitatively interpret the imposed sinusoidal stress field ? in a real material that is nominally free of dislocation structures , some type of length scale will naturally emerge as a function of macroscopic plastic strain due to the evolution of a growing and interacting dislocation structure that eventually leads to the phenomenon of patterning @xcite . although patterning is a term predominantly referring to the effects of latter stage ii and iii hardening regimes , the emergence of micro - structure length scales is expected to occur at all stages of plasticity ranging from slip , dipole and eventually cellular patterns @xcite . in fact , a micro - structural length - scale can equally well be defined for an undeformed as - grown material , where the mean dislocation spacing can be used to describe the initially present internal stress fluctuations a view point that is central to the early work of tinder and co - workers @xcite . from this perspective the imposed sinusoidal stress field , can be viewed as the simplest realisation of the internal stress field arising from such a structure . in the present model this stress field is time independent , implying it is constructed by that part of the dislocation population that is immobile , with the explicit dislocations and their dynamics , arising from the ( much smaller ) mobile component of the dislocation population . thus a typical loading simulation can be seen as the deformation of a model material that has a particular sample preparation or deformation history characterised by @xmath11 and @xmath12 , and a mobile dislocation density that is only a small part of the total dislocation density . much past work exists concerning the emergence of internal stress and length scales as a function of deformation history . in early work on the theory of cell formation , two relationships have emerged in which the total evolving dislocation density , @xmath109 , plays a central role . they are : @xmath110 and @xmath111 where @xmath20 is a representative ( not necessarily pure ) shear modulus and @xmath112 the burgers vector magnitude . in the above , @xmath113 is the evolving flow stress of the material and @xmath114 is an evolving internal length scale that can be referred to as a cell size . the first expression has its earliest origins in a taylor hardening picture in which the total dislocation density is seen as an immobile forest dislocation population . the second equation has its theoretical origins in the early work of holt @xcite who derived it for a dipolar population of screw - dislocations , showing that a uniform arrangement was unstable to fluctuations with one such length scale dominating , characterised by eqn . [ eqcell2 ] . this length scale , which could be related to a fixed self - screening distance of the dislocation network , was postulated to reflect an emerging cell size . the approach was based on an energy minimisation principle , however due to dislocation reactions , the more modern viewpoint is that the dynamics of cell formation lies in a statistical process involving dislocation reactions and that the screening length , and therefore cell - size , is an evolving variable @xcite . thus , the model parameter @xmath12 has a direct counterpart in cell formation theory , @xmath114 , which can represent quite generally , a mean - free path for a mobile dislocation , a dipolar screening length or a well evolved cell length scale . moreover , since an unloading / loading cycle will generally return a system to the flow stress before unloading , and that the present simulations have shown that the flow stress is partly controlled by @xmath11 , @xmath11 should be in some way related to @xmath113 . from this perspective , @xmath11 and @xmath12 are parameters that are not entirely independent from each other . in fact , eqns . [ eqcell1 ] and [ eqcell2 ] express that the cell size decreases inversely as a function of flow stress , a well known experimental observation that is refered to as `` similitude '' @xcite . although similitudity is generally confirmed by experiment , some experimental work does present a more complicated picture . early tensile / tem work on tapered cu single crystals finds an initially broad distribution of cell sizes that narrows and shifts to small lengths with increasing flow stress @xcite . this result suggests that a single structural length - scale might not always be a good statistical description of the evolving micro - structure . indeed , more modern viewpoints , in which dislocation structure evolution is a non - equilibrium process @xcite , tend to suggest a distribution of emerging length - scales leading to a scale - free fractal - like structure . although such micro - structures have been quantitatively established by tem investigations of latter - stage hardened single crystals of cu @xcite , their existence is not universal , depending strongly on material type and deformation history . the current work does not address this aspect . more general forms of an inhomogeneous internal stress field that capture such scale - free micro - structural features can be envisioned ; a direction which will be investigated in future work . whilst the dipolar mat geometry in an external field offers a platform with which to study the depinning transition and more generally the transition to extended plastic flow , when comparing to experiment , careful consideration has to be given to its regime of applicability . to do this , a typical simulation of secs . [ sec_results ] and [ sec_statistics ] is now broadly summarised . upon choosing numerical values for all model parameters , the @xmath82 dislocations are introduced to the system via a distribution of random positions . this unstable configuration is then relaxed to a local minimum energy in which the forces on each dislocation are below a small threshold value . the deformation simulation is then begun using one of the three loading modes of sec . [ sec_loading ] . as the stress increases , intermittent plasticity increasingly occurs until a stress is reached at which extended and overlapping strain events occur , which in the previous sections has loosely been referred to as the plastic flow regime . it is important to emphasise that no attempt has been made to obtain the global energy minimum of the starting configuration . such an initial state turns out to play a crucial role in the observed properties of the model , since many high energy configurations will exist , and it is these that dominate the early stages of plasticity . as a deformation simulation proceeds , such high energy configurations structurally transform eventually leading to a plastic flow regime and often to the homogenisation of the dislocation configuration . in other words , the extended plastic flow regime should be considered to be outside the applicability regime of the present model when a comparison to experiment is made , or equivalently , the present model is only suitable for the study of the micro plastic regime of the stress - strain curve . the rational behind the use of an initial high - energy dislocation configuration originates from the assumption that the explicit dislocation population of the model represents only the mobile dislocation network , which constitutes only a small part of the true population . thus , in the same way as @xmath11 and @xmath12 characterise the sample preparation or deformation history of the model material , so does the initial high energy ( explicit ) mobile dislocation content . this is quite compatible from the perspective of soc in which the dislocation structure reaches a critical configuration that is far from equilibrium , and that structural rearrangements correspond to the system transforming from one soc state to another . by construction , that part mediating the structural transformation will be the current mobile dislocation content . the central simplification of the present model , is that it separates the mobile and immobile populations , associating the former to an explicit mobile dislocation content that represents the non - equilibrium component of the network , and relegating the latter to an effective static internal stress field . that this internal stress field is unchanging and that the same explicit mobile dislocation population exists as a function of strain for the entire deformation simulation , is of course different from a real material , where the structure evolves with strain , and at any particular non - negligible strain interval , quite different dislocations might constitute the mobile dislocation population . this again emphasises that the present model should only be applied to the micro plastic regime , where significant structural evolution is minimal . experimental evidence for a lack of structural evolution in the mico - plastic regime is best seen in low amplitude cyclic deformation experiments of fcc metals , in which the plastic strain per cycle can be as low as @xmath115 leading to significant changes in load stress and internal length scale only after the occurrence of several tens - to - hundreds of thousands of cycles @xcite . it is further noted , that documented experimental studies on the micro - plasticity at room temperature primarily report on movements of edge or non - screw type dislocations , whereas a clear increase in dislocation density or the formation of dislocation structures as a result of multiplication remains absent @xcite . in the bulk case there are exceptions to this trend where in the case of a work - hardened al - mg alloy which exhibits dynamic strain ageing , emerging structural length - scales were already detected in the micro - plastic regime @xcite using high - resolution extensometry methods @xcite . the results of sec . [ sec_statistics_dist ] demonstrate that the developed model exhibits power - law behaviour in the distribution of strain burst magnitudes , and thus the scale - free avalanche phenomenon seen in experiments , either via the stress - strain curve of micro - compression tests @xcite or via in situ acoustic emission experiments @xcite , and in simulation , via two or three dimensional dislocation dynamics simulations in which the entire network is represented by an explicit dislocation population and individual dislocation reaction mechanisms are taken into account . with an exponent of approximately -1.96 , the model gives a value that is somewhat higher than that seen in simple metals @xcite and ice @xcite , but more comparable to that seen in lif crystals @xcite . that such a simple model can admit scale - free behaviour , is connected to the dependence of the intermittent plasticity on the extremal configurations of the explicit dislocation population . this was directly seen in the statistics of the first - burst shear - stress and also the distribution of strain burst magnitudes , where with a large enough increase of @xmath12 ( say from 2 @xmath59 m to 10 @xmath59 m ) the first burst statistics changes from being dominated by extreme value statistics to that being dominated by the statistics of the most probable ( fig . [ first_burst_fig]b ) corresponding to an increased presence of cut - off effects in the statistics of strain burst magnitudes ( fig . [ strain_burst_dist1_fig]a ) . this is a natural result of the observation that quantities that depend on extreme value statistics can exhibit power - law behaviour in their distributions , emphasising a connection with soc that is related to only the mobile dislocation population being in a non - equilibrium state and not to the characteristics of the present simplified immobile dislocation network a manifestation of a scenario referred to as `` nearly critical '' or `` robust critical '' @xcite . a simplified two dimensional dislocation modelling framework has been introduced in which the explicit interacting dislocation population , constrained to a simple dipolar mat geometry , represents only the mobile dislocation density component of the total dislocation density , and the much larger immobile dislocation population is described by a static internal sinusoidal shear - stress field defined by an internal shear - stress amplitude and wavelength . these model parameters , along with the initial non - equilibrium explicit mobile dislocation content characterise either the deformation or sample preparation history of the model material . because of the static nature of the internal field and the lack of dislocation - dislocation reactions , upon loading , the present model is restricted to the micro - plastic region of the stress - strain curve , and therefore to a deformation regime for a given material that involves negligible structural evolution . despite the simplicity of the model and the restriction to the micro - plastic regime , the deformation behaviour exhibits a rich variety of properties as a function of the model parameters . in particular , intermittent plasticity is observed whose strain burst magnitude distribution exhibits scale - free avalanche behaviour . p.m.d wishes to thank p. d. ispnovity for useful discussions . r.m . gratefully acknowledges the financial support of the alexander von humboldt foundation , his host g.m . pharr , as well as institutional support by j.r . greer at caltech . due to the long range nature of eqn . [ eqddint ] , all image contributions must be taken into account for the correct treatment of the periodic boundary conditions along the dipolar mat direction . the force per unit dislocation length on a dislocation @xmath116 ( at position @xmath117 ) due to a dislocation @xmath118 ( at position @xmath119 ) and it s images , is given by the infinite summation : @xmath120 due to the periodic boundary condition being one dimensional , an analytic solution to the above summation can be found . to do this , the summation in eqn . [ eq1app ] is re - written as @xmath121 where @xmath122 and @xmath123 , and @xmath124 . the above summation is explicitly convergent for all values of @xmath125 , and may be evaluated analytically in terms of poly - gamma functions . this is achieved by first expressing the summand in eqn . [ eq2app ] as an irreducible partial fraction : @xmath126 in this form , the summation to a finite @xmath125 may be straightforwardly obtained via a known series expansion of poly - gamma functions @xcite , giving @xmath127\ ] ] and @xmath128\ ] ] where @xmath129 is the @xmath130th derivative of the di - gamma function ( the logarithmic derivative of the gamma function @xcite ) . performing the substitutions and taking the leading order term as @xmath131 results in eqn . [ eq2app ] evaluating to @xmath132\right . \nonumber \\ \left .- i\pi^{2}\delta y\left[\cot^{2}\left(\frac{\pi(\delta x+\imath\delta y)}{d}\right)+\cot^{2}\left(\frac{\pi(\delta x-\imath\delta y)}{d}\right)\right]\right ] . \label{eq3app}\end{aligned}\ ] ] for a form that is more amenable to computation the trigonometric identity , @xmath133 is used , to obtain the final result for eqn . [ eq2app ] as @xmath134}{d^{2}\left(\cos\left(\frac{2\pi\delta x}{d}\right)-\cosh\left(\frac{2\pi\delta y}{d}\right)\right)^{2}}. \label{eq4app}\ ] ] eqn . [ eq4app ] with the prefactor @xmath135 replaces eqn . [ eqddint ] in eqn . [ eqtotalforce ] when used in evaluating the total force of each dislocation : eqn . [ eqddintimage ] . for dislocations on the same slip plane , @xmath136 , and eqn . [ eq4app ] reduces to the known result @xcite , @xmath137 using , eqn . [ eq4app ] with a dipolar mat periodicity length of @xmath65 @xmath59 m , fig . [ dd_interaction_fig ] displays the shear - stress dependence between two dislocations of opposite burgers vector and therefore on different slip planes ( see fig . [ schematic_fig ] ) , for different values of the dislocation density and thus different @xmath3 . inspection of this figure reveals the short range structure to be similar to that of the bare interaction given by eqn . [ eqddint ] in sec . [ sec_model ] , whereas at larger distances the interaction correctly limits to zero at @xmath138 . in this figure the chosen numerical values of the mobile dislocation density ( where @xmath44 ) will be typical of those used in the present work . inspection of this figure reveals that for the highest dislocation density , the restoring stress is at most of the order of 1 mpa and thus small when compared to the considered values of @xmath11 , and also typical values of the stress associated with intra - slip - plane dislocation interaction at sub - micron distances ( see the dashed line in fig . [ dd_interaction_fig ] , which displays the restoring shear - stress between two dislocations of the same burgers vector on the same slip plane ) . to investigate the regime in which the two slip planes strongly interact , higher mobile dislocation densities would need to be considered or a different relationship between @xmath3 and the dislocation density taken . 99 r. f. tinder and j. washburn , acta . * 12 * , 129 ( 1964 ) . r. f. tinder and j. p. trzil , acta metall . * 21 * , 975 ( 1973 ) . h. neuhuser , _ slip line formation and collective dislocation motion _ , in _ dislocations in solids , volume 6 , 1983 _ , edited by f. r. n. nabarro ( elsevier , amsterdam , 1983 ) , p. 319 m. d. uchic , d. k. dimiduk , j. n. florando and w. d. nix , science * 305 * , 986 ( 2004 ) . j. r. greer , w. c. oliver and w. d. nix , acta mater . * 53 * , 1821 ( 2005 ) . d. k. dimiduk , c. woodward , r. lesar and m. d. uchic , science * 312 * , 1188 ( 2006 ) . d. k. dimiduk , e. m. nadgorny , c. woodward , m. d. uchic and p. a. shade , phil . mag . * 90 * , 3621 ( 2010 ) . c. a. volkert and e.t . lilleodden , phil . mag . * 86 * , 13 ( 2006 ) . r. maa and m. d. uchic , acta mater . * 60 * , 1027 ( 2012 ) . r. maass , l. meza , b. gan , s. tin , j. r. greer , small * 8 * , 1869 ( 2012 ) . m. zaiser , j. schwerdtfeger , a. s. schneider , c. p. frick , b. g. clark , p. a. gruber and e. arzt , phil . mag . * 88 * 3861 ( 2008 ) . c. fressengeas , a. j. beaudoin , d. entemeyer , t. lebedkina , m. lebyodkin , and v. taupin , phys . b * 79 * , 014108 ( 2009 ) . r. n. mudrock , m. a. lebyodkin , p. kurath , a. j. beaudoin and t. a. lebedkinac , scripta materialia * 65 * 1093 ( 2011 ) . k. a. dahmen , y. ben - zion , and j. t. uhl , phys . * 102 * , 175501 ( 2009 ) n. friedman , a. t. jennings , g. tsekenis , j .- y . kim , m. tao , j. t. uhl , j. r. greer and k. a. dahmen , phys . lett . * 109 * , 095507 ( 2012 ) . sethna , k. a. dahmen and c. r. myers , nature * 410 * , 242 ( 2001 ) . b. a. sun , h. b. yu , w. jiao , h. y. bai , d. q. zhao , and w. h. wang , phys . lett . * 105 * , 035501 ( 2010 ) . c . miguel , a. vespignani , s. zapperi , j. weiss and j.r . grasse , nature * 410 * , 667 ( 2001 ) . j. weiss and d. marsan , science * 299 * , 89 ( 2003 ) . see for example , _ dislocations in solids , volume 11 , 2002 _ , edited by f. r. n. nabarro and j. hirth ( elsevier , amsterdam , 2002 ) . r. j. amodeo and n. m. ghoniem , phys . b * 41 * , 6958 ( 1990 ) ; phys . rev . b * 41 * , 6968 ( 1990 ) . b. devincre and v. pontikis , mat . . proc . * 291 * , 555 ( 1993 ) . r. fournet and j. m. salazar , phys . b * 53 * , 6283 ( 1996 ) . b. bak , i. groma , g. gyrgyi , g. zimnyi , comput . * 38 * , 22 ( 2006 ) . e. van der giessen and a. needleman , modelling simul . eng . * 3 * , 689 ( 1995 ) . d. ispnovity , i. groma , w. hoffelner and m. samaras , modelling simul . * 19 * , 045008 ( 2011 ) . v. s. deshpande , a. needleman and e. van der giessen , j. mech . * 53 * , 2661 ( 2005 ) . p. d. ispnovity , i. groma , g. gyrgyi , f. f. csikor and d. weygand , phys . * 105*. 085503 ( 2010 ) . d. weygand , l . h. friedman , e. van der giessen and a. needleman , mod . sim . mat . * 10 * , 437 ( 2002 ) . v. v. bulatov and w. cai , _ computer simulations of dislocations _ , oxford series on materials modelling , edited by a. p. sutton and r. e. rudd , ( oxford university press , oxford , 2006 ) . f. f. csikor , c. motz , d. weygand , m. zaiser and s. zapperi , science * 318 * , 251 ( 2007 ) . rao , d.m . dimiduk , t.a . parthasarathy , m.d . uchic , m. tang , c. woodward , acta . mater . * 56 * , 3245 ( 2008 ) . b. devincre , t. hoc and l. kubin , science * 320 * , 1745 ( 2009 ) . m. zaiser , adv . phys . * 55 * , 185 ( 2006 ) . p. bak , c. tang and k. wiesenfeld , phys . rev . lett . * 59 * , 381 ( 1987 ) . d. kuhlmann - wilsdorf , _ the les theory of solid plasticity _ , in _ dislocations in solids , volume 11 , 2002 _ , edited by f. r. n. nabarro and j. hirth ( elsevier , amsterdam , 2002 ) , p. 211 . p. moretti , m .- carmen miguel , m. zaiser and s. zapperi , phys . b * 69 * , 214103 ( 2004 ) . f. leoni and s. zapperi , phys . lett . * 102 * , 115502 ( 2009 ) . young , j. appl . phys . * 32 * , 1815 ( 1961 ) . g. vellaikal , acta metall . * 17 * , 1145 ( 1969 ) . n. hansen and d. kuhlmann - wilsdorf , mat . 81 * , 141 ( 1986 ) . d. hull and d. j. bacon , _ introduction to dislocations _ , ( butterworth heinemann , oxford , 2001 ) , chapter 4.6 . see chapter 10.3 of ref . @xcite b. devincre , thse de doctorat , paris xi orsay , 1993 . a. abel , acta metall . * 21 * , 99 ( 1973 ) . a. s. argon , _ strengthening mechanisms in crystal plasticity _ , oxford series on materials modelling , edited by a. p. sutton and r. e. rudd , ( oxford university press , oxford , 2008 ) . h. mughrabi , acta metall . * 31 * , 1367 ( 1983 ) l. p. kubin and g. canova , scripta metall . * 27 * , 957 ( 1992 ) . d. l. holt , j. appl . phys . * 41 * , 3197 ( 1970 ) . f. prinz and a. s. argon , phys . ( a ) * 57 * , 741 ( 1980 ) . p. hhner , acta mater . * 44 * , 2345 ( 1996 ) . p. hhner , k. bay and m. zaiser , phys . * 81 * , 2470 ( 1998 ) . m. zaiser , k. bay and p. hhner , acta mater . * 47 * , 2463 ( 1999 ) . h. mughrabi , mat . eng . * 33 * , 207 ( 1978 ) . s. basinski and s. j. basinski , prog . mat . 36 * , 89 ( 1992 ) . t. j. koppenaal , acta metall . * 11 * , 86 ( 1963 ) . a. roth , t.a . lebedkina , m.a . lebyodkin , materials science and engineering a * 539 * , 280 ( 2012 ) . lebyodkin , t.a . lebedkina , a. roth and s. allain , acta physica pol . a * 122 * , 478 ( 2012 ) . m. abramowitz and i. a. stegun , eds . _ handbook of mathematical functions with formulas , graphs , and mathematical tables . _ new york : dover . 258 - 260 .

we map the ground - state ensemble of antiferromagnetic ising model of spin- on a triangular lattice to an interface model whose entropic fluctuations are proposed to be described by an effective gaussian free energy , which enables us to calculate the critical exponents of various operators in terms of the stiffness constant of the interface . monte carlo simulations for the ground - state ensemble utilizing this interfacial representation are performed to study both the dynamical and the static properties of the model . this method yields more accurate numerical results for the critical exponents . by varying the spin magnitude in the model , we find that the model exhibits three phases with a kosterlitz - thouless phase transition at and a locking phase transition at . the phase diagram at finite temperatures is also discussed .

cond-mat9609007

over the years it has been found that there exist many two - dimensional classical spin models , discrete and continuous alike , whose ground - state manifolds are macroscopically degenerate and , more interestingly , also exhibit critical behaviours , i.e. , spin - spin correlation functions within the ground - state ensembles decay with distance as power laws . the classification of universality class for these models has always been a challenging problem@xcite an earlier example of this kind is the antiferromagnetic ising model on the triangular lattice . the exact solution for this model by stephenson@xcite showed that although this model remains paramagnetic at nonzero temperature , its ground state is critical . later works by blte _ et al _ revealed yet another remarkable property of the ground - state ensemble of this model , namely , it permits a solid - on - solid ( sos ) representation in which spin fluctuations are subsequently described by the fluctuating interface in the sos model@xcite . recent studies also demonstrated that this interfacial representation provides a valuable avenue for studying the ground - state ordering of quantum magnets@xcite and the ground - state roughness of oriented elastic manifolds in random media@xcite . other recently studied models with critical ground states include three - state antiferromatic potts model on the kagom lattice@xcite , the @xmath3 model on the honeycomb lattice@xcite , the four - coloring model on the square lattice@xcite , and the square - lattice non - crossing dimer model and dimer - loop model@xcite . on the other hand , some very similar models with degenerate ground states exhibit long - range order , such as the constrained 4-state potts antiferromagnet@xcite . in this article we study the ground - state properties of antiferromagnetic ising model of general spin on a triangular lattice which also belongs to the class of models mentioned above . recent numerical studies of this model include monte carlo simulations@xcite and transfer matrix calculations@xcite . here we revisit this model by performing monte carlo simulations . the motivation of the present work is two - fold : ( 1)unlike previous simulations , we utilize the interfacial representation directly in analyzing the simulation results , for example , we compute the stiffness constant of the fluctuating interface which , in turn , yields more accurate critical exponents of various operators ; and ( 2 ) we also study the dynamical properties of this model for the first time making use of the interfacial representation . the body of the this paper is organized as follows . section [ model ] describes the model hamiltonian and maps it onto a spin-1 problem whose interfacial representation is then described . in section [ height - rep ] , we propose an effective continuum theory for the long - wavelength fluctuations of the interface . here we also show how to relate scaling dimensions of various operators to the stiffness constant of the interface , and derive some other analytical results based on this `` height representation . '' this allows analytical understanding of the phase diagram ( sec . [ phasediag ] ) . details of monte carlo simulations and numerical results on both dynamical and static properties are presented in section [ mc - results ] , including a comparison of the new and old approaches to determining the exponents . as a conclusion , the paper is summarized and various possible extensions are outlined , in section [ conc - disc ] . the antiferromagnetic ising model of spin-@xmath0 on a triangular lattice can be described by the following hamiltonian : @xmath4 where the spin variable @xmath5 defined on lattice site @xmath6 of the triangular lattice can take any value from a discrete set @xmath7 $ ] , and the sum over @xmath8 runs over three nearest - neighbor vectors @xmath9 , @xmath10 and @xmath11 as shown in fig . [ fig1 ] . here the coupling constant @xmath12 is positive describing the antiferromagnetic exchange interaction between two nearest - neighbor spins : @xmath5 and @xmath13 . one important reason for interest in this model is that the @xmath14 limit@xcite is the same as the ising limit of the ( classical or quantum ) heisenberg antiferromagnet on the triangular lattice with ising - like anisotropic exchange . that model was shown to exhibit a continuous classical ground state degeneracy and unusual features of the selection by fluctuations of ground states@xcite . the ground - state configurations of the above model given by eq . ( [ eq1 ] ) consist of entirely of triangles on which one spin is @xmath15 , another is @xmath16 , and the third can be anything in @xmath17 $ ] . thus , if spin @xmath18 takes an intermediate value @xmath19 , this forces the six surrounding spins to alternate @xmath15 and @xmath16 ; exactly which intermediate value @xmath18 takes does not matter in determining whether a configuration is allowed . therefore , this allows us to reduce each state @xmath20 to a state @xmath21 of a _ spin-1 _ model , by mapping @xmath22 to @xmath23 , @xmath24 to @xmath25 , and intermediate values @xmath26 to @xmath27 . in this _ spin-1 _ representation of the model , the rules for allowed configurations are exactly the same as for the @xmath28 model ; however instead of being equal , the statistical weights have a factor @xmath29 for each spin with @xmath27 . it should be noted that in the @xmath30 case , @xmath31 simply maps to @xmath32 . it can also be shown that the expectation of any polynomial in @xmath33 , in the ground - state ensemble of the spin-@xmath0 model , can be written in terms of a similar expectation in the spin-1 model . specifically , one must simply replace @xmath34 , & m even \cr}\ ] ] where ( e.g. ) @xmath35 . thus there is no loss of information in this mapping . indeed , in some sense , the extra freedom to have @xmath5 vary from @xmath36 to @xmath37 is trivial : once given that @xmath5 and @xmath38 are intermediate spin values , there is no correlation between these values . so we henceforth restrict ourselves to the spin-1 mapped model whose partition function for its ground - state ensemble can be written as : @xmath39 where @xmath40 denotes the number of zero spins in a ground - state configuration @xmath41 . by varying the weight factor continuously in the spin-1 model , it would possible to give a precise meaning to _ any _ real value of @xmath0 , and to simulate such an ensemble . however , in this article we perform monte carlo simulations for an ensemble in which @xmath42 takes only integer values . the spin-1 representation could be further reduced to a spin-1/2 representation @xmath43 as described in refs . . they let @xmath44 here @xmath45 if @xmath32 and if @xmath27 , @xmath46 or @xmath47 according to whether the surrounding spins are @xmath48 or the reverse . note this mapping is not invertible . the spin-@xmath49 representation is less satisfactory in that is arbitrarily breaks the up - down symmetry of correlation functions , but it was desirable for the transfer - matrix calculations of lipowski _ et al_@xcite since it reduced the number of degrees of freedom . we define a _ microscopic _ , discrete - valued height function @xmath50 living on the vertex of the triangular lattice such that the step in @xmath50 between adjacent vertices is a function of the adjacent spins : @xmath51 where @xmath52 is the spin-1 operator and @xmath53 can be any of the three nearest - neighbor vectors @xmath54 . it is easy to show that the total change in height function , when traversed along any smallest loop , i.e , an elementary triangle , is zero . therefore , @xmath50 is well - defined everywhere for the ground - state configurations , but it is not well - defined in any excited state . this prescription generalizes that originally introduced by blte et al for the case @xmath30@xcite ( the prescriptions agree in that case ) . this type of height mapping differs from other sorts of mapping ( e.g. dualities ) in a crucial way : since the spin microstates of the spin-1 model are mapped essentially one - to - one to the height microstates , it is possible to perform monte carlo simulations and construct configurations @xmath50 after each time step . we have found that analysis of the @xmath50 correlations is much more efficient for extracting critical exponents than analysis of the spin correlations directly as was done in previous monte carlo simulations@xcite . in this section we propose an effective continuum theory which describes the long - wavelength fluctuations of the interface . we also demonstrate how the critical exponents of various operators are determined by the stiffness constant of the interface . to describe the interface in the rough phase , we must define a smooth height field @xmath55 by coarse - graining the discrete field @xmath50 . as a first stage , on every triangular plaquette formed by sites @xmath56 , define a new discrete height @xmath57 where @xmath58 is the center of a triangle . the possible values of the @xmath59 are @xmath60 , for any integer @xmath61 . ( for the case @xmath30 , the only possible values are integers . ) to each of these values corresponds a _ unique _ ground - state spin configuration of the spin-1 model on that triangle , i.e. , @xmath62 where @xmath63 is any vector from a site to the center of any adjoining triangle . the mapping is many - to - one : the function @xmath64 has period 6 . notice that the r.h.s . of eq.([eq5 ] ) turns out to be independent of @xmath65 , but the periodic dependence on @xmath66 is phase - shifted by a function @xmath67 which takes different values on each of the three @xmath68 sublattices . essentially , we have mapped the @xmath69 ensemble of the spin-1 problem into an equivalent interface problem . note that , given a configuration of @xmath70 , each @xmath52 is specified ( via eq . ( [ eq5 ] ) ) , once for each adjoining triangle . the requirement that these six values of @xmath52 coincide translates to a somewhat complicated set of contraints between pairs @xmath59 and @xmath71 on adjoining triangles ; the difference @xmath72 may be 0 , @xmath73 , or @xmath74 , but some of these are disallowed ( depending on which @xmath75 values are integer or half - odd - integer , and on the orientation of @xmath76 ) . the weight of each configuration is given , as in ( [ eq2 ] ) , by by @xmath77 . [ fig1 ] shows the @xmath59 mapping explicitly where the spins @xmath52 take values from @xmath78 . the twelve states are arranged in a circle because the pattern repeats when @xmath79 . there are certain special `` flat states '' in which @xmath59 is uniform on all triangles . each of these is periodic with a @xmath80 unit cell in effect it is a repeat of one of the triangles in fig . we shall name these states by writing the spins on the three sublattices,``@xmath81 '' and `` @xmath82 '' ; here `` @xmath83 '' stands for @xmath84 . it should be noted that there are two non - equivalent species of flat state corresponding to integer , and half - integer valued @xmath59 respectively . they are non - equivalent in the sense that they are not related by _ lattice _ symmetries . one of the species that is favored by the _ locking potential _ ( see eq . ( [ eq - vlock ] ) below ) is what is previously called `` ideal '' states@xcite . thus we can imagine that all states can be described as domains of uniform @xmath59 separated by domain walls . finally , by coarse - graining @xmath59 over distances large compared to the lattice constant , one obtains @xmath55 which enters the conjectured continuum formula for the free energy , which is entropic in origin@xcite , @xmath85 where @xmath86 is the stiffness constant of the fluctuating interface . a lattice shift by one lattice constant leaves the free energy invariant , but induces global shifts in height space @xmath87 ; hence the potential @xmath88 in ( [ eq6 ] ) must have period one . it is typically approximated as @xmath89 such a periodic potential , usually referred as the _ locking term_@xcite , favors the heights to take their discrete values one of the two types of flat state , depending on the sign of @xmath90 . for large @xmath0 we expect @xmath91 , favoring the @xmath82 states , in view of the large entropy of flippable spins ; it is not so sure which state is favored at smaller @xmath0 , but this does not matter for the critical exponents ( see sec . [ scaling ] and [ operators ] , below . in the _ rough phase _ , by definition , the locking term is irrelevent , and so the long - wavelength fluctations of height variable @xmath55 are governed by the gaussian term of eq . ( [ eq6 ] ) : @xmath92 where we have performed the fourier transform . hence by equipartition theorem , @xmath93 similarly , we can also measure the _ height - height difference function _ in the real space as : @xmath94 ^ 2\rangle \nonumber \\ & = & \frac{1}{2\pi k } \ln(\pi r / a ) + ... \;\ ; ( r \gg 1 ) \;\ ; , \label{eq9 } \end{aligned}\ ] ] where @xmath95 is the lattice spacing cutoff . ( [ eq9 ] ) , we can compute the scaling dimension @xmath96 of any _ local _ operator @xmath97 , which is defined as in the correlation function , @xmath98 by local operator , we mean that @xmath97 is a local function of spin operators in the vicinity of @xmath99 . now , the same spin configuration is recovered when the height variable @xmath59 is increased by 6.@xcite thus any local operator @xmath97 is also a periodic function in the height space , and can consequently be expanded as a fourier series : @xmath100 where @xmath101 runs over height - space reciprocal - lattice vectors ( i.e. multiples of @xmath102 ) . the last step of simplification in ( [ eq11 ] ) follows because the scaling dimension @xmath96 of the operator @xmath97 is determined by the leading relevant operator in the above expansion , i.e. , @xmath103 is the smallest @xmath101 with nonzero coefficient in the sum . inserting eq . ( [ eq11 ] ) into eq . ( [ eq10 ] ) and making use of eq . ( [ eq9 ] ) , we obtain the following : @xmath104 therefore , the critical exponent @xmath105 associated with the operator @xmath97 is given by : @xmath106 in this paper , besides the usual spin operator @xmath52 , we also study the bond - energy operator @xmath107 for the reason that will become clear in the next section : @xmath108 where @xmath8 denotes one of the three nearest - neighbor vectors as before . as discussed already , the spin operator on a given site has a periodicity of @xmath109 in the height space , from which a simple inspection shows that the bond - energy operator is also periodic in the height space with a periodicity of @xmath110 . therefore , the reciprocal lattice vectors of the most relevant operator in the fourier expansion in eq . ( [ eq11 ] ) are @xmath111 for spin and bond - energy operators respectively . if a magnetic field is implemented by adding a term @xmath112 to the hamiltonian , then our dimensionless uniform `` magnetic field '' is defined by @xmath113 . the exponents associated with @xmath114 ( and with the uniform magnetic susceptibility ) , are easily related to the correlation exponents of the uniform magnetization operator , @xmath115 where @xmath58 is the center of a triangle formed by sites @xmath56 . a simple inspection of fig . [ fig1 ] shows that such an operator has a periodicity of @xmath116 in the height space , thus yielding : @xmath117 observe that the microscopic height variable @xmath50 in any flat state is not uniform but is rapidly modulated with the wave vector @xmath118 . the amplitude of modulation itself is a periodic function of the _ coarse - grained _ height field @xmath55 which in turn implies that the correlation function decays with distance as a power - law , and consequently that its structure factor has a power - law singularity at @xmath119 . such a zone - corner singularity is also directly connected to the singularity in the structure factor of the bond - energy operator . to see this , recall that there is a linear relation between the microscopic height variables and the bond - energy operator given by eqs . ( [ eq3 ] ) and ( [ eq14 ] ) , i.e. , @xmath120 then it is interesting to note that the fourier transform @xmath121 of bond - energy operator given above turns out to be @xmath122 in other words , as a byproduct of measuring @xmath123 , we have at the same time measured the structure factor of , say , the bond - energy operator of the same orientation specified by the nearest - neighbor vector @xmath8 : @xmath124 we will utilize this relation in sec . [ structure ] to extract the exponent of bond - energy operator from the monte carlo simulations . the @xmath30 triangular ising antiferromagnet is exactly soluble , by the same techniques which solve the ferromagnetic two - dimensional ising model , and was immediately recognized to have critical behavior as @xmath125 . the spin and energy correlation functions were computed exactly by stephenson ; it transpires that @xmath126 and @xmath127 exactly , implying through the arguments of blte et al ( see sec . [ scaling ] and [ operators ] ) that the effective stiffness in eq . ( [ eq7 ] ) is @xmath128 exactly . the exponents implied by the interface scenario@xcite in particular , the magnetic field exponent @xmath129 are fully confirmed by numerical transfer - matrix computations.@xcite the coulomb gas picture of kondev _ et al_@xcite , wherein the @xmath30 triangular ising antiferromagnet is viewed as a fully - packed loop model@xcite with fugacity 1 , also predicts the exact exponents . in this section , we collect some consequences of the height representation for the phase diagram and the nature of the various phases within it.@xcite the locking potential @xmath88 in ( [ eq6 ] ) favors the flat states . in view of ( [ eq - vlock ] ) , its leading reciprocal - lattice vector is @xmath130 , corresponding to a scaling index @xmath131 for the corresponding conjugate field @xmath90 . it is well known that if @xmath132 , then @xmath90 becomes relevant ( under renormalization and the interface locks into one of the flat states.@xcite since @xmath86 grows monotonically with @xmath0 , such a locking transition occurs at a critical @xmath133 where @xmath134@xcite . in this `` smooth '' phase , any spin operator @xmath97 has long - range order , by arguments as in sec . [ scaling ] . one of our aims in this paper was to pinpoint the locking transition @xmath133 , which demands that we have a criterion to distinguish these phases . we must supplement eq . ( [ eq8 ] ) , which shows the expected qualitative behavior of height fluctuations @xmath135 in the rough phase , with a parallel understanding of the smooth phase . in the smooth state , the symmetry ( of height shifts ) is broken and a fully equilibrated system has long - range order , such that @xmath136 is well defined and uniform throughout the system . fluctuations around this height , then , have at most short - range ( exponentially decaying ) correlations . thus we expect them to have a spatial `` white noise '' spectrum : @xmath137 for small @xmath138 . a phase with `` hidden '' order was suggested by lipowski and horiguchi@xcite . numerical transfer - matrix calculations@xcite using the spin-1/2 representation indicated @xmath139 for @xmath140 , which is impossible if the spin correlations are derived from height fluctuations,@xcite as we reviewed in sec . [ height - rep ] . an exotic phase to reconcile these facts was to postulate a phase in which the interface was smooth and @xmath141 , yet for the real spins @xmath142 as suggested by spin correlation measurements . what does this imply for our height variable @xmath59 , which has a one - to - one correspondence with the real spin configuration @xmath143 ? if the interface is smooth , then the probability distribution of height values on a given plaquette , @xmath144 , is well defined . in order to `` hide '' the order , it is necessary that @xmath145 correspond to zero expectations of the spins . now , reversing @xmath5 on all three sites in the plaquette requires @xmath146 , as seen from fig . one can convince oneself that , to have ensemble average @xmath147 , the distribution @xmath145 must be at least as broad as @xmath148 , with @xmath149 , implying the bound @xmath150 \equiv \langle h({\bf r})^2 \rangle - \langle h({\bf r } ) \rangle ^2 \ge ( 3/2)^2 . \label{eq - dhbound}\ ] ] at @xmath151 , plaquettes with non - minimal energy are present and they correspond to vortices in the function @xmath55 . thus , unfortunately , the height approach of analyzing simulations more or less breaks down . nevertheless , one can still predict the @xmath151 phase diagram from knowledge of the @xmath69 stiffness constant derived from our simulations . the shape of this phase diagram has already been explained in ref . @xcite ; here we note some additional interesting behaviors which can be predicted ( following ref . ( b ) ) using the exponents associated with vortices . the other exponents in kosterlitz - thouless ( kt ) theory are associated with elementary defects ( often called vortices ) . indeed , it is easy to check ( in this system ) that the excess energy of a non - ground - state plaquette is directly proportional to its vortex charge ( a burgers vector in height - space ) , so the effect of nonzero temperature is simply to make the vortex fugacity nonzero . the vortex exponent is @xmath152 , so as usual the vortex fugacity becomes relevant and defects unbind , destroying the critical state , at the kt transition defined by a spin exponent taking the critical value @xmath153 . if @xmath154 at zero temperature , i.e. @xmath155 , then defects unbind as soon as @xmath151 . thus a zero - temperature kt transition occurs at @xmath156 defined by @xmath157.@xcite ref . did not , however , address the critical exponents of the correlation length @xmath158 and the specific heat @xmath159 as a function of temperature , which are also controlled by vortex exponents . naively , if the energy cost creating one vortex is @xmath160 , and if the minimum excitation is a vortex pair , then one would expect the low - temperature specific heat to behave as @xmath161 and at @xmath30 this is indeed true@xcite . however , the renormalization group@xcite shows the singular specific heat behaves as @xmath162 where @xmath163 is the vortex fugacity ; consequently when @xmath164 , the true behavior is @xmath165 with @xmath166 . ( physically , part of the excitation energy is cancelled by the large entropy due to the many places where the vortex pair could be placed . ) this behavior has been observed in the 3-state potts antiferromagnet on the kagom lattice@xcite , and should occur in the present system for all @xmath167 . it is interesting to consider the effect of a nonzero magnetic field @xmath114 . it is known already that at @xmath30,@xcite such a field is an irrelevant perturbation , so that the system remains in a critical state , yet at sufficiently large @xmath168 it undergoes a locking into a smooth phase,@xcite approximated by any of the three symmetry - equivalent flat states of type `` @xmath169 '' with magnetization @xmath170 as also already noted@xcite , there is a critical value @xmath171 defined by @xmath172 , beyond which @xmath173 so that the system locks into long - range order as soon as @xmath114 is turned on . within this regime , there are still two subregimes with different behavior of @xmath174 near @xmath175 . for @xmath176 , the initial slope is zero , i.e. , the susceptibility is not divergent ; when @xmath177 , as occurs for @xmath178 , there is a divergent susceptibility and correspondingly there should be a singularity at @xmath179 in the spin structure factor @xmath180 . what do we expect in the locked phase at @xmath181 ? here the difference between the two kinds of flat states becomes crucial . the @xmath114 field favors the @xmath169 type of flat state , but entropy favors the @xmath82 type of flat state . thus we expect a transition to the @xmath169 state only at a nonzero critical field @xmath182 . on reducing @xmath114 through @xmath182 , a twofold symmetry breaking occurs , in which one of the @xmath183 sublattices becomes the @xmath184 ( disordered ) sublattice ; hence , this transition should be in the ising universality class . presumably the line @xmath185 meets the @xmath186 axis at @xmath187 . there must also be line of locking transitions @xmath188 , which terminates on the @xmath186 axis at @xmath171 . for @xmath30 , the effect of the magnetic field was confirmed numerically in ref . . in this section we describe the implementation details of monte carlo simulations performed for spin-1 model in which @xmath42 takes only integer values from @xmath189 to @xmath190 . we then present numerical results for the relaxation times of slow modes in the monte carlo dynamics . two different methods of compute the critical exponents of the spin , bond - energy , and uniform - magnetization operators are described in different sub - sections : one in terms of the extrapolated stiffness constants of the interface and the other in terms of the singularities of the corrsponding structure factors . a spin is called _ flippable _ if its six surrounding nearest - neighbor spins alternate between @xmath191 and @xmath47 . clearly , changing the value of this flippable spin results in another new spin configuration in the ground - state ensemble , provided that we start with a spin configuration in the ensemble . moreover , such an update maintains the global tilt of the interface due to the _ local _ nature of this update . this update will be used as our monte carlo update in this paper . two slightly different cases arise for different values of @xmath42 : ( 1 ) for @xmath192 , the local update is precisely equivalent to a spin flip i.e. , @xmath193 due to the absence of zero spin ; and ( 2 ) for all other values of @xmath42 , a random choice must be made in the local update : for example , @xmath194 or @xmath47 . ( recall @xmath0 denotes the spin magnitude of the original model . ) let @xmath40 and @xmath195 denote the number of zero - spins and flippable spins of configuration @xmath196 . if an attempted single - spin update for @xmath196 results in a new configuration @xmath197 with @xmath198 and @xmath199 , then the transition probability @xmath200 in accordance with the detailed balance principle is : @xmath201 where @xmath202 denotes the _ bare _ transition probability : @xmath203 for @xmath192 , and @xmath204 for @xmath205 which reflects the random choice to be made in the local update as discussed above . with the transition probability given in eq . ( [ eq16 ] ) , it is straightforward to show that the detailed balance principle is satisfied , i.e. , @xmath206 , where @xmath207 denotes the probability for configuration @xmath196 to occur and @xmath208 since each spin configuration in the original spin-@xmath0 model has equal probability to occur . note also that @xmath209 for large @xmath210 , so this rule is important only because of the finite system size . to implement in practice the transition probability given above , we randomly select a site out of a list of the @xmath195 flippable sites , and randomly update this spin to one of the two possible new spin values if @xmath211 or simply flip this spin if @xmath192 . the total numbers of zero spins @xmath198 and flippable spins @xmath199 in the resulting configuration are then computed . this update is subsequently accepted with a probability : @xmath212 . a practical implementation of the transition probability given in eq . ( [ eq16 ] ) is thus achieved . throughout this paper , a unit time or one monte carlo sweep ( msc ) is defined such that there are @xmath213 attempts of updating within this unit of time ( or one attempt per spin on average ) . here @xmath213 denotes the total number of spins in the simulation cell . the simulation cell always contains @xmath214 spins in this paper unless explicitly mentioned otherwise . periodic boundary conditions are adopted . since we always start with a flat state , the simulations are thus performed in the sector with a zero global tilt of the interface . we now discuss the correlations between the configurations generated sequentially in the monte carlo simulations by studying the relaxation time of the slow modes in the model , namely , the fourier modes @xmath216 which play the role of an order parameter@xcite . the linear - response dynamics of such a mode is usually formulated as a langevin equation , @xmath217 where @xmath218 is the dissipation constant , and the static free energy functional @xmath219 is given by eq . ( [ eq6 ] ) . here @xmath220 is a stochastic noise generated in the markov chain of monte carlo simulations . as it is expected that the correlation time of the slow mode under consideration is much longer than that of the noise , and since the update steps are local and independent , it is proper to model @xmath220 as gaussian noise , uncorrelated in space or time : @xmath221 in which the choice of @xmath222 ensures that the steady - state of the interface under the langevin equation ( [ eq17 ] ) agrees with its equilibrium state under the free energy ( [ eq6 ] ) . this linear stochastic differential equation can be solved easily by performing fourier transform . ( [ eq17 ] ) thus reduces to @xmath223 which implies an exponentially decaying correlation function of @xmath224 with the relaxation time @xmath215 given by @xmath225 therefore , the dynamical scaling exponent for the monte carlo dynamics , defined by @xmath226 , is always @xmath227 in the rough phase . to check this prediction on the dynamical scaling exponent in practice where the above continuum theory is regularized on a lattice , we compute the following auto - correlation function @xmath228 of the _ microscopic _ height variable @xmath229 : @xmath230 here @xmath231 stands for the dynamical average , and the time @xmath232 is measured in unit of mcs . for each interger - valued @xmath233 , we perform @xmath234 mcs s with a flat initial configuration and compute the auto - correlation functions upto @xmath235 for modes that correspond to the five smallest @xmath236 values . in fig . [ fig2 ] , we display the results so obtained for @xmath192 . other cases of @xmath42 are found to have very similar features . it is clear from fig . [ fig2 ] that @xmath237 can be fitted very well by @xmath238 where @xmath95 and the relaxation time @xmath215 are the fitting parameters . in other words , the relaxation is strictly exponential in all cases . note that we used a cutoff @xmath239 in our fitting . the same fitting procedure is carried out for other cases of @xmath42 . the final results of the relaxation time @xmath215 as a function of @xmath236 for @xmath240 are shown in fig . [ fig3 ] ; and for @xmath241 as an insert . the fact that @xmath215 scales as @xmath236 for @xmath242 as indicated by the fitting in fig . [ fig3 ] thus shows that the ground - state ensembles for @xmath242 are in the rough phase . on the other hand , it is indeed clear from the insert that for @xmath243 and @xmath190 , @xmath215 curves downward as @xmath244 which is in sharp constrast to those of @xmath242 . from this , we conclude that ground - state ensembles for @xmath243 and @xmath190 are in the flat phase . as for @xmath245 , it is not conclusive from the data available whether @xmath215 scales as @xmath236 or curves downward as @xmath246 . nonetheless , the fact that the relaxation time of the slowest mode for @xmath245 is longer than for any smaller _ or larger _ value of @xmath0 , suggests that @xmath245 is very close to the locking transition . further support for this phase diagram is also obtained by explicit calculations of stiffness constants and critical exponents which is discussed in the next section . as implied by sec . [ fluctuations ] , the stiffness constant of the fluctuating interface can be directly measured by studying the long - wavelength fluctuations of the height variables , i.e. , their structure factor as given by eq . ( [ eq8 ] ) . it should be noted that concerning the task of calculating the fourier components @xmath247 in eq . ( [ eq8 ] ) , it can be replaced by the approximation in terms of the _ microscopic _ height variables @xmath229 given by @xmath248 where @xmath99 labels a lattice site of the finite triangular lattice of total @xmath213 lattice sites used in the simulation . here @xmath249 is the _ weight _ of a lattice site , i.e. , the area of its voronoi region , which is introduced so that the _ microscopic _ height variable @xmath229 coincides with the _ coarse - grained _ height variable @xmath247 in the long - wavelength limit ( @xmath250 ) . but unlike @xmath247 , @xmath229 still contains features such as zone - corner singularities discussed in sec . [ zone - corner ] that are only manifested in miscroscopic height variables . starting with a flat state , we perform @xmath251 mcs s as the equilibrium time ; subsequent measurements of physical quantities are carried out at intervals of @xmath252 mcs s . this separation is a compromise between the correlation times of small @xmath138 modes and of larger @xmath138 modes , which are respectively much longer and somewhat shorter than 20 mcs see fig . each run consisted of @xmath253 mcs , i.e. @xmath254 measurements were taken ; these were subdivided into @xmath252 independent groups for the purpose of estimating statistical errors . the same procedure is used for all @xmath233 reported in this paper . in fig . [ fig4 ] , we plot @xmath255 vs. @xmath256 for @xmath192 , including all @xmath257 in the first brillouin zone . from the plot , we observe that @xmath255 is remarkably isotropic up to about @xmath258 . this comes about because of the 6-fold rotational symmetry of the triangular lattice which ensures that _ anisotropy _ occurs only in @xmath259 and higher order terms , assuming that the function is analytic . this is in constrast to other models defined on the square lattice where anisotropy already sets in at the order of @xmath260@xcite . the lower envelope of the data points in fig . [ fig4 ] corresponds to the line of @xmath261 in the @xmath262-vector space . other cases of @xmath42 are found to have very similar features as illustated in the insert of fig . [ fig4 ] where we plot the lower envelope for all @xmath233 . the structure factor of the height variables appears to diverge in the long - wavelength limit @xmath263 for all @xmath0 values , even for the largest @xmath0 values . ( in the latter case , however , we believe one would see the plot asymptote to a constant value , in a sufficiently large system ; see below . ) two other interesting features of the structure factor are also revealed in the insert in fig . [ fig4 ] : ( 1 ) for @xmath211 , it appears to indicate yet another singularity at the zone corner @xmath264 in the thermodynamic limit @xmath265 ; and ( 2 ) for @xmath192 , it approaches a constant instead . as already discussed in sec . [ zone - corner ] , the appearance of zone - corner singularities is expected , the precise nature of such singularities , however , is discussed in the next section . in the remaining of this section , we analyze the zone - center singularity to check if height variables behave as required by eq . ( [ eq8 ] ) for the rough phase and consequently extract the stiffness constants . to further study the nature of zone - center singularity in terms of how @xmath123 scales as a function of @xmath256 in the long - wavelength limit , we show the log - log plot of @xmath266 vs. @xmath256 for @xmath267 in fig . [ fig5 ] . comparing the simulation results for different systems sizes of @xmath268 , @xmath269 and and @xmath270 , we notice that data are well converged down to accessible small @xmath257 vectors except for the case of @xmath245 and @xmath271 , where the finite size effect is still discernible . this is , of course , consistent with the fact that @xmath245 and @xmath271 are close to the locking transition where the correlation length diverges ; it is interesting , however , to notice that their finite - size trends are different . in the case @xmath245 , the data plot for @xmath272 curves upwards less than that for @xmath273 , while in the case @xmath243 , the @xmath272 data show _ more _ upwards curvature than the @xmath273 data . by fitting @xmath266 to a function @xmath274 with @xmath275 being the fitting parameter , we obtain , using the data of system size @xmath272 and a cutoff @xmath276 , the exponent @xmath277 and @xmath278 respectively for @xmath279 , and @xmath109 . apart from the case of @xmath245 , these values agree with @xmath280 as in the predicted @xmath281 power - law singularity of the structure factor in the rough phase , eq . ( [ eq8 ] ) . as for @xmath243 and @xmath190 , @xmath266 clearly deviates from a power - law scaling and instead curves upwards to level off , which indicates that models with @xmath243 and @xmath190 are in the smooth phases where @xmath123 remains finite as @xmath282 , as discussed in sec . [ smooth ] . this conclusion is in excellent agreement with that inferred from dynamical scaling analysis presented in sec . [ dynamical ] . it should be noted that in fig . [ fig5 ] , as a general procedure adopted throughout this paper in extracting numerical values of some physical quantities , we have averaged the data corresponding to the same magnitude of @xmath236 to further reduce the effect due to statistical errors . the relative statistical error on each individual data point @xmath123 of small @xmath257 , which is measured directly from the variance among the 20 groups , is found to range from @xmath283 to @xmath284 . this is indeed consistent with the estimates of such relative errors from the relaxation times of the slowest modes of models with different values of @xmath42 already given in sec . [ dynamical ] . it is perhaps also worth noting that another good check on the statistical errors on each data point is to compare the values of @xmath123 for three @xmath257 vectors which are related by @xmath285 rotations in reciprocal space , which ought to be equal by symmetry . for example , in the case of @xmath192 , the values of @xmath123 for the three @xmath257 vectors of the same smallest magnitude @xmath286 of system size @xmath272 are , respectively , @xmath287 , @xmath288 , and @xmath289 , from which one thus also obtain the relative error of about @xmath283 . this observation therefore motivates the averaging procedure used in this paper . the stiffness constants can be subsequently determined by fitting @xmath290 to the function @xmath291 for the isotropic part of the data in which the stiffness constant @xmath86 and @xmath292 are the fitting parameters . the final fitting on the averaged data is shown in fig . [ fig6 ] where we used a cutoff @xmath276 in the fitting . we also tried other different cutoffs of @xmath293 and @xmath294 , and found as expected that the stiffness is not sensitive to the value of cutoff as long as it falls into the isotropic part of the data . for example , we obtain , in the case of @xmath192 , @xmath295 , and @xmath296 for cutoff @xmath297 , and @xmath298 respectively . therefore , taking into account of the uncertainty introduced due to the cutoff , our final estimate for the stiffness constant is then @xmath299 which is in excellent agreement with the exact value @xmath300 . similar procedure is carried out for other cases of @xmath42 and the results are tabulated in table i. in the same table , we also give the value for the critical exponents of spin , bond - energy and uniform magnetization operators which are obtained straightforwardly according to eqs . ( [ eq13 ] ) , ( [ eq15 ] ) and ( [ eq - gm ] ) . the agreement of our @xmath301 values with the `` @xmath302 '' values from transfer - matrix eigenvalues ( see table i of ref . , is quite close and becomes better as @xmath0 grows ( until @xmath245 . ) as already discussed in sec . [ finitet ] , a kosterlitz - thouless ( kt ) transition occurs at a critical value @xmath156 where @xmath153 , such that for @xmath303 algebraic correlations persist even at small finite temperatures . it is clear from our data that @xmath304 . as for @xmath245 , the value of @xmath305 at the smallest nonzero @xmath306 is already larger than @xmath307 . that is , even if the system may have a `` rough '' behavior at the length scales probed in the simulation , the stiffness constant is such that the locking potential is relevant and must dominates at sufficiently large length scales , as discussed in sec . [ locking ] . a similar observation has already been used to argue that the constrained potts antiferromagnet is in a smooth phase@xcite . this fact together with the poor fitting using the formula suitable for the rough phase ( see the top curve of fig . [ fig6 ] ) leave us little choice but to conclude that the ground - state ensemble for @xmath245 also falls into the smooth phase , or possibly is exactly at the locking transition . just as the finite - size effect for @xmath245 was severe both for the spin - spin correlations ( measured via monte carlo@xcite ) and also in spin - operator eigenvalues ( measured via tranfer - matrix,@xcite ) we similarly find it is severe for height fluctuations . however , in view of the exponential relationship between the exponents and the stiffness constant , the latter measurements are much more decisive as to the true phase of the system . to sum up , based on the analysis on the nature of the singularity in the height structure factor at the long - wavelength limit and the numerical results on the stiffness constants , we thus conclude that the model exhibits three phases with a kt phase transition at @xmath1 and a locking phase transition at @xmath308 . another more traditional approach@xcite in calculating the critical exponents of various operators is to compute the corresponding structure factors and analyze the power - law singularities at the appropriate ordering wave vectors . namely , if the correlation function of an operator @xmath309 decays with distance as power - law ( thus critical ) @xmath310 then its structure factor near the ordering vector @xmath119 shows a power - law singularity @xmath311 from which the critical exponent @xmath312 can be numerically extracted . here in this section , we adopt this approach to calculate the critical exponents of spin , bond - energy , and uniform - magnetization operators so as to compare with those obtained from the stiffness constant . as given by eq . ( [ eq - se ] ) , @xmath313 . here @xmath118 is the ordering vector of the bond - energy operator . therefore the interesting feature of structure factor of height variables , namely , the appearance of zone - corner singularity as shown in fig . [ fig4 ] , is not only expected but also very useful in extracting the critical exponent @xmath314 . of course , such a zone - corner singularity can also be understood within the framework of interfacial representation , as in sec . [ height - rep ] , particularly subsec . [ zone - corner ] . ( similar zone - corner singularities have been studied in refs . and . ) finally , according to the exact result @xmath127 ( @xmath315 ) in the case of @xmath192 , i.e. , @xmath316 , the puzzling absence of the zone - corner singularity for @xmath192 as shown in fig . [ fig4 ] is also resolved . in fig . [ fig7 ] , we plot @xmath317 vs. @xmath318 where we have averaged data points with the same magnitude of @xmath319 . fitting @xmath320 to the function @xmath321 where @xmath322 and @xmath323 are the fitting parameters , we obtain the critical exponents @xmath324 which are tabulated in table i. in practice , we used two different cutoffs in the fitting : @xmath325 and @xmath326 . the fitting for the latter is shown in fig . [ fig7 ] , and the final quoted errors take into account the uncertainty due to the cutoffs . similarly , we also computed the structure factor for the spin operator @xmath327 using fast fourier transform while computing the height - height correlation function within the same monte carlo simulations . results are shown in fig . [ fig8 ] and the extracted exponents are also tabulated in table i. fitting precedure used is exactly the same as that for bond - energy except that we fit @xmath328 to the function @xmath329 with @xmath292 and @xmath330 being the fitting parameters . from table i , we note that the critical exponents extracted in this way are in good agreement with those obtained from stiffness constant utilizing the interfacial representation , however , the latter yields much better statistical errors by an order of magnitude using the same monte carlo simulation data . this clearly demonstates the superiority of the interfacial representation in extracting critical exponents from numerical data . similar points were made regarding other models , but based on much less extensive simulation data , in refs . and . similar fits were attempted for @xmath245 , yielding @xmath331 and @xmath332 . while the statistical error on @xmath333 is too large to render the fitting meaningful , the increase in the value of @xmath334 when compared with @xmath335 is added evidence that @xmath245 is _ not _ in the rough phase ; if it were still rough at that value of @xmath0 , we would have expected a continuation of the decreasing trend of @xmath336 with @xmath0 . as for the cases of @xmath243 and @xmath190 , the structure factors of both the spin and bond - energy operators show _ weaker _ than power - law behavior as @xmath337 , as in figs . [ fig7 ] and [ fig8 ] , but they increase to a larger value ( not seen in these logarithmic plots ) right _ at _ @xmath338 . this is indeed consistent with the @xmath339-function singularity . expected if these cases fall into the smooth phase with long - ranged order of the spin and bond - energy operators . finally , we consider the uniform magnetization correlation exponent @xmath129 . when @xmath340 , it can be predicted ( see @xmath341 in table i ) that @xmath342 , implying a divergent ( ferromagnetic ) susceptibility and a divergent structure factor @xmath343 as @xmath344 now , due to the linear relation ( [ eq - m ] ) between @xmath345 and @xmath143 , we immediately obtain @xmath346 near @xmath179 , just as @xmath347 near @xmath348 ( see sec . [ zone - corner ] and eq . ( [ eq - se ] ) ) thus , a singularity at @xmath179 is expected in the structure factor of spin operator which is plotted in fig . [ fig9 ] . from this figure , it appears that only for @xmath349 , @xmath350 , and @xmath109 does @xmath343 show a power - law singularity indicated by a straight line in this log - log plot . this confirms the prediction based on the stiffness constant ; however , the numerical values of @xmath129 extracted this way ( see table i ) differ considerably from those calculated from the stiffness constant in the case of @xmath351 and @xmath109 . it is also apparent from table i that @xmath352 is systematically overestimated as compared with the more accurate value derived from height fluctuations . we suspect that a similar overestimation affected the values of @xmath302 that were deduced from the finite - size scaling of the susceptibility of the staggered magnetization@xcite ( this obviously measures the same fluctuations seen in @xmath353 near @xmath338 . ) those data ( also quoted in ref . ) have quoted errors about four times as large as ours for @xmath301 . their exponent values are all noticeably larger than the accurate value ( @xmath301 or @xmath354 from ref . ) becoming _ worse _ as @xmath0 grows ( for @xmath355 the difference is twice their their quoted error . ) clearly the systematic contribution to their errors was underestimated . the transfer - matrix method@xcite ought to provide the effective exponent @xmath302 for spin correlations on length scales comparable to the strip width , and hence is likewise expected to overestimate @xmath302 ; indeed , every @xmath302 value found in ref . is slightly larger than our corresponding @xmath301 value . which type of flat state is actually selected in the smooth phase ? [ fig10 ] shows the measured expectation of @xmath40 , the number of zero spins in the spin-1 representation , for @xmath356 . as @xmath0 grows , it is found that @xmath357 approaches its maximum allowed value @xmath358 as in the @xmath82 state , rather than zero , as in the @xmath169 state . thus , the flat states with half - integer valued @xmath59 in fig . [ fig1 ] are being selected in the smooth phase . translating back to the spin-@xmath0 model , this means that spins on two sublattices of the triangular lattice take the extremal values , @xmath15 and @xmath16 respectively , while spins on the third sublattice remain disordered . it is perhaps more illuminating to study the distribution of height variables to probe the height fluctuations in the smooth phase . to this end , we also show , in fig . [ fig10 ] , the histogram of height variable @xmath59 in the cases of @xmath359 and @xmath360 , which is measured for a _ typical _ configuration generated in the monte carlo simulations.@xcite . the broad distribution observed in the case of @xmath359 ( @xmath361 ) evolves to a narrowly peaked distribution in the case of @xmath360 ( @xmath362 ) . ( it decays as @xmath363 . ) this supports the intuitive picture presented in sec . [ smooth ] . furthermore , the center of this peaked distribution is half - integer valued . ( numerically , the mean is @xmath364 for the distribution plotted in fig . [ fig10 ] . ) in other words , the locking potential @xmath365 favors the @xmath366 type of flat state , in which one sublattice is flippable , rather than the @xmath169 type of flat state . ( see fig . [ fig1 ] ) . this kind of flat state was also expected analytically in the limit of large @xmath0 @xcite . we have also computed @xmath367 for each value of @xmath0 , in two ways . first , @xmath368 is just normalization factors times @xmath369 , which we accumulated throughout the monte carlo run , as described earlier in this section ; then it can be shown that @xmath370 exactly . for @xmath371 this gives @xmath372 and @xmath373 for @xmath359 and @xmath360 , respectively , showing the contrast of the rough and smooth behavior . secondly , we can compute @xmath374 directly from the histogram ( from one snapshot ) seen in fig . [ fig10 ] ; this gives respective values @xmath375 and @xmath376 , in satisfactory agreement with the first method . the exotic `` hidden order '' phase@xcite ( see sec . [ smooth ] ) can be ruled out on the basis of these data : according to eq . ( [ eq - dhbound ] ) the variance of @xmath59 should be at least @xmath377 in the hidden - order phase , while our measurements indicate it is at most only @xmath373 . furthermore , for @xmath243 and @xmath190 , the structure factor @xmath378 at the zone - corner wave vector @xmath338 ( not plotted ) was much larger than at nearby @xmath138 ; that direct suggests a @xmath339-function singularity in the thermodynamic limit , i.e. , existence of long - ranged spin order in which @xmath379 on at least two of the sublattices . additionally , the spin structure factor @xmath353 near the zone - corner wave vector @xmath338 ( fig . [ fig8 ] ) showed a striking curvature in the `` smooth '' cases @xmath243 and @xmath190 , quite different from the behavior at smaller @xmath0 . this makes it plausible that @xmath380 , so that spin fluctuations have short - range rather than power - law correlations for @xmath362 . ( it was not emphasized in ref . , but power - law correlations are implied if one takes seriously their measured values @xmath381 for @xmath382 . ) we propose , then , that actually @xmath383 for @xmath384 , as in the simplest picture of the smooth phase , and that the observed nonzero values are simply finite - size effects due to the very slow crossover from rough to smooth behavior near a roughening transition ( see sec . [ disc - crossover ] , below , for further discussion . ) to conclude , in this article , we have investigated the ground - state properties of the antiferromagnetic ising model of general spin on the triangular lattice by performing monte carlo simulations . utilizing the interfacial representation , we extrapolated the stiffness constants by studying the long - wavelength singularity in the height variables , which in turn lead to straightforward calculation of critical exponents of various operators within the framework of height representation . the results so obtained are further compared with those extracted from a more tranditional method , and demonstrate that the method in terms of height representation method is by far the preferable one for extracting the critical exponents . we also analyzed both the dynamical and static properties of the model in order to map out the phase diagram which consists of three phases with a kosterlitz - thouless phase transition at @xmath1 and a locking phase transition at @xmath308 . even in the smooth state , analysis of the height fluctuations ( as in @xmath374 was helpful in resolving questions which are made difficult by the strong finite - size effects near the locking transition . one of our initial motivations for this study was the possibility of finding rational exponents even for @xmath167 . we believe the results in table i are the first which are accurate enough to rule out this possibility . indeed , @xmath385 and @xmath386 , with differences similar to the error ( 0.001 ) . but _ any _ random number differs from a rational number with denominator @xmath387 by the same typical error . the exception is that @xmath388 is quite close to @xmath389 , but we have given other reasons to be suspicious of this value . another intriguing question was whether the critical values @xmath391 and @xmath392 are exactly integers . previous data@xcite suggested that @xmath393 exactly , and had large enough errors that @xmath394 could not be excluded . since @xmath395 and @xmath396 , this question was answered by the preceding subsection : we find that definitely @xmath397 . furthermore , we suspect @xmath398 as concluded in sec . [ stiffness ] since the effective stiffness at the length scale we access is more than enough to drive the system to the locked phase . the question of the value of @xmath133 suggests paying closer attention to the behavior of systems near the locking transition . it has been noted previously how the locked phase tends to behave qualitatively like the rough phase in a finite - size system , since the crossover is a very slow function of size.@xcite this is consistent with the apparent power - law behaviors observed at @xmath399 in previous studies@xcite and with the tendency of those studies to overestimate the exponents @xmath400 and @xmath401 ( as compared with our more accurate estimates . ) this would suggest that , if extensive finite - size corrections were included in our analysis , they would reduce our estimate of @xmath390 a bit further , i.e. we would more definitely conclude that @xmath245 is in the locked phase . our analysis near the locking transition at @xmath390 suffers from our ignorance of the expected functional form of the critical behavior as a function of @xmath402 . a study of the roughening transition@xcite used the kosterlitz - thouless ( kt ) renormalization group to derive analytic approximations for the total height fluctuation ( closely analogous to @xmath374 in our problem ) , which made it possible to overcome very strong finite - size effects and fit the roughening temperature precisely . use of kt finite - size corrections was also essential in extracting meaningful numbers from transfer - matrix calculations near the locking transition induced by a magnetic field in ref . . thus , a similar adaptation of the kt renormalization group to give expressions for the behavior of @xmath403 , as a function of ( small ) @xmath404 and @xmath402 , or the functional form of @xmath405 near @xmath390 , could make possible a more conclusive answer as to whether @xmath406 exactly . since the long - wavelength behavior in this model ( in its rough phase ) is purely gaussian with @xmath227 ( see sec . [ dynamical ] ) , the critical slowing down is particularly transparent . it seems feasible to take advantage of the existence of a height representation to develop an acceleration algorithm . for example , it might be possible to extend the cluster algorithms which are known for the @xmath30 triangular ising antiferromagnet.@xcite these are well - defined at @xmath151 , but their effectiveness seems to depend in a hidden fashion on the existence of the height representation when @xmath125 . an intriguing alternative approach starts from the observation that at long wavelengths the system obeys langevin dynamics ( see sec . [ dynamical ] and ref . ) . fourier acceleration@xcite , a nonlocal algorithm ( efficient owing to use of the fast fourier transform algorithm ) , is known to be effective in such cases . the key is to replace the uncorrelated noise function @xmath407 of eq . ( [ eq18 ] ) with a new correlated noise function having @xmath408 . this might be implemented by first constructing a random function with such correlations , and then updating flippable spins with probabilities determined by that function , in such a fashion as to satisfy detailed balance . additionally , it may be possible to analyze transfer - matrices using the height representation . quite possibly this would yield an order - of - magnitude improvement in the accuracy of the numerical results , for the same size system , similar to the improvement in analysis of monte carlo data . the transfer matrix breaks up into sectors corresponding to the step made by @xmath50 upon following a loop transverse to the strip ( across the periodic boundary conditions . then the stiffness can be extracted directly from the ratio of the dominant eigenvalues of two such sectors ; such an analysis is already standard for quasicrystal random tilings , for which the long - wavelength degree of freedom is also an effective interface @xcite . .stiffness constant and critical exponents . here @xmath409 , @xmath410 and @xmath411 are the estimates for the critical exponents of spin and bond - energy operators calculated from the stiffness constant @xmath86 as done in sec . v(c ) , while @xmath412 , @xmath413 , and @xmath414 stand for the same critical exponents , but extracted from the singularities of their respective structure factors in sec . estimated errors are also given in the parenthesis . [ cols="^,^,^,^,^,^,^,^",options="header " , ] j. stephenson , j. math . * 5 * , 1009 ( 1964 ) ; j. math . * 11 * , 413 ( 1970 ) . h. w. j. blte and h. j. hilhorst , j. phys . a * 15 * , l631 ( 1982 ) ; ( b ) . b. nienhuis , h.j . hilhorst , and h.w . blte , j. phys . a * 17 * , 3559 ( 1984 ) . the limit of @xmath416 is well - defined and has been simulated by ref . and by y. yamada , s. miyashita , t. horiguchi , m. kang , and o. nagai , j. mag . * 140 - 144 * , 1749 ( 1995 ) . it is interesting as a model which has an infinite number of continuous degrees of freedom ( changes of @xmath417 for loose spins ) , under which the energy is degenerate ; in such systems it is expected that a particular average pattern of order is selected when fluctuations are taken into account , but it is nontrivial to compute the selecting terms of free energy , since the potential minimum is flat rather than quadratic as a function of the spins . s. miyashita and h. kawamura , j. phys . 54 * , 3385 ( 1985 ) ; b. kleine , e. mller - hartmann , k. frahm , and p. fazekas , z. phys . b * 87 * , 103 ( 1992 ) ; q. sheng and c. l. henley , j. phys . cond . mat . * 4 * , 2937 ( 1992 ) . g. murthy , d. arovas , and a. auerbach , preprint ( cond - mat/9607217 ) . t. ogawa and y. nakajima , prog . ( suppl . ) * 87 * , 90 ( 1986 ) , independently identified the height representation , but considered the height as defined modulo 6 ( so roughness could not be clearly defined , and the connection to power - law correlations was not evident . ) although we only describe the interfacial representation for the spin-1 model , a very similar procedure can be carried out for the original spin-@xmath0 model straightforwardly . contrary to what is suggested in ref . , p. 3890 , it is also possible to construct @xmath50 directly from @xmath33 for general @xmath0 , _ without _ first mapping it to a spin-1/2 or spin-1 model . the prescription is simply to replace @xmath418 on the right - hand side of ( [ eq3 ] ) . the long wavelength height fluctuations would be the same under the new definition ; it would be useful only at nonzero temperature , to confirm that excitations have burgers vector the it had long been noted that the triangular ising model could be mapped apporoximately to a 6-state clock model model ( see e. domany , m. schick , j. s. walker , and r. b. griffiths , phys . rev . b * 18 * , 2209 ( 1978 ) , particularly in connection with the effect of adding a second - neighbor ferromagnetic interactionn the height mapping of the _ microstates _ , is more powerful , in that it shows _ every _ excitation must correspond to vortex / antivortex pairs ; this is important in predicting the @xmath151 behavior . j. kondev , j. de gier , and b. nienhuis , preprint cond - mat/9603170 ; j. kondev , to appear _ proc . exactly soluble models in statistical mechanics _ ( march 1996 , northeastern university ) , cond - mat/9607181 . p. chandra , p. coleman , and l. b. ioffe , phys . b * 49 * , 12897 ( 1994 ) used a height representation to understand the phase diagram of another triangular antiferromagnet ; the behavior is reminiscent of the @xmath419 limit in our system ( discussed in footnote ) . a plaquette has half - integer - valued @xmath59 if and only if it has a zero spin at exactly one corner ( fig . [ fig1 ] ) . from this it follows that the total weight of the half - integer bars in the in the histogram is @xmath420 ; this clarifies the relation between the upper and lower panels of fig . [ fig10 ] . h. g. evertz , m. hasenbusch , m. marcu , and k. pinn , physica a * 199 * , 31 ( 1993 ) .

we present an analysis of the data produced by the manga prototype run ( p - manga ) , aiming to test how the radial gradients in recent star formation histories , as indicated by the 4000 - break ( d(4000 ) ) , h absorption ( ew(h ) ) and h emission ( ew(h ) ) indices , can be useful for understanding disk growth and star formation cessation in local galaxies . we classify 12 galaxies observed on two p - manga plates as either centrally quiescent ( cq ) or centrally star - forming ( csf ) , according to whether d(4000 ) measured in the central spaxel of each datacube exceeds 1.6 . for each spaxel we generate both 2d maps and radial profiles of d(4000 ) , ew(h ) and ew(h ) . we find that csf galaxies generally show very weak or no radial variation in these diagnostics . in contrast , cq galaxies present significant radial gradients , in the sense that d(4000 ) decreases , while both ew(h ) and ew(h ) increase from the galactic center outward . the outer regions of the galaxies show greater scatter on diagrams relating the three parameters than their central parts . in particular , the clear separation between centrally - measured quiescent and star - forming galaxies in these diagnostic planes is largely filled in by the outer parts of galaxies whose global colors place them in the green valley , supporting the idea that the green valley represents a transition between blue - cloud and red - sequence phases , at least in our small sample . these results are consistent with a picture in which the cessation of star formation propagates from the center of a galaxy outwards as it moves to the red sequence .

1502.07040

our understanding of galaxies in the local universe has improved dramatically over the past decade , in significant part due to large optical spectroscopic surveys such as the two - degree field galaxy redshift survey ( 2dfgrs ; * ? ? ? * ) and the sloan digital sky survey ( sdss ; * ? ? ? * ) . one of the major findings of the sdss is the discovery of ` galaxy bimodality ' , through which local galaxies are divided into two distinct populations , termed ` red sequence ' and ` blue cloud ' , according to their rest - frame colors or specific star formation rate ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? red - sequence galaxies typically present bulge - dominated morphology , red optical / uv color , and little gas and star formation , while blue - cloud galaxies are usually gas - rich with blue colors , ongoing star formation and disky morphology . the smaller number of galaxies found lying in the ` green valley ' between these two main populations in color space are widely believed to be caught in a transition phase from the blue cloud to the red sequence ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , although this picture has recently been disputed by some authors ( e.g. * ? ? ? narrow - field deep surveys reveal that this color bimodality persists out to redshift of at least @xmath4 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , though the prevalence of the red population has significantly increased since a redshift of unity @xcite . the cessation of star formation is thus an important process which has been contributing to galaxy evolution over the past @xmath5 . a complete picture of the way in which the star formation in galaxies gets shut down remains elusive . however , recent studies of the scaling relations of galaxy properties and their dependence on local environment have clearly established that , in addition to stellar mass , both internal structural properties and external environment are key indicators , or may even be drivers , of the star formation cessation processes in galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? for instance , when studying the central galaxies in groups or clusters , the presence of a prominent bulge - like structure is found to be a necessary ( but not sufficient ) condition for stopping star formation @xcite . in addition , studies of the relationship between galaxy morphology and color have revealed a significant population of red - sequence galaxies with disc - dominated spirals at both low - z ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and @xmath6 ( e.g. * ? ? ? * ) , which are preferentially found in galaxies with large bulges @xcite . these findings support the ` morphological quenching ' mechanism proposed by @xcite , although studies of cold gas in massive galaxies indicate that a reduction in gas content is also required ( e.g. * ? ? ? * ) . for central galaxies in massive dark matter halos above a critical mass of @xmath7 , `` radio - mode '' agn feedback and shock - heating may effectively reduce the gas cooling efficiency , thus also playing an important role in preventing further star formation @xcite . powerful agn feedback can also happen in the so - called `` quasar mode '' , which is predicted to be triggered by major mergers of gas - rich galaxies with comparable mass ( e.g. * ? ? ? * ; * ? ? ? * ) , and is commonly adopted as one of the quenching processes in semi - analytic models . for satellite galaxies , star formation quenching seems to be driven primarily by external processes occurring within their host group / cluster , such as gas stripping by ram - pressure @xcite and tidal interactions @xcite . the ` smoking gun ' of observations showing hi gas being removed by ram - pressure stripping have been obtained for spirals in nearby clusters of galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? studies of color profiles and surface brightness profiles for satellite galaxies in sdss group systems prefer more gentle processes such as `` starvation '' or `` strangulation '' ( e.g. * ? ? ? * ) . on the other hand , a recent study of the specific star formation rate stellar mass relation for galaxy groups / clusters at higher redshift ( @xmath8 , @xcite ) suggests that galaxy mergers play a primary role in quenching satellites in galaxy groups , while strangulation is a process more important in cluster - scale environment . there have also been statistical studies of the cold gas content and star formation in galaxies residing in different environments , using either direct atomic gas mass measurements from large blind surveys of hi emission such as the hi parkes all sky survey ( hipass ; * ? ? ? * ) and the arecibo legacy fast alfa ( alfalfa ; * ? ? ? * ) survey , or indirect estimates of gas content from the optical photometry in surveys like sdss ( e.g. * ? ? ? * ; * ? ? ? these studies have revealed that gas - related quenching depends not only on the stellar mass of the satellite galaxies , but also on their surface mass density . a full understanding of star formation cessation in galaxies thus requires spatially - resolved measurements of stellar and gaseous components to be obtained for a large sample of galaxies covering wide ranges in mass and color , probing a range of environmental conditions . previous studies for this purpose have been mainly based on multi - wavelength broadband photometry , for both nearby galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and those at high redshifts ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , while some authors also made use of long - slit spectroscopy ( e.g. * ? ? ? * ; * ? ? ? significant improvements have been made in the past decade , thanks to many integral field unit ( ifu ) surveys which have obtained spatially resolved spectroscopy for samples of galaxies . however , to - date , the samples have been relatively small ( @xmath9 galaxies ) : sauron @xcite , diskmass @xcite , atlas@xmath10 @xcite . the ongoing califa survey ( calar alto large integral field area * ? ? ? * ) is observing a sample of 600 galaxies , making a big step forward in 2d spectroscopic studies of nearby galaxies . for example , this survey has , for the first time , quantified the spatially - resolved history of stellar mass assembly for galaxies beyond the local group , demonstrating how massive galaxies grow their stellar mass inside - out @xcite . such an inside - out picture of galaxy formation is supported by further measurements of gradients in oxygen abundance of hii regions @xcite , in stellar metallicity of face - on spirals @xcite , and in stellar age and local surface mass density of galaxies of different morphologies and masses @xcite , all from the califa survey . however , the sample is still relatively modest , limiting the range of environments and galaxy properties that can be explored . the more recently launched sami survey @xcite takes a further step with a goal of obtaining ifu observations for 3400 galaxies by 2016 . in the meantime , the kmos@xmath11 survey is pushing forward the ifu observations for high redshift galaxies , observing a sample of 600 galaxies at @xmath12 using kmos at the very large telescope @xcite . as a next step at low - z , the study of these phenomena is a key goal of the upcoming manga ( mapping nearby galaxies at apache point observatory , * ? ? ? * ) survey . as one of the major programs of the fourth - generation of sdss , manga will obtain two - dimensional , integral - field spectroscopy ( ifs ) for 10,000 galaxies in the nearby redshift range @xmath13 , optimally selected for uniform coverage and resolution using the single ( central ) spectra from sdss . each galaxy will be spectroscopically observed with an ifu to obtain multiple high signal - to - noise ratio ( s / n ) spectra across the full optical wavelength range from 3600 to 10300 . manga will thus provide 2d maps of stellar populations and recent star formation histories for a large number of quiescent and star - forming galaxies , thus enabling extensive exploration of the various processes by which star formation ceases in the local universe . in this paper , to assess the potential use of the manga survey in studying the cessation of star formation , and to derive some initial results in this area , we present an analysis of the manga prototype ( p - manga ) data for a set of 12 galaxies , which were observed in the pilot run in 2012 december and 2013 january . these observations were made through time granted by the sloan digital sky survey - iii ( sdss - iii , * ? ? ? the prototype data are described in more detail in @xcite , which also provides a general overview of the manga project . for each of the p - manga galaxies and for each spectrum in the associated data cube , we have performed full spectral fitting using the methodology developed in @xcite , in order to separate out emission line components from the underlying absorption - line stellar spectrum . from these decomposed spectra , we have obtained full 2d maps of three key parameters d@xmath0(4000 ) ( the depth of the 4000 break ) , ew(h@xmath1@xmath2 ) ( the equivalent width of the h@xmath1 absorption line ) and h@xmath3 ( the equivalent width of the h@xmath3 emission line)which are known to be sensitive indicators of stellar populations of different ages @xcite . we then investigate any radial variations in the recent star formation histories of the galaxies by plotting the values of these parameters for different subregions of each galaxy , to see how they change with position . as we will show , the p - manga sample , though small in size , reveals intriguing systematic trends in the three diagnostic parameters across these galaxies , demonstrating the potential for using manga ( and similar ifu surveys ) to understand disk growth and star formation cessation in local galaxies . throughout this paper we assume a @xmath14 cold dark matter cosmology model with @xmath15 , @xmath16 and @xmath17 . a @xcite stellar initial mass function ( imf ) is adopted . llccccccccc ma003_9 & p9 - 127a & 8 - 188794 & j093457.30 + 214220.9 & 0.013 & 10.7 & -21.3 & 0.70 & 23.7 & @xmath18 & 0.7 + ma008_9 & p9 - 127b & 8 - 131835 & j093506.31 + 213739.5 & 0.013 & 9.1 & -18.5 & 0.51 & 6.8 & 1 & 2.4 + ma002_9 & p9 - 61a & 8 - 188807 & j093712.30 + 214005.0 & 0.019 & 10.1 & -20.5 & 1.30 & 9.3 & @xmath18 & 1.2 + ma005_9 & p9 - 19b & 8 - 131893 & j093109.60 + 224447.4 & 0.051 & 10.6 & -22.2 & 0.86 & 4.0 & 1 & 1.6 + ma007_9 & p9 - 19d & 8 - 131577 & j093109.07 + 205500.5 & 0.034 & 10.3 & -21.2 & 0.79 & 3.2 & 1 & 2.0 + ma001_9 & p9 - 19e & 8 - 131821 & j094030.22 + 211513.7 & 0.024 & 9.7 & -20.0 & 0.73 & 2.9 & 1 & 2.2 + ma003_4 & p4 - 127a & 8 - 109661 & j105555.26 + 365141.4 & 0.022 & 10.7 & -22.1 & 0.84 & 10.3 & @xmath18 & 1.6 + ma008_4 & p4 - 127b & 8 - 109682 & j105259.05 + 373648.2 & 0.042 & 11.0 & -22.7 & 0.77 & 13.9 & @xmath18 & 1.2 + ma002_4 & p4 - 61a & 8 - 113576 & j105746.61 + 361657.8 & 0.030 & 9.7 & -20.0 & 0.71 & 3.4 & @xmath18 & 3.4 + ma004_4 & p4 - 19a & 8 - 113557 & j110012.10 + 362313.8 & 0.027 & 9.4 & -19.5 & 0.49 & 2.5 & 1 & 2.5 + ma005_4 & p4 - 19b & 8 - 113506 & j104958.69 + 362454.0 & 0.023 & 9.5 & -19.5 & 0.53 & 4.6 & 1 & 1.4 + ma006_4 & p4 - 19c & 8 - 109657 & j105605.68 + 365736.1 & 0.022 & 9.5 & -19.4 & 0.72 & 4.8 & 1 & 1.3 [ tab : sample ] in the present work , we use data obtained using the manga engineering prototype instrument ( hereafter p - manga ) in 2013 january . p - manga was designed to explore a variety of instrument design options , observing strategies , and data processing algorithms . these p - manga data offer valuable insights into manga s potential , but differ substantially from the manga survey data in several ways . first , p - manga used just one of the two boss spectrographs , with only 560 total fibers and 8 ifus with three sizes : 19 fibers ( n@xmath19 ) , 61 fibers ( n@xmath20 ) , and 127 fibers ( n@xmath21 ) . the p - manga ifu complement was dramatically different from the manga survey instrument , with @xmath22n@xmath19 ( instead of just two ) , @xmath23n@xmath20 ( instead of 4 ) , @xmath24n@xmath21 ( instead of five ) , and no 37-fiber or 91-fiber ifus ( see * ? ? ? * for detailed description of manga instrumentation ) . finally , we also note that the masses and sizes of the p - manga target galaxies were not selected to be representative of the full manga sample . p - manga observations were obtained of three galaxy fields ( field 9 , 11 , 4 ) using sdss plates 6650 , 6651 , and 6652 , each containing 6 galaxies . some of the p - manga targets were drawn from early versions of the full manga sample ( wake et al . , in prep ) , but in many cases p - manga targets were chosen for specific reasons . in each of the three plates , one n@xmath21 ifu was allocated for comparison purposes to a galaxy observed by the califa survey @xcite , even if the galaxy would not otherwise satisfy the manga selection cuts . additionally , the non - optimal ifu complement of the p - manga instrument required some targets to be selected manually . for each plate , observations were obtained in sets of three 20-minute exposures , which were dithered by approximately a fiber radius along the vertices of an equilateral triangle to provide uniform coverage across each ifu . these three plates were observed to varying depths and in varying conditions , as required by the p - manga engineering tasks . although plate 6650 ( field 9 ) was observed to a depth comparable to what will be regularly achieved during manga operations , plates 6651 ( field 11 ) and 6652 ( field 4 ) are both significantly shallower than manga survey data , and plate 6651 was intentionally observed at high airmass ( @xmath25 ) resulting in particularly poor image quality . in this paper we use data obtained with plates 6650 and 6652 ( see table [ tab : sample ] ) , while excluding plate 6651 . the raw data were reduced using a prototype of the manga data reduction pipeline ( drp ) , which is described in detail by @xcite . in brief , individual fiber flux and inverse variance spectra were extracted using a row - by - row algorithm , wavelength calibration was obtained from a series of neon - mercury - cadmium arc lines , and the spectra were flatfielded using internal quartz calibration lamps . sky - subtraction of the ifu fiber spectra was performed by constructing a cubic basis spline model of the sky background flux as seen by the 41 individual fibers placed on blank regions of sky , and subtracting the resulting composite spectrum shifted to the wavelength solution of each ifu fiber . flux calibration of the p - manga data is performed by fitting kurucz model stellar spectra to the spectra of calibration standard stars covered with single fibers at each of the three dither positions . the flux calibration vectors derived from these single - fiber spectra were found to vary by @xmath26 10% from exposure to exposure , depending on the amount of light lost from the fiber due to atmospheric seeing and astrometric misalignments . while this uncertainty is acceptable for the present science purposes , the flux calibration uncertainty of the single fibers ultimately drove the design decision of the manga survey to instead use 7-fiber ifu `` mini - bundles '' for each calibration standard star , which results in a few percent photometric uncertainty ( see * ? ? ? flux - calibrated spectra from the blue and red cameras were combined across the dichroic break using an inverse - variance weighted basis spline function . astrometric solutions were derived for each individual fiber spectrum that incorporate information about the ifu bundle metrology ( i.e. , fiber location within an ifu ) , dithering , and atmospheric chromatic differential refraction , among other effects . fiber spectra from all exposures for a given galaxy were then combined into a single datacube ( and corresponding inverse variance array ) using these astrometric solutions and a nearest - neighbor sampling algorithm similar to that used by the califa survey . for the p - manga datacubes , a spaxel size of 05 was chosen . the typical effective spatial resolution in the reconstructed datacubes can be described by a gaussian with a full width at half maximum @xmath275 . when binning the datacubes , we scale the resulting error vectors to account , at least approximately , for wavelength and spatial covariance in the p - manga error cubes . all our galaxies have photometry in the sdss - i five bands , as well as archival spectroscopy from the sdss - i 30-diameter fiber . figures [ fig : sample_properties ] and [ fig : sdss_d4k_hda_hae ] show the properties of this sample based on the sdss data . with these two figures we aim to show how our galaxies compare to the larger sdss - i sample from which they are drawn . figure [ fig : sample_properties ] presents the galaxies on the plane of stellar mass versus nuv@xmath28 color and the bpt diagram @xcite . we use estimates of stellar mass and nuv@xmath28 from the new york university value - added galaxy catalog ( nyu - vagc ; * ? ? ? * ) and nasa sloan atlas ( nsa ) . nsa is a catalog of images and parameters of local galaxies based on data from sdss @xcite , galex @xcite and 2mass @xcite ; for details see @xcite . the stellar mass for each galaxy is estimated based on its spectroscopically - measured redshift and the five - band petrosian magnitudes from sdss photometric data , as described in detail in @xcite . the nuv@xmath28 color is defined by the integrated light in the nuv - band from galex and the petrosian magnitude in the @xmath29-band from sdss , both corrected for galactic extinction . measurements of the emission line ratios for the bpt diagram are taken from the mpa / jhu sdss database @xcite . figure [ fig : sdss_d4k_hda_hae ] displays the galaxies in the d@xmath0(4000 ) versus ew(h@xmath1@xmath2 ) and ew(h@xmath1@xmath2 ) versus @xmath30ew(h@xmath3 ) planes , where d@xmath0(4000 ) is the 4000 break in the optical spectrum as defined by @xcite , ew(h@xmath1@xmath2 ) the lick / ids index of the h@xmath1 absorption line defined by @xcite , and @xmath30ew(h@xmath3)the logarithm of h@xmath3 emission line equivalent width . we use measurements of these parameters from the mpa / jhu database , which were obtained from the sdss 30-diameter spectra , thus probing the central @xmath31 @xmath32 for the p - manga galaxies . in both these figures , we use squares and circles for galaxies in fields 4 and 9 respectively , and label each galaxy by the bundleid used in those fields . we divide these galaxies into two subsets according to the 4000 break in the central spaxel of their datacube ( see [ sec : fitting ] ) : _ centrally quiescent _ ( cq ) with d@xmath0(4000)@xmath33 and _ centrally star - forming _ ( csf ) with d@xmath0(4000)@xmath34 . the two subsets are highlighted in the figures with red solid and blue open symbols respectively . we will use the same symbols and colors for subsequent figures unless otherwise stated . for comparison , we have selected a volume - limited sample of 21,328 galaxies with stellar mass above @xmath35m@xmath36 and redshift in the range @xmath37 from the nsa . distributions of this sample are plotted in gray - scale maps in figures [ fig : sample_properties ] and [ fig : sdss_d4k_hda_hae ] . as can be seen from figure [ fig : sample_properties ] , the p - manga sample ( though small in size ) spans a wide range in both the color mass and bpt diagrams , similar to that of the general population . in the literature , galaxies are usually classified into three types according to their nuv@xmath28 color : red - sequence ( nuv@xmath28@xmath38 ) , green - valley ( @xmath39 nuv@xmath28 @xmath40 ) and blue - cloud ( nuv@xmath28@xmath41 ) . accordingly , half of the sample galaxies fall in the blue - cloud region , indicative of ongoing star formation and relatively rich cold gas content . for all the galaxies except ma002_9 and ma003_9 , the color - based classification and the d@xmath0(4000)-based classification are consistent with each other , in the sense that the blue - cloud galaxies are centrally star forming , while the objects in the green valley and red sequence combine to form the centrally - quiescent galaxy class . this result indicates that green - valley and red - sequence galaxies share similar properties in their central regions , and that their different global color can be mainly attributed to the different stellar population and gas content in their outer parts . one of the two exceptions , ma003_9 , is a strong seyfert - type agn according to its location in the bpt diagram , and is an outlier in the color mass diagram due to its relatively blue color at its stellar mass . as shown below , this galaxy presents a red core and a blue outer disk in the optical image , such that its classification would be more reasonable if the measurement of nuv@xmath28 color were confined to its central region . the strong contradiction between the central and global classification in this case highlights the importance of spatially - resolved measurements . the other exception , ma002_9 , is a dusty , inclined spiral . it is a centrally star - forming galaxy according to its small value of d@xmath0(4000 ) , but is classified as a green - valley galaxy based on the relatively red nuv@xmath28 color , which in this case is likely caused by dust . the d@xmath0(4000 ) and h@xmath1@xmath2 indices are known to be indicators of the recent star formation history of galaxies , with d@xmath0(4000 ) being a good proxy for the luminosity - weighted stellar age , and h@xmath1@xmath2 a sensitive tracer of the star formation that occurred 0.1 @xmath42 ago ( * ? ? ? * hereafter bc03 ) . as shown in @xcite , the location of a galaxy in the plane of d@xmath0(4000 ) and h@xmath1@xmath2 provides a powerful diagnostic of whether the galaxy has experienced a continuous star formation history or bursts over the past 1 @xmath43 . this property is illustrated in the left - hand panel of figure [ fig : sdss_d4k_hda_hae ] , where we present the d@xmath0(4000 ) and h@xmath1@xmath2 indices for the bc03 models of solar metallicity that follow exponentially declining star formation histories ( @xmath44 ) , either continuous star formation decline with long e - folding times ( @xmath45yr , solid lines ) , or bursts of star formation with fairly short e - folding times ( @xmath46yr , dashed lines ) . different colors indicate the adopted values of the characteristic timescale @xmath47 . the right - hand panel presents the same sets of models on the h@xmath1@xmath2 index versus @xmath30ew(h@xmath3 ) plane . the h@xmath3 luminosity is computed from the output of the bc03 models . we convert lyman continuum photons to h@xmath3 photons following . we take the recombination coefficients and h@xmath3/h@xmath48 ratios from @xcite . for the lmc , solar , and super - solar metallicity models we assume nebular temperatures of @xmath4915,000 k , 10,000 k , 5000 k respectively . the p - manga galaxies occupy roughly the same region as the general population in both panels , with the csf galaxies located in the low - d@xmath0(4000 ) and high - h@xmath1@xmath2 region and the cq galaxies in the opposite corner . it is , however , notable that the p - manga sample lacks galaxies of intermediate values in these indices , and consequently the current sample is somewhat biased to a more bimodal distribution on both planes compared to the full population of sdss galaxies . in the rest of this paper we will use the d@xmath0(4000)ew(h@xmath1@xmath2)and ew(h@xmath1@xmath2)@xmath30ew(h@xmath3 ) planes to investigate the maps and radial profiles of recent star formation histories of the p - manga galaxies . in this case we use relative measurements of the three indices as indicators of recent star formation history , but we do not present quantitative measurements of mean stellar age , which depends on the sensitivity of these indices to stellar metallicity , element abundance and dust attenuation ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in a parallel paper @xcite we attempt to simultaneously obtain the stellar age , metallicity and dust content for the p - manga galaxies , by fitting stellar population models to the full spectrum . in addition , in the present paper we assume that h@xmath3 emission is just a signature of recent star formation . however , ew(h@xmath3 ) lower than 6 was found to be more compatible with light from post - agb stars @xcite . one should keep these caveats in mind when interpreting the observational results presented in this paper . for each spectrum in the datacubes generated as described in [ sec : p - manga ] , we have performed a decomposition of the emission - line component and the continuum plus absorption - line component ( hereafter called the ` stellar component ' ) , using both the public spectral fitting code starlight @xcite and our own code described in @xcite . we have carefully masked out the emission - line regions when doing the spectral fitting . figure [ fig : spectral_fitting ] displays the results of this procedure when applied to two representative spectra from the manga datacubes , one with a high signal - to - noise ratio ( @xmath50 ) at 5500 in the continuum and one with a more modest @xmath51 . we measure the d@xmath0(4000 ) and h@xmath1@xmath2 indices from each best - fit spectrum , defining the indices in the same way as in the previous subsection . we have also measured the flux and equivalent width of the emission lines ( both balmer and forbidden ) by fitting a gaussian profile to these lines in the stellar component - subtracted spectrum . the emission - line parameters from starlight were compared to those obtained from our code . we have also performed the same procedure of spectral fitting and parameter measuring on voronoi - binned versions of the datacubes to assess the effects of spatial averaging before fitting . all these analyses were found to return indistinguishable results , and so in what follows we will only present the values we obtain when applying our code to the unbinned datacubes . we have also applied our code to the archival sdss spectra of the centers of the p - manga galaxies , and in figure [ fig : compare_d4k_hda_hae ] we compare the resulting measurements of d@xmath0(4000 ) , ew(h@xmath1@xmath2 ) and @xmath30ew(h@xmath3 ) to those obtained from the central spaxel of the p - manga datacubes ( plotted as blue squares in the figure ) . in the same figure , we also compare our measurements , from both sdss ( black crosses ) and the central spaxel of p - manga datacubes ( red triangles ) with the measurements taken from the mpa / jhu database . overall , the values from different observations and pipelines agree well with each other . we have visually examined the best - fit spectrum for the few outliers , and find that our code provides a reasonably good fit to both sdss and p - manga spectra even in these cases . the relatively large discrepancy in one of the outliers ( the blue square and black cross located in the left - bottom corner in the right - hand panel ) is caused by the bad data pixels over the h@xmath3 band in the sdss spectrum . we note that , when fitting the spectra and measuring the diagnostic parameters , we do not take into account the covariance between wavelengths , nor the covariance between pixels . methods for tracking such covariances have been studied in depth ( e.g. * ? ? ? * for the sami survey ) . as pointed out in @xcite , when compared to p - manga , manga will provide more accurately estimated spectral errors ( thus accurate error propagation through to the final datacubes ) , allowing construction of covariances between pixels . we will apply a detailed treatment of covariance in manga data in future work . before investigating the spatially - resolved properties of these galaxies star - formation histories , we take a first look at any spatial variations by comparing the central values of the diagnostic parameters to their integrated values across the field of view . to this end , we generated a global spectrum for each galaxy by stacking the entire datacube , with each spaxel at given wavelength being weighted by @xmath52 following @xcite , where @xmath53 is the signal - to - noise ratio at 5500 in the continuum . spaxels with @xmath54 were excluded from the stacking . we then measure the diagnostic parameters d@xmath0(4000 ) , ew(h@xmath1@xmath2 ) and @xmath30ew(h@xmath3 ) in the same manner as above . figure [ fig : cen_vs_glb ] compares the global measurements to the same parameters obtained from the central spaxel of the galaxies . for csf galaxies , the global parameters differ little from the central ones . in contrast , however , most of the cq galaxies show significant difference between central and global parameters , with smaller d@xmath0(4000 ) and larger ew(h@xmath1@xmath2 ) and @xmath30ew(h@xmath3 ) in the latter . this result implies that the outer regions of cq galaxies contain younger stellar populations and have therefore experienced more recent star formation than their inner parts . the difference in ew(h@xmath3 ) is relatively weak compared to the other two parameters , suggesting that the central - to - global variation may be mainly driven by the radial variation in recent star formation history , but not the current star formation rate . it may also be the case that h@xmath3 is not a reliable tracer of star formation in galaxies dominated by older stellar populations ( see [ sec : summary ] ) . a comparison of the global parameters with the central parameters on the diagnostic diagrams of recent star formation history is shown in figure [ fig : d4k_hda_hae ] , where we plot the galaxies on the d@xmath0(4000)-ew(h@xmath1@xmath2 ) and ew(h@xmath1@xmath2)-@xmath30ew(h@xmath3 ) planes , using the central and global parameters for the upper and lower panels , respectively . comparing the upper and lower panels , we identify two noticeable effects . first , the gap between the csf and cq galaxies as seen in the upper panels is largely filled in the lower panels , and this in - fill is made up exclusively of cq galaxies due to the broad scatter in their global values of d@xmath0(4000 ) and ew(h@xmath1@xmath2 ) . second , when globally measured , both the csf galaxies and the cq galaxies appear to deviate from the continuous star - formation models , moving by varying degrees toward the regime of recent bursts . in figures [ fig : maps1 ] and [ fig : maps2 ] we present the 2d maps of the three diagnostic parameters , d@xmath0(4000 ) , ew(h@xmath1@xmath2 ) and @xmath30ew(h@xmath3 ) , for all the 12 galaxies in the sample , with the csf galaxies in one figure and the cq galaxies in the other . in each case , the galaxies have been ordered such that the central values of d@xmath0(4000 ) , as measured from the central spaxel of the datacubes , increases from left to right . the sdss @xmath55 image is also shown for each galaxy . the effective radius , length and direction of major and minor axes are indicated in the images and maps . when generating the maps , we only employ spaxels where the s / n in the continuum at 5500 is greater than three . consistent with what was found from the global measurements in the previous subsection , the overall impression from the 2d maps is that all the galaxies show variation within the probed radius in all three parameters , but to varying degrees and differing radial extents , with the variations more pronounced in the cq galaxies . as noted above in [ sec : sdss_data ] , galaxy ma003_9 shows a blue outer disk in the optical image , which causes it to be classified as a blue - cloud galaxy in the color mass diagram , although its red core suggests that this is intrinsically a cq galaxy . in agreement with this interpretation , the galaxy shows strong contrast between the central and outer parts in the maps of all the three diagnostic parameters . similar behavior is seen in galaxy ma008_4 , the most massive galaxy in the p - manga sample with @xmath56 , which is also a face - on spiral with a relatively blue outer disk ( although less prominent than in ma003_9 ) and strong radial variation in the diagnostic parameters . the galaxy is classified as a green - valley galaxy on the color mass diagram ( see figure [ fig : sample_properties ] ) due to its intermediate color , nuv@xmath28@xmath57 , which is unusually blue for its large stellar mass . the radial variations of the galaxies are shown more clearly in figure [ fig : profiles ] , where we plot the radial profiles of the three parameters for all 12 galaxies . to save space we display two galaxies in every panel , with one csf galaxy ( blue symbols ) and one cq galaxy ( red symbols ) . this paired format also highlights the highly different behavior of the two broad classes . in constructing these profiles , we have corrected for the effects of inclination on the radius for each spaxel in the datacubes , using the minor - to - major axis ratio from nsa , as determined from the sdss @xmath29-band photometry . when estimating the radial profile , we bin all the spaxels with the continuum @xmath58 at 5500 into a set of radial intervals with a constant width of @xmath59@xmath60@xmath61 , according to the deprojected radius of the spaxel ( @xmath29 ) and the effective radius of the galaxy ( @xmath60 ) . the value of the profile at given radius is then estimated by the median of the spaxels falling in the radial interval , and the error is given by the @xmath62 scatter between the spaxels . figure [ fig : profiles ] clearly shows that cq galaxies generally present significant radial gradients in all the parameters , in the sense that the outer part of the galaxies show weaker d@xmath0(4000 ) ( thus a larger fraction of stellar populations with ages of @xmath63 ) , stronger h@xmath1 absorption ( thus more recent star formation in the past a few myr ) , and stronger h@xmath3 emission ( thus likely stronger ongoing star formation ) . in contrast , csf galaxies possess quite flat profiles in all cases , revealing little by way of gradient in star - formation properties . at least out to @xmath60 , the radial profiles are close to linear for all the galaxies . for each galaxy we apply a linear fit to the radial profile within @xmath60 and show the best fit as a solid line in the figure . the slope of the best - fit line is indicated in each panel . figure [ fig : compare_slopes ] compares the slope index for the three parameters , @xmath64d@xmath0(4000)@xmath65 , @xmath64ew(h@xmath1@xmath2)@xmath65 and @xmath64@xmath30ew(h@xmath3)@xmath65 . these plots confirm that the radial gradients in the csf galaxies are small , but also shows that the different parameters are well correlated across galaxies of both types . this correlation is particularly strong for @xmath64d@xmath0(4000)@xmath65 and @xmath64ew(h@xmath1@xmath2)@xmath65 , which are beautifully ( albeit non - linearly ) correlated . however , given the small sample size , these apparent correlations should probably not be over - interpreted . figure [ fig : profiles_in_a_panel ] compares the slopes of the radial profiles in a different way , where we plot all the profiles in the same panel for a given parameter , normalizing each profile by the inner - most radial bin and scaling the radii by the effective radius . the differing behavior of the two classes of galaxies is striking . the csf galaxies show only weak or no radial variation , while the cq galaxies display steeper profiles , spanning a wide range in slope . it is , however , interesting to note that although the average properties of the two types differ dramatically , there is a degree of overlap between the classes in all panels , suggesting that a continuum of the same underlying physical processes has driven star formation in all these galaxies . as indicated from the 2d maps , the galaxies with a blue outer disk ( ma003_9 and ma008_4 ) indeed show the strongest radial variation in all parameters . these two galaxies manifest their peculiarity in the profile of @xmath30ew(h@xmath3 ) , which is flat ( for ma008_4 ) or even negative ( for ma003_9 ) at @xmath66@xmath60 , before rapidly increasing at larger radii . when compared to these two galaxies , the other cq galaxies present much shallower profiles ( though still steeper than that of csf galaxies ) , with @xmath30ew(h@xmath3)increasing smoothly at @xmath67@xmath60 and becoming flat at larger radii . the large values of @xmath30ew(h@xmath3 ) at large radii indicates that these two galaxies are strongly forming stars in their outer disks . this measurement is consistent with blue nuv@xmath28 color in these outer disks , which is known to be a sensitive indicator of the cold gas mass fraction in galaxies ( e.g. * ? ? ? * ; * ? ? ? the negative @xmath30ew(h@xmath3 ) profile at @xmath66@xmath60 seen in ma003_9 is also found in ma003_4 , and may be caused by the presence of a bulge or agn in the galactic center . indeed , according to the bpt diagram ( figure [ fig : sample_properties ] ) , the two galaxies are the only agn candidates in the sample . it is likely that an agn contributes to the h@xmath3 emission in the central region , thus enhancing the central @xmath30ew(h@xmath3 ) and leading to a dip in the profile as seen at @xmath68@xmath60 . interestingly , apart from this feature , the radial profiles of the two probably agn hosts are similar to those of the other cq galaxies , suggesting that current , low levels of agn activity may not have a strong effect on recent star formation activity , at least for these two examples . it is interesting to note that the three galaxies discussed above , ma003_9 , ma008_4 and ma003_4 , share a number of common properties : they are all spiral galaxies with a quiescent center , at the massive end of our sample with m@xmath69@xmath70m@xmath36 , and showing the largest gradients in d@xmath0(4000)indicative of young stellar populations formed within the past a few gyr . they all share some properties with the population of `` passive red spirals '' , identified recently as an interesting set of possible transition objects with disk - like morphologies but red colors , at both low - z ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and high - z ( e.g. * ? ? ? statistical analyses on a larger sample from manga ( or other ifu surveys ) is needed in order to better understand the physical link between our galaxies and the passive / quiescent spiral population . the 2d maps in figures [ fig : maps1 ] and [ fig : maps2 ] , as well as the radial profiles in figures [ fig : profiles ] and [ fig : profiles_in_a_panel ] , are in broad agreement with maps / profiles of stellar population age estimated from both califa @xcite and p - manga @xcite , in the sense that most galaxies in the nearby universe show flat to negative gradients in stellar age , although in those studies the detailed behaviors of the age gradients depend on galaxy mass or morphology . in this work we only consider the dependence of our measurements on the d@xmath0(4000)of the galactic center , which is an indicator of the prominence of stellar populations younger than 1 - 2 gyr , rather than the mean stellar age as investigated in those studies . despite these differences , the ifu - based results obtained so far are all consistent with the `` inside - out '' growth of galactic disks @xcite , a picture that was supported previously by both numerical simulations ( e.g. * ? ? ? * ) and abundant studies of broadband color gradients pioneered more than 50 years ago by @xcite . we now return to the comparison of diagnostic diagrams of recent star formation history , based on the radial profiles obtained in the previous subsection . figure [ fig : diagnostics ] presents the d@xmath0(4000 ) vs. ew(h@xmath1@xmath2 ) ( left panels ) and ew(h@xmath1@xmath2 ) vs. @xmath30ew(h@xmath3 ) ( right panels ) planes , with each panel once again contrasting two galaxies , one csf system ( blue symbols ) and one cq system ( red symbols ) . panels are ordered such that d@xmath0(4000 ) increases from top to bottom . for each galaxy , the radial profiles within the effective radius @xmath60 as shown in the previous figure are plotted here , with larger data points corresponding to larger radii . the set of bc03 models in figure [ fig : d4k_hda_hae ] are also repeated here as the solid and dashed lines , but for clarity these lines are not colored ; they provide a reference grid to register the relative position of the different radial bins on the diagnostic diagrams . both the weak radial variations for csf galaxies and the strong variations for cq galaxies are readily apparent in this figure . comparing the location of the radial bins on the two diagrams , as well as their relative location with respect to the model curves , presents a comprehensive picture of how and where star formation cessation is occurring within the galaxies . the diagrams demonstrate that all the cq galaxies behave in the same manner : d@xmath0(4000 ) decreases , while ew(h@xmath1@xmath2 ) and @xmath30ew(h@xmath3 ) increase , as one moves outward in the galaxies . this consistency is even clearer in figure [ fig : diagnostics_1panel ] where we overplot the radial profiles of all 12 galaxies in a single set of diagnostic diagrams . the profiles are plotted with cyan triangles for csf galaxies , and magenta triangles for cq galaxies . we highlight the two peculiar galaxies discussed above , ( ma003_9 and ma008_4 ) , which show the strongest radial gradients and unusual shape in the profile of @xmath30ew(h@xmath3 ) , by identifying them with yellow triangles . we also pick out the green - valley galaxies ( @xmath71nuv@xmath28@xmath72 ) as green triangles . for comparison , the central spaxels are plotted with blue open symbols and red solid symbols for the two classes of galaxies . the cq galaxies indeed follow a tight track on the d@xmath0(4000 ) versus ew(h@xmath1@xmath2 ) plane . these spatially - resolved data , reaching out to @xmath60 for all 12 galaxies , also show how the clear distinction between the two classes of galaxies becomes blurred when one also considers their outer parts . figure [ fig : diagnostics_1panel ] shows that the gaps in the parameter space between the regions occupied by the centers of the galaxies are populated by these same galaxies outer parts . this continuous sequence forms a tight relation in the d@xmath0(4000)-ew(h@xmath1@xmath2 ) plane , closely following the continuous star formation models . in detail , though , the intermediate region on the diagrams is mainly occupied by the two galaxies ( ma003_9 and ma008_4 ) that have strong radial variations and unusual profiles in @xmath30ew(h@xmath3 ) . in order to see this dominance more clearly , we have repeated the same panels of figure [ fig : diagnostics_1panel ] , but excluding these two systems , and we find the gap between the galaxy types once again starts to emerge . it is clear , however , that the cq galaxies do still extend substantially into the intermediate region , and that the effect appears to be stronger for green - valley galaxies than for the other objects in the cq galaxy class . ma008_4 : : this system , cgcg 184 - 033 , is the central galaxy of a group in the catalog of @xcite . the system comprises 8 member galaxies with spectroscopically - measured redshift from sdss , including the central galaxy itself . this is the most massive galaxy in the sample with a stellar mass of @xmath56 . the sdss spectroscopy and the central spaxel of the p - manga datacube reveal that the central region of this galaxy shows relatively strong absorption in h@xmath1 for its ew(h@xmath3 ) , implying that a starburst event has happened within the galaxy in the past 1 @xmath43 . considering the fact that the galaxy has a very close companion , a strongly star - forming spiral at the same redshift ( z=0.042 ) , the post - starburst feature may be the product of a tidal interaction with the companion , which caused the central part of the galaxy to have ceased star formation recently . the p - manga data also reveal strong radial variation in all the diagnostic parameters in this galaxy : d@xmath0(4000 ) , ew(h@xmath1@xmath2 ) and ew(h@xmath3 ) . in particular the profile of @xmath30ew(h@xmath3 ) shows a very strange shape when compared to the other cq galaxies . ma003_9 : : this galaxy is a spiral with a red core but very blue outer disk in the optical image ( see figure [ fig : maps2 ] ) . it is classified as a blue - cloud galaxy because of its blue nuv@xmath28 color , but it is classified as a centrally quiescent galaxy due to its large central d@xmath0(4000 ) . the red core with old stellar populations and a blue global color seems to suggest that the galaxy might be a ` bluedisk ' galaxy , a class of system with an unusually high fraction of hi gas mass , as recently studied in detail by @xcite . however , the hi gas mass is @xmath73 according to the hi global profile from @xcite , homogenized in @xcite . this is comparable to the value expected from the nuv@xmath28 and stellar surface mass density for a normal spiral @xcite . the bpt diagram shown in figure [ fig : sample_properties ] reveals strong nuclear activity in the galactic center . a @xmath74 mjy source coincident with this galaxy is detected by the nvss survey @xcite , but no source is detected by the first survey @xcite , suggesting that the radio emission from this galaxy is variable . this is consistent with an agn origin , albeit at a low luminosity ( @xmath75 w / hz ) . see the related paper by @xcite for a spatially - resolved bpt diagram of this galaxy , also based on the p - manga data . ma003_4 : : this system is one of the two agns in the sample ( the other one is ma003_9 above ) . the agn component is likely responsible for the strange profiles of the @xmath30ew(h@xmath3 ) diagnostic in these systems , as discussed above . ma006_4 : : with a stellar mass of @xmath76 , this galaxy is the only low - mass cq galaxy in the p - manga sample . it has unusually low ew(h@xmath3 ) for its ew(h@xmath1@xmath2 ) ( see figure [ fig : diagnostics ] ) . this might be a metallicity effect , as the galaxy is much closer to the continuous models for a lower metallicity of z=0.008 . using datacubes for 12 galaxies produced during the manga prototype ( p - manga ) observations , we have obtained maps and radial profiles of the d@xmath0(4000 ) , ew(h@xmath1@xmath2 ) and ew(h@xmath3 ) spectral diagnostics . these parameters were derived for each spaxel by decomposing the spectrum into an emission - line component and a continuum plus absorption - line component , and measuring the relevant indices from these separate components . we study the spatially - resolved recent star formation histories of these galaxies by displaying their radial profiles on the d@xmath0(4000 ) versus ew(h@xmath1@xmath2 ) and ew(h@xmath1@xmath2 ) versus @xmath30ew(h@xmath3 ) planes . we have classified the galaxies as either ` centrally quiescent ' ( cq ) or ` centrally star - forming ' ( csf ) according to the value of d@xmath0(4000 ) in the central spaxel , and we compare the 2d maps and radial profiles for the two classes . cq galaxies present significant radial gradients in all the three parameters , in the sense that d@xmath0(4000 ) decreases , while both ew(h@xmath1@xmath2 ) and ew(h@xmath3 ) increase as one goes from the galactic center toward the outer region . in contrast , csf galaxies show very weak or no radial variations . the outer parts of the galaxies show greater scatter on the diagnostic diagrams than their central regions . in particular , the separation between cq and csf galaxies in these diagnostic planes is largely filled in by the outer parts of galaxies whose global colors place them in the green valley . the three spectral diagnostics are known to be sensitive indicators of stellar populations of different ages . spatially resolving a galaxy on the d@xmath0(4000)ew(h@xmath1@xmath2)ew(h@xmath3 ) diagrams probes the radial variations in recent star formation histories , thus revealing the way in which the galaxy grows and its star formation ceases . when spatially resolved , the p - manga galaxies are found to form a tight sequence on the d@xmath0(4000 ) versus ew(h@xmath1@xmath2 ) plane , although the distribution of their central regions is strongly bimodal ( figure [ fig : diagnostics_1panel ] ) . this sequence closely follows the continuous star formation locus predicted by current stellar population models , and covers a very similar area to the distribution of the large sample of galactic centers from sdss ( figure [ fig : sdss_d4k_hda_hae ] ) . this relation suggests that , at least for the systems being studied here , galaxy growth has been a smooth process , and is likely regulated by a common set of physical drivers . this result is consistent with the ` inside - out ' picture of galaxy growth , where the stellar mass assembly starts in the galactic center and gradually extends to the outer regions @xcite . in this picture , the shutdown of star formation also first occurs in the central region and slowly propagates out to ever larger radii . however , this result should not be overemphasized given the small size of our sample . the tight sequence on the d@xmath0(4000)ew(h@xmath1@xmath2 ) diagram might be just a result of small number statistic . it could also be the case that there are a variety of physical processes behind this sequence , which can be discriminated only when a larger sample becomes available . radially resolved stellar populations have been studied previously to some extent , based on multi - wavelength broadband photometry at both low - z ( e.g. * ? ? ? * ; * ? ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and high - z ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , as well as long - slit spectroscopy for nearby galaxies ( e.g. * ? ? ? * ; * ? ? ? most spirals are found to have stellar population gradients , with the inner regions being older and more metal - rich than their outer regions , supporting the inside - out scenario of disk galaxy formation . the majority of the early - type galaxies show flat color profiles , with only a small fraction being `` blue - core '' systems with negative color gradients ( i.e. bluer outer parts ) . the star formation history of galactic disks is found to be primarily driven by its structural properties , as usually quantified by the surface stellar mass density ( @xmath77 , * ? ? ? * ) and stellar light concentration @xcite , while the total stellar mass is a secondary parameter , correlated more closely with metallicity than with age . for massive galaxies , recent short - lived episodes or bursts of star formation occurring in their outer regions are required to interpret the radially resolved spectral indices as measured from long - slit spectroscopy @xcite , a result that is consistent with the existence of many early - type galaxies in the nearby universe which display extended star - forming discs , as originally discovered by ultraviolet imaging ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the overall radial dependence of the diagnostic parameters presented in our work is apparently consistent with these previous findings , although the small sample size does nt allow the correlation of the radial profiles with galaxy morphology , stellar mass and structural properties to be examined in a statistical manner . more recent studies using data from califa have provided the most comprehensive results so far regarding the radial variations of the stellar population and star formation history of nearby galaxies @xcite . @xcite analyzed the first 105 galaxies of the califa survey , quantifying their spatially resolved history of stellar mass assembly and demonstrating how massive galaxies grow their stellar mass inside - out . oxygen abundances estimated from hii regions in 306 califa galaxies present a common negative gradient in the oxygen abundance between 0.3 and 2 times @xmath60 , independent of morphology , the presence of bars , absolute magnitude or mass , provided that the analysis is limited to those galaxies without clear evidence of an interaction @xcite . this result also supports the inside - out picture for the growth of galactic disks , and is consistent with the stellar age and metallicity gradients obtained by @xcite based on califa data for a smaller sample of 62 nearly face - on spirals . using a sample of 107 galaxies from califa , @xcite performed a detailed study of the radial structure of the stellar mass surface density ( @xmath77 ) and stellar population age , as well as their dependence on the total stellar mass and morphology . in support of the inside - out formation picture , negative gradients of the stellar age are present in most of their galaxies . the behaviors of the d@xmath0(4000 ) profiles in the p - manga galaxies ( see fig . [ fig : profiles_in_a_panel ] ) are in good agreement with this result , although the d@xmath0(4000 ) is not a direct measure of stellar age , but rather an indicator of the prominence of young stellar populations . confirming the previous broadband photometry - based findings , the califa data reveals that @xmath77 is more important than m@xmath69 in driving the star formation history in disks , while m@xmath69plays a more fundamental role in spheroidal systems . the califa results , together with the results from broadband photometry and long - slit spectroscopy , strongly suggest that galaxy growth and death are driven by multiple parameters ( at least the local surface density , total stellar mass and morphology ) , and that further studies of much larger samples are needed in order to isolate the role of each parameter , and the physical process behind it . the increasing manga sample is expected to provide much better results in the near future . another interesting result from our work is that , when spatially resolved , the distribution of galaxies on the diagnostic diagrams is no longer bimodal , and the separation between the cq and csf populations is filled in by those galaxies that are classified as ` green - valley ' according to their optical / uv color . in a recent complementary study , @xcite selected from sdss a large sample of ` recently quenched ' galaxies that have a significant fraction of young stars within the central @xmath78 region , and found them to populate the region of green - valley galaxies in d@xmath0(4000 ) and optical colors , so their central parts seem to be at the same evolutionary stage as the outer parts of the p - manga galaxies . the two studies combine to suggest that the green valley does , indeed , pick out a transition phase between star - forming and passive populations , but that different parts of different galaxies enter the green valley at different times . there are a few caveats to this analysis that we will address in future work . we have used relative measurements of d@xmath0(4000 ) , ew(h@xmath1@xmath2 ) and ew(h@xmath3 ) as indicators of recent star formation history . we have not presented quantitative measurements of mean stellar ages due to the sensitivity of these indices to stellar metallicity , element abundance and dust attenuation @xcite . throughout this paper we use solar metallicity models for comparison . we have compared the data to models with three different metallicities ( z=0.4 , 1 , 2.5z@xmath36 ) , finding that the dependence on metallicity is very small for younger stellar populations , but become more important at older ages . in @xcite we explore the possibility of simultaneously deriving the age and metallicity of p - manga galaxies using stellar population model fits to the full spectrum . in the present paper , we use the h@xmath3 emission line to characterize the instantaneous star formation rate . however , for galaxies with dominant old stellar populations , ob stars may not be the only source of ionizing photons ( e.g. * ? ? ? * ; * ? ? ? indeed , ew(h@xmath3 ) lower than 6 was found to be more compatible with light from post - agb stars @xcite , and it would be interesting to evaluate the contribution of such stars to the emission - line properties of the spectra once a larger sample is available . in a recent sdss - based study @xcite has nicely demonstrated that a combination of d@xmath0(4000 ) , ew(h@xmath1@xmath2 ) and the specific star formation rate ( sfr / m@xmath69 ) provides stringent constraints on the recent star formation histories of the _ central region _ of galaxies in the stellar mass range @xmath79m@xmath36 . in particular , this methodology is powerful in separating out the galaxies with ongoing or past starbursts from those with a continuous star formation history . by applying this new methodology to the spatially resolved spectroscopy for a large sample from manga , we may well expect to better quantify the fractions of the mass produced by ongoing and recent bursts , as well as their radial variation across the whole galaxy , thus leading to more reliable constraints on the maps and radial profiles of the recent star formation history . previous studies have proposed a variety of galaxy properties as indicators of the termination of star formation , including both color indices ( e.g. nuv@xmath28 ) and structural parameters like the central surface mass density and stellar velocity dispersion ( e.g. * ? ? ? * ) , the effective radius @xmath60 and the ratio of stellar mass relative to @xmath60 , @xmath60@xmath80 , or @xmath60@xmath81 ( e.g. * ? ? ? * ) , and sersic profile index @xmath82 ( e.g. * ? ? ? since different indicators may trace different physical processes , it is important to ask which galaxy property is most closely linked to the end of star formation . again , once larger samples of galaxies become available , it will be very interesting to correlate the kind of spectral diagnostic gradient study presented here with these other parameters , to see which is the trigger , and even how the trigger varies between galaxies of different types or in different environments . despite these limitations , it is heartening to see how much can be learned about the cessation of star formation from even a sample of 12 galaxies , thanks to the resolved spectroscopic observations that p - manga ifus have provided . the full manga survey began on 2014 july 1 , offering new opportunities to extend these kinds of studies , as observations of its 10,000 galaxies continue . cl acknowledges the support of the 100 talents program of chinese academy of sciences ( cas ) . this work is supported by nsfc ( 11173045,11233005,11325314,11320101002 ) , by the strategic priority research program `` the emergence of cosmological structures '' of cas ( grant no . xdb09000000 ) , and by world premier international research center initiative ( wpi initiative ) , mext , japan . we have made use of data from hyperleda for our work . funding for sdss - iii and sdss - iv has been provided by the alfred p. sloan foundation and participating institutions . additional funding for sdss - iii comes from the national science foundation and the u.s . department of energy office of science . further information about both projects is available at www.sdss3.org . sdss is managed by the astrophysical research consortium for the participating institutions in both collaborations . in sdss - iii these include the university of arizona , the brazilian participation group , brookhaven national laboratory , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . the participating institutions in sdss - iv are carnegie mellon university , colorado university , boulder , harvard - smithsonian center for astrophysics participation group , johns hopkins university , kavli institute for the physics and mathematics of the universe , max - planck - institut fuer astrophysik ( mpa garching ) , max - planck - institut fuer extraterrestrische physik ( mpe ) , max - planck - institut fuer astronomie ( mpia heidelberg ) , national astronomical observatory of china , new mexico state university , new york university , the ohio state university , penn state university , shanghai astronomical observatory , united kingdom participation group , university of portsmouth , university of utah , university of wisconsin , and yale university .

let be a classical split real lie group of type , , or and a maximal parabolic subgroup of . the homogeneous space is a minimal flag manifold of and may be realized , respectively , as an even orthogonal , isotropic and odd orthogonal real grassmannian . in this paper we compute the cellular-homology group of such manifolds by a lie theoretic approach which naturally furnish them with a cellular structure given by the schubert varieties . the results are well understood in terms of what we call half - shifted young diagrams whose construction is based on the the works of buch - kresch - tamvakis , ikeda - naruse and graham - kreiman .

1604.02177

let @xmath0 be a classical split real lie group of type @xmath1 , @xmath2 , or @xmath3 and let @xmath7 be its split real semi - simple lie algebra . let @xmath8 be a maximal parabolic subgroup of @xmath0 of type @xmath9 given as the complementary set of a root @xmath10 of the dynkin diagram of @xmath7 . the homogeneous space @xmath11 is called a minimal flag manifold of @xmath0 . in the context of type @xmath1 ( resp . @xmath3 ) , @xmath11 may be realized as the orthogonal grassmannians , i.e. , submanifolds of isotropic subspaces in an odd ( resp . even ) dimensional vector space equipped with a non - degenerate symmetric bilinear form . in the context of type @xmath2 , @xmath11 may be realized as the isotropic grassmannians , i.e. , submanifolds of isotropic subspaces in an even dimensional vector space equipped with a symplectic form . in this paper we compute the cellular @xmath6-homology group of such manifolds developing the half - shifted diagrams that appears as a fundamental tool to describe the cellular structure given by the schubert varieties . in the general context of flag manifolds , the topology of the complex flag manifolds of semi - simple lie groups is well developed and also a very ramified area of research nowadays ( see brion @xcite ) . yet the first results about the topology of real flag manifolds dates to 1970 s by burghelea hangan moscovici verona @xcite . in the 1980 s , the work of duistermaat - kolk - varadarajan @xcite computes the mod 2 homology . only in the 1990 s a complete description of its integral homology was done by kocherlakota @xcite . the morse theory appears as the main tool of these works . the first explicit cellular approach was described by wiggerman @xcite to get the fundamental groups of the real flag manifolds . those ideas where generalized in rabelo - san martin @xcite where a @xmath12-dimensional cellular version is obtained to improve the integral homology computations elucidating some ambiguities left in @xcite . a more sophisticated point of view is the infinite dimensional representation theory of the real semi - simple lie group described by casian - stanton @xcite . the euler characteristic was addressed in the paper by altomani medori nacinovicht @xcite . a complete description of the orientability was described by patro san martin seco @xcite . notice that in the real context there are cells in all dimensions while in the complex one it appears only in even dimensions and hence torsion may occur in the real case whilst in the complex case the homology boundary maps are trivial elsewhere . furthermore , the description of the cohomology ring over the integers of real flag manifolds is open in general . some results have been done for the classical real grassmannians , i.e. , grassmannians of type a by casian - kodama @xcite . in this context , the homology groups were described by burghelea hangan moscovici verona in the book @xcite . now , if we consider other grassmannians ( isotropic and orthogonal ) , the only specific results are the computations of its euler characteristic ( @xcite ) and its orientability ( @xcite ) . in rabelo @xcite , we have a first step towards solving the problem , whose result was obtained only for the maximal case . it is remarkable that the young diagrams has been largely used in the context of cohomology rings of complex grassmannians mainly because its proven product properties related to the comohology rings of such manifolds . instead , in the former paper of the second author @xcite , the strict young diagrams were the main tool to deal with homology of some real grassmannians . in this work , we extend those ideas about using adapted young diagrams to lead with the additive structure of the homology . since the homology groups of real flag manifolds are completely described by what is called by ikeda - naruse as the beta - sequence of a weyl group element , one of the main contribution of this paper lies in the presentation of the beta - sequences of the weyl group elements that parametrizes all grassmannians of types b , c and d in terms of the half - shifted young diagrams . for type a , this was done by ikeda - naruse @xcite and also by graham - kreiman @xcite . we also hope that these diagrams may provide a way to get some results about the cohomology in the same sense that was done by casian - kodama @xcite for type a. we now give a very brief introduction to these diagrams ( for details , see the sections [ secyoung ] and [ secroots ] ) and the results for homology . let @xmath13 be the @xmath14-module freely generated by the schubert varieties @xmath15 , @xmath16 , the main discussion in this work is how to compute the integer coefficients @xmath17 of the boundary map @xmath18 . if @xmath19 , the coefficient has the formula @xmath20 , where @xmath21 is the integer defined by @xmath22 , @xmath23 and @xmath24 is the unique root such that @xmath25 . in general , a element @xmath16 , either of type b , c or d , can be written as a permutation of the form @xmath26 where @xmath27 , @xmath28 and @xmath29 . using the definition in pragacz - ratajski @xcite , the set @xmath30 is bijective to set of double partitions @xmath31 and each double partition can be represented as a `` half '' shifted young diagram ( which for groups of type b and c , means that the young diagram of @xmath32 fits inside a @xmath33 square and @xmath34 fits inside a @xmath35 staircase shape ) . the benefit to use the half shifted form for the young diagrams is demonstrated when we need to compute the set @xmath36 of all positive roots sent to negative by @xmath37 , which give us a easy way to get the coefficient @xmath38 only drawing its half shifted young diagrams . given a schubert variety @xmath39 with partition @xmath40 such that @xmath19 , the diagram associated to @xmath41 is obtained by removing one box from the diagram of @xmath42 . the partition @xmath41 is called 1 . a _ @xmath32-removing of @xmath42 _ if @xmath43 for some @xmath44 ; 2 . a _ @xmath34-removing of @xmath42 _ if @xmath45 for some @xmath46 . our main theorem is stated below : given @xmath47 in @xmath30 such that @xmath48 , this can be either a @xmath32-removing or a @xmath34-removing of @xmath42 and consider @xmath49 the row where the box is removed . the coefficient @xmath17 in the boundary map is obtained from @xmath50 , which depends on the group @xmath0 and which box is removed : @xmath51 the article is organized as follows . section [ secpre ] presents the preliminaries about the flag manifolds and the bruhat decomposition . section [ sechomflag ] summarizes the basic results about the cellular homology of real flag manifolds as obtained in @xcite . section [ secyoung ] shows the description of the double young s diagram of all grassmannians and how it is possible to write the weyl element from the diagram . section [ secroots ] presents the formulas to get the set @xmath36 of all positive roots sent to negative by @xmath37 . section [ secmares ] states the main results describing the boundary maps of the cellular homology of grassmannians of type b , c and d. section [ secproofs ] explicitly presents the computations . we thanks to david anderson for helpful suggestions and valuable discussions . the main facts about semi - simple lie groups and their flag manifolds may be found in helgason @xcite , knapp @xcite , warner @xcite and san martin @xcite . furthermore , @xcite is the reference for the notation and for some specific results is the article . flag manifolds are defined as homogeneous spaces @xmath5 where @xmath0 is a non - compact semi - simple lie group and @xmath4 is a parabolic subgroup of @xmath0 . let @xmath7 be a non - compact real semi - simple lie algebra . the flag manifolds for the several groups @xmath0 with lie algebra @xmath7 are the same . take a cartan decomposition @xmath52 and let @xmath53 be a maximal abelian sub - algebra contained in @xmath54 . we denote by @xmath55 the set of roots of the pair @xmath56 and fix a simple system of roots @xmath57 . denote by @xmath58 the set of positive and negative roots respectively and by @xmath59 the weyl chamber @xmath60 . let @xmath61 be the direct sum of root spaces corresponding to the positive roots . the iwasawa decomposition of @xmath7 is given by @xmath62 . the notations @xmath63 and @xmath64 are used to indicate the connected subgroups whose lie algebras are @xmath65 and @xmath66 respectively . a sub - algebra @xmath67 is said to be a cartan sub - algebra if @xmath68 is a cartan sub - algebra of @xmath69 . if @xmath70 is a cartan sub - algebra of @xmath7 we say that @xmath7 is a split real form of @xmath69 . a minimal parabolic sub - algebra of @xmath7 is given by @xmath71 where @xmath72 is the centralizer of @xmath53 in @xmath65 . let @xmath4 be the minimal parabolic subgroup with lie algebra @xmath73 . note that @xmath4 is the normalizer of @xmath73 in @xmath0 . we call @xmath74 the maximal flag manifold of @xmath0 and denote by @xmath75 the base point @xmath76 in @xmath5 . associated to a subset of simple roots @xmath77 there are several lie algebras and groups . we write @xmath78 for the semi - simple lie algebra generated by @xmath79 , @xmath80 . let @xmath81 be the connected group with lie algebra @xmath78 . moreover , let @xmath82 be the sub - algebra generated by the roots spaces @xmath83 , @xmath84 and put @xmath85 . the normalizer @xmath86 of @xmath87 in @xmath0 is a standard parabolic subgroup which contains @xmath4 . the corresponding flag manifold @xmath88 is called a partial flag manifold of @xmath0 or flag manifold of type @xmath9 . we denote by @xmath89 the base point @xmath90 in @xmath91 . in this paper the classical split real lie groups considered are the orthogonal groups @xmath92 , @xmath93 , @xmath94 , @xmath95 , and the symplectic group @xmath96 , @xmath97 , where @xmath7 is the respective lie algebra given by @xmath98 , @xmath99 and @xmath100 . according to @xcite , @xmath98 , @xmath99 are real forms of type bdi and @xmath100 of type ci . given @xmath101 , we will work with the minimal flag manifolds of the form @xmath11 where @xmath102 is a maximal proper subset of the simple set of roots . let @xmath103 be the set of simple roots of @xmath7 ordered as below : for each choice of group @xmath0 , the corresponding flag manifold has a geometric realization . the orthogonal grassmannian of type @xmath1 is the set of @xmath104-dimensional isotropic subspaces in the vector space @xmath105 equipped with a nondegenerate symmetric bilinear form . it will be denoted by @xmath106 . the isotropic grassmannian of type @xmath2 is the set @xmath104-dimensional isotropic subspaces in the symplectic vector space @xmath107 . this set will be denoted by @xmath108 . the orthogonal grassmannian of type @xmath3 is the set of @xmath109-dimensional isotropic subspaces in the vector space @xmath110 equipped with a nondegenerate symmetric bilinear form . it will be denoted by @xmath111 . a central role in our context will be played by the weyl group @xmath112 associated to @xmath53 . this is the finite group generated by the reflections over the root hyperplanes @xmath113 contained in @xmath53 , @xmath114 . alternatively , it may be given as the quotient @xmath115 where @xmath116 and @xmath117 are respectively the normalizer and the centralizer of @xmath53 in @xmath63 ( the lie algebra of @xmath118 is @xmath72 ) . viewing the elements of @xmath112 as product of simple reflections @xmath119 , @xmath120 , the length @xmath121 of @xmath122 is the number of simple reflections in any reduced decomposition of @xmath42 which is equal to the cardinality of @xmath123 , the set of positive roots sent to negative roots by @xmath37 . if @xmath124 is a reduced decomposition of @xmath42 then we put @xmath125 it follows that @xmath126 . this is called the @xmath24-sequence of @xmath42 . there is a partial order in the weyl group called the bruhat - chevalley order . we say that @xmath127 if given a reduced decomposition @xmath128 then @xmath129 for some @xmath130 . for a subset @xmath131 , the subgroup @xmath132 is defined to be the stabilizer of @xmath133 . alternatively , @xmath134 may be seen as the subgroup of the weyl group generated by the reflections with respect to the roots @xmath84 . we also define the subset @xmath135 of @xmath112 by @xmath136 since there exists a unique element @xmath137 of minimal length in each coset @xmath138 , @xmath135 is called the subset of minimal representatives of the cosets of @xmath139 in @xmath112 . the bruhat decomposition presents the flag manifolds as a union of @xmath64-orbits , namely , @xmath140 where @xmath141 if @xmath142 . each @xmath64-orbit through @xmath42 is diffeomorphic to a euclidean space . such an orbit @xmath143 is called a bruhat cell . its dimension is given by the formula @xmath144 where @xmath145 is the multiplicity of the root space @xmath146 and @xmath147 denotes the roots in @xmath148 generated by @xmath9 . given any @xmath149 , in order to establish a relation between the dimension @xmath150 and the length @xmath121 , we must choose the minimal representative for @xmath42 in @xmath135 . in this case , for @xmath151 , @xmath152 ( see @xcite , corollary 2.6 ) . in the case of a split real form , i.e. , @xmath153 and @xmath154 , it follows that @xmath155 a schubert variety is the closure of a bruhat cell , i.e. , @xmath156 for a maximal flag manifold we avoid the superscript @xmath157 and write @xmath158 the bruhat - chevalley order defines an order between the schubert varieties by @xmath159 a well known characterization of the schubert varieties is given by @xmath160 i.e. , schubert varieties endow the flag manifolds with a cellular decomposition . we summarize the main results of @xcite . for our purpose , we are supposing that @xmath0 is a group of type b , c or d. consider firstly a maximal flag manifold @xmath161 . the cellular homology of a cw complex is defined from a cellular decomposition of the complex provided in our context by the schubert varieties . given a schubert variety @xmath162 , we fix once and for all reduced decompositions @xmath124 as a product of simple reflections , for each @xmath163 , and @xmath164 , @xmath120 . let @xmath165 be the @xmath6-module freely generated by @xmath166 , @xmath163 . the boundary maps @xmath167 are defined by @xmath168 for some coefficients @xmath169 . in case @xmath170 then @xmath171 . if @xmath172 then @xmath173 is computed as the degree of a map between cells of codimension one ( see @xcite , page 140 ) . [ propappendix ] let @xmath174 . the following statements are equivalent . 1 . @xmath175 and @xmath172 ; 2 . if @xmath124 is a reduced decomposition of @xmath163 as a product of simple reflections , then @xmath176 is a reduced decomposition and @xmath177 . the first main result is that the coefficient @xmath17 is the sum of the degree of two sphere homeomorphisms which has degree one . for any choice @xmath178 such that @xmath172 , the coefficient @xmath179 . it is possible to get a more accurate expression for the coefficients @xmath17 in terms of roots . for @xmath124 and @xmath180 , reduced decompositions of @xmath42 and @xmath41 respectively ( see proposition [ propappendix ] ) , we define @xmath181 and @xmath182 [ teo : mainrs ] [ teoforcw ] let be @xmath183 be defined as in ( [ forsigmasoma ] ) . then @xmath184 for @xmath122 , let @xmath185 be the sum of roots in @xmath186 , i.e. , the sum of the @xmath24-sequences of @xmath42 . given @xmath42 and @xmath41 such that @xmath187 , let @xmath188 be the unique root ( not necessarily simple ) satisfying @xmath189 , that is , @xmath190 . then it is possible to show that @xmath191 where @xmath192 is the sum ( [ forsigmasoma ] ) ( see @xcite , proposition 4.8 ) . we get immediately the following formula for @xmath193 . [ teoforcw1 ] @xmath194 where @xmath195 is the integer defined by @xmath196 and @xmath188 is the unique root such that @xmath189 . this result was first obtained by kocherlakota ( see theorem a , @xcite ) by a morse homology approach . the advantage here is working with reduced decompositions to compute the numbers @xmath197 . in the context of the partial flag manifolds @xmath198 , the schubert varieties are @xmath199 , @xmath200 . [ teo : mainrs2 ] the cellular homology of @xmath198 is isomorphic to the homology of @xmath201 which is the boundary map of the free module @xmath202 generated by @xmath162 , @xmath203 , obtained by restricting @xmath204 and projecting onto @xmath202 . hence the coefficients @xmath205,[w^{\prime } ] ) $ ] for the boundary map @xmath206 of the cellular homology of the partial flag manifolds @xmath207 is @xmath208,[w^{\prime } ] ) = c(w , w^{\prime } ) .\ ] ] and the computation of @xmath205,[w'])$ ] reduces to a computation of @xmath17 on @xmath161 . the goal of this section is to provide the main tool for dealing with the combinatorics of isotropic grassmannians . the complexity of the weyl group that parametrizes the cellular decomposition of these manifolds has a very nice description in terms of what we call _ half shifted _ young diagrams . the main source to construct such diagrams is the work of buch - kresch - tamvakis @xcite , where the model of @xmath209-strict partition and young diagrams is presented together with properties of grassmannians of all types . another reference is given by ikeda - naruse @xcite and also by graham - kreiner @xcite in the context of maximal grassmannians where its found how to fill in each young diagram with the positive roots sent to negative by the respective weyl group element . here we propose a model where it is generalized by showing how to fill in the diagrams defined by bkt for all types of grassmannians with the positive roots sent to negative by the respective weyl group element . as far as we know , such a description has not been done yet elsewhere . in particular , this will be very useful to compute boundary maps for cellular homology . consider the grassmannian @xmath210 which parametrizes @xmath104-dimensional isotropic subspaces of a real @xmath211-dimensional symplectic vector space . let @xmath212 denote the set of integer partitions @xmath213 with @xmath214 so that the young diagrams of @xmath32 fits inside a @xmath215 rectangle . also define @xmath216 the set of all strict partitions @xmath217 with @xmath218 . the weyl group @xmath219 for the root system @xmath220 are permutations in @xmath221 with a sign attached to each entry , that means it is isomorphic to the semidirect product @xmath222 ; we will write these elements as barred permutations of the form @xmath223 using the bar to denote a negative sign , and we take the natural order on them , as above . if @xmath224 is the parabolic subgroup generated by @xmath225 then the set @xmath226 of minimal length coset representatives of @xmath224 parametrizes the schubert varieties in @xmath210 . this indexing set @xmath227 can be identified by a set of barred permutations of the form @xmath228 where @xmath229 with @xmath230 , @xmath231 and @xmath232 . there is a description of elements of @xmath30 by means of young diagrams , i.e. , each element @xmath233 corresponds to a pair of partitions @xmath31 where the `` top '' partition @xmath234 is defined slightly different from pragacz - ratajski @xcite @xmath235 for @xmath236 and @xmath237 ; the `` bottom '' partition is @xmath34 . note that the length of @xmath42 is @xmath238 , where @xmath239 and @xmath240 . the schubert varieties @xmath241 are then parametrized by the set @xmath242 of pairs @xmath243 with @xmath244 , @xmath229 and such that @xmath245 . we may arrange the top and the bottom partition in the same diagram . we propose here an alternative way of arrange the diagram of @xmath246 which corresponds just to a `` half '' shifted diagram in the sense presented by tamvakis @xcite . we consider the top diagram @xmath32 left justified in the @xmath33 rectangle and , for each @xmath247 , the @xmath248-th row of @xmath34 is shifted to the right @xmath249 units . with that shift , the bottom diagram may be seen inside a staircase partition and for reasons that will be clear soon we consider the staircase partition with @xmath12 rows . the condition @xmath245 implies that the number of rows in the bottom diagram does not exceed the number of boxes in the last row of the top diagram . let us denote by @xmath250 and by @xmath251 the set of square boxes with coordinates @xmath252 the top and the bottom ( shifted ) diagrams , respectively . each diagram @xmath250 and @xmath253 is arranged in a plane with matrix - style coordinates . the figure [ fig : schub_shift_554_8741 ] presents the half shifted diagram of the schubert variety @xmath254 in @xmath255 and how to read a @xmath256 box in each diagram . in @xmath255 . ] for each row in the top diagram , construct a diagonal line ( indicated as a dashed line in the figure [ fig : schub_related_554_8741 ] ) from the center of the last box at right of the respective row to the center of the first box in the first row of the bottom diagram ( it means that these boxes are @xmath209-related in this half shifted diagram as defined by tamvakis @xcite ) . consider the boxes of the staircase partition that are outside @xmath34 organized into columns . the columns that contain a such @xmath209-related box are called _ related _ and the remaining columns are _ non - related_. the _ length _ of a column is the number of blank boxes of this column in the staircase @xmath35 shape . we may recover the permutation associated to such diagram by taking the length of the related and non - related columns . namely , the permutation element for @xmath257 is defined by @xmath258 in the equation , where @xmath259 are the length of the related columns , @xmath260 and @xmath261 are the length of the non - related columns . for example , the partition @xmath262 corresponds to the element @xmath263 as illustrated in the figure [ fig : schub_related_554_8741 ] . in @xmath255 . ] this description represents a slight modification with respect to that is given in @xcite in terms of the model of @xmath209-strict partitions . the choice for representing schubert varieties by half shifted diagrams instead of a diagram associated to a @xmath209-strict partition will be clarified in the next section . we now give a notion of row - reading expression for each @xmath264 , @xmath265 . with respect to the top partition @xmath32 , let @xmath266 be defined , for @xmath267 and @xmath268 , by @xmath269 for a given partition @xmath244 associated to @xmath265 , the row - reading map is a bijection @xmath270 defined by assign the numbers increasingly to the boxes of @xmath250 from right to left starting from the bottom row to the top row . then , we can form a word @xmath271 where @xmath272 , for all @xmath273 . with respect to the bottom partition @xmath34 , let @xmath274 , for @xmath275 , @xmath276 and @xmath277 , by @xmath278 as before , for a given partition @xmath229 associated to @xmath265 , the row - reading map is a bijection @xmath279 defined by assign the numbers increasingly to the boxes of @xmath250 from right to left starting from the bottom row to the top row . then , we can form a word @xmath280 where @xmath281 , for all @xmath282 . the concatenation of expressions and gives @xmath283 . the maps @xmath284 , @xmath285 , @xmath286 and @xmath287 for the schubert variety given by the partition @xmath254 of @xmath255 are illustrated in figure [ fig : schub_rowreading_554_8741 ] . then , @xmath288 and @xmath289 . in @xmath255 . ] consider the odd orthogonal grassmannian @xmath290 which parametrizes @xmath104-dimensional isotropic subspaces of a real @xmath291-dimensional vector space equipped with a nondegenerate symmetric bilinear form . the weyl group for the root system @xmath292 is the same as that for @xmath220 . hence , they have the same diagrams . the only distinction will appear in the level of roots . this will be described in the next section . consider the grassmannian @xmath294 which parametrizes @xmath109-dimensional isotropic subspaces of a real @xmath295-dimensional vector space equipped with a nondegenerate symmetric form . we denote by @xmath296 the weyl group of type @xmath297 . the elements of @xmath296 are signed permutations of @xmath298 with an even number of sign changes which will be denoted by barred numbers . if @xmath299 is the subgroup of @xmath296 generated by @xmath300 , then the set @xmath301 of minimal length coset representatives of @xmath299 parametrizes the schubert varieties in @xmath302 . the set @xmath303 consists of barred permutations of the form @xmath304 where @xmath305 , @xmath306 , @xmath307 , @xmath308 and @xmath309 is equal to @xmath310 or @xmath311 so that the expression has an even number of bars . the elements in @xmath303 correspond to a set @xmath312 of partition pairs , which involve a partition @xmath313 and a @xmath229 such that @xmath245 . more precisely , define the partition @xmath32 as @xmath314 for @xmath267 . then , @xmath312 consists of * pairs @xmath315 such that @xmath316 ; these are elements of @xmath303 of the form @xmath317 which is referred as _ of type 0 _ ; * two types of pairs @xmath318 and @xmath319 $ ] such that @xmath320 ; these correspond to elements of @xmath303 of the form @xmath321 which is referred as _ of type 1 _ and @xmath322 } = ( \widehat{u}_{k } , \ldots , u_1 , \overline{\lambda_1 + 1 } , \ldots , \overline{\lambda_r+1 } , \overline{1 } , v_{n - k - r } , \ldots , v_1)\ ] ] which is referred as _ of type 2_. with such description we may represent these elements by means of young diagrams , i.e. , the pair of partitions @xmath323 $ ] ( where @xmath324 $ ] means a generic partition of any type ) provides the `` top '' partition @xmath32 and the `` bottom '' partition is @xmath34 . the schubert varieties @xmath241 are then parametrized by the set @xmath312 of pairs @xmath325 $ ] with @xmath313 , @xmath229 and such that @xmath245 . we also propose likewise in the type c case an alternative way of arrange the diagram of @xmath246 which corresponds just to a `` half '' shifted diagram in the sense presented by tamvakis ( see @xcite , page 19 ) . we consider the top diagram @xmath32 left justified in the @xmath326 rectangle and , for each @xmath247 , the @xmath248-th row of @xmath34 is shifted to the right @xmath248 units . with that shift , the bottom diagram may be seen inside a staircase partition with @xmath327 rows . also denote by @xmath250 and by @xmath251 the set of square boxes with coordinates @xmath252 the top and the bottom ( shifted ) diagrams , respectively . each diagram @xmath250 and @xmath253 is arranged in a plane with matrix - style coordinates . for example , consider the schubert variety @xmath328 and @xmath329 in @xmath330 . notice that @xmath331 which means that it is of type 0 . the diagram and its respective half shifted diagram are illustrated in figure [ fig : schub_d_543_761 ] . in @xmath330 . ] for each row in the top diagram , construct a diagonal line ( indicated as a dashed line in the figure [ fig : schub_related_54x_763d ] ) from the center of the last box at right of the respective row to the center of the first box in the first row of the bottom diagram ( it means that these boxes are @xmath209-related in this half shifted diagram as defined by tamvakis @xcite ) . consider the boxes of the staircase partition that are outside @xmath34 organized into columns . the columns that contain a such @xmath209-related box are called _ related _ and the remaining columns are _ non - related_. the _ length _ of a column is the number of blank boxes of this column in the staircase @xmath35 shape . we may recover the permutation associated to such diagram by taking the length of the related and non - related columns . namely , the permutation element for @xmath332 $ ] is defined by @xmath258 in the equation , where @xmath259 are the length of the related columns , @xmath333 are the length of the non - related columns . for example , the schubert variety @xmath334 of type 0 in @xmath330 corresponds to the permutation @xmath335 as illustrated in figure [ fig : schub_related_54x_763d ] ( left ) . notice that one must fill in the diagonal boxes of the first @xmath336 rows that appear in the shifting procedure . if the diagram is of type 0 , we have a unique diagram since @xmath316 . if the diagram is of type 1 or 2 , since @xmath337 , we will have two possible diagrams : one of them with the diagonal box in the @xmath338-row not filled in and another with such box filled in . by definition , they will correspond to the elements of type 1 and of type 2 , respectively . consider , for example , the schubert varieties given by @xmath339 and @xmath340 $ ] of type 1 and 2 , respectively , inside @xmath330 as illustrated in figure [ fig : schub_related_54x_763d ] . the permutations are , respectively , @xmath341 and @xmath342 . ( left ) , @xmath343 ( center ) and @xmath344 $ ] ( right ) in @xmath330 . ] the diagrams presented here of type @xmath3 are also slight modifications of the model of the so called typed k - strict partitions presented by buck - kresch - tamvakis . for details , see @xcite , section 6 . we notice that the diagrams presented here have the advantage of incorporating the @xmath345 factor that appears in the bkt - model . finally , we hope that the distinction between the type 1 and 2 weyl group elements were clarified in terms of its diagrams . as it was made for groups of type c and b we have a row - reading expression for each @xmath346 , @xmath332 \in \mathcal{p}(k , n)$ ] . we start with the maps @xmath347 that says how the boxes of a diagram are filled in with simple reflections . * type 0 : * with respect to the top partition @xmath32 , let @xmath348 be defined , for @xmath267 and @xmath268 , by @xmath349 with respect to the bottom partition @xmath34 , let @xmath350 be defined , for @xmath275 , @xmath276 and @xmath277 , by @xmath351 * types 1 and 2 : * the case of types @xmath352 and @xmath353 also depends on the parity of the length of @xmath34 . with respect to the top partition @xmath32 , @xmath348 is defined , for @xmath267 and @xmath268 , by @xmath354 with respect to the bottom partition @xmath34 , @xmath355 is defined , for @xmath275 , @xmath276 and @xmath277 , by @xmath356 [ obs : t12 ] observe , in the equation , that for @xmath357 and @xmath358 , i.e. , for the first box at the last row of @xmath34 , the reflection is @xmath359 if @xmath42 has type 1 and @xmath360 if @xmath42 has type 2 . then , we can easily find out the type of @xmath42 if we know which one appears first in the word : @xmath361 or @xmath362 , implying it is of type 1 or 2 , respectively . now , we define the reading row map for all types @xmath363 and @xmath353 in the same way as made for groups of type c : define @xmath364 and @xmath365 by assign the numbers increasingly to the boxes of @xmath250 and @xmath251 , respectively , from right to left starting from the bottom row to the top row . then , we can form the words @xmath366 @xmath272 and @xmath367 . the concatenation of the equations gives @xmath368 . the maps @xmath284 , @xmath285 , @xmath286 and @xmath287 for the schubert variety given by the partition of type 0 @xmath369 of @xmath330 are illustrated in the figure [ fig : schub_rowreading_543_763 ] . in @xmath330 . ] for the schubert variety given by the partition of type 1 @xmath370 and type @xmath353 @xmath371 $ ] of @xmath330 , this maps are described in the figure [ fig : schub_rowreading_544_763_t1_t2 ] . and @xmath371 $ ] in @xmath330 . ] an important data needed to compute homology of groups is the description of the set @xmath372 of positive roots sent to negative roots by @xmath37 . this section is devoted to show the relationship between @xmath372 and the respective half shifted young diagram of @xmath42 . more specifically , since the number of roots in @xmath373 is equal to length @xmath121 , there is a rule that shows how each box in the diagram of @xmath42 is filled in with a root of @xmath373 . our purpose is to describe the @xmath24-sequences of @xmath42 using the position @xmath256 of a box in the diagram of @xmath42 . let @xmath241 be the schubert variety parametrized by @xmath243 with @xmath244 and @xmath229 . for a box @xmath374 of @xmath34 define @xmath375 , where @xmath376 is the @xmath24-sequence for @xmath377 . for a box @xmath378 of @xmath32 define @xmath379 , where @xmath380 is the @xmath24-sequence for @xmath381 . observe that @xmath382 is a permutation obtained by reordering @xmath42 according to the order of indexes given by @xmath383 . this @xmath382 also can be seen as the element of the set @xmath384 of lagrangian grassmannians @xmath385 related to the shifted diagram @xmath34 . [ propbetaseqbc ] the @xmath24-sequences can be described as follows : * the top part will be the same for both types c and b : @xmath386 * the bottom part is different for each case : 1 . * type c : * @xmath387 ; 2 . * type b : * @xmath388 ; + for @xmath374 by definition , @xmath389 is the @xmath24-sequence for @xmath377 and this formula is done in @xcite . to proof for the `` top '' part , it is enough to apply ( * ? ? ? * lemma 3 ) to the definition of @xmath390 . this proposition give us an easier way to compute the set @xmath36 of all @xmath24-sequences by placing such roots into each box of the young diagram . [ ex : roots_554_8741 ] consider @xmath262 in @xmath255 . in this case , @xmath263 and @xmath391 . first of all , we must label the rows and the columns of top diagram using @xmath42 , starting from the bottom row to the top and , afterwards , from the left column to the right . then , the @xmath24-sequence associated to each box @xmath378 is the root @xmath392 , where @xmath393 and @xmath394 are the label of @xmath248-th row and @xmath395-th column , respectively . for example , the box @xmath396 has labels 6 and @xmath397 for row and column , respectively , and it implies that @xmath398 as we can see in the figure [ fig : exemppioftypeb ] . for the bottom diagram , we must label the rows using the reflection of @xmath377 @xmath399 from the bottom row ( the @xmath12-th row at @xmath35 staircase ) to the top and using @xmath382 to label the columns from the left to the right . the @xmath24-sequences are obtained likely to the top part . the figure [ fig : exemppioftypeb ] shows all the @xmath24-sequences of the partition @xmath246 . -sequences associated to the partition @xmath254 . ] let @xmath241 the schubert variety parametrized by pairs @xmath325 $ ] with @xmath313 , @xmath229 . for a box @xmath374 of @xmath34 we define @xmath400 , where @xmath376 is the @xmath24-sequence for @xmath377 . for a box @xmath378 of @xmath32 we define @xmath401 , where @xmath380 is the @xmath24-sequence for @xmath381 . we denote by @xmath402 the permutation associated to @xmath403}$ ] where we remove the hat from the first entry @xmath404 ( observe that @xmath402 might not lie in @xmath405 ) . notice that @xmath382 is a permutation obtained by reordering @xmath42 according to the order of indexes given by @xmath406 . this @xmath382 also can be seen as the element of the set @xmath384 of the grassmannian of @xmath407-dimensional isotropic subspaces @xmath408 related to the shifted diagram @xmath34 . define @xmath409 by reordering @xmath402 according to the order of indexes @xmath406 . [ propbetaseqd ] the @xmath24-sequences can be described as follows : * the top part will be given by : @xmath410 * the bottom part will be given by : @xmath411 considering each type separately , this proof if similar to proposition [ propbetaseqbc ] . consider @xmath412 of type 1 and @xmath413 $ ] of type 2 in @xmath330 . in this case , we have @xmath414 and @xmath415 ; @xmath416 and @xmath417 . repeating the same process as in the example [ ex : roots_554_8741 ] but now using the permutation @xmath402 to label the rows and columns instead of use the original @xmath42 , the figures [ fig : exemppioftypedi ] and [ fig : exemppioftypedii ] show the @xmath24-sequences for the partitions @xmath418 and @xmath419 , respectively . -sequences of the partition @xmath370 of type @xmath352 . ] -sequences of the partition @xmath371 $ ] of type @xmath353 . ] let @xmath0 be either an orthogonal or a symplectic group and let @xmath8 be the parabolic subgroup , where @xmath9 is the complementary simple roots of @xmath420 of the respective dynkin diagram . let @xmath165 be the @xmath14-module freely generated by the schubert varieties @xmath15 , @xmath421 . the coefficients @xmath17 of the boundary map @xmath422 are defined by the equation . the cellular homology of the isotropic grassmannian @xmath210 , odd and even orthogonal grassmannians @xmath290 and @xmath423 is obtained applying the theorem [ teo : mainrs ] which provides a formula to compute the coefficients @xmath17 . this proof will be done in next section . let @xmath15 be a schubert variety ( of any isotropic or orthogonal grassmannian ) and the associated double partition @xmath424 $ ] , where @xmath425 and @xmath426 which @xmath427 . given a schubert variety @xmath39 with partition @xmath40 such that @xmath187 , the partition @xmath41 is called 1 . a _ @xmath32-removing of @xmath42 _ if @xmath43 for some @xmath44 ; 2 . a _ @xmath34-removing of @xmath42 _ if @xmath45 for some @xmath46 . when we look at the diagram , removing a box from a row of the partition @xmath246 means that we are removing the last box of the respective row . in respect to the diagram , there are three different kinds of a @xmath34-removing of @xmath42 : 1 . @xmath34-removing of @xmath42 _ from the diagonal _ ( only happens for type c and b ) ; 2 . @xmath34-removing of @xmath42 _ from a related column _ ; 3 . @xmath34-removing of @xmath42 _ from a non - related column _ ; observe that for any @xmath34-removing for groups of type c and b , by symmetry , we have @xmath428 unfilled boxes in the staircase pattern below the removed box . being this a related column or not , @xmath429 or @xmath430 , is equal to @xmath428 . the same idea holds for type d. this implies the following lemma : [ lemaremovnondiagbox ] suppose that @xmath41 is a @xmath34-removing of @xmath42 and @xmath431 for some @xmath46 . recall the formulas , for type c and b , and , for type d , of @xmath42 in terms of the lengths @xmath432 s and @xmath433 s . * if it is a @xmath34-removing from a related column , then @xmath434 * if it is @xmath34-removing from a non - related column , then @xmath435 [ teo : main ] given @xmath47 in @xmath30 such that @xmath48 , this can be either a @xmath32-removing or a @xmath34-removing of @xmath42 and consider @xmath49 the row where the box is removed . the coefficient @xmath17 in the boundary map of the cellular homology of the respective @xmath11 is given by : @xmath436 where @xmath437 and @xmath50 depends on the group @xmath0 and which box is removed : @xmath438 therefore , we have four different possibilities to remove a box from @xmath246 as listed below . in these computations we only need to know what happens to the top part since in @xcite we can find the proof for the bottom part . remember that the permutation of @xmath42 can be written as @xmath441 and these values are obtained by counting the length of related and non - related columns of the bottom diagrams . suppose that we are removing a box from the row @xmath442 for some @xmath443 . looking through an imaginary diagonal line , the row @xmath442 is related to the @xmath444-th column in the bottom diagram @xmath253 . clearly , the @xmath445-th column must be non - related . after remove the box @xmath450 , the related @xmath444-th and non - related @xmath445-th columns of @xmath42 swicht their role in the permutation and they become , respectively , non - related and related columns of @xmath41 . then , the permutation of @xmath41 is obtained from @xmath42 just by swapping @xmath429 and @xmath430 @xmath451 denote by @xmath452 the @xmath24-sequences associated to @xmath41 . using proposition [ propbetaseqbc ] , we can observe that the @xmath24-sequences in the top diagram are equal to @xmath42 and @xmath41 unless for those box at @xmath453-th row or @xmath442-th column . by definition , @xmath454 consider @xmath262 in @xmath255 and its @xmath24-sequences in figure [ fig : exemppioftypeb ] . suppose that @xmath463 , where @xmath464 . the @xmath24-sequences for @xmath41 can be described in the diagram as same as the ones for @xmath42 . comparing both diagrams , we can see in figure [ fig : exempcase1c ] ( left ) that the @xmath24-sequences filled in the boxes are only different for some in the top part which have a dot . subtracting the diagram of @xmath24-sequence of @xmath41 from the diagram of @xmath42 , we get an integer multiple of the root @xmath465 in the dotted boxes and zero in all other . in the figure [ fig : exempcase1c ] ( right ) , the multiplicity of such root is 1 for the removed box and for all in the same row or same column in the top part . then , @xmath439 can be easily computed just using the diagram by adding all coefficients in it . for this example , @xmath466 . the above process can be applied in general for any @xmath32-removing of @xmath42 : put 1 s into all boxes of the top part in the same column and row of the removed box . adding all this numbers we get exactly the coefficient @xmath468 . although this whole work is devoted to establish the relation between the half - shifted young diagrams and grassmannians for groups of type b , c and d , we can point out that the schubert varieties of the usual grassmannian ( of type a ) are paremetrized by the standard young diagram , a particular case of half - shifted when the bottom part @xmath470 . then , the homology coefficients for the usual grassmannians can be computed considering only the @xmath32-removing case as made above . suppose that we are removing a box from the row @xmath471 for some @xmath472 such that this is in the diagonal of the diagram @xmath253 of @xmath42 . so , we must have @xmath473 and @xmath474 . since the new `` free '' of @xmath41 box is in the first non - related column , the permutations @xmath42 and @xmath41 are @xmath475 by proposition [ propbetaseqbc ] , we can observe that the @xmath24-sequences in the top diagram are equal to @xmath42 and @xmath41 unless for those in the @xmath476-th column . observe that , since @xmath427 then all the boxes in the column @xmath476 must belong to the diagram @xmath250 . then , @xmath477 consider @xmath262 in @xmath255 ( type c ) and suppose that @xmath483 , where @xmath484 . subtracting the diagram of @xmath24-sequence of @xmath41 from @xmath42 , we get an integer multiple of the root @xmath485 into the dotted boxes , which is given in the figure [ fig : exempcase2c ] ( center ) . then , the sum of such multiplicities is @xmath486 . if @xmath246 is in @xmath487 ( type b ) , the integer multiple of the root @xmath488 is given in the figure [ fig : exempcase2c ] ( right ) and @xmath489 . the above process can be applied in general for any @xmath34-removing of a diagonal column of @xmath42 : for groups of type c , put 1 s into all boxes in the same column of the removed box , in both top and bottom parts . adding all this numbers we get exactly the coefficient @xmath490 . in the other hand , for groups of type b , put 1 only into the removed box and 2 s into all boxes above and in the same column of the removed box . adding all this numbers we get exactly the coefficient @xmath491 . suppose that we are removing a box from the row @xmath471 for some @xmath472 such that this belongs to a related column of the diagram of @xmath42 . using lemma [ lemaremovnondiagbox ] , there exist a @xmath492 such that @xmath493 . when the box @xmath494 is removed from the bottom diagram of @xmath42 , it will increase by 1 the length @xmath429 and the permutation @xmath41 is @xmath495 using proposition [ propbetaseqbc ] , we can observe that the @xmath24-sequences in the top diagram are equal to @xmath42 and @xmath41 unless for those box in @xmath496-th row or @xmath453-th column . then , @xmath497 consider @xmath262 in @xmath255 and suppose that @xmath505 , where @xmath464 and @xmath506 . subtracting the diagram of @xmath24-sequence of @xmath41 from @xmath42 , we get an integer multiple of the root @xmath507 into the dotted boxes , which is given in the figure [ fig : exempcase3c ] ( center ) . then , the sum of such multiplicities is @xmath508 . if @xmath246 is in @xmath487 ( type b ) , the integer multiple of the root @xmath507 is given in the figure [ fig : exempcase3c ] ( right ) and @xmath509 . the above process can be applied in general for any @xmath34-removing of a related column of @xmath42 : for type c , fill the diagram in the same way ; for type b , just change the 2 in the diagram by 1 . adding all this numbers we get exactly the coefficient @xmath511 for type c and @xmath512 for type b. using lemma [ lemaremovnondiagbox ] , there exist a @xmath513 such that @xmath514 . removing the box @xmath494 from the bottom diagram of @xmath42 , it will increase by 1 the length of @xmath430 and the permutation @xmath41 is @xmath515 where @xmath516 is in the position @xmath517 in the permutation . now , if @xmath532 , we must analyze the diagram differently . first of all , two any consecutive rows @xmath395 and @xmath533 are related , respectively , to the @xmath534-th and @xmath535-th columns in the bottom part . there are exactly @xmath536 non - related columns between them . so , we can relate each box @xmath537 , for @xmath538 , to a non - related column between @xmath534-th and @xmath535-th columns by drawing a diagonal line from the box @xmath537 to a first box in the @xmath539-th column . the figure [ fig : proptransp ] shows this relation to non - related columns using dashed lines . in the other hand , we can observe that each column @xmath453 of the top part , for @xmath540 , is related , by the above process , to some non - related column in the bottom part , and such column is exactly the @xmath541-th column of @xmath34 . consider @xmath262 in @xmath255 and suppose that @xmath548 , where @xmath549 and @xmath550 . subtracting the diagram of @xmath24-sequence of @xmath41 from @xmath42 , we get an integer multiple of the root @xmath551 into the dotted boxes , which is given in the figure [ fig : exempcase4c ] ( center ) . then , the sum of such multiplicities is @xmath552 . if @xmath246 is in @xmath487 ( type b ) , the integer multiple of the root @xmath551 is given in the figure [ fig : exempcase4c ] ( right ) and @xmath553 . the above process can be applied in general for any @xmath34-removing of a non - related column of @xmath42 : for type c , fill the diagram in the same way ; for type b , just change the 2 in the diagram by 1 . adding all this numbers we get exactly the coefficient @xmath555 for type c and @xmath556 for type b. the boundary operator is obtained once we compute @xmath439 where @xmath440 is the sum of the elements in the @xmath24-sequence of @xmath42 . the approach for the type @xmath3 is almost the same as for the types @xmath1 , @xmath2 . in some cases , the calculations are exactly the same . however , the existence of elements of type @xmath363 and @xmath353 asks for a careful analysis in some specific situations which turn out to be our focus in this section . we use the results obtained by @xcite for the bottom part when needed and compute here the top part . suppose that we are removing a box from the row @xmath442 for some @xmath443 . we first observe that the same pattern over the diagrams verified in the types @xmath558 occur here , namely , this removing transforms a related column and a non - related column , respectively , into a non - related and related column . explicitly , if @xmath429 , @xmath559 , and @xmath430 , @xmath560 , are the length of related column @xmath446 and non - related column @xmath447 , respectively , we have that @xmath561 observe also that @xmath448 and @xmath562 . in terms of the beta - sequences associated to the diagrams , the proposition [ propbetaseqd ] shows that the formula for the roots in the top diagram is the same as for the type @xmath558 given by the proposition [ propbetaseqbc ] . 1 . if @xmath566 then we only have that @xmath563 , otherwise , if @xmath564 and @xmath566 then @xmath567 is minimal such that it is not possible to remove a box in the last row . since @xmath448 , we have that @xmath310 is unchanged which means that the type is preserved . 2 . if @xmath568 then we must have @xmath565 ( both @xmath563 and @xmath564 cases are included ) . but since @xmath562 , @xmath565 implies that @xmath569 , i.e. , the term @xmath570 which characterizes the type of @xmath42 as @xmath352 or @xmath353 is preserved . these cases are characterized by the fact that elements of type @xmath352 and @xmath353 are changed to a type @xmath573 element after removing the box . indeed , if @xmath574 @xmath575 is the partition for @xmath576 , with @xmath572 and @xmath260 , then @xmath577 will be the permutation given by the partition @xmath578 , i.e. , @xmath42 is an element of type @xmath579 and @xmath41 is an element of type @xmath573 . the beta sequences of @xmath24 and @xmath452 of @xmath42 and @xmath41 , respectively , are represented in the corresponding diagram according to the proposition [ propbetaseqd ] . we describe below only the boxes that are changed and contribute to the homology coefficient , namely , the @xmath209-row and the @xmath584-column of the top diagram . at these rows and columns we have the following roots : @xmath585 implying that @xmath586 . for sake of illustration , let us consider the case when @xmath592 and @xmath336 is odd , i.e. , @xmath593 and @xmath594 . there are @xmath476 roots @xmath595 given by the difference of roots at the @xmath476-th row of the top diagram . there are also @xmath476 roots @xmath596 given by the difference of the roots at the @xmath476-column of the bottom diagram . adding them up we have @xmath476 roots @xmath597 which correspond to the same roots given by the difference of roots at the @xmath584-th column of the top diagram . analogous considerations will hold when @xmath598 and @xmath336 is even . first of all , consider @xmath412 in @xmath330 of type 1 and suppose that @xmath600 of type 0 , where @xmath549 . subtracting the diagram of @xmath24-sequence of @xmath601 from @xmath42 , we get : ( i ) for the top part , the sum of the roots is @xmath602 and ( ii ) for the bottom part , the sum of the roots is @xmath603 . this implies that @xmath604 . in the other hand , considering @xmath413 $ ] of type 2 and the same @xmath605 , then ( i ) for the top part , the sum of the roots is @xmath606 and ( ii ) for the bottom part , the sum of the roots is @xmath607 , implying that @xmath608 . even though the above process does not give exactly the root @xmath611 as subtraction of @xmath24-sequences , the coefficient @xmath21 is given in the same way as the @xmath32-removing case for groups of type c and b : put 1 s into all boxes of the top part in the same column or row of the removed box . adding all this numbers we get exactly the coefficient @xmath468 . suppose that we are removing a box from the row @xmath471 for some @xmath472 such that this belongs to a related column of @xmath42 . by the lemma [ lemaremovnondiagbox ] , there is a @xmath612 such that @xmath613 . lets considering two cases : when @xmath614 and when it is not . after remove the unique box at the row @xmath616 , we have @xmath617 , which implies that @xmath41 does not have type 0 . moreover , the simple permutation of the map @xmath285 at the removed box of @xmath42 depends on the parity of @xmath260 : it is @xmath361 if @xmath476 is even and @xmath362 if @xmath476 is odd . then , the simple permutation at the first box of the row @xmath618 is @xmath362 if @xmath476 is even and @xmath361 if @xmath476 is odd , implying that @xmath41 has type 2 and 1 , respectively , by remark [ obs : t12 ] . we can write the permutation of @xmath42 and @xmath41 as @xmath619 since this new `` free '' box is in the first related column and has length 2 . by proposition [ propbetaseqd ] , we can observe that the @xmath24-sequences in the top diagram are equal to @xmath42 and @xmath41 unless for those in the last row and in the @xmath476-th column . therefore , @xmath620 then , @xmath621 . now , the @xmath24-sequence for the bottom diagram change only for the @xmath476-th and @xmath584-th columns . in fact , the permutations are @xmath622 and the @xmath24 sequences are : @xmath623 if @xmath566 then @xmath627 and @xmath41 still has type @xmath573 because remove a box from @xmath628 can not change the number of rows of @xmath629 . since @xmath630 , the permutations @xmath42 and @xmath41 are @xmath631 observe that the hat at the first element wo nt be a problem to compute the @xmath24-sequences using the proposition [ propbetaseqd ] because @xmath627 . now , if @xmath632 or 2 then @xmath41 still has the same type of @xmath42 and the permutations are @xmath633 again , the hat at the first position wo nt affect the @xmath24-sequences because the proposition [ propbetaseqd ] uses , for type 1 and 2 , the permutations @xmath402 and @xmath634 , which are , by definition , the same permutations but without the hat at the first position . consider @xmath334 in @xmath330 of type 0 and suppose that we are removing the element @xmath637 . since the permutation of @xmath42 can be written , by the row - reading , as @xmath638 then @xmath639 and , by remark [ obs : t12 ] , @xmath41 has type 1 and it s partition is @xmath640 , where @xmath549 and @xmath641 . then , we get : ( i ) for the top part , the sum of the roots is @xmath642 and ( ii ) for the bottom part , the sum of the roots is @xmath643 . this implies that @xmath644 . the above process gives exactly the root @xmath645 as a subtraction of @xmath24-sequences , the coefficient @xmath21 is given in the same way as the @xmath34-removing case of a related column for groups of type c and b : fill the diagram in the same way of above example . adding all this numbers we get exactly the coefficient @xmath646 . suppose that we are removing a box from the row @xmath471 for some @xmath472 such that this belongs to a non - related column of @xmath42 . by the lemma [ lemaremovnondiagbox ] , there is a @xmath647 such that @xmath648 . lets considering two cases : when @xmath614 and when it is not . this case means we are removing the unique box at the last row of @xmath34 and @xmath42 must have type 1 or 2 otherwise if @xmath42 has type 0 , the box lies in related column . clearly , @xmath473 and @xmath649 , as @xmath650 . by the row - reading map , if @xmath632 then @xmath651 . in fact , @xmath632 implies that the removed box is labeled by @xmath359 and hence the diagonal of the last row of @xmath41 is labeled by @xmath360 , which means that @xmath651 . for the same reason , if @xmath652 then @xmath653 . the permutation @xmath42 is @xmath654 and when we remove the box @xmath655 , the permutation of @xmath41 becomes @xmath656 by proposition [ propbetaseqd ] , we can observe that the @xmath24-sequences in the top diagram are equal to @xmath42 and @xmath41 unless for those in the @xmath476-th and @xmath584-th columns and both columns must have @xmath209 box each because @xmath657 . then @xmath658 then , @xmath659 . now , the @xmath24-sequence for the bottom diagram change only for the @xmath476-th and @xmath584-th columns . in fact , the permutations are @xmath660 and the @xmath24 sequences are : @xmath661 if @xmath566 then @xmath41 still has type @xmath573 because remove a box from @xmath628 can not change the number of rows of @xmath629 . since @xmath664 , the permutations @xmath42 and @xmath41 are @xmath665 now , if @xmath632 or 2 then @xmath666 and @xmath41 still has the same type of @xmath42 . the permutations are @xmath667 where @xmath581 changes according to the type of @xmath42 . observe that the no matter the type of @xmath42 , it will only change the values in @xmath668-th and @xmath669-th positions . to compute the @xmath24-sequences of the top and bottom parts , we can use the same ideas as in case 4 of type c and the results in @xcite . hence , @xmath670 first of all , consider @xmath412 in @xmath330 of type 1 and suppose that we are removing the element @xmath637 . since the permutation of @xmath674 can be written , by the row - reading , as @xmath675 then @xmath676 and , by remark [ obs : t12 ] , @xmath601 has type 2 and it s partition is @xmath677 $ ] , where @xmath549 and @xmath678 . we get : ( i ) for the top part , @xmath679 and ( ii ) the sum of the roots is @xmath680 . this implies that @xmath681 . in the other hand , considering @xmath413 $ ] of type 2 , the same argument shows that @xmath682 has type 1 and @xmath683 . the above process gives exactly the root @xmath685 as subtraction of @xmath24-sequences , the coefficient @xmath21 is given in the same way as the @xmath34-removing case of a non - related column for groups of type c and b : fill the diagram in the same way of above example . adding all this numbers we get exactly the coefficient @xmath512 .

the first off - lattice monte carlo kinetics model of interstellar dust - grain surface chemistry is presented . the positions of all surface particles are determined explicitly , according to the local potential minima resulting from the pair - wise interactions of contiguous atoms and molecules , rather than by a pre - defined lattice structure . the model is capable of simulating chemical kinetics on any arbitrary dust - grain morphology , as determined by the user - defined positions of each individual dust - grain atom . a simple method is devised for the determination of the most likely diffusion pathways and their associated energy barriers for surface species . the model is applied to a small , idealized dust grain , adopting various gas densities and using a small chemical network . hydrogen and oxygen atoms accrete onto the grain , to produce ho , h , o and ho . the off - lattice method allows the ice structure to evolve freely ; ice mantle porosity is found to be dependent on the gas density , which controls the accretion rate . a gas density of , appropriate to dark interstellar clouds , is found to produce a fairly smooth and non - porous ice mantle . at all densities , h molecules formed on the grains collect within the crevices that divide nodules of ice , and within micropores ( whose extreme inward curvature produces strong local potential minima ) . the larger pores produced in the high - density models are not typically filled with h . direct deposition of water molecules onto the grain indicates that amorphous ices formed in this way may be significantly more porous than interstellar ices that are formed by surface chemistry .

1310.2512

chemistry on dust - grain surfaces is crucial to the chemical evolution and enrichment of the interstellar medium . the simplest and most abundant interstellar molecule , h@xmath0 , has long been known to form exclusively on dust - grain surfaces ( gould & salpeter 1963 ) before rapidly returning to the gas phase , where it participates in ion - molecule reactions that may increase the chemical complexity of the gas phase ( dalgarno & black 1976 ) . however , in both quiescent clouds and in protostellar envelopes , the accretion onto the grains of gas - phase atoms and molecules also leads to the build - up of molecular ice mantles ( e.g. whittet et al . 1989 ; chiar et al . 1995 ; gibb et al . 2000 ; boogert et al . the observed dust - grain ices are composed primarily of water ice ( h@xmath0o ) , as well as other simple hydrides such as methane ( ch@xmath3 ) and ammonia ( nh@xmath4 ) , and other organic species including carbon monoxide ( co ) and carbon dioxide ( co@xmath0 ) . these ices are understood to form primarily at low dust temperatures ( @xmath510 20 k ) , with the hydrides forming mainly by the surface diffusion and repetitive addition of atomic hydrogen to other atoms ( e.g. hasegawa et al . atomic h remains significantly mobile even at low temperatures ( e.g. manic et al . 2001 ) , which allows it to diffuse around the surface of the grain ( or the surface of the ice mantle ) , thermally hopping from one binding site ( i.e. potential well ) to another , until it meets a reactive species and forms a new molecule . solid - phase water , thus formed , may account for the majority of the oxygen budget in many dense interstellar regions . the porosity of both the dust - grain ices and of the grains themselves has recently been suggested as an important influence on grain - surface chemistry . chemical models that treat the formation of h@xmath0 within pores in the grains ( perets & biham 2006 ) and the formation of other molecules in pores within the ices ( taquet et al . 2012 ) have demonstrated significant albeit moderate effects in each case . these models combine rate equation - based treatments of the chemical kinetics with generic assumptions regarding the characteristics of the pores . experiments on ice mixtures ( collings et al . 2004 ; fayolle et al . 2011 ) have also shown that pores in laboratory water ice mixtures may absorb more volatile species , such as co and co@xmath0 , to be released at higher temperatures . amorphous solid water ( asw ) ices formed in the laboratory by deposition of water molecules directly onto a cold substrate show a large degree of microporosity , defined as pores with diameter @xmath620 ( raut et al . 2007 and references therein ) ; however , it is unclear whether ices formed by such means are representative of the structure of interstellar dust - grain ices formed largely by surface - chemical processes over many thousands of years . the large - scale morphology of interstellar dust grains is also likely to be important to the ice chemistry and structure . interplanetary dust grains appear to be irregular in shape , and interstellar dust models imply `` fluffy '' morphologies ( mathis 1996 ) ; interstellar dust coagulation simulations also appear to confirm this view ( ossenkopf 1993 ; ormel et al the chemical implications of such morphologies can not be investigated using current dust - grain chemical models , which do not specify actual grain shapes . various models currently exist for the simulation of interstellar dust - grain surface chemistry , which may be divided into broad two groups . the most common is that which uses generalized reaction rates for surface species , based on representative values for the binding energies and surface - diffusion barriers for each atom or molecule . this chemical system may be simulated using either numerical solver routines to integrate ordinary differential equations ( e.g. hasegawa et al . 1992 ; garrod & herbst 2006 ; garrod 2013 ) , or using monte carlo ( mc ) techniques ( charnley 2001 ; vasyunin et al . 2009 ; vasyunin & herbst 2013 ) . however , in either case , little structural information about the ice can be obtained , because the _ positions _ of atoms and molecules are not explicitly modeled , other than ( in some cases ) by depth . furthermore , the effects of local binding properties of the surface ( caused by its shape or molecular composition ) can not be accounted for by this approach ; the models implicitly assume a perfectly smooth grain with uniform binding . unfortunately , heterogeneous surface effects appear to be important , even in very simple systems such as that of h@xmath0 formation on the simplest non - smooth surface ( cuppen & garrod 2011 ) . the other type of model uses a microscopically - exact kinetic monte carlo approach ( chang , cuppen & herbst 2005 ; cuppen & herbst 2007 ; cuppen et al . 2009 ) , which allows the behavior of each individual surface particle to be simulated . these models begin with a square section of grain surface , with periodic boundary conditions to imitate the closed surface of a dust grain . using a so - called _ on - lattice _ approach , atoms and molecules on the surface are allowed to diffuse between pre - defined lattice positions , with diffusion barriers and binding energies dependent on the pair - wise interations between species in adjacent binding sites . every diffusion , desorption , accretion and reaction event is treated explicitly ; consequently , the method is more computationally demanding than those based on simple rate equations , and requires a simpler network of species and reactions . while the latter method produces valuable structural information , it retains several limitations . the use of an on - lattice approach means that the surface species are not fully free to adapt their positions to local conditions , and the sizes of individual molecules are not taken into account in any way . furthermore , the method does not treat chemistry on a three - dimensional grain surface , and so the effects of large - scale grain morphologies on the surface chemistry and ice structure can not be investigated . an _ off - lattice _ approach requires information concerning the possible diffusion pathways of a surface particle , as well as its final position , as these considerations are no longer guided by the fixed lattice . off - lattice techniques have been used in other fields ( e.g. konwar et al . 2011 , and references therein ) , using pre - calculated , generic diffusion pathways . such information may be obtained through intensive molecular dynamics calculations , and incorporated into kinetic monte carlo models of diffusion on regular surface structures . however , for an amorphous ice surface , composed of multiple different chemical species , the number of calculations and data to be stored using such a method would be extremely large . the long timescales involved in interstellar chemistry would also likely make such intensive approaches prohibitively slow . here , a new off - lattice astrochemical kinetic monte carlo model of interstellar grain - surface chemistry is presented . a simple method is devised for the _ ad hoc _ calculation of the post - hop positions of diffusing particles , under the assumption that thermal hopping occurs through easily - identified surface - potential saddle points . particle positions are determined by local minima in the surface potentials felt by individual particles , as defined by the binding partners of each surface species . in this way , the positions and chemical kinetics of every particle on a dust grain may be traced as atoms accrete from the gas phase and an ice mantle is formed . the model allows chemistry to be simulated on a three - dimensional dust grain of arbitrary size and shape , whose surface is explicitly defined by the positions of its constituent atoms . the morphology and porosity of the resultant ice structure is determined purely by the chemical kinetic processes occurring on the grain / ice surface . results are presented for a preliminary version of the new model , as applied to a range of gas densities , using a simple chemical network and a small , idealized dust grain . future work will consider the parameter space in greater depth . section 2 details the new method ; the results are presented in section 3 . a discussion of the model and its initial results are provided in section 4 . the conclusions of this work are outlined in section 5 . the model presented here is used to trace the microscopic chemical kinetics that occur on an interstellar dust - grain / ice surface as the result of the accretion and diffusion of atoms derived from the gas phase . the model starts with a three - dimensional dust grain ( see fig . 1 ) , whose surface is explicitly defined by the positions of the constituent atoms . the arrival of a particle from the gas phase results in its binding in a local potential minimum , whose position and strength are determined by the combined pair - wise interaction potentials between the particle and the nearest surface atoms . the number of surface binding partners is dependent on the morphology of the surface , including the presence of other accreted surface particles . the strength of the pair - wise interaction potential between specific chemical species is defined explicitly within the model ( see table 1 ) . binding is treated as a `` physisorption '' or van der waals - type interaction . no _ chemical _ bonding to the grain surface itself is considered . in general , a surface particle may undergo thermal desorption into the gas phase , or may diffuse to a binding site ( i.e. surface potential minimum ) adjacent to its current location , via thermal hopping . the interaction potentials felt by the particle determine the rates at which such processes may occur , and thus determine the competition with similar processes for other surface species or with further accretion from the gas phase . details of each of these model elements are presented in the subsections below . in this preliminary model , all atoms and molecules including atoms that constitute the dust grain itself are assigned a uniform radius , with an assumption of spherical symmetry . experimentally - determined charateristic radii for water molecules , which comprise the majority of the dust - grain surface ice in the models presented here , are dependent on the overall ice structure . measured values for the intermolecular distance in amorphous solid water ( asw ) ice can range to as high as 3.3 (angell 2004 ) , while static , non - polarizable water models produce values around 3.16 (vega et al . 2009 ) , with radii centered on the oxygen atom . an intermolecular spacing of @xmath7 is chosen for the current model , which for practical purposes defines the hard - sphere radius of a water molecule to be 1.6 ; this value is applied to all other chemical species . canonical van der waals radii for oxygen and hydrogen atoms are 1.52 and 1.20 , respectively ( bondi 1964 ) . the choice of a water - specific generic value ensures that the sizes of the dust - grain ice mantles produced by the models are appropriate . future models will incorporate van der waals radii from the literature , allowing the ice structure to reflect more accurately the varying sizes of the constituent atoms and molecules . the model adopts the numerical monte carlo ( mc ) approach suggested and proved by gillespie ( 1976 ) , and outlined below ; the general method has been successfully used in the past for astrochemical simulations by several authors ( e.g. charnley 2001 ; chang , cuppen & herbst 2005 , cuppen & herbst 2007 , vasyunin et al . 2009 ) . in common with the work of cuppen , the new model takes a microscopic view of the grain - surface chemistry , but otherwise uses a new methodology for the treatment of the chemistry and the physical positions of atoms and molecules on the grain . gillespie ( 1976 ) treats the time evolution of a chemical kinetic system as a series of consecutive events . the two key quantities to be determined by the probabilistic approach at any time , @xmath8 , are : ( i ) which event occurs next , and ( ii ) how much time elapses between this and the previous event . to determine this information , each possible chemical process , @xmath9 , is assigned a rate , @xmath10 ( in units of s@xmath11 ) . in the new model , such processes include an accretion event , a desorption event , or an individual thermal hop from one binding site to another ; in the case of desorption and diffusion , specific rates are calculated for each individual surface particle . following gillespie , each process is then assigned a probability equal to its own rate divided by the sum of all rates , @xmath12 . a computer - generated random number , @xmath13 , then chooses from these probabilties which process will occur next . another random number , @xmath14 , determines the time elapsed since the last event , equal to @xmath15 . the populations , positions , rates , or any other relevant information for each affected chemical species is then updated and the above steps are repeated until the user - defined end - time of the simulation is reached . in the model , this method is used to determine the overall sequence of thermal hopping , reaction , desorption or accretion of every surface atom / molecule , according to their individual rates . the new model takes explicit account of the positions of particles on the grain surface , without using a pre - determined position lattice . however , it _ does _ require the grain - surface structure to be defined initially . the model will allow chemical simulations on any arbitrary grain size and structure . for this purpose , a simple code has been constructed to produce coordinates for every atom in an approximately spherical dust - grain of radius 5 atoms ; see fig . 1 . the grain may be simplistically considered to be composed of carbon atoms , although the purpose of the generated grain is to provide a simple surface structure upon which the new model may be initially tested . atoms are positioned in a simple cubic structure , and assigned the representative radius of 1.6 ( which varies only slightly from the canonical van der waals radius for carbon atoms of 1.70 ; bondi 1964 ) . the grain has a diameter of approximately 32 . future applications of the generating code will be used to produce alternative / arbitrary grain sizes and morphologies , as well as more specific surface structures , compositions , and atomic sizes . the total rate of accretion of gas - phase material onto the grain is dependent on the cross - section presented by the grain and its surface ice mantle . in simpler models , this rate is typically approximated by assuming a spherical grain , using a mean radius corresponding to either a bare grain ( hasegawa et al . 1992 ) or the combined grain and mantle ( acharyya et al . 2011 ) . however , unlike most past models , the new method explicitly tracks the position of each surface species , and allows the non - uniform build - up of surface structures . thus , the treatment of accretion adopted in this model must account for the _ trajectory _ of the incoming particle as well as the rate . in the new model , accretion from the gas phase is handled in several stages ; firstly , a basic accretion rate is constructed for each gas - phase chemical species , based on the cross - section of the smallest sphere that can contain the entire grain and ice mantle . following hasegawa et al . ( 1992 ) , @xmath16 for the sphere bounding a single dust grain , where @xmath17 and @xmath18 are the thermal velocity and concentration of the accreting gas - phase species , @xmath9 , and @xmath19 , where @xmath20 is the radius of the sphere . ( because the bounding sphere is by definition larger than the combined mantle and dust grain , this basic rate is larger than the effective rate ultimately produced by the model , as explained below ) . an independent accretion rate for each gas - phase species is constructed in this way . at such time that the monte carlo routine selects , according to the rate given by equation ( 1 ) , the accretion of a particular species to be the next process in the sequence , a randomized entry point into the bounding sphere is then generated in spherical coordinates : @xmath21 where @xmath22 and @xmath23 are random numbers between 0 and 1 . equations ( 2 ) and ( 3 ) result in a uniform distribution of points over the surface of the bounding sphere . expressions of the same form as equations ( 2 ) and ( 3 ) , using new random numbers , are then used to determine a trajectory for the incoming particle that is randomized over solid angle . however , only _ inward _ trajectories are considered ( outward trajectories are reversed to point inward ) , thus avoiding the double - counting of cases where a particle enters from an alternative point on the sphere . the randomly - generated trajectory is traced through the interior of the sphere until a grain / ice particle is met by the incoming particle . at this stage , the accretion is judged to be successful , and another routine finds the nearest potential well to the contact point , assuming lennard - jones potentials for all interactions between the accreting particle and the grain / ice particles ( see section 2.3.2 ) . if the trajectory of the incoming particle does not intersect with any atoms / molecules on the grain or surrounding ice mantle , then the accretion was unsuccessful . in such a case , the clock is nevertheless advanced , and the monte carlo selection process continues as usual . the entry of a particle into the bounding sphere is still a valid process , even if no accretion ultimately results . the choice of the radius of the bounding sphere could be made arbitrarily large ( producing very low accretion efficiency for particles entering the bounding sphere ) , with no loss of accuracy to the model results . the purpose of the adoption of the smallest bounding sphere possible is simply to avoid unnecessary and time - consuming calculations . in the results presented , only atomic h and o are allowed to accrete from the gas phase , with an assumed sticking coefficient of unity . fractional abudances of @xmath24 are adopted for each species . the overall gas density is varied in the models , but the fractional abundances are assumed to remain constant over time , and no gas - phase chemistry is treated in the current model . each of these restrictions may be lifted in future work . monte carlo treatments of _ gas - phase _ chemistry need consider only the evolution of , and rates associated with , the _ total _ population of each gas - phase chemical species . however , a microscopically - exact treatment of ice structure requires the positions and behavior of individual grain - surface atoms and molecules to be simulated explicitly . in this model , rates for diffusion and desorption are calculated for each individual grain - surface atom or molecule . diffusion and desorption are treated as thermal processes , with individual rates of the form : @xmath25 where @xmath26 is a specific atom or molecule , @xmath27 is the characteristic vibrational frequency of the species in question , @xmath28 is the dust temperature , and @xmath29 is either the binding energy or diffusion barrier , depending on which process is being considered . in this paper , the terms `` thermal desorption '' and `` evaporation '' are considered synonymous . in practice , the monte carlo procedure initially selects a _ particle _ to undergo a thermal process , using the rate of the fastest thermally - activated process allowed for each surface particle in order to make the selection ( although these rates also compete with the accretion rates described above ) . from here , a random selection is made between desorption and each and every diffusion process that the chosen particle may undergo , weighted according to each individual rate . this two - stage approach is necessary because , for example , the probability that a particle attains sufficient energy to diffuse also contains the small probability that it attains sufficient energy to desorb entirely ; equation ( 4 ) represents the rate at which a particle reaches an energy of @xmath30 _ or higher_. these processes are therefore in competition with one - another and must be treated as such . in the same way , for the case of a particle that has , for example , four available diffusion paths with equal energy barriers , the diffusion rate is not increased four - fold , but split four ways . the atoms that constitute the dust - grain surface itself are not allowed to diffuse or desorb , but do contribute to the binding energies of surface - bound species . the treatments applied to desorption and diffusion processes are described in more detail below . when an atom has just accreted onto the grain surface , it initially resides in a potential well whose position and binding energy are calculated at the time of accretion . ( it should be noted that the position assigned to each particle represents only its mean position ; in reality , all particles would be vibrating in their potential wells ) . in the model , a particle is considered to be bound to a surface if it is bonded to 3 or more other atoms / molecules these could be atoms of the dust - grain itself , or other particles that are themselves bound to the grain surface . two particles , a and b , are considered to be bonded to each other if their center - to - center separation , @xmath31 , is approximately equal to @xmath32 ( specifically , in the range @xmath33 ; closer bonding is explicitly prohibited in the model , in consideration of van der waals repulsion ) . for each individual atom / molecule , the binding energy ( i.e. the energy required to desorb / evaporate into the gas phase ) , @xmath34 , is equal to the sum of the interactions , @xmath35 , with the other particles to which it is bound ( n.b . equation ( 5 ) , below , is _ not _ used for these calculations ) . this value is used to evaluate the evaporation rate given by eq . ( 4 ) , and in the calculation of @xmath27 , following hasegawa et al . any particles that are spatially `` boxed - in '' , under the hard - sphere approximation , are not allowed to evaporate or diffuse . thus , molecules within bulk ice mantles are fixed in place unless the outer - surface molecules are removed . interaction potentials are defined for the pairing of every type of chemical species in the model , as shown in table 1 . potentials between any chemical species and a particle in the dust grain itself are chosen to reproduce approximately the binding energies used in previous , rate - based gas - grain chemical models , under the assumption that binding to the bare grain involves bonding with 4 dust - grain atoms . for example , the binding energy of atomic hydrogen assumed by garrod & herbst ( 2006 ) of 450 k may be compared with a value in this model of 400 k , which would be obtained for bonding to a flat face of the dust grain used in the model ( see figs . 1 & 2 ) . bonding to only three grain atoms , or to as many as 5 , is possible in some positions on the bare grain surface , which would produce a binding energy of 300 k , or 500 k , respectively . for atomic oxygen , the 200 k interaction potential with a dust - grain atom adopted in this model reproduces the commonly - used binding energy of 800 k ( tielens & allamandola , 1987 ) , for most positions on the grain . interaction potentials between particles that are not a part of the grain itself are nevertheless guided by the grain - binding values . in general , the interaction potential of a pairing a b is assigned the lower of the values for a gr and b gr ( where `` gr '' represents a dust - grain atom ) . an exception is permitted for species expected to demonstrate strong hydrogen - bonding properties . thus , the interaction potentials involving h@xmath0o , h@xmath0o@xmath0 , and oh ( when bound to one of those two molecules ) are augmented by several hundred kelvin . for h@xmath0o h@xmath0o and h@xmath0o@xmath0h@xmath0o@xmath0 bonding , the a gr values are doubled . this produces a 4-partner binding energy for water of 4000 k , although in practice the model may yield water - ice that is sufficiently amorphous to produce many surface water molecules that are bound to 5 or 6 other waters . the generic value adopted by garrod & herbst ( 2006 ) was 5700 k. the binding treatment assumed throughout the model may be characterized as a `` hard - sphere , nearest - neighbor '' approach . the interaction potentials assumed in the model are isotropic ; polar molecules may in reality exhibit more directionality , but such considerations are currently outside the scope of this model . because the model uses an `` off - lattice '' approach , the positions of the adjacent potential wells into which a particle could diffuse are not initially specified , and must be calculated as needed from the current physical state of the system , as defined by the positions of all nearby particles . the method set out for this model allows the energy barrier for each possible diffusion pathway of a given surface particle to be calculated at the same time as the total binding energy , prior to any calculation of a final _ post - diffusion _ position associated with it . the putative final position resulting from a thermal hop need therefore only be calculated if that particular hop has actually been selected as the next process . to determine diffusion barriers , the model requires information about the possible _ pathways _ that a diffusing particle may take from the current binding site to another . for the illustrative case of an h atom bound to a flat face of the grain ( see fig . 2 ) , the atom is bound to four surface - binding partners . the atom may also be seen to have four possible diffusion pathways , corresponding to diffusion along the surface _ between pairs of binding partners_. while in theory there is an infinite number of paths that a diffusing atom may take , these four constitute those that pass through a saddle - point in the surface potential : they are , by far , the most probable paths . for each saddle - point pair , a diffusion barrier is defined equal to the total binding energy , _ minus _ the interaction potentials of the diffusing particle with each of the pair of atoms through which the saddle - point would pass . in other words , diffusion requires the breaking of two of the four bonds , while the other two remain intact , even after the hop is finished ( fig . 2 ) . in a different case where , for example , the atom were bound to only 3 binding partners , only 1 bond would need to be broken . this approach is consistent with typical values of the ratio of diffusion barriers to binding energies , estimates of which can range from @xmath50.3 0.8 in current models ( see e.g. garrod & pauly 2011 ) . such values may be achieved through surface binding to different numbers of binding partners ; the minimum of 3 partners ( of equal interaction potential ) would produce a value @xmath36 , while values closer to unity may be attained for binding to highly irregular ( rough ) surfaces . the calculated diffusion barriers are stored and used to determine the selection of a particle and a thermal process as described in sections 2 and 2.3 . if diffusion is selected , the next step is to determine the final position of the particle after diffusion through the selected saddle point . in order to determine this position , the diffusion process is treated as a _ rotation _ around the saddle - point pair ( with which the bonds remain intact following diffusion ) . the particle is rotated at a fixed radius around the axis joining the saddle - point pair , until it is close enough to be considered `` bonded '' to the atoms on the other side of the saddle - point . at this stage , the position of the particle is optimized iteratively , to find the deepest part of the combined interaction potentials of the particles to which it is now bound . each of these interaction potentials is treated as a lennard - jones 612 potential : @xmath37\ ] ] where @xmath38 is the current center - to - center separation , @xmath35 is the depth of the potential , and @xmath39 is the separation at which the potential minimum is reached , which is set to 3.2 for the present model . ( this expression is used wherever distance - dependent pair - wise potentials are required in the model ; for example , the binding positions of newly - accreted particles are also calculated iteratively using this expression . ) the derivatives of the potentials are used to calculate the net force vector resulting from the combined field at the current position . the position of the particle is advanced incrementally according to the force vector , to yield the position of the potential minimum to a tolerance of 0.005 . after the position is optimized in this way , the new binding energy is , again , taken simply as the sum of the strengths of the interaction potentials , @xmath35 , with all the species to which the particle is bound ( i.e. @xmath40 ) ; new diffusion barriers for all _ viable _ saddle - point pairs are also calculated at this time . the viability of saddle - point pairs is defined in the model as those pairs for which a rotation would be unhindered by other particles to which the diffusing species is _ already _ bound . for example , in the case shown in fig . 2 , while there are four pairs that would allow rotation ( and thus diffusion ) , a pairing of _ diagonally - arranged _ particles from this group ( fig . 2c ) is disallowed , because rotation / diffusion around the axis between the diagonal particles would be obstructed in either direction by another binding partner . all rotation - pairings of particles to which the species in question is bound ( i.e. @xmath41 pairings , for @xmath42 bonds ) are initially allowed . each pairing is then assessed according to the following test : a plane is defined by the initial positions of the diffusing particle and each of the particles of the rotation pairing . if the _ other _ particles to which the surface species is bound are either _ all above _ or _ all below _ the plane , then diffusion is allowed ( in a direction _ away _ from those particles ) . otherwise , diffusion is disallowed , as a collision would result ( as for the diagonal pairing shown in fig . 2c ) . typically , only pairings between adjacent ( although not necessarily _ bonded _ ) particles are allowed by this test , although irregular surface morphologies may allow other cases to be permitted . in this way , the possible diffusion paths on an arbitrary surface may be determined . species for which all rotation pairs are forbidden are considered `` boxed in '' , and are thence also prohibited from desorption , until such time as this condition changes due to diffusion or evaporation of other particles . it should be noted that , following accretion , desorption , or diffusion , the diffusion and desorption properties of all affected species such as binding partners both prior to and following a hop are re - calculated , and the rates and diffusion viabilities are re - assessed . the rotation itself should not be considered an accurate description of the _ motion _ of the particle , but rather a convenient method for finding the potential minimum to which the particle hops , via a potential saddle - point . actual diffusion paths for arbitrary binding structures would require intensive molecular dynamics or quantum chemical calculations . in the case given in the example ( fig . 2 ) , the diffusion barrier associated with each of the four possible hops is the same . however , if the particle were bound to different chemical species , differences in the magnitudes of the interaction potentials would lead to differences in the probabilities of diffusion along each path . on an irregular surface , the particle could be bound to 3 or more other particles ; the _ number _ of binding partners as well as their chemical species is therefore important to the energy barrier to diffusion ( i.e. the energy required to break all but two of the bonds ) . the model is also optimized such that previous hops of the currently - active particle are stored in memory , so long as no other particles diffuse , desorb or accrete in the meantime . this decreases run times by several orders of magnitude ; the calculations are often dominated by the diffusion of a single weakly - bound atom ( until it finds a stronger binding site ) . in the case where either a diffusion or accretion event leads to a particle being bonded to a species with which it is allowed to react , reaction is assumed to occur immediately . the reaction product inherits the position of the _ stationary _ reactant , but the other physical characteristics of the new particle are re - calculated . both the so - called `` langmuir - hinshelwood '' ( diffusive ) and `` eley - rideal '' ( prompt ) surface - reaction mechanisms are treated automatically in this model . the current network includes the species h , o , h@xmath0 , o@xmath0 , oh , h@xmath0o and h@xmath0o@xmath0 , commonly referred to as the `` water system '' . no activation energy barrier - mediated reactions are currently considered in the network ; see table 2 . many models include a number of other grain - surface processes not included here , such as photo - desorption and photo - dissociation . the latter is often suggested as an important mechanism for the formation of complex organic molecules in the interstellar medium ( e.g. garrod & herbst 2006 ) . all such processes are expected to be included in future versions of the model , but are omitted here so that the basic approach may be tested . the inclusion in the model of photo - induced desorption based on estimated or measured rates ( berg et al . 2009 ) would be technically trivial . photo - dissociation would be somewhat more complex , as it would require the positioning of two surface products that originated from a single molecule . consideration of the formation of complex organic molecules using the new model is anticipated , but will depend on the adoption of a significantly larger reaction network that includes co and its related products . furthermore , only _ surface _ processes are considered in the current model ; photo - processing of the sub - surface ice - mantle material would involve a considerable increase in technical complexity . figure 1 shows a simulation of grain - surface chemistry using the method described above . the simulation uses a gas density of @xmath43 @xmath2 , and dust and gas temperatures of 10 k. the images show accurately the positions of each simulated particle ; however , the size and appearance of each particle in the image is chosen purely for ease of visual identification . as described above , every atom and molecule in the present model is treated as a uniform sphere of radius 1.6 ; no inference as to the polar orientation of molecules should be drawn . the images are constructed using the free - ware ray - tracing software _ pov - ray_. panel ( b ) shows an instance wherein 2 oxygen atoms and 1 hydrogen atom have been accreted from the gas phase . all are free to diffuse , but the hydrogen atom is significantly more mobile than the oxygen atoms . following several thermal hops , the hydrogen atom meets one of the oxygen atoms , with which it reacts to form oh . the further accretion of an oxygen and a hydrogen atom from the gas phase results in the formation of o@xmath0 and h@xmath0o . a video of this sequence may be found in the online version of the journal . figure 3 shows another simulation with a gas density of @xmath43 @xmath2 , extended for a much longer period ; 1000 water molecules are formed over 350 yr . ( an identical simulation is shown in figs.[f_surface]d and [ f_cross]d , extended to 200,000 water molecules , or 8,471 yr ) . the resultant ice mantle is composed primarily of water molecules , as expected . some molecular oxygen ( o@xmath0 ) and molecular hydrogen ( h@xmath0 ) are also present ; panel ( b ) shows an h@xmath0 molecule a little below the center of the image . to the left of this may be seen both an o@xmath0 molecule and a hydrogen peroxide molecule ( h@xmath0o@xmath0 ) . h@xmath0 and h@xmath0o are in fact formed in approximately equal quantities , but the molecular hydrogen evaporates rapidly unless trapped in a strong binding site , typically one with @xmath444 binding partners . panel ( c ) shows that the h@xmath0 indicated in panel ( b ) has left that binding site . in rate equation - based models , the formation of any significant quantity of h@xmath0o@xmath0 by the direct addition of two oh radicals at low temperature would be unlikely , due to the low diffusion rate of oh . such methods treat only the direct diffusion of each stated reactant . however , because the model presented here considers the explicit positions of all particles at all times , h@xmath0o@xmath0 may instead be formed by the diffusion of an o atom to a position that is in contact with an oh radical , followed by the surface diffusion of an h atom onto the oxygen atom . this results in an immediate two - stage reaction : h + o + oh @xmath45 oh + oh @xmath45 h@xmath0o@xmath0 . the occurrence of this process is dependent only on the relatively fast diffusion of hydrogen and oxygen , and not oh . such effects , which are essentially three - body surface reactions , are not generally taken into account in rate - based models . however , a similar process for the surface reaction of h + o + co was considered by garrod & pauly ( 2011 ) , who found this mechanism to be the primary formation route for co@xmath0 ice at low temperatures , due to the immobility of co and oh . because the new model takes explicit account of physical positions and structure , all such multi - stage processes are included by default . the formation of the ice mantle shown in fig . 3 is not uniform ; panels ( b ) and ( c ) in particular show the formation of a `` hole '' in the ice , corresponding to a region on the grain whose geometry results in weaker surface potential minima . in the center of the hole shown in panel ( c ) is a single dust - grain atom , around which six other dust - grain atoms reside . this arrangement produces six contiguous potential minima that allow bonding to only 3 dust - grain atoms , rather than the typical 4 . moderately mobile species that are momentarily bound in these weak sites rapidly diffuse out again . thus , the geometry of the dust grain alone may be seen to influence the consequent ice - mantle structure . there exist 8 such regions on the grain used in these models , where similar behavior is found . in this particular model , these holes eventually close over , as molecules build up at the edges of the holes , presenting positions of stronger binding within . under other conditions , such holes may progress to form pores within the ice structure ( see section 3.1.1 ) . in order to investigate the formation of structure in the ices , simulations have been run to produce ice mantles several times thicker than the diameter of the underlying grain . four sets of models have been produced , corresponding to gas densities in the range @xmath46 @xmath47 @xmath2 . for each density , the model has been run three times , adopting a different initial random number seed . this is done to demonstrate that the behavior of the models is a general feature and not dependent on the specific random numbers used in each run . all the simulations are run until 200,000 water molecules have been formed on the grain . as before , dust and gas temperatures are held at 10 k and the gas - phase fractional abundances of h and o atoms are fixed at @xmath24 . in addition , a single model has been run in which water molecules are accreted directly from the gas phase , with no accretion of h and o atoms . an arbitrary gas density of @xmath47 @xmath2 is used ; however , for the consideration of ice structure , the precise rate of accretion is unimportant , due to the lack of surface chemistry and the neglible rate of water diffusion caused by the low temperature . figure 4 shows the final state of each density model ( run 1 is shown in each case ) , including pure accretion ( panel a ) . hydrogen molecules ( h@xmath0 ) are highlighted in blue to aid the eye . a number of features are immediately apparent ; firstly , the large- and small - scale structure of the ice is strongly dependent on the gas density . the most irregular structure results from the higher - density simulations , culminating in the accretion - only model in this case , water molecules remain wherever they land as the result of accretion . in the simulations with active chemistry , higher densities result in more rapid hydrogenation of surface oxygen atoms , before they have the opportunity to find alternative , stronger binding sites . the higher - density ice mantles exhibit a highly porous , or `` creviced '' structure . panels ( b ) and ( c ) in particular show a `` cauliflower - like '' structure , wherein the crevices act to isolate large - scale nodules of ice . in the lower density simulations , the nodules become broader and the crevices less deep . the lowest - density model indeed shows a fairly smooth ice structure with no deep crevices or voids . in the accretion - only simulation , the ice appears irregular on all scales . the small - scale porous structure is largely the result of the lack of surface diffusion , such that new species are unable to move even a short distance into the stronger binding sites that are provided by areas of extreme small - scale curvature of the ice surface . however , the large - scale irregularity of the ice is caused by the combination of fully - randomized trajectories with a three - dimensional grain ; the growth of randomly - produced surface irregularities , or `` bulges '' , is amplified by their protrusion into the accretion field , picking up more material from a greater solid angle , while blocking out those trajectories for other regions of the grain / ice surface . such effects are still prevalent in the active - chemistry models panel ( c ) shows an ice mantle whose underlying dust grain is centered in the image ; a large protrusion may be seen on the right . while a moderate degree of surface diffusion may act to close over the small - scale pores or crevices in the ice , atoms of oxygen are required to diffuse a significant distance over the ice surface to counteract the large - scale irregularities . consequently , while smooth on the small scale , the lowest - density simulation ( panel e ) nevertheless shows an overall irregular shape . figure 5 shows cross sections of the same final ice mantles . each cross section passes through the dust - grain center at an arbitrary angle , showing molecules that sit within a `` slice '' @xmath48 deep this allows space for up to two molecules along a line of sight . videos of panels ( a ) , ( b ) and ( e ) are available in the online version of the journal , showing cross sections over 360 degrees . the detail of the porosity within the ice mantles is clearly visible in figure 5 . panel ( a ) shows significant and fairly uniform porosity , with pores or crevices that pass deep within the mantle almost down to the dust grain itself . the highest - density simulation with active chemistry is not as extreme , but also shows very deep pores / crevices in places , and exhibits some closed pores . the degree of porosity falls away at lower gas densities , until , for @xmath49 @xmath2 , there are essentially no open pores and the ice is extremely compact . one of the most striking features of fig . 5 is the arrangement of hydrogen molecules ( h@xmath0 ) . shown in blue in figs . 4 and 5 for ease of identification , h@xmath0 is almost completely segregated from the water molecules , especially in the higher - density simulations . furthermore , the h@xmath0 molecules are arranged in `` veins '' that are similar in appearance to the empty porous structures . as may be seen three - dimensionally in fig . 4 , the h@xmath0 indeed fills in crevices between larger structures . the veins of h@xmath0 are typically no more than around 5 molecules across ( commensurate micro - porous structure ) , while many of the unfilled pores are significantly larger . following its formation on the grains , an h@xmath0 molecule may diffuse around the surface . if it finds a weak binding site , it may evaporate entirely , but if it finds a strong binding site such as may be found in a pore , where multiple binding potentials converge then it may remain bound in place . the addition of further hydrogen molecules could then contain it , and allow a pore to fill up with hydrogen . the presence of larger , unfilled pores in the higher - density simulations suggests that these pores have an insufficient degree of inward curvature to produce binding potentials great enough to retain h@xmath0 for long enough before the deposition of new material can lock it into place . figure [ bar_coord ] shows a bar chart of the number of surface bonds of each molecule ( of any species ) within the ice mantle , for each of the simulations shown in figs . 4 and 5 . the accretion - only model shows a very different distribution from the active - chemistry models . approximately 35% of all molecules in this simulation experience only 3 5 bonds , making them ( typically ) surface molecules . conversely , only around 6% of molecules in even the highest - density active - chemistry simulation are so weakly bound , and the number of species with only 3 bonds is negligible . however , this does not imply that potential minima that would allow only 3 bonds do not exist only that they are unoccupied . pores , and the associated h@xmath0 veins , may begin to form in such positions , with ice structure building up around them . the porosity visible in figs . 4 and 5 is reflected in the bonding distributions of fig . 6 , with the most compact ice ( produced by the lowest - density model ) showing the greatest bias toward 10 12 bonds , which are necessarily bulk ice molecules . such distributions may therefore be useful as a measure of porosity in these simulations . figure 7 shows the fraction of bonds shared with an h@xmath0 molecule , averaged over all h@xmath0 molecules . the highest density model shows the greatest degree of clustering of h@xmath0 molecules . based on a purely statistical comparison of abundances , the expected value would be around 13 14% for this density ( based on values in table 3 ) . for the lowest density model , the statistical expectation would be @xmath58% . in each case , the actual fractional bonding of h@xmath0 to h@xmath0 is close to 3 times higher than the statistical expectation . it should also be borne in mind that the binding potentials used for h@xmath0 ( table 2 ) are the same for all binding partners . the h@xmath0 clustering is thus the result of the structural arrangement of the ice mantle to produce pores or concentrations of strong binding sites in which h@xmath0 molecules may become trapped , rather than a preference for h@xmath0 molecules to bind to each other , per se . figure 8 shows time - dependent grain - surface chemical abundances for each of the stable molecules included in the model , for all four gas - density values . results for all three runs at each density are plotted in the same panel . the end - time of each model corresponds to the time at which 200,000 water molecules have been formed ; the precise value varies according to the gas density and , to some degree , according to random variation between runs at the same density . for o@xmath0 and h@xmath0o@xmath0 , which have generally low abundances , there is some random fluctuation apparent at early times , and there are small systematic differences between runs of the same density . however , as a proportion of the total quantities , these variations fall away as the ice mantles grow . table 3 shows the final values for each model run , as well as the total amount of h@xmath0 formed , much of which evaporates back into the gas phase . h@xmath0o@xmath0 shows the largest variation between same - density runs , of around 7 - 8% , but variations for the other species are typically 1 - 2% . the elevated value for h@xmath0o@xmath0 may be due to its more intricate formation mechanism in this model ( see section 3 ) . for reference , table 3 also indicates end - times of the models , @xmath50 , and the greatest radial distance of any ice - mantle particle from the centroid of the grain , @xmath51 , achieved by the end of each run . abundances of species other than h@xmath0o are seen to be somewhat lower for the lower density models , including h@xmath0 . however , the total amount of h@xmath0 formed ( see table 3 ) is much less affected than the quantity that is stored on the grain at each gas density . accreted atomic hydrogen is typically able to remain on the grain for long enough for another reactive species to accrete and react with it , for all density models . in the case where the product is h@xmath0 , it is more likely to evaporate for the lower density models , which have a more regular surface that provides fewer strong - binding positions or pores . in each of the simulations , the amount of h@xmath0 formed is only a small amount larger than the total water formed . the running of the new model over various density conditions allows an assessment of the computational efficiency to be made . each simulation was run with an _ intel xeon x5482 _ cpu , using a single thread . the run - time for all of the three higher - density simulations ( @xmath52 , @xmath53 , and @xmath47 @xmath2 ) is approximately 90 95 hours , depending on which random - number seed run is considered . only the lowest - density simulations show an apparent density dependence , taking @xmath5180 hours to run . the majority of cpu - time is used up on the calculation of diffusion pathways and on the assignment of position and pairing information associated with each surface hop . because the model keeps diffusion paths in memory , the most intensive calculations are typically done only once per particle , providing the positions of surface potential minima for an individual , mobile particle for large numbers of subsequent , identical hops , prior to reaction , desorption or trapping in a strong binding site . it appears that the larger drain on cpu - time in the low - density case is caused by the relative smoothness of the ice surface . in this case , a diffusing particle ( typically atomic hydrogen ) on the grain has a smaller probability of becoming trapped in a strong binding site ( thus ceasing its diffusion ) , while the probability of another particle being accreted onto the grain , and subsequently reacting with the first particle , is smaller ( due to lower gas density ) . this results in a longer sequence of diffusion before a reaction occurs . it is to be expected that the use of larger grains would ameliorate this effect , as the current model dust grain is extremely small , providing low overall rates of accretion . conversely , higher temperatures would most likely make it worse , due to the resulting faster diffusion , although desorption rates would also increase . the inclusion of a broader network of surface species is expected to make the surface more heterogeneous , such that the smoothness of the surface potential would be reduced , minimizing the possibility that surface species may spend long periods of cpu - time without being trapped in a strong binding site . this model allows porous structures within the ice mantle to form naturally , as the result of chemical and physical conditions . however , the consideration of porosity in simulations of interstellar dust - grain chemistry is not new ; perets & biham ( 2006 ) used a rate - based model to study the effects of porosity within the dust - grain structure itself . the models indicated that porosity could increase the efficiency of h@xmath0 formation over a wider range of temperatures than would otherwise be possible . taquet et al . ( 2012 ) extended the treatment of perets et al . to a full gas - grain chemical model that included the formation of a porous ice mantle , using pre - defined values for the number and sizes of pores . they found only moderate increases in the production of certain key ice - mantle species . cuppen & herbst ( 2007 ) used a monte carlo technique to treat the formation of ice on a surface with periodic boundary conditions , using a fixed lattice for the positions of particles on the grain or within the ice mantle . depending on the physical conditions , ice structures of varying compactness and porosity were produced , including some with tower - like structures of apparently single - molecule thickness . the models that show the most extreme porosity , however , were obtained assuming temperatures rather higher than the 10 k used here . in the present model , only a limited reaction set is considered , which does not include any activation barrier - mediated reactions . nevertheless , the basic reactions that allow the formation of the most important species , such as h@xmath0 and h@xmath0o , are indeed present . the use of an off - lattice approach in this model requires that the binding of atoms or molecules to the surface involve at least 3 other binding partners . thus , the `` sky - scraper '' structures produced by the models of cuppen & herbst are not found here , both because binding to a single partner can not occur and because the direction of the binding is not pre - determined , but depends on the arrangement of the binding partners . perhaps most importantly , the off - lattice approach of the present model allows the effects of the curvature of the surface to be automatically taken into account . inward curvature , such as may be found within the pores , produces stronger binding conditions resulting from the ability to bind with a greater number of partners ; likewise , the outward curvature found on the bare grain , and on the `` nodules '' or `` bulges '' that form on the ice , tend to minimize the number of binding partners . these curvature effects drive mobile particles into the pores as the pores are forming , so that they become filled with h@xmath0 , producing veins . larger pores which are formed in the higher - density models , and which experience less extreme curvature tend not to fill up in this way . the kind of micro - pores investigated by taquet et al . ( 2012 ) , which appear to be comparable in size to the h@xmath0 veins that form here , may not therefore be long - lived enough to affect the chemistry significantly . the present model indicates that such pores would be filled up , although others may develop in their place . in general , however , the low density model ( which perhaps provides the most appropriate comparison ) tends to predict a compact , albeit segregated , ice structure . the gas - phase chemical abundances used in the model presented here represent dense - cloud conditions , under which most hydrogen takes the form of h@xmath0 , resulting in similar h and o gas - phase abundances . however , higher values for h are entirely plausible , which would result in greater h fluxes onto the grains , and therefore a shorter time period before surface oxygen atoms could find alternative binding sites . it is therefore likely that a cloud with large quantities of h not yet converted to h@xmath0 should produce more porous dust - grain ices . it is also probable that , other model quantities being equal , the smallest of these pores would rapidly fill up . the consideration here of a model that involves the direct accretion of water onto the grain allows a basic comparison with laboratory amorphous solid water ( asw ) ices , which are also formed by the deposition of water molecules directly onto a cold surface . it is well known that the angle of deposition has a strong influence on the degree of porosity in the ice ( e.g. kimmel et al . 2001 ; raut et al . the present model does not investigate the effect of deposition angle directly , but uses a completely randomized field of accreting particles . the result is an extremely porous ice , which is qualitatively different from the forms produced by active grain - surface chemistry , because of the minimal diffusion of the water molecules following accretion . in the active - chemistry models , the slow formation of the ice from its constituent atoms allows voids and strong binding sites to be filled more effectively , so that the lowest - density simulations show an absence of pores , and a relatively small quantity of h@xmath0 veins as compared with higher gas - density ( i.e. faster - accretion ) models . it therefore seems appropriate to suggest that interstellar ice analogs formed in the laboratory may be significantly more porous than actual interstellar ices . this would imply that effects observed in laboratory ices , relating to the absorption of other molecules into pores and their subsequent release at higher temperatures ( e.g. collings et al . 2004 ) , may be of less importance in interstellar ice mantles . in mitigation , the present model does not include any mechanism for super - thermal surface diffusion of the accreted particle , or indeed of newly - formed surface molecules . the acceleration of the accreting particle into a surface potential well , or the release of chemical energy from the surface formation of a molecule , could provide sufficient energy for diffusion . this could allow an otherwise immobile particle to find a nearby binding site with stronger binding properties . however , in view of the experimental evidence for significant porosities in laboratory asw ice ( e.g. westley et al . 1998 , and references therein ) , the importance of this effect may be small . as shown in table 3 , for decreasing density the h@xmath0:h@xmath0o production ratio approaches 1:1 . this may be explained through a simple analysis : taking the 1:1 gas - phase abundance ratio for h : o used in the present model , the ratio of accretion rates is 4:1 , due to the 16 times higher mass of oxygen , as entered into eq . ( 1 ) . under conditions where atomic - hydrogen mobility dominates all surface processes , and accretion dominates evaporation , this rate should indeed produce a 1:1 formation rate for h@xmath0:h@xmath0o . ( for every 1 oxygen that accretes , 4 h - atoms will accrete ; 2 are used up , as they rapidly find the o / oh on the surface , to form water . the other two will on average meet each other and form h@xmath0 . ) this analysis ignores o@xmath0 and h@xmath0o@xmath0 formation , each of which would otherwise remove 4 and 2 more hydrogen atoms , respectively , if their oxygen atoms had instead formed surface water . this difference accounts for around 40% of the h@xmath0 excess over h@xmath0o in the low - density models . it may be seen , therefore , that for the lowest density simulations in particular , the production of h@xmath0 is very close to what may be called a `` smooth - grain '' limit , in which hydrogen - atom mobility is not significantly affected by the presence of strong binding sites . with increasing density , there are more strong binding sites ( due to the more irregular ice structure ) , while the waiting period for the accretion of a new particle is also shorter . these effects heighten the probability that an accreted hydrogen atom will wait for another h to react with it , rather than to diffuse to find a free oxygen atom elsewhere on the dust / ice surface , thus producing a greater proportion of molecular hydrogen . for gas densities less than @xmath5@xmath46 @xmath2 , one should expect that the ice surface would be similarly smooth , and perhaps yet more spherical , while the waiting period for accretion would be greater , all of which should produce h@xmath0:h@xmath0o ratios even closer to unity . under such circumstances , the use of rate - based models that employ generalized binding and diffusion characteristics should not produce strongly divergent results for h@xmath0 production from those obtained here , providing appropriate characteristic values were chosen . such an approach would also require that the stochastic behavior of the chemistry be treated adequately ( e.g. the use of the modified rate equations of garrod 2008 and garrod et al . 2009 ) , as the scenario described clearly falls into the case where standard rate equations should fail . it is unclear , however , whether such an approach could accurately reproduce the quantity of the resultant h@xmath0 that is retained on the grain surface , which is still determined by heterogeneous structural considerations . the choice of physical conditions investigated in the present study has been limited to the gas density , which controls the rate of accretion onto the grain . the gas density demonstrates a clear effect on the resultant ice structure , and thence on the retention of h@xmath0 molecules on the grain . the relative formation rates of o@xmath0 and h@xmath0o@xmath0 are also seen to be affected by the density variation , due to the competition between chemical reactions whose rates are dependent on surface diffusion . each of the effects mentioned above concern the interplay between accretion , diffusion and desorption . the gas density is thus not the only parameter which may affect the chemistry and structure of the ice formed on the grain . future work will address the importance of dust temperature ; however , one may predict that temperatures higher than the 10 k used here would produce more diffusion of oxygen atoms , meaning more compact ices . likewise , lower temperatures would result in more porosity , and greater retention of h@xmath0 within those structures . the present , preliminary model uses a very small grain , of radius 16 , as compared to a canonical value of 0.1 @xmath54 m ( = 1000 ) . however , the final ice - mantle radii are significantly larger , at around 150 190 . future models will investigate chemistry on larger grains . the application of the model to a porous dust grain , of explicitly defined structure , would also be quite possible . the model presented in this paper is the first off - lattice monte carlo kinetic model of interstellar dust - grain chemistry . the model allows the full three - dimensional simulation of chemical kinetics and ice structure on a grain surface defined by the positions of its constituent atoms . the use of an off - lattice technique allows the precise positions , binding strengths and diffusion probabilities of grain - surface particles to be determined according to the interaction potentials with binding partners ; particles are not fixed in a pre - determined lattice structure . the model also allows arbitrary morphologies and structures to be used for the underlying dust grain . this includes not only the local surface roughness of a particle , but the degree of grain porosity , as well as the choice of a spherical , spheroidal , irregular , or any other grain shape . the model opens up a large new parameter space that will be investigated in future , in conjunction with a more extensive chemical network . 1 . the shape of the ice mantle is irregular , even for low gas - density models . the porosity of the ice is strongly dependent on the gas density , with higher densities producing greater porosity . 3 . the small - scale curvature of local regions of the ice / grain surface has a strong effect on the chemistry and structure of the ice formed on top . such considerations can only be treated using an off - lattice approach . 4 . inward curvature on small scales as found with micropores allows binding with a greater number of binding partners than on a relatively flat surface . this increases desorption and diffusion barriers , encouraging the trapping of mobile species in micropores and strong surface binding sites . micropores are formed in the ices , but are rapidly filled with h@xmath0 molecules , forming veins of h@xmath0 . thus , h@xmath0 shows significant segregation from water molecules in all models . larger pores remain unfilled , as they do not provide sufficient small - scale curvature of the surface to significantly increase binding potentials . 7 . for gas densities appropriate to dark interstellar clouds ( @xmath46 @xmath2 ) , the ice mantle is fairly smooth and non - porous , albeit with veins of h@xmath0 throughout . . direct deposition of water molecules , such as is used to produce laboratory amorphous water ice , results in far greater porosity than is achieved when the ice is formed by surface reactions between accreted atoms . , then another hydrogen atom has landed , to react with oh , forming h@xmath0o . _ a video of this sequence is availble in the online version of the journal , showing one thermal hop per frame._,title="fig:",scaledwidth=49.0% ] , then another hydrogen atom has landed , to react with oh , forming h@xmath0o . _ a video of this sequence is availble in the online version of the journal , showing one thermal hop per frame._,title="fig:",scaledwidth=49.0% ] , then another hydrogen atom has landed , to react with oh , forming h@xmath0o . _ a video of this sequence is availble in the online version of the journal , showing one thermal hop per frame._,title="fig:",scaledwidth=49.0% ] , then another hydrogen atom has landed , to react with oh , forming h@xmath0o . _ a video of this sequence is availble in the online version of the journal , showing one thermal hop per frame._,title="fig:",scaledwidth=49.0% ] @xmath2 . the `` hole '' in the ice structure is caused by the comparatively weak surface potential of the underlying grain surface in that region , resulting from its geometry . _ a video of this simulation is availble in the online version of the journal , showing the formation of 1 water molecule per frame._,title="fig:",scaledwidth=32.0% ] @xmath2 . the `` hole '' in the ice structure is caused by the comparatively weak surface potential of the underlying grain surface in that region , resulting from its geometry . _ a video of this simulation is availble in the online version of the journal , showing the formation of 1 water molecule per frame._,title="fig:",scaledwidth=32.0% ] @xmath2 . the `` hole '' in the ice structure is caused by the comparatively weak surface potential of the underlying grain surface in that region , resulting from its geometry . _ a video of this simulation is availble in the online version of the journal , showing the formation of 1 water molecule per frame._,title="fig:",scaledwidth=32.0% ] lrrrrrrrr h & 100 & 100 & 50 & 100 & 100 & 100 & 100 & 100 + h@xmath0 & 50 & 50 & 50 & 50 & 50 & 50 & 50 & 50 + o & 200 & 100 & 50 & 200 & 200 & 200 & 200 & 200 + o@xmath0 & 300 & 100 & 50 & 200 & 300 & 300 & 300 & 300 + oh & 400 & 100 & 50 & 200 & 300 & 400 & 500 & 600 + h@xmath0o & 500 & 100 & 50 & 200 & 300 & 500 & 1000 & 1000 + h@xmath0o@xmath0 & 600 & 100 & 50 & 200 & 300 & 600 & 1000 & 1200 + rcccccccccccc @xmath46 & 1 & & @xmath55 & 147.1 & & 3 & 1 & 3 & 1,238 & 240 & 18,284 & 207,313 + & 2 & & @xmath56 & 152.6 & & 6 & 3 & 3 & 1,241 & 261 & 18,399 & 206,096 + & 3 & & @xmath57 & 149.1 & & 1 & 4 & 0 & 1,260 & 257 & 18,407 & 208,248 + + @xmath52 & 1 & & @xmath58 & 153.9 & & 6 & 7 & 8 & 1,599 & 304 & 23,011 & 209,973 + & 2 & & @xmath59 & 148.3 & & 9 & 4 & 5 & 1,547 & 301 & 22,998 & 206,612 + & 3 & & @xmath60 & 160.0 & & 7 & 6 & 8 & 1,523 & 325 & 23,494 & 208,044 + + @xmath53 & 1 & & @xmath61 & 164.9 & & 37 & 30 & 19 & 3,073 & 620 & 28,872 & 217,227 + & 2 & & @xmath62 & 176.0 & & 27 & 23 & 19 & 2,974 & 618 & 29,216 & 217,463 + & 3 & & @xmath63 & 178.3 & & 27 & 20 & 18 & 3,031 & 596 & 29,620 & 216,183 + + @xmath47 & 1 & & @xmath64 & 191.9 & & 72 & 42 & 52 & 5,308 & 1,321 & 32,824 & 225,258 + & 2 & & @xmath65 & 177.3 & & 60 & 53 & 45 & 5,385 & 1,233 & 31,484 & 228,142 + & 3 & & @xmath66 & 183.6 & & 81 & 42 & 42 & 5,381 & 1,281 & 33,319 & 227,244 + gibb , e. l. , whittet , d. c. b. , schutte , w. a. , boogert , a. c. a. , chiar , j. e. , ehrenfreund , p. , gerakines , p. a. , keane , j. v. , tielens , a. g. g. m. , van dishoeck , e. f. , & kerkhof , o. 2000 , , 536 , 347

the reaction has been investigated over the center - of - momentum energy , , range from threshold up to 2.2 gev in a tree - level effective lagrangian model that incorporates most of the well - established baryon resonances with spins equal to or below . four less well - established nucleon resonances of higher mass are also included . the fitted parameters consist , for each resonance included , of the products of the coupling strengths at the electromagnetic and strong interaction vertices and , for the less - established nucleon resonances , the total decay width . for the well - established nucleon resonances , the energy and momentum dependence of the widths is treated within a dynamical model that is normalized to give the empirical decay branching ratios on the resonance mass shells . for the less - established resonances , the total decay width is treated as a single parameter independent of the reaction kinematics . the model is used to fit recent data for the unpolarized differential cross section ( clas ) , the induced hyperon polarization asymmetry , ( clas , graal , and saphir ) , the beam spin asymmetry , ( leps ) , and the double polarization observables and ( clas ) . two different fits were obtained : one that incorporates su(3 ) symmetry constraints on the born contributions to the reaction amplitude and one in which these constraints are relaxed . explicit numerical results are given only for the first fit since the two fits gave nearly identical results for the observables and the per degree of freedom obtained with the second fit was only marginally better than that of the first fit ( better ) . results are presented for the fitted observables at several different energies and center - of - momentum ( c.m . ) frame kaon angles .

0809.3805

interest in the electromagnetic production of strangeness from few nucleon targets , such as the proton and the deuteron , dates back to the 1960 s @xcite , but it is comparatively recently that high quality data , suitable for quantitatively testing theoretical models , has become available @xcite . the strangeness degree of freedom imparts to these reactions the potential to provide fundamental information concerning both the strong and electromagnetic interactions beyond that obtainable from reactions involving just the non - strange baryons . within an effective lagrangian model , one could use these reactions to search for baryon resonances that decay to strange particles and possibly test su(3 ) symmetry relations among the couplings of resonances within the same su(3 ) multiplets . by comparing photoproduction and electroproduction results , one might be able to extract information concerning the electromagnetic form factors of baryon resonances . finally , with the aid of a quantitative model for electromagnetic strangeness production from the proton , one could , within the impulse approximation , use results for strangeness production from the deuteron and other light nuclei to study final state interactions involving the @xmath9 and @xmath4 baryons . much of the theoretical work over the past 20 years or so has been based on effective lagrangian models @xcite . recently , there have been several coupled channel analyses @xcite that have revealed the need for resonances that had not been previously included in many of the effective lagrangian models . until recently , the fits and models were largely based on older data , and often combined photoproduction data and electroproduction data to generate the fits . more recent fits have made use of various combinations of recent data from the saphir @xcite , clas @xcite , leps @xcite , and graal @xcite collaborations . in ref . @xcite , it was suggested that the photoproduction data and the electroproduction data should _ not _ be fit together ; rather one should first generate a model for the basic reaction using photoproduction data alone and then use that model , in conjunction with electroproduction data , to obtain information concerning the electromagnetic form factors of the various resonances in the model . this consideration , along with the abundance of new data , particularly , polarization data , has motivated us to develop a new model for the reaction @xmath0 over the energy range from threshold up to a center - of - momentum ( c.m . ) energy of 2.2 gev . although the clas data extends up to 2.6 gev , the lack of @xmath10-channel resonances in our model with masses above 2.2 gev precludes a reliable treatment of the higher energy data . as discussed in sec . [ sec - results ] , the fits involve a subtle interference between @xmath10-channel contributions to the reaction on the one hand and @xmath11 and @xmath12-channel contributions on the other hand , which , to be effective , requires that the @xmath10-channel resonances included in a particular fit have masses spanning the whole energy range of that fit . the model is similar to that described in refs . @xcite , but has been expanded to include spin @xmath2 baryon resonances in both the @xmath10 and @xmath11-channels , in addition to the spin @xmath13 and spin @xmath14 resonances included in the earlier work . the present model also includes several higher energy , less well - established resonances in the @xmath10-channel that were not included in the earlier work . finally , the fits described here are much more elaborate than those described in refs . @xcite in that both single and double polarization data are included in the present fits . the reaction model is described in some detail in sec . [ sec - model ] . two separate fits were generated . the first fit has su(3 ) symmetry constraints imposed on the born terms in all three channels . these constraints , along with inputted empirical values for the baryon magnetic moments , require particular relationships between the various born terms and also , in conjunction with other considerations , provide a range of values for the @xmath15 coupling . this is discussed more fully in sec . [ sec - fitting ] . in the second fit , the @xmath15 coupling was allowed to move outside the su(3 ) symmetry range and assume whatever value yielded the best fit to the data . the second fit obtained is nearly identical to the first fit and yielded a @xmath7 per degree of freedom that differs from that of the first fit by less than @xmath16 . for these reasons , only results derived from the first fit are presented here . we have used all recently published results for the spin observables , @xmath3 @xcite , @xmath4 @xcite , @xmath5 and @xmath6 @xcite in our fits . to avoid difficulties associated with inconsistencies between different data sets @xcite , only the most recent cross section data from the clas collaboration @xcite were used in the fits . further details concerning the fitting procedure are contained in sec . [ sec - fitting ] . the resulting fit , along with several figures illustrating the quality of the fits , are presented and discussed in sec . [ sec - results ] . [ sec - results ] also contains some concluding remarks and a brief discussion of future work . the reaction model incorporates contributions in the @xmath10-channel , @xmath11-channel , and @xmath12-channel . these are illustrated in fig . [ fig - chan ] . the @xmath10-channel contributions include the born term with an intermediate proton and contributions in which an intermediate nucleon resonance is excited . similarly , the @xmath11-channel born contributions with the excitation of an intermediate @xmath9 or @xmath4 baryon are supplemented by contributions involving the excitation of an intermediate hyperon resonance . in the @xmath12-channel , contributions from both @xmath17 and @xmath18 exchange are included , as well as the born contribution involving ground state kaon exchange . in the various channels , the reaction amplitudes have the general forms @xmath19 @xmath20 and @xmath21 where @xmath22 , @xmath23 , and @xmath24 are the intermediate 4-momenta in the various channels , the @xmath25 s are the electromagnetic and strong interaction vertices , and the @xmath26 s are the associated intermediate baryon and meson propagators . in all channels the forms of both the electromagnetic vertices and the strong interaction vertices depend on the spin and the parity of the intermediate hadron that is excited . in the @xmath12-channel , the two vertices are given by @xmath27 and @xmath28 for an intermediate ground state kaon ( the @xmath12-channel born term ) , by @xmath29 and @xmath30 for an intermediate @xmath17 resonance and by @xmath31 and @xmath32 for an intermediate @xmath18 resonance . here , @xmath33 is the photon polarization 4-vector and @xmath34 is a scaling mass , set equal to 1000 mev , that is introduced to make the electromagnetic coupling strengths dimensionless . the corresponding kaon resonance propagators both have the same form : @xmath35 where the label @xmath36 now refers to either of the two resonances . we note that the propagator employed here , in contrast with that used in refs . @xcite , does not contain a width since the intermediate energies in the @xmath12-channel lie well below the thresholds of any possible decay channels . in the @xmath10 and @xmath11-channel born contributions and in those contributions arising from the excitation of intermediate spin @xmath13 resonances , we employ standard expressions for the electromagnetic vertices and use the pseudoscalar form for the strong interaction vertices . for positive parity intermediate baryons , this gives @xmath37 and @xmath38 with @xmath39 where @xmath40 is defined by its relation to the transition magnetic moment , @xmath41 @xmath42 is the mass of the incoming or outgoing baryon ( @xmath43 or @xmath44 ) , and @xmath45 is the mass of the intermediate baryon . the corresponding expressions for negative parity intermediate baryons just have the @xmath46 factor transposed from the strong interaction vertex to the electromagnetic vertex . for intermediate protons , there is an additional term , @xmath47 arising from the proton s charge . for the spin @xmath13 propagator , we employ , in agreement with other authors , a relativistic breit - wigner form , @xmath48 where the width @xmath49 is non - zero only in the @xmath10-channel resonance contributions . in the @xmath10 and @xmath11-channel contributions from intermediate spin @xmath14 baryons , a number of different forms have been employed for the vertices and the propagator . the authors of ref . @xcite introduced a form for the spin @xmath14 propagator in the @xmath10-channel in which the intermediate baryon mass appearing in the numerator and projection operator of the rarita - schwinger propagator was replaced by @xmath50 . this was motivated by the desire to ensure gauge invariance off - shell . however , use of the same prescription in the @xmath11-channel leads to unphysical singularities . moreover , as pointed out in ref . @xcite , the propagator employed in ref . @xcite does not satisfy the differential equation that defines the propagator as a green s function . for these reasons , we employ the standard rarita - schwinger form for the spin @xmath14 propagator and use forms for the corresponding vertices that are similar to those introduced in ref . however , in contrast to the work of ref . @xcite , we make no attempt to include off - shell terms in the vertices . the results reported in ref . @xcite suggest that these terms , as well as off - shell terms in the propagator , have a relatively modest effect on the calculated observables . the resulting vertices for positive parity spin @xmath14 intermediate baryons take the forms @xmath51 and @xmath52 \gamma_5 , \label{phv32}\ ] ] where @xmath53 is the 4-momentum of the incoming or outgoing ground state baryon . as for the spin @xmath13 contributions , the negative parity vertices just have the @xmath46 factor transposed from one vertex to the other vertex . note in these expressions that the strong interaction coupling has been divided by the pion mass , rather than the kaon mass as in ref . this makes it easier to compare our coupling strengths with the corresponding pion couplings for the purpose of testing su(3 ) symmetry relations among the couplings . the rarita - schwinger propagator is obtained by multiplying the spin @xmath13 propagator given by eq . ( [ rprop12 ] ) on the right by the spin @xmath14 projection operator @xmath54 with the exception of ref . @xcite , most of the earlier work on the photoproduction of strangeness does not include contributions from intermediate states with spin @xmath2 , even though there are several well - established baryon resonances with this spin below 2 gev . @xcite and some of the later work do include spin @xmath2 resonances in the @xmath10-channel but not in the @xmath11-channel . to our knowledge , the present work is the first analysis to include resonances with spin greater than @xmath14 in _ both _ the @xmath10 and @xmath11-channels . for the spin @xmath2 vertices , we employ forms similar to those given in ref . @xcite but modified so as to be consistent with the forms adopted here for the spin @xmath14 vertices . again , we do not include any off - shell terms in the spin @xmath2 vertices . for positive parity intermediate resonances , the resulting vertices are given by @xmath55 and @xmath56 \frac{p_{\gamma}^{\nu}}{m_{\pi}}. \label{phv52}\ ] ] as for the other @xmath10 and @xmath11-channel vertices , the corresponding negative parity vertices just have the @xmath46 factor transposed from one vertex to the other vertex . the corresponding propagator is constructed by multiplying the spin @xmath13 propagator on the right by the spin @xmath2 projector operator , @xmath57 [ p52 ] with @xmath58 where @xmath59 the intermediate nucleon resonances excited in the @xmath10-channel generally lie at energies above the thresholds for decay into various decay channels . thus , the propagators employed for the @xmath10-channel resonances need to include widths , and these widths are generally required rather far off the resonance mass shells . most previous studies have ignored the off - shell nature of the resonances and simply used the on - shell values of the widths . in ref . @xcite a model was proposed to dynamically generate widths off - shell by making use of partial width data summarized in the particle data tables @xcite . the full width is first decomposed into a number of different decay channels . in each such channel , the off - shell energy and momentum dependence of the partial width is then treated using an effective lagrangian model with the required coupling strength adjusted to yield the empirical on - shell branching ratio for decay into that channel . two types of decays are considered two - body decays in which both decay products are stable under the strong interaction , and decays in which one of the decay products is itself unstable , so that ultimately more than two decay products result . decays of the first type all involve the decay of a nucleon resonance into a pseudoscalar meson and a spin @xmath13 ground state baryon . in the resonance rest frame , the corresponding widths are given by the expressions @xmath60 , \label{gam12p}\ ] ] @xmath61 , \label{gam12m}\ ] ] @xmath62 , \label{gam32p}\ ] ] @xmath63 , \label{gam32m}\ ] ] @xmath64 and @xmath65 for the pseudoscalar meson decays , where @xmath66 is the channel momentum , and @xmath42 and @xmath67 are the mass and energy of the baryon decay product . to obtain the total contribution to the width from two - body channels at a particular energy , the partial widths are summed over all two - body decay channels open at that particular energy . any part of the full on - shell decay width not accounted for by the two - body decays discussed above is attributed to decays in which one of the decay products is itself unstable under the strong interaction . these latter decays are approximated as decays into either a ground state baryon and a meson resonance or as a decay into a ground state meson and a baryon resonance . in practice , for the low lying nucleon resonances , only decays into the @xmath68 , the @xmath69 , and the @xmath70 channels are considered . the @xmath68 channel is treated as a decay into a nucleon and a scalar meson of zero width . the corresponding decay widths are the same as those for two - body pseudoscalar decays of resonances with the opposite parity . the two remaining channels both involve a stable ground state hadron and an unstable resonance which itself has a width . to treat such decays , we employ the method developed in ref . @xcite , which involves an integration over the unstable decay product mass of the decay phase space factor multiplied by a breit - wigner distribution function . in particular , for decay into either channel , the partial width is given by the general expression @xmath71 where @xmath72 is the coupling strength , @xmath73 is the decay phase space factor , and the integration limits are defined by @xmath74 in the last expressions , @xmath75 is the threshold value of the center - of - momentum energy for decay into that channel , and @xmath76 is the stable decay product mass ( either @xmath77 or @xmath78 ) . the breit - wigner distribution function has the form @xmath79 where @xmath80 is the unstable decay product width , @xmath81 is the mass of the unstable decay product at the center of its mass distribution , and the parameter @xmath82 is defined by the normalization requirement @xmath83 the vertices for decays of spin @xmath13 resonances into the @xmath70 channel are related to those for decays of spin @xmath14 resonances into the @xmath84 channel by just the interchange of the initial and final baryon states . the vertices for decays of spin @xmath14 and spin @xmath2 resonances into the @xmath70 channel each involve two independent couplings , only one of which can be fixed by the on - shell partial width . to avoid this difficulty , we keep , in each case , only the coupling of lowest order in the channel momentum . with that proviso , the phase space factors for decays into the @xmath70 channel are given by @xmath85 @xmath86 @xmath87 @xmath88 @xmath89 and @xmath90 where @xmath91 is the mass that is integrated over in the mass distribution , @xmath66 is the channel momentum for that value of @xmath91 , and @xmath92 . like the decays of spin @xmath14 and spin @xmath2 resonances into the @xmath70 channel , decays into the @xmath69 channel generally involve vertices with two independent couplings . for the spin @xmath13 resonances decays , we adopt the same procedure as used in ref . @xcite . for the spin @xmath14 and spin @xmath2 resonance decays , we simply drop the couplings of higher order in the momenta and energies . this yields the phase space factors @xmath93 @xmath94 @xmath95 @xmath96 @xmath97 and @xmath98 where @xmath99 and @xmath100 with @xmath101 and @xmath102 . the dynamic width model described above was employed for all of the three and four star status @xmath10-channel resonances used in the fits . for these well - established resonances there are generally enough branching ratio data that reasonably good estimates for the partial widths on the resonance mass shells can be generated . for the higher energy , less well - established resonances , this is generally not the case . hence , for these resonances , we employ energy - independent widths which are treated as parameters to be varied in the fits . the matrix elements for the reaction @xmath0 have the general structure @xmath103u_{m_p}(p_p ) , \label{dme}\ ] ] where @xmath104 and @xmath105 are the 4-momentum and spin projection of the proton , and @xmath106 and @xmath107 the 4-momentum and spin projection of the @xmath9 . the operators @xmath108 , @xmath109 , @xmath110 , and @xmath111 depend upon the spin and parities associated with the particular contributions considered . detailed expressions for these operators are given in the appendix . ( [ dme ] ) can either be evaluated directly or converted to the equivalent pauli form , @xmath112 \chi_{m_p } \label{pme}\ ] ] where @xmath113 and @xmath114 the pauli matrix elements can be evaluated analytically but the procedure is rather tedious . instead , we have evaluated eq . ( [ pme ] ) numerically . as a check on the procedure , an independent code was written to evaluate the dirac matrix elements numerically without recourse to the pauli reduction given by eq . ( [ pme ] ) and the results compared with those of the numerical evaluation of eq . ( [ pme ] ) . table [ tab - resonances ] lists the @xmath10 and @xmath11-channel resonances with three or four star status in the particle data tables @xcite that are included in the present fits . .well - established resonances considered in the model [ cols="^,^,^",options="header " , ] these higher mass @xmath10-channel resonances were included to improve the fits to the data at the higher energy end of the kinematic region considered . aside from the empirical evidence for their existences , as reflected in their particle data table listings , they are predicted by quark models @xcite , and some of them have been included in another recent analysis of photoproduction data @xcite . in accord with our philosophy to incorporate only resonances for which there is independent empirical evidence , we have _ not _ included the so - called `` missing '' @xmath115 resonance at 1900 mev in our final results . although this resonance has been predicted in quark models @xcite and has been included in several other analyses of strangeness photoproduction @xcite , there has so far been little evidence for its existence in any other reactions . we will comment on this further in sec . [ sec - results ] . as discussed previously , width data for the resonances listed in table [ tab - res2 ] are extremely limited or non - existent . for this reason , no attempt has been made to extend our dynamical off - shell width model to these resonances . instead , their widths are treated as energy - independent parameters to be determined in the fits to the data . the complete set of varied parameters thus includes the coupling strength products in all three channels and the total widths of the resonances in table [ tab - res2 ] . the coupling strength products are defined by the relations @xmath116 for the ground state baryons and spin @xmath13 resonances in the @xmath10 and @xmath11-channels , by @xmath117 for the spin @xmath14 and spin @xmath2 resonances in the @xmath10 and @xmath11-channels , by @xmath118 for the ground state kaon in the @xmath12-channel , and by @xmath119 for the @xmath12-channel kaon resonances , where @xmath120 is the dimensionless electric charge . note in eqs . ( [ res12 ] ) , that the @xmath121 , @xmath122 , and @xmath123 subscripts refer to either the corresponding ground state baryon or a spin @xmath13 resonance . for the proton , we also need the charge coupling product . this is given by @xmath124 the various born term coupling products can be related to each other through su(3 ) symmetry relations , su(2 ) isospin coupling coefficients , and the well - established values for the baryon ground state magnetic moments . in particular , @xmath125 , @xmath126 , @xmath127 , and @xmath128 satisfy the simple relationships @xmath129 for the magnetic moment factors in these relations , we employ the values @xcite @xmath130 and @xmath131 ( note the definitions of the @xmath40 s as given by [ phcoup ] and [ mut ] ) . the two strong coupling strengths , @xmath132 and @xmath133 , can each be expressed as a product of an su(3 ) isoscalar factor and an su(2 ) clebsch - gordon coefficient , @xmath134 the su(3 ) isoscalar factors appearing here are related by su(3 ) symmetry @xcite . in particular , their ratio can be expressed in terms of an su(3 ) parameter @xmath135 , @xmath136 which , in turn , can be fixed by other couplings within the same su(3 ) multiplets . using the empirical values for the couplings of pions to the @xmath4 and @xmath9 baryons @xcite yields the value @xmath137 . combining eqs . ( [ strong ] ) and ( [ su3 ] ) with the coupling product definitions , eq . ( [ res12 ] ) , we obtain the ratio @xmath138 for the transition magnetic moment parameter , we use the particle data table value @xcite , @xmath139 , to get @xmath140 . the imposition of these relations reduces the number of born parameters to be varied to just one , which we choose to be the parameter @xmath128 . the parameter @xmath128 is also restricted by su(3 ) symmetry relations and by other considerations . a recent study by general and cotanch @xcite based on a generalized goldberger - treiman relation in conjunction with the dashen - weinstein sum rule , arrived at the pair of constraints , @xmath141 and @xmath142 . the second constraint yields an upper limit for @xmath128 of -1.98 . since the photoproduction data , analyzed within the model presented here , seem to favor a small magnitude for @xmath128 , we use this upper limit in the first of our fits , which has @xmath128 fixed . for the second fit , we allowed @xmath128 to assume whatever value yields the best fit to the data consistent with the born term relations given above . comparison of the two fits enabled us to study the extent to which the quality of the fit is affected by the constraint on @xmath128 . note that the choice @xmath143 yields the value @xmath144 , which is consistent with general and cotanch s constraint on the value of @xmath145 . the fitting procedure required many iterations starting with a fit of the most recent clas data @xcite for the unpolarized differential cross section , given in the c.m . by @xmath146 where @xmath147 is the outgoing 3-momentum in the c.m . and @xmath148 is the squared c.m . the resulting parameter values were then employed as starting values to fit the cross section data , the clas @xcite , saphir @xcite , and graal @xcite data for the hyperon polarization asymmetry @xmath3 , and the clas @xcite data for the double polarization observables @xmath5 and @xmath6 . here , @xmath3 is defined by @xmath149 where the superscripts @xmath150 and @xmath151 refer to spin projections above and below the scattering plane , and @xmath5 and @xmath6 are defined by @xmath152 where now the superscripts @xmath150 and @xmath151 refer to @xmath9 spin projections along and opposite to the @xmath153 or @xmath154 axes , and the incident photon is circularly polarized with positive helicity . at this juncture we generated a prediction for the photon - beam asymmetry , defined by @xmath155 where @xmath156 and @xmath157 refer to polarization vectors perpendicular and parallel to the scattering plane respectively . we obtain good agreement with the graal @xcite data if we interpret their definition of the observable @xmath4 to be the negative of the one defined above . our definition of this parameter is the standard one used in most theoretical analyses . the graal definition is given in terms of vertical and horizontal planes that are not specified relative to the scattering plane , so the possibility exists that there is a sign discrepancy between our definition of @xmath4 and that of the experimentalists . the last step of the fitting procedure was to include both the graal and leps @xcite @xmath4 data in the fit ( with the sign of eq . [ eq - sigma ] reversed ) . in carrying out the fits , we minimized the @xmath7 per degree of freedom defined by the relation @xmath158 where the sum is over all of the individual data points , @xmath159 and @xmath160 are the calculated and experimental values of the observable , and @xmath161 is the squared statistical uncertainty in @xmath160 . the number of degrees of freedom is given by @xmath162 , where @xmath163 is the number of data points and @xmath164 is the number of parameters in the fit . a code based on a modified marquardt prescription was employed in the fitting procedure . as well as the parameters themselves , this code generates the covariance matrix associated with the fit from which well - defined parameter uncertainties can be extracted . the parameters from the fit with all of the born terms constrained ( the first fit described in the previous section ) are presented in table [ tab - results ] . here the coupling constant products obtained for all resonances included in the final fit are listed together with the total widths obtained for the higher mass nucleon resonances included . listed also are the parameter uncertainties obtained from the covariance matrix of the fit . these uncertainties measure the sensitivities of the fit to the corresponding parameters . a small relative value for this uncertainty means that the corresponding parameter is well determined by the fit ; by contrast , a large relative value means that the parameter is poorly determined and is probably strongly correlated with other parameters in the fit . the @xmath7 per degree of freedom associated with this fit is 1.68 . we have not listed the parameters obtained in the second fit in which the parameter @xmath128 was allowed to vary since they are very similar to those obtained in the fit in which @xmath128 was fixed at -1.98 . in fact , with the exception of @xmath128 , the differences in the parameters obtained in the two fits all lie within the parameter uncertainties derived from the covariance matrices . as mentioned previously , the values obtained for the @xmath7 per degree of freedom in the two fits differ by less than @xmath16 , and the results obtained for the observables are nearly indistinguishable . in the second fit , the value obtained for @xmath128 increased to the value @xmath165 with an uncertainty of 0.16nearly within one standard deviation of the upper limit . ccc + @xmath166 & @xmath167 & [email protected] + @xmath169 & @xmath167 & [email protected] + @xmath170 & @xmath167 & [email protected] + @xmath171 & @xmath167 & [email protected] + @xmath172 & @xmath173 & 3.43 @xmath1685.10 + @xmath174 & @xmath173 & [email protected] + + @xmath175 & @xmath176 & [email protected] + & @xmath177 & [email protected] + @xmath178 & @xmath176 & [email protected] + & @xmath177 & [email protected] + @xmath179 & @xmath176 & [email protected] + & @xmath177 & [email protected] + @xmath180 & @xmath176 & [email protected] + & @xmath177 & [email protected] + & @xmath181 & [email protected] + @xmath182 & @xmath176 & [email protected] + & @xmath177 & [email protected] + & @xmath181 & [email protected] + @xmath183 & @xmath184 & [email protected] + & @xmath185 & [email protected] + @xmath186 & @xmath187 & [email protected] + & @xmath188 & [email protected] + @xmath189 & @xmath187 & [email protected] + & @xmath188 & [email protected] + + @xmath190 & @xmath176 & [email protected] + & @xmath177 & [email protected] + @xmath191 & @xmath176 & [email protected] + & @xmath177 & [email protected] + @xmath192 & @xmath176 & [email protected] + & @xmath177 & [email protected] + & @xmath181 & 1133@xmath168490 + @xmath193 & @xmath176 & [email protected] + & @xmath177 & [email protected] + & @xmath181 & [email protected] + @xmath194 & @xmath184 & [email protected] + & @xmath185 & [email protected] + @xmath195 & @xmath184 & [email protected] + & @xmath185 & [email protected] + @xmath196 & @xmath184 & [email protected] + & @xmath185 & [email protected] + @xmath197 & @xmath187 & [email protected] + & @xmath188 & [email protected] + @xmath198 & @xmath187 & [email protected] + & @xmath188 & [email protected] + + @xmath199 & @xmath200 & [email protected] + & @xmath201 & [email protected] + @xmath202 & @xmath200 & [email protected] + & @xmath201 & [email protected] + several observations are in order concerning the contents of table [ tab - results ] . first , it will be noted that the number of hyperon resonances appearing in this table is much smaller than the number appearing in table [ tab - resonances ] . we found that attempts to incorporate all of the hyperon resonances listed in table [ tab - resonances ] resulted in unacceptably large coupling products for many of these resonances accompanied by enormous parameter uncertainties . this clearly indicates that a model that incorporates all of the possible @xmath11-channel resonances is too rich ; i.e. , the photoproduction reaction is not sensitive to particular @xmath11-channel contributions and thus , can not be used to unambiguously determine individual @xmath11-channel coupling products . the reason for this is obvious once one notes that the kinematic variable @xmath11 in the photoproduction reaction is usually negative . this makes the denominators of the intermediate baryon propagators in the @xmath11-channel always large in magnitude and insensitive to the baryon mass . thus , individual contributions to the reaction amplitude in the @xmath11-channel are difficult to distinguish from one another and become highly correlated since many different combinations of the @xmath11-channel coupling products yield the same result in the reaction matrix element . two consequences of this are very large parameter uncertainties and the possibility of a run away effect in the fitting routine in which small increases in @xmath203 result from simultaneous huge increases in correlated couplings . to avoid this difficulty , we systematically removed those @xmath11-channel resonances with the largest parameter uncertainties and then refit the remaining parameters , accepting the result if the resulting @xmath203 did not increase by more than a few percent over the value obtained with all resonances included . this procedure led to reduced values of the remaining @xmath11-channel coupling products and greatly reduced parameter uncertainties , finally culminating in the values listed in table [ tab - results ] . attempts to further reduce the number of @xmath11-channel resonances incorporated in the model led to larger increases in @xmath203 ( @xmath204 or more ) than we deemed acceptable . as the procedure was carried out , the values obtained for the @xmath10 and @xmath12-channel coupling products did not shift significantly , indicating that these coupling products are not sensitive to the model employed for the @xmath11-channel . even with the reduction in the number of @xmath11-channel resonances included in the fit , the parameter uncertainties associated with the hyperon resonances are still rather large , often larger than the magnitudes of the couplings themselves . this indicates that the @xmath11-channel couplings are still highly correlated and rather poorly determined . evidently , the photoproduction reaction is of limited value as a means for studying the couplings of hyperon resonances . the quality of our fits is illustrated in figs . [ cscw]-[sigtheta ] . [ cscw ] shows the @xmath1 dependence of the differential cross section for different values of @xmath205 . over the range in @xmath1 that the data were fit ( up to 2.2 gev ) , our fit well reproduces the features of the data . the minor exception is in the two forward - angle bins at around @xmath206 gev where the data show a peak not indicated by our fit . this could well be an indication of the presence of the @xmath207 included by others @xcite . [ csctheta ] shows the same cross section plotted against @xmath205 for a selection of @xmath1 values . the figure also includes data from leps , which are generally consistent with the clas data . our model actually does a good job of matching the cross section data up to @xmath208 gev for forward angles , a trend seen for the other observables as well . for bins of @xmath205 as indicated . data above 2.2 gev ( red points ) were not included in the fit . the blue curve is from our fit and the data are from ref . @xcite . ] for bins of @xmath1 as indicated . the blue curve is from our fit , with black data points from clas @xcite and light blue points from leps @xcite . the highest two @xmath1 bins are clas data that were not included in the fit . ] the @xmath9 polarization fits are shown as a function of @xmath1 in fig . [ pw ] and as a function of @xmath205 for selected @xmath1 bins in fig . [ ptheta ] . as with the cross section , our fit reproduces the major features of the data quite well . polarization , @xmath3 , vs. @xmath1 for bins of @xmath205 as indicated . data above 2.2 gev ( red points ) were not included in the fit . the blue curve is from our fit with black data points from clas @xcite , magenta points from graal @xcite , and green points from saphir @xcite . ] polarization , @xmath3 , vs. @xmath205 for bins of @xmath1 as indicated . the blue curve is from our fit with black data points from clas @xcite . the magenta points from are from saphir @xcite with a similar value of @xmath1 , which is indicated in the parantheses . ] figs . [ cxw ] and [ cxtheta ] show the fits of the @xmath5 data . over most of the kinematic range of data , we again reproduce the general features of the data with our fit . the exception here is at low @xmath1 ( see especially fig . [ cxtheta ] , panel @xmath209 gev ) , where the data are systematically lower than our fit . , vs. @xmath1 for bins of @xmath205 as indicated . data above 2.2 gev ( red points ) were not included in the fit . the blue curve is from our fit and the data are from clas @xcite . ] vs. @xmath205 for bins of @xmath1 as indicated . the blue curve is from our fit and the data are from clas @xcite . the data in the highest @xmath1 bin were not included in the fit . ] [ czw ] and [ cztheta ] show the fits of the @xmath6 data . in the two most forward angle bins of the @xmath1 distributions , the fit is systematically lower than the data from about 1.9 to 2.1 gev . otherwise , the data are well represented by the fit . , vs. @xmath1 for bins of @xmath205 as indicated . data above 2.2 gev ( red points ) were not included in the fit . the blue curve is from our fit and the data are from clas @xcite . ] vs. @xmath205 for bins of @xmath1 as indicated . the blue curve is from our fit and the data are from clas @xcite . the data in the highest @xmath1 bin were not included in the fit . ] finally , fig . [ sigtheta ] shows the photon - beam asymmetry @xmath4 as a function of @xmath205 . the fit shows excellent agreement with the graal data ( @xmath210 gev ) . however , from @xmath211 gev up to 2.2 gev , disagreement between these leps data and the fit grows with @xmath1 . vs. @xmath205 for bins of @xmath1 as indicated . the blue curve is from our fit and the data are from graal @xcite ( magenta ) and leps @xcite ( light blue ) . ] detailed examination of the matrix elements in the model reveals that the @xmath10-channel matrix elements , the @xmath11-channel matrix elements , and the @xmath12-channel matrix elements , when taken individually , all increase monotonically with energy . this indicates that successful reproduction of the data results from a subtle interference between the matrix elements in different channels . because the mandelstam variables @xmath11 and @xmath12 are generally negative over the energy range of interest here , the individual resonance contributions in the @xmath11 and @xmath12-channel are only weakly dependent on the energy . by contrast the individual @xmath10-channel contributions are strongly energy dependent , especially at energies near the resonance masses . consequently , to successfully fit the data within a given energy range with reasonable parameter values , it is necessary to include nucleon resonances with masses that span the full energy range considered . since only nucleon resonances with masses up to 2.2 gev were included in the fits , attempts to fit the data beyond 2.2 gev resulted in a @xmath203 that increased rapidly with @xmath1 . furthermore , because of the subtle interferences among the three channels that are fine tuned by the fit , fits obtained within a given range of energies can not be extended significantly beyond that range . indeed , one finds that beyond the energy range of the fit , the calculated cross sections increase precipitously with energy , in marked contrast with the data . in summary , we have presented a new fit of the kaon photoproduction data from the clas collaboration using an effective lagrangian model . in this fit the born terms were all fixed at values imposed by su(3 ) symmetry constraints and other considerations . a second fit , in which the value of the born parameter @xmath128 was allowed to vary while requiring the other born parameters to satisfy su(3 ) relations among the various born couplings , did not differ significantly from the first fit . in both fits , data for the unpolarized cross section , the hyperon polarization asymmetry , and two double polarization observables from threshold up to 2.2 gev were included . in general , the new fit yields good representations of both the cross section data and the spin observables . one exception is the forward - angle cross section data around 1.9 gev . this suggests that our fit is perhaps missing one or more resonances in this energy range a subject of future work . further improvements to our model should result from including other polarization observables . the clas collaboration expects to produce multiple spin observables in the near future @xcite including data with a polarized target and a linearly polarized photon beam . such data should provide more constraints and potentially enable inclusion of resonances that are not necessary in our present fit . the results of this work can be employed to study the electroproduction of kaons from the proton , i.e. , the reaction @xmath212 . the latter reaction involves a virtual , rather than a real , photon in the strangeness production interaction and thus , requires electromagnetic form factors at the photon vertices . if one uses the photoproduction fits described here to represent the underlying reaction mechanism in the electroproduction reaction , then one can use electroproduction data to study the electromagnetic form factors associated with the intermediate hadrons in the different reaction channels . of course , the electroproduction reaction involves longitudinal , as well as transverse , photons , but the corresponding contributions to the reaction amplitude are related through the lorentz structure of the interaction vertices , i.e. , through the fact that the photon polarization vector is a lorentz 4-vector . thus , it should be possible to extract important information concerning the electromagnetic form factors of hadronic resonances using the fits described here in conjunction with electroproduction data . work in this direction is currently in progress . the operators @xmath108 , @xmath109 , @xmath110 , and @xmath111 appearing in eqs . ( [ dme ] ) and ( [ pme ] ) depend on the spin and parity of the particular intermediate hadron considered . they can all be expressed in terms of a set of @xmath4 and @xmath213 operators defined by the relations @xmath214 @xmath215 and @xmath216 where @xmath217 and @xmath218 are the time and space components of the 4-vector @xmath219 . in terms of these operators , the operators for intermediate baryons with positive parity and spin @xmath13 are @xmath220 in the @xmath10-channel and @xmath221 in the @xmath11-channel , where @xmath222 and @xmath33 are the photon 4-momentum and polarization , @xmath223 and @xmath66 are the mass and 4-momentum of the intermediate baryon , and @xmath26 is the propagator denominator defined by @xmath224 the coupling products @xmath225 are defined by eqs . ( [ res12 ] ) . note that the intermediate baryon width @xmath181 in eq . ( [ bigd ] ) is zero in the born terms . for an intermediate proton there are additional contributions to the operators from the charge coupling . these are given by @xmath226 for contributions with intermediate spin @xmath14 resonances , we define the coupling parameters @xmath227 with @xmath228 where @xmath42 is the mass of the ground state baryon at the photon vertex , and @xmath229 and @xmath230 are the couplings defined by eqs . ( [ res32 ] ) . with these definitions , the operators for intermediate resonances of positive parity and spin @xmath14 are given by @xmath231 , \nonumber \\ \hat{b}_s^{\frac{3}{2}^+ } & = & \frac{1}{3}[\beta_1\sigma(p_k , k_1 ) + 2 f_1(p_k\cdot p)\sigma(p_{\gamma},\epsilon)-3\sigma(p , q_1 ) -2 f_1\sigma_4(p , p_k , p_{\gamma},\epsilon)-3 f_2(p_k\cdot k_1 ) ] , \nonumber \\ \hat{c}_s^{\frac{3}{2}^+ } & = & \frac{1}{3m}[\beta_1\omega_3(p , p_k , k_1 ) + 2 f_1(p_k\cdot p)\omega_3(p , p_{\gamma},\epsilon ) + 3 f_2(p_k\cdot k_1)\sigma\cdot{\bf p } \nonumber \\ & & + 3m^2\sigma\cdot{\bf q}_1 - 2m^2 f_1\omega_3(p_k , p_{\gamma},\epsilon ) ] , \nonumber \\ \hat{d}_s^{\frac{3}{2}^+ } & = & \frac{1}{3m}[\beta_1\sigma_3(p , p_k , k_1 ) + 2 f_1(p_k\cdot p)\sigma_3(p , p_{\gamma},\epsilon ) -3 f_2(p_k\cdot k_1)e \nonumber \\ & & -3m^2 q_1 ^ 0 - 2m^2 f_1\sigma_3(p_k , p_{\gamma},\epsilon ) ] \label{sop32}\end{aligned}\ ] ] in the @xmath10-channel and @xmath232 , \nonumber \\ \hat{b}_u^{\frac{3}{2}^+ } & = & \frac{1}{3}[-\beta_3\sigma(k_1,p_k ) -2 f_1(p_k\cdot p)\sigma(\epsilon , p_{\gamma})+3\sigma(p , q_1 ) + 2 f_1\sigma_4(p,\epsilon , p_{\gamma},p_k)-3\beta_2(p_k\cdot k_1 ) ] , \nonumber \\ \hat{c}_u^{\frac{3}{2}^+ } & = & \frac{1}{3m}[-\beta_1\omega_3(p , k_1,p_k ) -2 f_1(p_k\cdot p)\omega_3(p,\epsilon , p_{\gamma } ) -3 f_2(p_k\cdot k_1)\sigma\cdot{\bf p } \nonumber \\ & & -3 m^2\sigma\cdot{\bf q}_2 + 2m^2 f_1\omega_3(\epsilon , p_{\gamma},p_k ) ] , \nonumber \\ \hat{d}_u^{\frac{3}{2}^+ } & = & \frac{1}{3m}[-\beta_1\sigma_3(p , k_1,p_k ) -2 f_1(p_k\cdot p)\sigma_3(p,\epsilon , p_{\gamma})+3 f_2(p_k\cdot k_1)e \nonumber \\ & & + 3 m^2 q_2 ^ 0 + 2m^2 f_1\sigma_3(\epsilon , p_{\gamma},p_k ) ] \label{uop32}\end{aligned}\ ] ] in the @xmath11-channel , where @xmath233 is the energy of the intermediate resonance , @xmath234 is the kaon 4-momentum , @xmath235 and @xmath236 for contributions with intermediate spin @xmath2 resonances , we define the coupling parameters @xmath237 where @xmath229 and @xmath230 are the coupling products given by eqs . ( [ res32 ] ) , and the linear combinations @xmath238 where @xmath239 with @xmath240 and @xmath241 four other useful combinations are @xmath242 with @xmath243 in terms of these quantities , we have for positive parity spin @xmath2 resonances @xmath244+\frac{1}{5}f_2[c_3\omega(q , p)-\omega(q,\xi_2 ) ] \nonumber \\ & & + \frac{1}{5}f_1[\frac{1}{m^2}\omega_4(p , q , p,\xi_1 ) + 2q\cdot p_{\gamma}\omega_4(p , q , p_{\gamma},\epsilon ) -\omega_4(p , q,\zeta , p_{\gamma } ) ] , \nonumber \\ \hat{b}_s^{\frac{5}{2}^+ } & = & f_2c_4+f_1[c_1\sigma(p , p_{\gamma } ) -c_2\sigma(p,\epsilon)]+\frac{1}{5}f_2[c_3\sigma(q , p)-\sigma(q,\xi_2 ) ] \nonumber \\ & & + \frac{1}{5}f_1[\frac{1}{m^2}\sigma_4(p , q , p,\xi_1 ) + 2q\cdot p_{\gamma}\sigma_4(p , q , p_{\gamma},\epsilon ) -\sigma_4(p , q,\zeta , p_{\gamma } ) ] , \nonumber \\ \hat{c}_s^{\frac{5}{2}^+ } & = & f_2\frac{c_4}{m}\sigma\cdot p + f_1m[c_1\sigma\cdot p_{\gamma}-c_2\sigma\cdot\epsilon]-\frac{1}{5m}f_2 [ c_3\omega_3(p , q , p)-\omega_3(p , q,\xi_2 ) ] \nonumber \\ & & -\frac{1}{5}f_1[\frac{1}{m}\omega_3(q , p,\xi_1)+2q\cdot p_{\gamma}m \omega_3(q , p_{\gamma},\epsilon)-m\omega_3(q,\zeta , p_{\gamma } ) ] , \nonumber \\ \hat{d}_s^{\frac{5}{2}^+ } & = & -f_2c_4\frac{e}{m}-f_1c_1me_{\gamma } -\frac{1}{5m}f_2[c_3\sigma_3(p , q , p)-\sigma_3(p , q,\xi_2 ) ] \nonumber \\ & & -\frac{1}{5}f_1[\frac{1}{m}\sigma_3(q , p,\xi_1)+2q\cdot p_{\gamma}m \sigma_3(q , p_{\gamma},\epsilon)-m\sigma_3(q,\zeta , p_{\gamma } ) ] \label{sop52}\end{aligned}\ ] ] in the @xmath10-channel and @xmath245+\frac{1}{5}f_2[c_3\omega(p , q)-\omega(\xi_2,q ) ] \nonumber \\ & & + \frac{1}{5}f_1[\frac{1}{m^2}\omega_4(\xi_1,p , p , q ) + 2q\cdot p_{\gamma}\omega_4(\epsilon , p , p_{\gamma},q ) -\omega_4(p_{\gamma},p,\zeta , q ) ] , \nonumber \\ \hat{b}_u^{\frac{5}{2}^+ } & = & f_2c_4+f_1[c_1\sigma(p_{\gamma},p ) -c_2\sigma(\epsilon , p)]+\frac{1}{5}f_2[c_3\sigma(p , q)-\sigma(\xi_2,q ) ] \nonumber \\ & & + \frac{1}{5}f_1[\frac{1}{m^2}\sigma_4(\xi_1,p , p , q ) + 2q\cdot p_{\gamma}\sigma_4(\epsilon , p , p_{\gamma},q ) -\sigma_4(p_{\gamma},p,\zeta , q ) ] , \nonumber \\ \hat{c}_u^{\frac{5}{2}^+ } & = & -f_2\frac{c_4}{m}\sigma\cdot p + f_1m[c_2\sigma\cdot\epsilon - c_1\sigma\cdot p_{\gamma}]+\frac{1}{5m}f_2 [ c_3\omega_3(p , p , q)-\omega_3(p,\xi_2,q ) ] \nonumber \\ & & + \frac{1}{5}f_1[\frac{1}{m}\omega_3(\xi_1,p , q)+2q\cdot p_{\gamma}m \omega_3(\epsilon , p_{\gamma},q)-m\omega_3(p_{\gamma},\zeta , q ) ] , \nonumber \\ \hat{d}_u^{\frac{5}{2}^+ } & = & f_2c_4\frac{e}{m}+f_1c_1me_{\gamma } + \frac{1}{5m}f_2[c_3\sigma_3(p , p , q)-\sigma_3(p,\xi_2,q ) ] \nonumber \\ & & + \frac{1}{5}f_1[\frac{1}{m}\sigma_3(\xi_1,p , q)+2q\cdot p_{\gamma}m \sigma_3(\epsilon , p_{\gamma},q)-m\sigma_3(p_{\gamma},\zeta , q ) ] \label{uop52}\end{aligned}\ ] ] in the @xmath11-channel . for intermediate baryons of negative parity , the @xmath108 and @xmath109 operators are given by the same expressions as for intermediate baryons of positive parity and the same spin ; whereas , the @xmath110 and @xmath111 operators are given by expressions that are the negatives of the corresponding positive parity expressions . for the @xmath12-channel , we define the coupling parameters @xmath246 where @xmath34 is the same scaling mass that appears in eqs . ( [ tphk * ] ) and ( [ tphk1]),and the @xmath247 are the coupling products defined by eqs . ( [ tchres ] ) . in terms of these parameters , the @xmath12-channel operators are given by @xmath248 for an intermediate kaon , @xmath249 for an intermediate @xmath17 resonance , and @xmath250 , \nonumber \\ \hat{b}_{k1}^t & = & \alpha^t[\epsilon\cdot p_k \sigma(p , p_{\gamma } ) + p_{\gamma}\cdot p_k\sigma\cdot{\bf p } \sigma\cdot{\bf \epsilon } ] , \nonumber \\ \hat{c}_{k1}^t & = & \alpha^v[p_{\gamma}\cdot p_k\sigma\cdot{\bf \epsilon } -\epsilon\cdot p_k\sigma\cdot{\bf p}_{\gamma } ] , \nonumber \\ \hat{d}_{k1}^t & = & \alpha^v\epsilon\cdot p_k e_{\gamma } \label{tk1}\end{aligned}\ ] ] for an intermediate @xmath18 resonance , where p and e are the 4-momentum and energy of the intermediate meson , @xmath251 and @xmath252 h. thom , phys . * 151 * , 1322 ( 1966 ) f.m . renard and y. renard , nucl . phys . * b1 * , 389 ( 1967 ) ; f.m . renard and y. renard , phys . 24b * , 159 ( 1967 ) ; f.m . renard and y. renard , nucl . b25 * , 490 ( 1971 ) ; y. renard , _ ibid . _ * b40 * , 499 ( 1972 ) . h. thom , e. gabathuler , d. jones , b.d . mcdaniel , and w.m . woodward , phys . lett . * 11 * , 433 ( 1963 ) ; m. grilli , l. mezzetti , m. nigro , and e. schiavuta , nuovo cim . * 38 * , 1467 ( 1965 ) ; d.e . groom and j.h . marshall , phys . 159 * , 1213 ( 1967 ) ; t. fujii et al . d * 2 * , 439 ( 1970 ) . b445 * , 20 ( 1998 ) ; s. goers _ et al . _ , _ ibid . _ * b464 * , 331 ( 1999 ) ; k.h . et al . _ , eur . j. a * 19 * , 251 ( 2004 ) . et al . _ , 91 * , 092001 ( 2003 ) ; m. sumihama _ et al . _ , c * 73 * , 035214 ( 2006 ) . g. niculescu _ et al . _ , lett . * 81 * , 1805 ( 1998 ) ; l. teodorescu _ et al . _ , a658 * , 362 ( 1999 ) ; r.m . et al . _ , c * 67 * , 052205 ( 2003 ) . et al . _ , lett . * 90 * , 131804 ( 2003 ) ; d.s . et al . _ , c * 79 * , 065205 ( 2009 ) . c * 69 * , 042201 ( 2004 ) . a. lleres _ et al . _ , eur . j. a * 31 * , 79 ( 2007 ) ; a. dangelo _ et al . _ , _ ibid _ * 31 * , 441 ( 2007 ) ; a. lleres _ et al . _ , arxiv:0807.3839v1 [ nucl - ex ] . r. bradford _ _ , phys . rev . c * 73 * , 035202 ( 2006 ) . r. bradford _ et al . _ , c * 75 * , 035205 ( 2007 ) . p. ambrozewicz _ c * 75 * , 045203 ( 2007 ) . r. nasseripour _ et al . c * 77 * , 065208 ( 2008 ) . adelseck , c. bennhold , and l.e . wright , phys . c * 32 * , 1681 ( 1985 ) . r. a. adelseck and l.e . wright , phys . c * 38 * , 1965 ( 1988 ) ; r.a . adelseck and b. saghai , _ ibid . _ * 42 * , 108 ( 1990 ) . robert a. williams , chueng - ryon ji , and stephen r. cotanch , phys . c * 46 * , 1617 ( 1992 ) . t. mart , c. bennhold , and c.e . hyde - wright , phys . c * 51 * , r1074 ( 1995 ) ; h. haberzettl , c. bennhold , t. mart , and t. feuster,_ibid . _ * 58 * , r40 ( 1998 ) ; t. mart and c. bennhold , nucl . phys . * a639 * , 237c ( 1998 ) ; h. haberzettl , c. bennhold , and t. mart , acta phys . pol . b * 31 * , 2387 ( 2000 ) ; h. haberzettl , c. bennhold , and t. mart , nucl . phys . * a684 * , 475c ( 2001 ) . t. mart and c. bennhold , phys . c * 61 * , 012201(r ) ( 1999 ) . cheoun , b.s . han , b.g . yu , and il - tong cheon , phys . rev . c * 54 * , 1811 ( 1996 ) ; bong son han , myung ki cheoun , k.s . kim , and il - tong cheon , nucl . * a691 * , 713 ( 2001 ) . david , c. fayard , g.h . lamot , and b. saghai , phys . c * 53 * , 2613 ( 1996 ) . t. mizutani , c. fayard , g .- h . lamot , and b. saghai , phys . c * 58 * , 75 ( 1998 ) . hsiao , d.h . lu , and shin nan yang , phys . rev . c. * 61 * , 068201 ( 2000 ) . wen tai chiang , f. tabakin , t - s.h . lee , and b. saghai , phys . lett . * b517 * , 101 ( 2001 ) . stijn janssen , jan ryckebusch , wim van nespen , dimitri debruyne , and tim van cauteren , eur . j. a * 11 * , 105 ( 2001 ) ; stijn janssen , jan ryckebusch , dimitri debruyne , and tim van cauteren , phys . c * 65 * , 015201 ( 2001 ) ; stijn janssen , jan ryckebusch , dimitri debruyne , and tim van cauteren , _ ibid . _ * 66 * , 035202 ( 2002 ) ; s. janssen , j. ryckebusch , and t. van cauteren , _ ibid . _ , * 67 * , 052201(r ) ( 2003 ) ; s. janssen , d.g . ireland , and j. ryckebusch , phys . lett . * b562 * , 51 ( 2003 ) . oren v. maxwell , phys . c * 69 * , 034605 ( 2004 ) ; oren v. maxwell , phys . rev . c * 70 * 044612 ( 2004 ) oren v. maxwell , phys . c * 76 * , 014621 ( 2007 ) . a. usov and o. scholter , phys . c * 72 * , 025205 ( 2005 ) . sarantsev , v.a . nikonov , a.v . anisovich , e. klempt , and u. thoma , eur . j. a * 25 * , 441 ( 2005 ) ; a.v . anisovich , v. kleber , e. klempt , v.a . nikonov , a.v . sarantsev , and u. thoma , eur . j. a * 34 * , 243 ( 2007 ) ; v.a . nikonov , a.v . anisovich , e. klempt , a.v . sarantsev , and u. thoma , phys . b * 662 * , 245 ( 2008 ) . n. kaiser , t. waas , and w. weise , nucl . a * 612 * , 297 ( 1997 ) . t. feuster and u. mosel , phys . c * 59 * , 460 ( 1999 ) . b. julia - diaz , b. saghai , t. s. lee , and f. tabakin , phys . c * 73 * , 055204 ( 2006 ) . see , for example , the discussion in t. mart and a. sulaksono , phys . c * 74 * , 055203 ( 2006 ) ; and p. bydv zovsky and t. mart , phys . c * 76 * , 065202 ( 2007 ) . m. benmerrouche , r.m . davidson , and nimai c. mukhopadhyay , phys . c * 39 * , 2339 ( 1989 ) . w.m . yao et al . , j. phys . g * 33 * , 1 ( 2006 ) . simon capstick and w. roberts , phys . d * 58 * , 074011 ( 1998 ) . f.j . klein _ et al . _ , jefferson lab experiment e02 - 112 . o.dumbrajs , r. koch , h. pilkuhn , g.c . oades , h. behrens , j.j . de swart , and p. kroll , nucl . phys . * b216 * , 277 ( 1983 ) . deswart , rev . modern phys . * 35 * , 916 ( 1963 ) . ignacio j. general and stephen r. cotanch , phys . c * 69*,035202 ( 2004 ) .

the evidence for the accelerated expansion of the universe and the time - dependence of the fine - structure constant suggests the existence of at least one scalar field with a mass of order . if such a field exists , then it is generally assumed that its coupling to matter must be tuned to unnaturally small values in order to satisfy the tests of the equivalence principle ( ep ) . in this paper , we present an alternative explanation which allows scalar fields to evolve cosmologically while having couplings to matter of order unity . in our scenario , the mass of the fields depends on the local matter density : the interaction range is typically of order 1 mm on earth ( where the density is high ) and of order au in the solar system ( where the density is low ) . all current bounds from tests of general relativity are satisfied . nevertheless , we predict that near - future experiments that will test gravity in space will measure an effective newton s constant different by order unity from that on earth , as well as ep violations stronger than currently allowed by laboratory experiments . such outcomes would constitute a smoking gun for our scenario .

astro-ph0309411

there is growing evidence in cosmology for the existence of nearly massless scalar fields in our universe . on the one hand , a host of observations , from supernovae luminosity - distance measurements @xcite to the cosmic microwave background anisotropy @xcite , suggests that 70% of the current energy budget consists of a dark energy fluid with negative pressure . while observations are consistent with a non - zero cosmological constant , the dark energy component is more generally modeled as quintessence : a scalar field rolling down a flat potential @xcite . in order for the quintessence field to be evolving on cosmological time scales today , its mass must be of order @xmath0 , the present hubble parameter . on the other hand , recent measurements of absorption lines in quasar spectra suggest that the fine - structure constant @xmath2 has evolved by roughly one part in @xmath3 over the redshift interval @xmath4 @xcite . time - variation of coupling constants are generally modeled with rolling scalar fields @xcite , and the recent evidence for a time - varying @xmath2 requires the mass of the corresponding scalar field to be of order @xmath0 @xcite . in either case , the inferred scalar field is essentially massless on solar system scales , and therefore subject to tight constraints from tests of the equivalence principle ( ep ) @xcite . the current bound on the e@xmath5tv@xmath5s parameter , @xmath6 , which quantifies the deviation from the universality of free - fall , is @xmath7 , from the e@xmath5t - wash experiment @xcite . from a theoretical standpoint , massless scalar fields or moduli are abundant in string and supergravity theories . indeed , generic compactifications of string theory result in a plethora of massless scalars in the low - energy , four - dimensional effective theory . however , these massless fields generally couple directly to matter with gravitational strength , and therefore lead to unacceptably large violations of the ep . therefore , if the culprit for quintessence or time - varying @xmath2 is one of the moduli of string theory , some mechanism must effectively suppress its ep - violating contributions . for instance , damour and polyakov @xcite ( see also @xcite ) have proposed a dynamical mechanism to suppress the coupling constants @xmath8 between the various matter fields and the dilaton of string theory . alternatively , the suppression could be the result of approximate global symmetries @xcite . in a recent paper @xcite , we presented a scenario in which scalar fields can evolve cosmologically while having couplings to matter of order unity , _ i.e. , _ @xmath9 . this is because the scalar fields acquire _ a mass whose magnitude depends on the local matter density_. in a region of high density , such as on earth , the mass of the fields is large , and thus the resulting violations of the ep are exponentially suppressed . in the solar system , where the density is much lower , the moduli are essentially free , with a compton wavelength that can be much larger than the size of the solar system . finally , on cosmological scales , where the density is very low , the mass can be of the order of the present hubble parameter , thereby making the fields potential candidates for causing the acceleration of the universe or the time - evolution of the fine - structure constant . while the idea of density - dependent mass terms is not new @xcite , the novelty of our work lies in that the scalar fields can couple directly to baryons with gravitational strength . in our scenario , scalar fields that have cosmological effects , such as quintessence , do not result in large violations of the ep in the laboratory because we happen to live in a very dense environment . thus , the main constraint on our model is that the mass of the field be sufficiently large on earth to evade ep and fifth force constraints @xcite . the generation of a density - dependent mass for a given modulus @xmath10 results from the interplay of two source terms in its equation of motion . the first term arises from self - interactions , described by a monotonically - decreasing potential @xmath11 which is of the runaway form ( see fig . [ pot ] ) . in particular , we underscore the fact that the potential need not have a minimum ; rather , it must be monotonic . the second term arises from the conformal coupling to matter fields , of the form @xmath12 . the coupling constants @xmath8 need not be small , however , and values of order unity or greater are allowed . although these two contributions are both monotonic functions of @xmath10 , their combined effect is that of an effective potential which does display a minimum ( see fig . [ poteff ] ) . furthermore , since this effective potential depends explicitly on the local matter density @xmath13 , both the field value at the minimum and the mass of small fluctuations depend on @xmath13 as well , with the latter being an increasing function of the density . although the scalar fields are quite massive on earth , their behavior is strikingly different in the solar system where the local matter density is much smaller . thus our model makes a crucial prediction for near - future experiments that will test gravity in space . for example , consider the see project @xcite which among other things will measure newton s constant to an unprecedented accuracy . our scenario generically predicts that the see experiment should observe corrections of order unity to newton s constant compared to its measured value on earth , due to fifth - force contributions which are important in space but exponentially suppressed on earth . moreover , three satellite experiments to be launched in the near future , step @xcite , galileo galilei ( gg ) @xcite and microscope @xcite , will test the universality of free - fall in orbit with expected accuracy of @xmath14 , @xmath15 and @xmath16 , respectively . we predict that these experiments should observe a strong ep - violating signal . in fact , for a wide range of parameters , our model predicts that the signal will be larger than the ground - based e@xmath5t - wash bound of @xmath17 . if see does measure an effective newton s constant different from that on earth , or if step observes an ep - violating signal larger than thought permitted by the e@xmath5t - wash experiment , this will strongly indicate that a mechanism of the form proposed here is realized in nature . for otherwise it would be hard to explain the discrepancies between measurements in the laboratory and those in orbit . these new and surprising outcomes are a direct consequence of the fact that scalar fields in our model have drastically different behavior in regions of high density than in regions of low density . we refer to @xmath10 as a `` chameleon '' field , since its physical properties , such as its mass , depend sensitively on the environment . moreover , in regions of high density , the chameleon `` blends '' with its environment and becomes essentially invisible to searches for ep violation and fifth force . even though we predict significant violations of the ep in space , all existing constraints from planetary orbits @xcite , such as those from lunar laser ranging @xcite , are easily satisfied in our model . this is because of the fact that the chameleon - mediated force between two large objects , such as the earth and the sun , is much weaker than one would naively expect . to see this , we use calculus and break up the earth into a collection of infinitesimal volume elements . consider one such volume element located well - within the earth . since the mass of the chameleon is very large inside the earth , the @xmath10-flux from this volume element is exponentially suppressed and therefore contributes negligibly to the @xmath10-field outside the earth . this is true for all volume elements within the earth , except for those located in a thin shell near the surface . infinitesimal elements within this shell are so close to the surface that they do not suffer from the bulk exponential suppression . thus , the exterior field is generated almost entirely by this thin shell , whereas the bulk of the earth contributes negligibly . a similar argument applies to the sun . consequently , the chameleon - mediated force between the earth and the sun is suppressed by this thin - shell effect , which thereby ensures that solar system tests of gravity are satisfied . however , note that this only applies to large objects , such as planets . sufficiently small objects do not suffer from thin - shell suppression , and thus their entire mass contributes to the exterior field . in particular , a small satellite in orbit , such as see , may not exhibit a thin - shell effect . this is why the orbits of the planets are essentially unaffected by the @xmath10-force , whereas the fifth force between two test particles in the see capsule is significant . since @xmath10 couples directly to matter fields , all mass scales and coupling constants of the standard model depend on space and time . once again due to the thin - shell mechanism described above , spatial variations of constants are sufficiently small in our model to satisfy current experimental bounds , for instance from the vessot - levine experiment @xcite . time variation of coupling constants are also not a problem since , during most of the history of the universe , the various couplings actually change by very little . thus the bounds from big bang nucleosynthesis , for instance , are easily satisfied . this will be described in more detail in a separate paper dealing with the cosmological evolution in our model @xcite . in sec . [ setups ] , we describe the ingredients of the scenario , focusing on a single modulus @xmath10 for simplicity . we show how the dynamics of @xmath10 are governed by an effective potential that depends on the local matter density . in sec . [ compact ] , we derive approximate solutions for @xmath10 for a compact object such as the sun , for instance , and describe the thin - shell mechanism mentioned earlier . in sec . [ earth ] , we specialize the solution for @xmath10 to the case of the earth , and apply the results in sec . [ tests ] to derive constraints on the parameters of the theory based on laboratory tests of the ep and searches for a fifth force . we then show in sec . [ condmass ] that , for a potential of power - law form , @xmath18 , these constraints translate into the requirement that the energy scale @xmath19 be less than an inverse millimeter or so . curiously , this is also the scale associated with the cosmological constant today . in sec . [ solar ] , we argue that our model easily satisfies constraints from solar system tests of gr . it is showed ( sec . [ sep ] ) that the same holds true for bounds from spatial and time variation of coupling constants . we then predict ( sec . [ satellite ] ) that near - future experiments that aim at testing the ep and measuring a fifth force should observe a large signal , perhaps stronger than previously thought possible . finally , we conclude and summarize our results in sec . [ conclu ] . focusing on a single scalar field @xmath10 for simplicity , the action governing the dynamics of our model is given by @xmath20 where @xmath21 is the reduced planck mass , @xmath22 is the determinant of the metric @xmath23 , @xmath24 is the ricci scalar and @xmath25 are matter fields . the scalar field @xmath10 interacts directly with matter particles through a conformal coupling of the form @xmath12 . explicitly , each matter field @xmath25 couples to a metric @xmath26 which is related to the einstein - frame metric @xmath23 by the rescaling @xmath27 where @xmath8 are dimensionless constants @xcite . moreover , for simplicity , we assume that the different @xmath25 s do not interact with each other . note that eq . ( [ action ] ) is of the general form of low - energy effective actions from string theory and supergravity , where @xmath11 arises from non - perturbative effects . =3 in the potential @xmath11 is assumed to be of the runaway form . that is , it is monotonically decreasing and satisfies @xmath28 as well as @xmath29 where @xmath30 , _ etc_. see fig . for instance , a fiducial example is the inverse power - law potential @xmath31 where @xmath19 has units of mass and @xmath32 is a positive constant . the above conditions on the asymptotics of @xmath33 are generally satisfied by potentials arising from non - perturbative effects in string theory @xcite . note that it is also of the desired form for quintessence models of the universe @xcite . the equation of motion for @xmath10 derived from the above action is then @xmath34 where @xmath35 is the stress - energy tensor for the @xmath36th form of matter . for the purpose of this paper , it will suffice to approximate the geometry as minkowski space , that is , @xmath37 . this is valid provided that the newtonian potential is small everywhere , and that the backreaction due to the energy density in @xmath10 is also small . this latter assumption will be justified when we analyze post - newtonian corrections in sec . [ ppn ] . for non - relativistic matter , one has @xmath38 , where @xmath39 is the energy density . however , we shall find it convenient to express our equations not in terms of @xmath39 , but rather in terms of an energy density @xmath40 which is conserved in einstein frame . in other words , @xmath41 is defined so that it is independent of @xmath10 . equation ( [ eom0 ] ) thus reduces to @xmath42 from the right - hand side of eq . ( [ eom ] ) , we see that the dynamics of @xmath10 are not solely governed by @xmath11 , but rather by an effective potential @xmath43 which depends explicitly on the matter density @xmath41 . in particular , although @xmath11 is monotonic , @xmath44 does exhibit a minimum provided that @xmath45 . this is illustrated in fig . [ poteff ] for the case of a single component @xmath13 with coupling @xmath46 . ( one could equivalently consider the case @xmath47 and @xmath48 . ) unfortunately , known examples in string theory have @xmath49 and @xmath8 occurring with the same sign . if @xmath10 is the modulus of an extra dimension , for instance , then one expects that @xmath50 and @xmath51 in the decompactification limit @xcite ; in this limit , the various masses of particles also tend to zero . =3 in we will denote by @xmath52 the value assumed by @xmath10 at the minimum , that is , @xmath53 meanwhile , the mass of small fluctuations about @xmath52 is obtained as usual by evaluating the second derivative of the potential at @xmath52 : @xmath54 = 5 in equations ( [ phimin ] ) and ( [ mmin ] ) respectively imply that the local value of the field , @xmath52 , and its mass , @xmath55 , both depend on the local matter density . since @xmath49 is negative and monotonically increasing , while @xmath56 is positive and decreasing , it also follows that larger values of @xmath41 correspond to smaller @xmath52 and larger @xmath55 . this is illustrated in fig . [ lrhosrho ] . the denser the environment , the more massive is the chameleon . we will later see that it is possible for @xmath55 to be sufficiently large on earth to evade current constraints on ep violations and fifth force , while being sufficiently small on cosmological scales for @xmath10 to have interesting cosmological effects . the upshot of our model , from a theoretical standpoint , is that the potential @xmath11 need not have a minimum , nor does the coupling constant @xmath46 need be tuned to less than @xmath57 to satisfy ep constraints @xcite . quite the contrary , @xmath11 is assumed monotonic , while @xmath46 can be of order unity . in order to study the observable consequences of our model , in particular with regards to ep violations and fifth force mediation , we must first understand the profile that @xmath10 acquires on earth and in the solar system . therefore , in this section , we derive an approximate solution for @xmath10 in the case where the source is a compact object , which we idealize as being perfectly spherical and having homogeneous density . thus consider a static , spherically - symmetric body of radius @xmath58 , homogeneous density @xmath59 and total mass @xmath60 . in the case of the earth , the latter can be approximated by the characteristic terrestrial density : @xmath61 . we treat the object as isolated , in the sense that the effect of surrounding bodies is neglected . it is not in vacuum , however , but is instead immersed in a background of homogeneous density @xmath62 . in the case of solar system objects , this models the fact that our local neighborhood of the galaxy is not empty , but rather filled with an approximately homogeneous component of baryonic gas and dark matter with density @xmath63 . in the case of a baseball in the earth s atmosphere , @xmath62 denotes the surrounding atmospheric density : @xmath64 . with these assumptions , eq . ( [ eom ] ) reduces to @xmath65 where @xmath66 note that we temporarily focus on the case where all @xmath8 s assume the same value @xmath46 . this is done for simplicity only , and the following analysis remains qualitatively unchanged when these are taken to be different . moreover , this assumption will be dropped when we derive the resulting violations of the ep in sec . [ tests ] . in other words , in the end we are not assuming that the theory is brans - dicke @xcite . throughout the analysis , we denote by @xmath67 and @xmath68 the field value which minimizes @xmath44 for @xmath69 and @xmath70 , respectively . that is , from eqs . ( [ phimin ] ) and ( [ rho ] ) , we have @xmath71 similarly , we denote by @xmath72 and @xmath73 the mass of small fluctuations about @xmath67 and @xmath68 , respectively . that is , @xmath72 ( @xmath73 ) is the mass of the chameleon field inside ( outside ) the object . equation ( [ sun1 ] ) is a second order differential equation and as such requires two boundary conditions . since the solution must be non - singular at the origin , we require @xmath74 at @xmath75 , as usual . moreover , since @xmath76 at infinity , it is natural to impose @xmath77 as @xmath78 . this latter condition is physically sensible as it implies @xmath79 as @xmath78 , and thus that the @xmath10-force between the compact object and a test particle tends to zero as their separation becomes infinite . to summarize , eq . ( [ sun1 ] ) is subject to the following two boundary conditions : @xmath80 before solving this problem explicitly , it is useful to give a heuristic derivation of the solution . far outside the object , @xmath81 , we know that the chameleon tends to @xmath68 , as required by the second boundary condition . there are then two types of solution , depending on whether the object is large ( _ e.g. _ , the earth ) or small ( _ e.g. _ , a baseball ) . the distinction between large and small will be made precise below . small objects do not generate large variations in @xmath10 . thus , their solution can be thought of as a perturbation on the background solution @xmath82 . hence , one has @xmath83 everywhere in this case , including the interior of the object . large objects , on the other hand , are strongly perturbing . within the object , @xmath69 , one finds that the chameleon nearly minimizes the effective potential , and thus @xmath84 . hence , the solution essentially extrapolates between @xmath85 within the core and @xmath82 far outside . to be more precise , let us describe the exterior solution ( @xmath70 ) for large bodies , assuming @xmath86 for simplicity . for this purpose , it is convenient to break up the object into infinitesimal volume elements @xmath87 and consider their individual contribution to the @xmath10-field . well - within the object , one has @xmath84 , and the mass of the chameleon is large , @xmath88 . thus , the contribution from a volume element @xmath87 within the core is proportional to @xmath89 and is therefore exponentially suppressed . in other words , it contributes negligibly to the @xmath10-field outside . this holds for all infinitesimal volume elements within the object , except for those lying within a thin shell of thickness @xmath90 near the surface @xcite . see fig . [ neman ] . thus , the exterior solution is obtained by summing over all elements within this shell : @xmath91 where @xmath92 is the distance from the center of the object . =3 in we will find in sec . [ rigor ] that @xmath93 where @xmath94 is the newtonian potential at the surface of the object . the derivation of eq . ( [ thinsoln ] ) implicitly assumed that the shell was thin , that is , @xmath95 we shall henceforth refer to this as the thin - shell condition . small objects , in the sense that @xmath96 , do not have a thin shell . rather , their entire volume contributes to the @xmath10-field outside , and thus the exterior solution is @xmath97 which is recognized as the yukawa profile for a scalar field of mass @xmath73 . note that eq . ( [ thinsoln ] ) and ( [ warmup ] ) differ only by a thin - shell suppression factor of @xmath98 . to summarize , the exterior solution for a compact object is given by @xmath99 with @xmath98 defined in eq . ( [ dr ] ) . the ratio @xmath100 , which appears in eq . ( [ dr ] ) and determines whether or not an object has a thin shell , can be interpreted physically as follows . given a background profile @xmath101 , it is straightforward to show from the action ( [ action ] ) that the resulting chameleon - force on a test particle of mass @xmath19 and coupling @xmath46 is given by @xmath102 it follows that @xmath10 should be thought of as a potential for this `` fifth '' force . thus , @xmath100 is the ratio of the difference in @xmath10 potential to the newtonian potential , and effectively quantifies how perturbing the object is for the @xmath10 field . to get an intuition for the boundary value problem at hand , it is useful to think of @xmath92 as a time coordinate and @xmath10 as the position of a particle , and treat eq . ( [ sun1 ] ) as a dynamical problem in classical mechanics . this is akin to the familiar trick performed in bubble nucleation calculations @xcite . in this language , the particle moves along the inverted potential @xmath103 , and the second term on the left - hand side of eq . ( [ sun1 ] ) , proportional to @xmath104 , is recognized as a damping term . an important difference here is that @xmath103 is `` time''-dependent since @xmath13 depends on @xmath92 . more precisely , the effective potential undergoes a jump at time @xmath105 , as illustrated in fig . [ potr ] . = 5 in the particle begins at rest , since @xmath74 at @xmath75 , from some initial value which we denote by @xmath106 : @xmath107 for small @xmath92 , the friction term is large , and thus the particle is essentially frozen at @xmath108 . it remains stuck there until the damping term , proportional to @xmath104 , is sufficiently small to allow the driving term , @xmath109 , to be effective . in other words , the amount of `` time '' the particle remains stuck near @xmath108 depends on the slope of the potential , @xmath109 , at @xmath108 . once friction is negligible , the particle begins to roll down the potential . see fig . [ potr]a ) . it rolls down until , at some later time @xmath105 , the potential suddenly changes shape as @xmath110 undergoes a jump from @xmath59 to @xmath62 ( see eq . ( [ rho ] ) ) . but @xmath10 and @xmath111 are of course continuous at the jump , and the particle keeps rolling , this time climbing up the inverted potential . see fig . [ potr]b ) . if the initial position @xmath106 is carefully chosen , the particle will barely reach @xmath68 in the limit @xmath78 , as desired . ( it is easy to prove that such @xmath106 always exists . ) thus the problem is reduced to determining the initial value @xmath106 . in the end @xmath106 will depend on the physical properties of the compact object , such as its density @xmath59 and its radius @xmath58 , as well as on the various parameters of the theory , such as @xmath46 and the shape of the potential . rather than first choosing a set of values for @xmath59 , @xmath58 , _ etc . _ and then solving for @xmath106 , we will instead choose a range of @xmath106 and determine the corresponding region in the @xmath112 parameter space . more precisely , we shall consider the two regimes @xmath113 and @xmath114 , and we will show that these correspond respectively to @xmath115 and @xmath116 . anticipating this result , we refer to these two regimes as _ thin - shell _ and _ thick - shell _ , respectively . * thin - shell regime : @xmath117 . * this corresponds to @xmath10 being released from a point very close to @xmath67 . since @xmath67 is a local extremum of the effective potential , the driving term @xmath109 is negligible initially , and the dynamics are strongly dominated by friction . consequently , the field remains frozen at its initial value @xmath118 for a long time , until the friction force is sufficiently small to allow the particle to roll . we shall denote by @xmath119 the `` moment '' at which this occurs . hence , we have @xmath120 when @xmath121 , the field is still near @xmath67 but has now begun to roll . since @xmath122 as soon as @xmath10 is displaced significantly from @xmath67 , as illustrated in fig . [ poteff ] , we may approximate eq . ( [ sun1 ] ) in the regime @xmath123 by @xmath124 where we have also assumed @xmath125 . the solution to eq . ( [ approx1 ] ) with boundary conditions @xmath85 and @xmath74 at @xmath126 is @xmath127 the full solution for @xmath128 is thus approximated by eqs . ( [ frozen ] ) and ( [ rlrc2 ] ) . the approximation of separating the solution into the two regions @xmath129 and @xmath123 only makes sense , however , if @xmath130 . for otherwise there is no clear separation between the two regions , and one needs a solution valid over the entire range @xmath128 . at @xmath105 , the energy density undergoes a jump from @xmath59 to @xmath62 , as described by eq . ( [ rho ] ) . for @xmath70 , the effective potential is shown in fig . [ potr ] b ) , and the particle is climbing up the hill . since the speed of the particle is initially large compared to the curvature of the potential , eq . ( [ sun1 ] ) can be approximated by @xmath131 whose solution , satisfying @xmath132 as @xmath78 , is given by @xmath133 where @xmath134 is a constant and where we have used the fact that the potential is approximately quadratic near @xmath82 . the two unknowns , @xmath119 and @xmath134 , are then determined by matching @xmath10 and @xmath111 at @xmath105 using eqs . ( [ rlrc2 ] ) and ( [ better2 ] ) . with the approximation that @xmath130 , it is straightforward to show that the exterior solution is @xmath135 with @xmath136 where we have substituted the newtonian potential @xmath137 . * thick - shell regime : @xmath114 . * in this case , the field is initially sufficiently displaced from @xmath67 that it begins to roll almost as soon as it is released at @xmath75 . hence there is no friction - dominated regime in this case , and the interior solution for @xmath10 is most easily obtained by taking the @xmath138 limit of eq . ( [ rlrc2 ] ) and replacing @xmath67 by @xmath106 . matching to the exterior solution as before , we obtain @xmath139 and @xmath140 moreover , equating these two equations at @xmath105 , we find @xmath141 . in particular , since @xmath114 and using the definition @xmath142 , this implies @xmath143 we conclude this section with a word on how the above solutions , which assumed homogeneous @xmath13 , can be generalized to the more realistic case of spatially - varying matter density . in most cases of interest , such as the interior of the earth for instance , we will find that the matter density varies on scales much larger than the compton wavelength @xmath144 of the chameleon field in that region . more precisely , it is generally the case that @xmath145 within dense objects . if so , one can make an adiabatic approximation which consists of treating @xmath146 as a constant in the equations of motion . in other words , in this case one may simply substitute @xmath146 in the expressions above . since the most stringent constraints on possible violations of the ep derive from experiments performed on earth , it is important to discuss in some detail the profile for @xmath10 inside and in the vicinity of the earth . admittedly , the model for our planet described below is rather crude , but is sufficiently accurate to derive order - of - magnitude estimates of resulting violations of the ep . more realistic descriptions can be obtained for instance by using the adiabatic approximation discussed at the end of sec . [ compact ] , or through numerical analysis . the earth is modeled as a solid sphere of radius @xmath147 cm and homogeneous density @xmath148 . surrounding it is an atmosphere which we approximate as a layer 10 km in radius with homogeneous density @xmath149 . moreover , we treat our planet as an isolated body , neglecting the effect of surrounding compact objects such as the sun and the moon . furthermore , far away from the earth , the matter density is approximated by the density of homogeneous gas and dark matter in our local neighborhood of the galaxy : @xmath150 . the set - up is thus almost identical to that of eq . ( [ sun1 ] ) , except that the matter density now has three phases instead of two : @xmath151 where @xmath152 10 km . we henceforth denote by @xmath153 , @xmath154 and @xmath155 the field value which minimizes the effective potential for the respective densities . similarly , @xmath156 , @xmath157 and @xmath158 are the respective masses . following the discussion in sec . [ compact ] , the solution depends on whether or not the earth and its atmosphere have a thin shell . as we will prove below , it is necessary that the atmosphere has a thin shell , for otherwise unacceptably large violations of the ep will ensue . in this case , one has @xmath159 in the bulk of the atmosphere . moreover , since the earth is much denser than the atmosphere , it follows that the earth itself has a thin shell , in which case @xmath160 inside the earth . from eqs . ( [ dr ] ) and ( [ thincond ] ) , the thin - shell condition for the atmosphere reads @xmath161 where @xmath162 . we can refine this bound by noting that , in order for the atmosphere to have a thin shell , clearly the thickness of the shell must be less than the thickness of the atmosphere itself , which is @xmath163 . hence this requires @xmath164 . using the fact that @xmath165 , and thus @xmath166 , we can write this as @xmath167 this condition , which ensures that the atmosphere has a thin shell , will play a crucial role in the analysis of tests of gravity in the following sections . the exterior solution , @xmath168 , is then given by eq . ( [ rgrc2fin ] ) with @xmath169 and @xmath170 : @xmath171 to summarize , the solution for @xmath10 for the earth and its atmosphere is well - approximated by @xmath172 where @xmath173 is defined in eq . ( [ condatm ] ) . it remains to show that tests of the ep require the atmosphere to have a thin shell . the proof proceeds by contradiction . suppose that condition ( [ condatm ] ) is violated and instead we have @xmath174 we can therefore ignore the atmosphere altogether , and thus the @xmath10-profile in the laboratory is given by eq . ( [ earthext ] ) : @xmath175 where we have neglected the exponential factor since @xmath176 , as we will see in sec . [ condmass ] . from eq . ( [ fifth ] ) , this profile results in a fifth force on a test particle of mass @xmath19 and coupling @xmath8 of magnitude @xmath177 supposing that the @xmath8 s are all of order @xmath46 but assume different values for different matter species , then the resulting difference in relative free - fall acceleration for two bodies of different composition will be @xmath178 where @xmath6 is the e@xmath5tv@xmath5s parameter , and where the numerical coefficient is appropriate for cu and be test masses @xcite , as used in the e@xmath5t - wash experiment @xcite . for @xmath46 of order unity , we see from eq . ( [ condatmviolated ] ) that this violates the bound @xmath179 . it follows that the atmosphere must be have a thin shell . the tightest constraints on our model derive from laboratory tests of the ep and searches for a fifth force @xcite . since these experiments are usually done in vacuum , we first need to derive an approximate solution for the chameleon inside a vacuum chamber . for simplicity , we model the chamber as a perfectly empty , spherical cavity of radius @xmath180 . in the absence of any other parts within the chamber , and ignoring the effect of the walls , the equation for @xmath10 is given by eq . ( [ sun1 ] ) : @xmath181 where we have assumed @xmath182 as we did throughout sec . [ compact ] , and where @xmath183 the boundary conditions are the same as in sec . [ compact ] : @xmath74 at @xmath75 and @xmath184 as @xmath78 . the solution within the vacuum chamber is analogous to the solution for a compact object with thin - shell ( see sec . [ compact ] ) . in both cases , due to the large density contrast between the object or the vacuum cavity and their environment , @xmath10 must start at @xmath75 from a point where it can remain almost frozen for the entire volume . that is , for @xmath128 in the case of the overdense object and @xmath185 for the vacuum cavity . for a compact object , this freezing point lies naturally near the local extremum @xmath85 of the effective potential . for the vacuum chamber , the effective potential has no extremum for @xmath186 ( since @xmath187 there ) , and thus the only way @xmath10 can remain still is by starting from a point where it will be friction - dominated for almost the entire range @xmath185 . in other words the curvature of the potential at that point must be of order @xmath188 , that is , this freezing point corresponds to a value @xmath189 where the mass of small fluctuations , @xmath190 , is equal to @xmath191 . while the precise solution to ( [ vacc ] ) depends of course on the details of the potential , numerical analysis confirms the above qualitative discussion : * the chameleon assumes the value @xmath192 within the vacuum chamber , where @xmath193 satisfies @xmath194 that is , @xmath193 is the field value about which the compton wavelength of small fluctuations equals @xmath180 , the radius of the chamber . * throughout the chamber , @xmath10 varies slowly , with @xmath195 . * outside the chamber the solution tends to @xmath154 within a distance of @xmath196 from the walls . these generic properties are all we need for our analysis . the potential energy associated with fifth force interactions is generally parameterized by a yukawa potential : @xmath197 where @xmath198 and @xmath199 are the masses of two test bodies , @xmath92 is their separation , @xmath2 is the strength of the interaction ( with @xmath200 for gravitational strength ) , and @xmath201 is the range . null fifth - force searches therefore constrain regions in the @xmath202 parameter space ( see fig . 2.13 of @xcite ) . as discussed above , the range @xmath201 of @xmath10-mediated interactions inside a vacuum chamber is of the order of the size of the chamber . that is , @xmath203 . for @xmath204 , the tightest bound on the coupling constant @xmath2 from laboratory experiments is from hoskins _ et al . _ @xcite : @xmath205 now consider two identical test bodies of uniform density @xmath59 , radius @xmath58 and total mass @xmath206 . if these have no thin shell , then they each generate a field profile given by eq . ( [ warmup ] ) with @xmath207 and @xmath208 : @xmath209 dropping the irrelevant constant , the resulting potential energy is @xmath210 comparison with eq . ( [ fifthpot ] ) shows that the coupling strength is @xmath211 in this case , which clearly violates the bound in eq . ( [ hos ] ) for @xmath212 . hence it must be that the test masses have a thin shell , that is , @xmath213 in this case , their field profile is given by eq . ( [ thinsoln ] ) : @xmath214 and the corresponding potential energy is @xmath215 once again comparing with eq . ( [ fifthpot ] ) , we find that the bound in eq . ( [ hos ] ) translates into @xmath216 note that , @xmath217 , this constraint implies that the thin - shell condition in eq . ( [ condbodyvac ] ) is satisfied . to make the condition ( [ condbodyvac2 ] ) more explicit , note that a typical test body used in hoskins _ _ had mass @xmath218 g and radius @xmath219 cm , corresponding to @xmath220 . substituting in eq . ( [ condbodyvac ] ) and assuming @xmath221 , we obtain the constraint @xmath222 which ensures that the current bounds from laboratory searches of a fifth force are satisfied . turning our attention to the magnitude of ep violations inside our vacuum cavity , we recall that the solution for @xmath10 inside the chamber satisfies @xmath223 from eq . ( [ fifth ] ) , this yields an extra component to the acceleration of magnitude @xmath224 . for cu and be test masses , as used in the e@xmath5t - wash experiment @xcite , this yields an e@xmath5tv@xmath5s parameter of @xmath225 substituting @xmath226 cm and @xmath212 , the e@xmath5t - wash constraint of @xmath179 translates into @xmath227 which is a much weaker constraint than eq . ( [ condvacfin ] ) and thus shall henceforth be ignored . in this section we summarize the constraints derived in the previous sections and apply them to a general power - law potential @xmath228 where @xmath19 has units of mass and @xmath32 is a positive constant . as mentioned earlier , potentials of this form have the desired features for quintessence models of the universe @xcite . we will find that the energy scale @xmath19 is generally constrained to be of the order of @xmath229 . we then discuss the resulting bounds on the interaction range in the atmosphere , in the solar system and on cosmological scales . broad considerations of ep violation lead us to conclude in sec . [ earth ] that both the earth and its atmosphere must have a thin shell . this resulted in the constraint ( see eq . ( [ condatm ] ) ) : @xmath230 recall that , by definition , @xmath155 is the value of @xmath10 which minimizes the effective potential with @xmath231 , _ i.e. _ , @xmath232 . assuming @xmath233 and substituting the power - law potential of eq . ( [ vtypical ] ) gives @xmath234 with @xmath235 and @xmath236 , it is then straightforward to show that eq . ( [ condition2 ] ) can be rewritten as a bound on @xmath19 : @xmath237 then , in sec . [ tests ] , we studied laboratory tests of the ep and fifth force , including the fact that these are performed in vacuum , and derived the condition @xmath238 where @xmath193 is the field value about which the compton wavelength of small fluctuations is of the order of the size of the vacuum chamber , @xmath180 . applying eq . ( [ mv ] ) to the power - law potential yields @xmath239 where we have assumed @xmath240 for concreteness . it is easy to see that smaller values of @xmath180 result in weaker constraints , and this is therefore a conservative choice . it is then straightforward to show that eq . ( [ condvacfin ] ) reduces @xmath241^{-1/(4+n ) } \cdot 10^{3n/(4+n ) } \cdot ( 1\;{\rm mm})^{-1}\ , . \label{condm2}\ ] ] we thus see that , for @xmath32 and @xmath46 of order unity , eqs . ( [ condm1 ] ) and ( [ condm2 ] ) both constrain @xmath19 to be less than approximately an inverse millimeter , or @xmath242 ev . it is remarkable that this is also the mass scale associated with the cosmological constant or dark energy causing the present acceleration of the universe @xcite . granted , this is tiny compared to typical particle physics scales , such as the weak or the planck scale , and thus our potential ( [ vtypical ] ) suffers from fine - tuning . nevertheless , it is our hope that whatever mechanism suppresses the scale of the cosmological constant from its natural value of @xmath243 gev down to @xmath242 ev might also naturally account for the energy scale characterizing our potential . the above constraints can be translated into bounds on the range of chameleon - mediated interactions in the atmosphere ( @xmath196 ) , in the solar system ( @xmath244 ) and on cosmological scales today ( @xmath245 ) . from eq . ( [ mmin ] ) , these are given by @xmath246 where @xmath247 is the current energy density of the universe and @xmath248 is the corresponding value of @xmath10 on cosmological scales . substituting the above bounds on @xmath19 , it is straightforward to show that , for @xmath249 and @xmath46 of order unity , @xmath250 where the numbers on the right - hand depend somewhat on the value of @xmath32 and @xmath46 ( recall that @xmath251 cm and @xmath252 cm ) . thus , while @xmath10-interactions are short - range in the atmosphere ( short compared to the size of the atmosphere ) , they are rather long - range in the solar system . in particular , it is possible for the scalar field to be essentially free on solar system scales . this striking difference in behavior of the field in space compared to earth is an original ingredient of our scenario and , as we will show in sec . [ satellite ] , can lead to unexpectedly large signals for ep and fifth force experiments to be performed in orbit in the near future . on cosmological scales , we see that a power - law potential implies an interaction range which is smaller than the present size of the observable universe , @xmath253 . it follows that @xmath254 is too large for @xmath10 to be rolling on cosmological time scales today . a general study of potentials and their viability for quintessence models of the universe will be presented in a future paper @xcite . in this section , we discuss the constraints on scalar - tensor theories from planetary orbits , in particular from solar system tests of the ep and fifth force ( sec . [ solarep ] ) , as well as post - newtonian corrections ( sec . [ ppn ] ) . in short , these constraints are all satisfied because large bodies , such as the sun and the earth , all have a thin shell which greatly suppresses the @xmath10-force between them . precise measurements of the lunar orbit from laser ranging constrains the difference in free - fall acceleration of the moon and the earth towards the sun to be less than approximately one part in @xmath255 @xcite . that is , denoting their respective acceleration by @xmath256 and @xmath257 , we have @xmath258 where @xmath259 is the newtonian acceleration . in our model , this difference is naturally very small since the sun , earth and moon are all subject to the thin shell effect . we have already imposed that the earth ( and its atmosphere ) have a thin shell . so must the sun , therefore , since its newtonian potential is larger than that of the earth . hence we only need to show that the same holds true for the moon . but , assuming @xmath260 , this trivially follows from eq . ( [ condatm ] ) : @xmath261 where we have used @xmath236 and @xmath262 . hence the @xmath10 profile outside each of these bodies is given by eq . ( [ rgrc2fin ] ) with @xmath169 and @xmath170 . assuming @xmath263 , since this yields maximal violation of the ep , it is then straightforward to show that the acceleration of the earth towards the sun is given by @xmath264 while for the moon @xmath265 substituting @xmath266 , @xmath236 and @xmath262 , this gives a difference in free - fall acceleration of @xmath267 where we have used eq . ( [ condatm ] ) in the last step . this satisfies the bound in eq . ( [ diffacc ] ) for reasonable values of @xmath46 . we next consider solar - system tests of the existence of a fifth force . deviations from a @xmath268 force law , for instance due to the exponential factor in eq . ( [ fifthpot ] ) , contribute an anomalous component to the perihelion precession of planetary orbits in comparison with the predictions of gr . for instance , lunar laser - ranging measurements lead to the constraint @xmath269 for a fifth - force with range @xmath270 @xcite . a similar analysis for the orbits of mercury and mars gives @xmath271 for the range @xmath272 au @xcite . ( not surprisingly , these tests are most sensitive to a fifth force whose range is of the order of the distance between the sun and the orbiting body . ) in our model , these celestial objects are all subject to the thin - shell effect , and , just as with the ep analysis above , the screening mechanism makes the constraints from perihelion precession trivial to satisfy . to estimate the constraints from post - newtonian corrections , consider the @xmath10-profile due to the earth given by eq . ( [ earthext ] ) : @xmath273 where we have neglected the exponential factor . comparison with the expected profile if there were no thin - shell suppression , given by eq . ( [ sol3 ] ) , we see that the exterior solution above corresponds to that of a massless scalar with effective coupling @xmath274 where in the last step we have used the condition that the atmosphere has a thin - shell ( see eq . ( [ condatm ] ) ) . treating our model as a brans - dicke theory with effective coupling constant @xmath275 given above , which is a good approximation in the solar system since the chameleon behaves essentially as a free field , it is straightforward to show that the corresponding effective brans - dicke parameter , @xmath276 , is given by @xcite @xmath277 the tightest constraint on brans - dicke theories comes from light - deflection measurements using very - long - baseline radio interferometer @xcite : @xmath278 . we see that this is easily satisfied in our model . similarly , one can show that the constraint from the decay of the orbital period of binary pulsars , @xmath279 , is trivially satisfied . ( note that density - dependent effective couplings were previously noted in a different context @xcite . ) our discussion of the ep has so far been restricted to the weak ep , which essentially states that the laws of gravity are the same in any inertial frame . general relativity , however , satisfies a stronger version of the ep @xcite , in the sense that all laws of physics , including non - gravitational interactions , assume the same form in any inertial frame ( local lorentz invariance ) , and the various parameters describing these non - gravitational forces , such as the fine - structure constant , are independent of space and time ( local position invariance ) . at any space - time event , one can find a coordinate system in which the einstein - frame metric @xmath23 equals the minkowski metric @xmath280 . since all other metrics @xmath26 appearing in the action ( [ action ] ) are conformally - related to @xmath23 , all @xmath26 are proportional to @xmath280 in this frame . all laws of physics are therefore lorentz invariant at that space - time event , and thus our model satisfies local lorentz invariance . the most stringent constraint on spatial variations of couplings comes from the vessot - levine experiment @xcite which measured the redshift between a hydrogen - maser clock flown at an altitude of @xmath281 km and another one on the ground . in a theory which has lpi , as in gr , the redshift @xmath282 is given by the difference in newtonian potential @xmath283 between the emitter and the receiver @xcite . as shown in will @xcite , lpi violations generate an extra contribution @xmath284 to the redshift , of the form @xmath285 where @xmath286 is a constant that depends on the newtonian potential of the emitter , with @xmath287 corresponding to the case of no lpi violation . the bound from the vessot - levine experiment is @xmath288 . to estimate @xmath286 in our case , recall from eq . ( [ conformal ] ) that a test particle of the matter field @xmath25 follows geodesics of the metric @xmath26 related to the einstein - frame metric @xmath23 by @xmath289 . thus , a constant mass scale @xmath290 in the @xmath25-frame is related to a @xmath10-dependent mass scale @xmath291 in einstein frame by the rescaling @xmath292 . similarly , a @xmath25-clock with frequency @xmath293 is measured in einstein frame to have a frequency @xmath294 . assuming @xmath125 , the @xmath10-dependence of @xmath295 therefore yields the following extra contribution to the redshift : @xmath296 where we have dropped the superscript @xmath297 , and where @xmath298 ( @xmath299 ) is the distance between the emitter ( receiver ) and the center of the earth . for the vessot - levine experiment , we have @xmath300 and @xmath301 , and it follows from eqs . ( [ earthsummary ] ) that @xmath302 and @xmath303 , respectively . since @xmath304 in this case , massaging eq . ( [ delz1 ] ) gives @xmath305 from which we can read off the corresponding value of @xmath286 : @xmath306 this comfortably satisfies the vessot - levine bound of @xmath307 . time - variation of coupling constants are constrained by geophysical measurements ( such as the oklo nuclear reactor @xcite ) , by the study of absorption lines in quasar spectra and by nucleosynthesis . recent analysis indicates that the fine - structure constant has evolved by more than one part in @xmath3 over the cosmological redshift range @xmath308 @xcite , thus suggesting that lpi does not hold in the universe . in our scenario , the various coupling constants are determined by @xmath10 , whose value depends on the local density . thus , even though the coupling constants may vary on cosmological scales today , the fact that the density of the earth is constant in time implies that the various couplings measured on earth do not vary significantly . in particular , this implies that constraints from the oklo nuclear reactor are easily evaded in our model . ( see @xcite for a related discussion of a density - dependent fine - structure constant . ) in any case , we find that the time - variation of coupling constants and masses on cosmological scales is very small in the case of the power - law potential of eq . ( [ vtypical ] ) . to see this , consider once again a constant mass scale @xmath290 in the matter frame , with corresponding @xmath292 in the einstein frame . hence , the time variation of @xmath291 between nucleosynthesis and the present epoch , say , is given by @xmath309 where @xmath310 is the value of @xmath10 at nucleosynthesis . for the power - law potential considered earlier , @xmath311 , the bound of @xmath312 derived in sec . [ condmass ] gives @xmath313 which satisfies all current astrophysical and cosmological bounds . in particular , it appears that the power - law potential can not account for the variation of the fine - structure constant reported by webb _ et al . _ a more detailed analysis of time - variation of coupling constants in our model , including more general potentials , will be presented elsewhere @xcite . our scenario has the remarkable feature that the physical characteristics of the scalar field can be very different in the laboratory than in space . we have seen , for instance , that the range of the interactions it mediates is of order 1 mm in the atmosphere while being greater than 10 au in the solar system . hence , as we will show in this section , the predicted strength of ep violations and fifth force can be strikingly different in orbit than in the laboratory . this is particularly interesting as this decade should witness the launch of several satellite experiments that aim at making these measurements . three of them , namely step , gg and microscope , will test the universality of free - fall with respective expected sensitivity of @xmath14 , @xmath15 and @xmath16 for the e@xmath5tv@xmath5s parameter @xmath6 . another satellite , the see project , will further attempt to measure a fifth force between two test bodies in orbit . in this section we will show that , for a wide range of parameters , our model predicts that these satellite experiments will see a strong signal . it is in fact likely that the signal will be stronger than previously thought possible based on laboratory measurements . for instance , the see capsule could detect corrections to the effective newton s constant ( including fifth force contribution ) of order unity compared to the value measured on earth or inferred from planetary motion . moreover , step , gg and microscope could measure violations of the ep with @xmath314 , which is larger than the current bound from the ( ground - based ) e@xmath5t - wash experiment . such outcomes would constitute a smoking gun for our model , for it would otherwise be difficult to reconcile the results in space with those on earth . let us begin with the see satellite , which will orbit the earth at an altitude of approximately @xmath315 km . this multi - faceted experiment will test for deviations from @xmath268 in the force law , search for violations of the ep and measure newton s constant @xmath316 to one part in @xmath317 . this will be achieved by accurately determining the orbit of two tests masses as they interact gravitationally with each other and with the earth . for our purposes we shall focus on the expected value of @xmath316 , including the contribution from the fifth force mediated by @xmath10 . we first show that it is possible for the satellite _ not _ to have a thin shell . this is desirable in this case in order to maximize the signal . that is , we derive the conditions under which @xmath318 holds for the see capsule . according to the current design , the satellite will have a total mass of @xmath319 kg and an effective radius of @xmath320 m ( although it has cylindrical symmetry , we approximate the capsule as a sphere of equal volume for simplicity ) . thus its newtonian potential is @xmath321 . moreover , at an altitude of 1000 km , the chameleon field assumes the value @xmath322 , as seen from eq . ( [ earthext ] ) . therefore , the condition @xmath318 requires @xmath323 combining this with the condition that the atmosphere have a thin shell , eq . ( [ condition2 ] ) , we find that the allowed range for which the see satellite will _ not _ have a thin shell is @xmath324 this constitutes a wide region in parameter space . following the analysis of sec . [ condmass ] , for a potential @xmath18 with @xmath325 and @xmath326 , this requires the range of interactions in the atmosphere , @xmath196 , to be anywhere between @xmath327 m and 1 mm . therefore , if the inequality in eq . ( [ noscreen ] ) is satisfied , the scalar field @xmath10 is only slightly perturbed by the satellite , and thus the field value within the capsule is @xmath328 . moreover , since @xmath244 is much larger than the size of the satellite , @xmath10 behaves as a massless field . hence , if the test bodies have mass @xmath329 and coupling @xmath8 , @xmath330 , they will experience a total force , gravitational plus @xmath10-mediated , given by @xmath331 where we have restored newton s constant @xmath332 . this implies , therefore , that the see satellite will measure an effective newton s constant , @xmath333 , _ which differs by order unity from newton s constant measured on earth_. since this is an important prediction , let us summarize how it was obtained . first of all , the key ingredient is that the see experiment is performed in orbit , where the range of @xmath10-mediated interactions is greater than 10 au . thus , small perturbations in @xmath10 are essentially massless on the scale of the capsule , and therefore the force law between two test bodies is proportional to @xmath268 to a very good approximation . a second crucial condition is that there is no thin - shell effect within the satellite , as expressed in eq . ( [ noscreen ] ) . this ensures that there is no suppression of the fifth force , and the coupling strength is therefore of order @xmath334 . hence , for @xmath46 of order unity , the effective newton s constant receives large corrections from the fifth force mediated by @xmath10 . as mentioned earlier , such an outcome would constitute strong evidence that a mechanism of the kind we are proposing is at play in nature . our model also has surprising implications for satellite experiments that aim at measuring ep violations in orbit , such as microscope , gg and step . these experiments will attempt to measure a difference in free - fall acceleration between two concentric cylinders of different composition with expected sensitivity of one part in @xmath335 , @xmath336 and @xmath243 , respectively . we focus on the step satellite for concreteness . although eq . ( [ noscreen ] ) pertains to the see capsule , a similar condition is obtained for step , since its physical characteristics are not too different from those of see . hence , there is no thin - shell effect if eq . ( [ noscreen ] ) holds , and the @xmath10 profile within the satellite is well - approximated by eq . ( [ earthext ] ) . it is then straightforward to show that the e@xmath5tv@xmath5s parameter for be and nb test cylinders , as appropriate for step , is given by @xmath337 combining with eq . ( [ atmgg ] ) , we find the allowed range @xmath338 which falls almost entirely within the range of sensitivity of step . moreover , we see that @xmath46 can be larger than @xmath17 , the current bound from experiments performed on earth @xcite . in other words , it is possible that microscope , gg and step will measure violations of ep stronger than currently thought to be allowed by laboratory measurements . this is another striking manifestation of the fact that scalar field dynamics in our model are very different on earth than in orbit . in this paper , we have presented a novel scenario in which both the strength and the range of interactions mediated by a scalar field @xmath10 depend sensitively on the surrounding environment . in a region of high density , such as on earth , the mass of the scalar field is sufficiently large , typically of order 1 @xmath339 , to evade constraints on ep violation from laboratory experiments . meanwhile , the field can be essentially free on solar system scales , with a typical compton wavelength of 100 au . we have argued that our scenario satisfies all existing bounds from ground - based and solar - system tests of gravity . it will be left for future work to study our model in the context of strongly gravitating systems , such as black holes @xcite . implications for cosmology will be analyzed in a separate paper @xcite , where we will show in detail that chameleons are consistent with cosmological constraints on the existence of non - minimally coupled scalars , such as the bound on the time - variation of @xmath316 from nucleosynthesis , for example . moreover , we will describe how the field dynamically reaches the minimum of the effective potential , an element that was assumed _ a priori _ in the present work , and show how our scenario naturally provides a solution to the old moduli problem @xcite . the striking difference in interaction range in the solar system versus on earth can lead to unexpected outcomes for experiments that test gravity in space . it was argued that the see project could measure an effective newton s constant drastically different than that on earth . meanwhile , the microscope , gg and step satellites could detect violations of the ep stronger than currently allowed by ground - based experiments . either outcome would effectively constitute a proof of the existence of chameleons . more importantly , these possible surprises for space experiments will hopefully strengthen the case for launching these satellites . also , the door is open for conceiving of new table - top experiments to probe for the existence of chameleons , for instance by exploiting the fact that the properties of chameleons are sensitive to the surrounding matter density . for example , one could imagine doing atomic physics experiments inside or in the vicinity of a massive oscillating shell of matter . these oscillations would induce a time - dependence in the profile of the chameleon and thus in the frequency of emission lines @xcite . it would also be interesting to investigate the behavior of the field at intermediate altitude , for instance at about 40 km where balloon experiments take place . this would require a more realistic modeling of the atmosphere than presented here . such an analysis could reveal that it may be possible to detect chameleons by performing gravity experiments in balloons @xcite . we thank n. arkani - hamed , j.r . bond , r. brandenberger , p. brax , c. van de bruck , s. carroll , t. damour , a .- c . davis , g. esposito - farse , g. gibbons , b. greene , d. kabat , a. lukas , j. murugan , b.a . ovrut , m. parikh , s .- j . rey , k. schaalm , c.l . steinhardt , n. turok , c.m . will , t. wiseman , and especially n. kaloper and p.j . steinhardt for insightful discussions . this work was supported by the columbia university academic quality fund , the ohrstrom foundation ( jk ) , doe grant de - fg02 - 92er40699 and the university of cape town ( aw ) . for a review of experimental tests of the equivalence principle and general relativity , will , _ theory and experiment in gravitational physics _ , 2nd ed . , ( basic books / perseus group , new york , 1993 ) ; c.m . will , living rev . rel . * 4 * , 4 ( 2001 ) . t. barreiro , b. de carlos and e.j . copeland , phys . d * 57 * , 7354 ( 1998 ) ; p. bintruy , m.k . gaillard and y .- y . wu , phys . b * 412 * , 288 ( 1997 ) ; nucl . b493 * , 27 ( 1997 ) ; j.d . barrow , phys . b * 235 * , 40 ( 1990 ) .

this paper describes a multiwavelengh optical study of chromospheres in two x - ray / euv selected active binary stars with strong h emission , and . the goal of the study is to determine radial velocities and fundamental stellar parameters in chromospherically active binary systems in order to include them in the activity - rotation and activity - age relations . we carried out high resolution echelle spectroscopic observations and applied spectral subtraction technique in order to measure emission excesses due to chromosphere . the detailed study of activity indicators allowed us to characterize the presence of different chromospheric features in these systems and enabled to include them in a larger activity - rotation survey . we computed radial velocities of the systems using cross correlation with the radial velocity standards . the double - line spectral binarity was confirmed and the orbital solutions improved for both systems . in addition , other stellar parameters such as : spectral types , projected rotational velocities ( ) , and the equivalent width of the lithium li i.8 absorption line were determined .

0901.1634

why each particular star reaches a certain level of activity , and why seemingly similar stars may have drastically different chromospheres ? it is believed that differential rotation and convection are the key factors responsible for the magnetic activity in the star . a significant fraction of stellar population , especially at later evolutionary stages , is found in close binaries , where tidal forces and mass transfer become additional factors controlling rotation and atmospheric properties of companions . in order to understand the physical origin of activity , the time - scale of variability , and to be able to extrapolate to other systems , we initiated a multiwavelengh optical survey of a large sample of binaries with various levels of activity . we carried out the study of the chromosphere of active binary systems using the information provided by several optical and near infrared spectroscopic features that are formed at different heights in the chromosphere ( see montes et al . 1997 , 1998 , 2000 ; glvez et al . 2002 , 2007 ) , and applied it to several known and newly discovered binaries , most of which are by dra or rs cvn type with various levels of activity . the ultimate goal of the project is to establish a reliable activity - rotation relation in binary stars . to achieve high accuracy one needs to investigate all possible causes of activity for each particular system and for each component in the system . it is essential to take into consideration dispersion and signal - to - noise of the spectral data obtained with different instrumental set - ups . furthermore , the study of binaries as compared to single stars is complicated by the interaction between companions , line blending ( sb2 case ) , reflected light , etc . active stars , which are recognized by the presence of emission lines in one or both components , present the most challenging case and need to be studied on the individual basis . to obtain an accurate orbital solution and to study the behavior of activity , one needs to separate radial velocities , equivalent widths ( @xmath3 hereafter ) of emission lines , and fluxes between two components . it is also vital to carry out a precise stellar classification due to the importance of the subgiant and giant presence in the activity - rotation relation ( e.g. in rs cvn case ) . here we concentrate on two recently discovered , x - ray / euv selected , chromospherically active double - line spectroscopic binaries included in this survey : v789 mon ( 2re j0725 - 002 , bd-00 1712 ) and gz leo ( 2re j1101 + 223 , hd 95559 , bd+23 2297 , hip 53923 ) . v789 mon , @xmath4 , was classified by jeffries et al . ( 1995 ) as a double - line spectroscopic binary ( sb2 ) with almost identical k5v components . they gave a rotational velocity , @xmath1 , of 25 kms@xmath5 and an orbital period of 1.40 days . the photometric period ( 1.412 days ) reported by cutispoto et al . ( 1999 ) and robb & gladders ( 1996 ) indicates synchronous rotation . jeffries et al . ( 1995 ) suggest that this system could be eclipsing , but robb & gladders ( 1996 ) did not find any evidence of eclipses in their photometric observations . cutispoto et al . ( 1999 ) derived a k3v + m0:v classification for this system from colors and spectral signatures . regarding activity , jeffries et al . ( 1995 ) detected the h@xmath0 line in emission above the continuum level for both components . gz leo is a sb2 binary with @xmath6 mag . jeffries et al . ( 1995 ) obtained a first orbital solution with a period of 1.528 days using eleven radial velocity measurements . later , fekel & henry ( 2000 ) reported an improved orbital solution using nine new radial velocity determinations and one additional value given by strassmeier et al . the system consists of two similar rapidly rotating stars ( @xmath1 @xmath7 30 kms@xmath5 ) . the spectral type of the components is not well established . jeffries et al . ( 1995 ) classified it as g5 while popper ( 1996 ) as k2 , based on the strength of the na i d@xmath8 & d@xmath9 lines . the latter is closer to the k1v type determined by fekel & henry ( 2000 ) from the temperature and luminosity - sensitive line ratios in the 6430 - 6465 region . popper ( 1996 ) reported photometric variability , but could not confirm whether it is due to eclipses . strassmeier et al . ( 2000 ) determined a period of 2.944 days from their photometric observations , but noted that a period of 1.514 days was also possible . finally , fekel & henry ( 2000 ) and pandey et al . ( 2002 , 2005 ) confirmed that the photometric period ( 1.526 days ) is nearly identical to its orbital period . this binary system shows the typical signatures of chromospheric activity . a filled - in h@xmath0 absorption line in both components was reported by jeffries et al . ( 1995 ) , while ca ii h & k emissions were detected in the spectra by strassmeier et al . in addition , gz leo is a young system as indicated by its large lithium abundance ( strassmeier et al . 2000 ; fekel & henry 2000 ) . a preliminary analysis of these two systems can be found in glvez et al . ( 2003 ) and glvez ( 2005 ) . in @xmath10 2 we summarize our observations . in @xmath10 3 we derive stellar parameters and the orbit elements of the systems . the behavior of the different chromospheric activity indicators is described in @xmath10 4 . finally @xmath10 5 gives a summary . in this paper we analyzed high resolution echelle spectra of both systems that we obtained during two observing runs detailed in table [ tab : reduc1 ] . the spectra were extracted using the standard reduction procedures in the iraf echelle package ( bias subtraction , flat - field division and optimal extraction of the spectra ) . we obtained the wavelength calibration by taking spectra of a th - ar lamp . finally , we normalized the spectra by a polynomial fit to the observed continuum . we obtained nine spectra of v789 mon in one observing run and fifteen spectra of gz leo in two . the signal to noise ratio ( @xmath11 ) achieved in the h@xmath0 line region in all the spectra is @xmath7 60 for v789 mon and @xmath7 100 for gz leo . we list the adopted stellar parameters of both systems in table [ tab : par ] . spectral type and the photometric data ( @xmath12 , @xmath13 , @xmath14 ) are taken from simbad and cutispoto et al . ( 1999 ) for v789 mon and from simbad and fekel & henry ( 2000 ) for gz leo . the astrometric data ( parallax , @xmath15 ; proper motions , @xmath16@xmath17@xmath18 and @xmath16@xmath19 ) are from @xmath20 ( esa 1997 ) and @xmath21 ( h@xmath22 g et al . 2000 ) catalogues . for v789 mon no parallax value is available , so we give the spectroscopic parallax determined here ( see @xmath10 3.4 ) . finally , the orbital period ( @xmath23 ) and the projected rotational velocities ( @xmath1 ) have been determined in this paper as well ( see below ) . in order to obtain the spectral type of these binary systems we compared our high resolution echelle spectra with spectra of inactive reference stars of different spectral types and luminosity classes observed during the same observing run . to avoid contamination by chromospheric emission , we only used spectral orders that are free of lines sensitive to chromospheric activity . this analysis makes use of the program starmod developed at penn state university ( barden 1985 ) and modified by us . starmod allows to rotationally broaden spectra and shift them in the velocity space . we ran it on the spectra of the reference stars and combined with the appropriate weights to create composite spectra that were used as template for comparison with our targets . for v789 mon we obtained the best fit between observed and synthetic spectra using a k5v reference star ( ) for both components , contributing respectively 60% and 40% of the total flux in the continuum . the fits were carried out in all the phases where the spectra of the components could be separated . our spectral classification is in agreement with a k5v type reported by jeffries et al . ( 1995 ) , while the relative contribution of the companions is compatible with cutispoto et al . ( 1999 ) result that assigned a later spectral type for the secondary component . in the case of gz leo , we obtained the best fit between observed and synthetic spectra using a k0v reference star ( ) for both components , with the equal contribution from each component . similarly to v789 mon , the fits were performed in all the phases where the components could be separated , and several spectral types and luminosity class reference stars were used . our spectral classification is very close to the k2v type reported by popper ( 1996 ) and the k1v type by fekel & henry ( 2000 ) . jeffries et al . ( 1995 ) estimated the projected rotational velocity ( @xmath1 ) for v789 mon as 25kms@xmath5 . for gz leo they found 27 kms@xmath5 , while fekel & henry ( 2000 ) reported values of 32.4 and 31.6 kms@xmath5 for each component . fekel & henry ( 2000 ) noted that the actual @xmath1 may slightly smaller , because spots cause the lines to be shallower than in an unspotted star , making fwhm appear larger . strassmeier et al . ( 2000 ) report similar values of 31 and 26 kms@xmath5 respectively . using starmod we obtained for each observing run @xmath1 values about 30 kms@xmath5 for both components of v789 mon and gz leo . in order to determine a more accurate rotational velocities we applied a cross - correlation technique to our high resolution echelle spectra of these stars using iraf task fxcor . the method is described in detail in our previous papers ( see glvez et al . 2002 ; lpez - santiago et al . 2003 ) . in short , the method is based on the fact that when a stellar spectrum with rotationally broadened lines is cross - correlated against a narrow - line spectrum , the width of the cross - correlation function ( ccf ) will depend on the amount of rotational broadening of the former spectrum . for v789 mon , we used as a template a slowly rotating k5v star hr 1614 . we obtained a @xmath1 of [email protected] and [email protected] kms@xmath5 for the primary and the secondary component respectively . in the case of gz leo , we used a k0v star hr 166 and a k2v star as templates , in mcdonald and foces observing runs respectively . we obtained averaged values of [email protected] and [email protected] kms@xmath5 for each component ( see table [ tab : par ] ) . heliocentric radial velocities were also obtained by using the cross - correlation technique ( see e.g. glvez et al . the spectra of the target were cross - correlated order by order , using the routine fxcor in iraf , against spectra of radial velocity standards with similar spectral type taken from beavers et al . we derived the radial velocity for each order from the position of peak of the cross - correlation function ( ccf ) and calculated the uncertainties based on the fitted peak height and the antisymmetric noise as described by tonry & davis ( 1979 ) . since both systems are sb2 , we could see two peaks in the ccf associated with two components ; we fitted each peak separately . when the component peaks were too close , we used de - blending technique . in the tables [ tab : bvr ] and [ tab : vr ] we list , for each spectrum , the heliocentric radial velocities ( @xmath25 ) and their associated errors ( @xmath26 ) . the latter are obtained as weighted means of the individual values deduced for each order in v789 mon and gz leo spectra . we note that the uncertainties returned by fxcor for sb2 binaries are overestimated ; when fitting each component , the presence of the other component will increase the antisymmetric noise , thereby biasing the error . to compute the orbital solution for these systems we combined our radial velocity data ( for both components ) with the data of jeffries et al . ( 1995 ) , fekel @xmath27 henry ( 2000 ) and strassmeier et al . ( 2000 ) , see tables [ tab : bvr ] and [ tab : vr ] . the radial velocity data are plotted in fig . [ fig : orb2n ] for v789 mon and gz leo on left and right side respectively . solid symbols represent the primary and open symbols represent the secondary . the curves represent a minimum @xmath28 fit orbit solution . the orbit fitting code uses the _ numerical recipes _ ( press et al . 1986 ) implementation of the levenberg - marquardt method of fitting a non - linear function to the data and weights each datum according to its associated uncertainty ( see e.g. glvez et al . 2002 , 2007 for further details ) . we found a nearly circular ( @xmath29 ) for v789 mon system , with an orbital period of 1.4021 days which confirms it as a synchronous system ( @xmath14 @xmath7 1.412 days , cutispoto et al . the mass ratio ( @xmath30 ) indicates that both components are very similar . the obtained parameters are in agreement with jeffries et al . see table [ tab : orb2 t ] for details . we arrived at a similar conclusions for gz leo : a nearly circular orbit ( @xmath31 ) with an orbital period of 1.5260 days , which is very similar to the photometric rotational period ( @xmath14 @xmath7 1.5264 days , fekel & henry 2000 ) , indicating a synchronized rotation . the obtained minimum masses ( @xmath32 ) and the mass ratio ( @xmath33 ) are in agreement with fekel & henry ( 2000 ) . see table [ tab : orbt ] for computed parameters . adopting spectral types derived in @xmath10 3.1 for the primary components and adopting masses and temperatures from landolt - brnstein tables ( schmidt - kaler 1982 ) , we derived masses and spectral types for the secondaries . we also estimated minimum radii ( @xmath34 ) , luminosities and minimum masses ( @xmath35 ) of the components using photometric periods ( from the literature , see @xmath10 3 . ) and rotational velocities ( from @xmath10 3.2 ) . the results , given in table [ tab : parb ] , confirm our previous classification . furthermore , since @xmath20 parallax of v789 mon is not known , we calculated a spectroscopic parallax . using bolometric magnitude of a k5v star ( + 6.7@xmath36 ) , the bolometric correction ( @xmath37 ) corresponding to a k5v , and the luminosity ratio , we obtain the bolometric and the visual total absolute magnitudes for the system . comparing the total visual absolute magnitude ( 6.76@xmath36 ) , with the measured v magnitude ( 9.33@xmath36 ) , we derive a distance of @xmath38 @xmath240.13 pc ( @xmath39 @xmath240.22 mas ) for v789 mon . due to possible activity - induced variations in the v , we expect a 10% additional error in this distance measure . since the system is relatively close , the interstellar reddening is neglected . using the total luminosity based on the measured distance ( 54.26 pc ) for gz leo , we obtained the total magnitude for the system @xmath40 . the difference between the observed @xmath13 magnitude ( 8.92 ) and the calculated one can be explained by the presence of photospheric cool spots over the stellar surfaces of both components . we computed the galactic space - velocity components ( @xmath41 , @xmath13 , @xmath42 ) of these systems using as radial velocity the center of mass velocity ( @xmath43 ) and accurate proper motions and parallax taken from @xmath20 ( esa 1997 ) and @xmath21 ( h@xmath22 g et al . 2000 ) catalogues ( see table [ tab : par ] ) , except for spectroscopic parallax of v789 mon that has been calculated here . the obtained values and associated errors are given in table [ tab : uvw ] . the velocity components lie in the ( @xmath41 , @xmath13 ) diagram obviously inside the young disk population boundaries ( eggen 1984 , 1989 ; montes et al . 2001a , 2001b ) indicating that both systems are young disk stars . in addition , the eggen kinematic criteria predict that gz leo may be a member of hyades supercluster ( see montes et al . 2001a and glvez 2005 ) . the resonance doublet of li i at @xmath26707.8 is a spectroscopic feature very important in the diagnostic of age in late - type stars , since it is destroyed easily by thermonuclear reactions in the stellar interior . the spectral region of this line is included in all our spectra for both systems . at the spectral resolution of our spectra and with the rotational velocity , we have determined for the components of both binaries , the li i line is blended with the nearby fe i @xmath26707.41 line . in the case of v789 mon the li i line is very weak and we could not measure the @xmath3 with the required precision . for gz leo , we clearly see the li i+fe i absorption feature in all the spectra ( see fig . [ fig : li ] left side ) , but due to the blending with other photospheric lines of both components , we have only measured the @xmath3 in the spectrum of the third night of the foces02 observing run ( see fig . [ fig : li ] right ) which is very close to conjunction ( @xmath44 ) . the obtained value @xmath3(li i + fe i ) = 60.7 m contain the contribution from both components , but taken into account that they have the same spectral type and contribution to the continuum we can assume that this is also the @xmath3(li i+fe i ) of each component . we have corrected this total @xmath3 by subtracting the @xmath3(fe i ) = 19.8 m calculated from the empirical relationship with ( @xmath45@xmath13 ) given by favata et al . ( 1993 ) resulting @xmath3(li i ) = 40.9 m . one additional estimation of the @xmath3(li i ) in this spectrum can be obtained by using the spectral subtraction technique . in the subtracted spectrum the contribution from the photospheric lines is eliminated , therefore the @xmath3 = 51.2 m measured in this spectrum is the total @xmath3(li i ) and , as both components are equal , also the @xmath3(li i ) of each one . the difference in the @xmath3 obtained from both methods is likely due to the influence of the stellar metallicity in the calibration used in the first one to obtain the @xmath3(fe i ) and in the standard star used as reference in the subtraction technique that are not taken into account here . the @xmath3(li i ) of each component we have determined ( 40.9 m with the first method and 51.2 m with the second one ) is smaller than the 63 m estimated by fekel & henry ( 2000 ) and the 60 m given by strassmeier et al . the small differences between these values can be attributed to the different method used by these authors to estimate the fe i contribution and the error in the @xmath3 determination . this small @xmath3(li i ) is in agreement with the values observed in hyades cluster stars of this spectral type ( see montes et al . 2001b ) and with the membership to the hyades supercluster ( see @xmath10 3.5 ) . that situated the system age in about 600 myr the echelle spectra analyzed in this work allowed us to study the behavior of the different chromospheric activity indicators , from the ca ii h & k to the ca ii irt lines , which are formed at different atmospheric heights . the chromospheric contribution in these features were determined by using the spectral subtraction technique described in detail by montes et al . ( 1995 , 1997 , 1998 , 2000 ) and glvez et al . ( 2002 , 2007 ) . the excess emission in @xmath3 is measured in the subtracted spectra , after correcting for the relative contribution of each component in the total continuum . this contribution is determined by using the radii and temperatures adopted in @xmath10 3 . for instance , in the h@xmath0 line region the relative contributions were @xmath46 for the primary component and @xmath47 for the secondary component in v789 mon and for gz leo we had the 0.50 contributions for both components . in table [ tab : ew ] we give the corrected @xmath3 for the h@xmath0 and ca ii irt lines for v789 mon . in table [ tab : ewf ] we list corrected @xmath3 for the ca ii h & k , h@xmath48 , h@xmath49 , h@xmath43 , h@xmath50 , h@xmath0 , and ca ii irt ( @xmath28498 , @xmath28542 , @xmath28662 ) lines for gz leo in foces run and for h@xmath0 and ca ii irt in mcdonald run . the corrected @xmath3s is given separately for each component ( p / s ) , unless the lines of both components overlap , in which case the combined @xmath3 is given . the final estimated errors in @xmath3 are in the range 10 - 20% ( see e.g. glvez et al . 2007 for details ) . finally , corrected @xmath3s were converted to the absolute surface fluxes by using the empirical stellar flux scales calibrated by hall ( 1996 ) as a function of the star color index . in our case , we used the @xmath12 index and the corresponding coefficients for ca ii h & k , h@xmath0 and ca ii irt . for h@xmath48 we used the same coefficients as for ca ii h & k , while for h@xmath49 , h@xmath43 and h@xmath50 the coefficients were obtained by interpolating between the values for ca ii h & k and h@xmath0 . in tables [ tab : ew ] and [ tab : fluxf ] , we give the logarithm of the calculated absolute flux at the stellar surface ( log@xmath51@xmath52 ) for different chromospheric activity indicators . fig : ha3 ] - [ fig : irt ] show a representative regions around the h@xmath0 , ca ii h & k and ca ii irt @xmath2@xmath53 , @xmath2@xmath54 lines . the observed ( solid line ) and the synthesized spectra ( dashed line ) are shown in the left panel , while subtracted spectrum ( dotted line ) in the right . the observing run and the orbital phase ( @xmath55 ) of each spectrum are also shown in these figures . in fig . [ fig : hbgd ] we plot a representative subtracted spectra in h@xmath50 , h@xmath43 and h@xmath49 region for gz leo . in v789 mon h@xmath0 is observed in emission above the continuum in all the observed spectra ( see fig . [ fig : ha3 ] , left panel of the left side ) . after applying the spectral subtraction technique , one can see that h@xmath0 emission from the primary and secondary components looks very much alike . in all the spectra , except one which is very close to conjunction , we measured the emission coming from both components by using a two - gaussian fit to the subtracted spectra ( see fig . [ fig : ha3 ] , right panel of the left side ) . the combined @xmath3 is given when it was not possible to deblend the two components . the persistence of h@xmath0 emission in v789 mon spectra indicates that it is a very active binary system similar to some rs cvn and by dra systems , which always show h@xmath0 emission above the continuum . the @xmath3 measured in the subtracted spectra by gaussian fitting gives an average value of @xmath3(h@xmath0 ) = 0.65/0.62 for primary and secondary components ( which is higher than reported by jeffries et al . ( 1995 ) , @xmath3(h@xmath0 ) = 0.09/0.25 ) . in gz leo , we analyzed 15 spectra including the h@xmath0 line region . in the first run ( mcdonald98 ) , the h@xmath0 line appears completely filled by emission in both components ( see fig . [ fig : ha3 ] , left panel of the right side ) . as in the case of v789 mon , the emission from both components is nearly identical and was measured by applying a two - gaussian fit to the subtracted spectra ( see fig . [ fig : ha3 ] , right panel of the right side ) . the spectrum at phase 0.46 is close to conjunction , so the measured @xmath3 given in table [ tab : ewf ] is the combined @xmath3 . the mean values of @xmath3 are @xmath56(h@xmath0 ) = 0.32 and @xmath57(h@xmath0 ) = 0.36 . we note however that a variation with phase is observed in both components , with a remarkable tendency to behave in the opposite way for ca ii irt @xmath3s ; unfortunately the phase coverage is not enough for a reliable conclusion ( table [ tab : ewf ] ) . when we analyze the foces02 run , we see again that the h@xmath0 line is completely filled - in by emission for both components . after fitting the subtracted spectra we measured the excess emission @xmath56(h@xmath0 ) = 0.46 and @xmath57(h@xmath0 ) = 0.47 , i.e. , a 35% increase compared to mcdonald98 run . in gz leo system the secondary star usually shows slightly larger values of h@xmath0 @xmath3 and no variation with phase was detected ( see table [ tab : ewf ] ) . the other three balmer lines ( h@xmath50 , h@xmath43 and h@xmath49 ) were included only in some of our spectra for gz leo during the foces02 . the absorption lines were clearly filled - in with emission in the observed spectra . after applying the spectral subtraction , we could extract the excess emission from both components ( see representative spectra in fig . [ fig : hbgd ] ) . when the @xmath11 was high enough we deblended the emission coming from both components by applying a two - gaussian fit to the subtracted spectra ( see table [ tab : ewf ] ) . these three lines show similar variations with orbital phase to the h@xmath0 line in both components , their mean values are @xmath3(h@xmath50 ) = 0.13/0.13 , @xmath3(h@xmath43 ) = 0.09/0.09 and @xmath3(h@xmath49 ) = 0.07/0.09 . we also measured the ratio of excess emission @xmath3 in the h@xmath0 and h@xmath50 lines , @xmath58 , and the ratio of excess emission @xmath59 with the correction : @xmath60 from hall & ramsey ( 1992 ) . it takes into account the absolute flux density in these lines and the color difference in the components ( @xmath61=1.45 for a k0v star ) . we obtained a mean values of @xmath59 @xmath62 for the primary and @xmath63 for secondary . these values indicate , according to buzasi ( 1989 ) and hall & ramsey ( 1992 ) the presence of prominence - like material at the stellar surface in both components of the system . the ca ii h & k line region was included only in the spectra of the foces run for gz leo . in these spectra we observed strong emission in the ca ii h & k lines and a clear emission in the h@xmath48 line arising from both components ( see fig . [ fig : hyk ] ) . in our spectra , the ca ii h & k lines are located at the end of the echellogram , where the efficiency of the spectrograph and the ccd decrease very rapidly , and therefore the @xmath11 ratio obtained is very low , and the normalization of the spectra are very difficult . in spite of this we could apply the spectral subtraction in this region , see fig . [ fig : hyk ] . as we can see in this figure , the h@xmath48 line arising from one of the component overlaps at some orbital phases with the ca ii h line from the other component , so their @xmath3 were measured with a gaussian fit when it was possible ( see comments for each spectrum in the footnotes of table [ tab : ewf ] ) . the mean measured @xmath3s in these spectra are @xmath3= 1.13/0.84 for each component in the ca ii k line and @xmath3= 0.82/0.65 for each component in the ca ii h line . an increase of the ca ii h & k emissions is observed with the orbital phase . the presence of an excess emission from h@xmath48 line ( @xmath3= 0.30/0.43 ) indicates that it is a very active system . the three lines of the ca ii infrared triplet ( irt ) were included in all our echelle spectra for both systems . in all of them we could observe a clear emission above the continuum in the core of the ca ii irt absorption lines ( see figs . [ fig : irt ] ) from both components . in v789 mon , ( fig . [ fig : irt ] left side ) , we measured an average @xmath3 of @xmath64 for the three lines , and we found small variations with the orbital phase that were anti - correlated with the balmer lines @xmath3 variations in both primary and secondary components . for gz leo binary , ( fig . [ fig : irt ] right side ) , we found that in foces run there were no significant variations with the orbital phase , but in mcdonald s one an evident orbital variation were observed which shows an anti - correlation with the @xmath3(h@xmath0 ) variation . we measured mean @xmath3s of @xmath65 for each component in mcdonald run and @xmath66 for both components in foces run . in addition , we calculated the ratio of excess emission @xmath3 , @xmath67 , which is also an indicator of the type of chromospheric structure that produces the observed emission . in solar plages , values of @xmath67 @xmath7 1.5 - 3 are measured , while in solar prominences the values are @xmath7 9 , the limit of an optically thin emitting plasma ( chester 1991 ) . in v789 mon , small values of the @xmath67 ratio were found , @xmath7 1.5 for the primary component and 1.4 for the secondary ( see table [ tab : ew ] ) . this indicates that the ca ii irt emission of this star arises from plage - like regions . in gz leo we found a @xmath67 value of @xmath7 2.4 and @xmath7 1.4 for primary and secondary in mcdonald run , but @xmath7 0.9 for primary and secondary in foces run . these values indicate that ca ii irt emission come from plage - like regions , in contrast to the balmer lines which seem to come from prominences . this markedly different behavior of the ca ii irt and the h@xmath0 emission has also been found in other chromospherically active binaries ( see arvalo & lzaro 1999 ; montes et al . 2000 ; glvez et al . 2003 ; glvez 2005 and references therein ) . the different behavior from one epoch to another is consistent with the different activity level measured in each observing run . the lack of variation with orbital phase of the gz leo emission lines in foces run could be due to that the hight level of activity is reflecting a large number of active regions maybe distributed along all surface . such , could produce an even spotted - plaged surface that would avoid to see the modulation clearly . in this paper we present a detailed spectroscopic analysis of two x - ray / euv selected chromospherically active binary systems v789 mon ( 2re j0725 - 002 ) and gz leo ( 2re j1101 + 223 ) . we analyzed high resolution echelle spectra that include the optical chromospheric activity indicators from the ca ii h & k to ca ii irt lines , as well as the li i @xmath26707.8 line and other photospheric lines of interest . using a large number of radial velocities we improved the orbital parameters of these systems . we obtained that both systems have nearly circular orbits and have an orbital period very close to photometric period , indicating that they have a synchronous rotation . the spectral classification derived using the comparison with spectra of reference stars , two k5v components for v789 mon and two k0v components for gz leo is in good agreement with the classification obtained from physical parameters ( @xmath34 ) . as both components are main sequence stars , we can classify these chromospherically active binaries as by@xmath68dra systems ( fekel et al . 1986 ) . by using the information provided by the width of the ccf we determined a projected rotational velocity , @xmath1 , of 28.28 and 25.09 kms@xmath5 for the primary and secondary components of v789 mon , and 26.23 and 26.98 kms@xmath5 for the components of gz leo . the presence of the li i line in both systems is in agreement with the kinematics results , i.e. , they belong to the young disk population . also , with a @xmath3(li i ) @xmath7 40 m for each component , gz leo could be a member of the hyades supercluster . we analyzed , using the spectral subtraction technique , all the optical chromospheric activity indicators . in v789 mon , both components show high levels of chromospheric activity . the emission in the h@xmath0 line is above the continuum and it is very similar in the two components , showing no variations with phase . gz leo shows a high level of chromospheric activity too . after applying spectral subtraction the chromospheric excess emission from both components are certainly detected in all the activity indicators during the observing runs . the ca ii h & k lines are in emission and h@xmath48 is also detected in emission for both systems . the level of chromospheric emission change , i.e. , we have observed a 35% increment in the @xmath3 of h@xmath0 emission line and a nearly 30% increment in the @xmath3 of ca ii irt emission lines from one epoch to another . furthermore , the variation of h@xmath0 and ca ii irt emission with phase is anti - correlated and the ratios @xmath69 and @xmath67 indicate that the emission of the balmer lines would arise from prominence - like material , and that the emission of ca ii irt lines arise from plage - like regions . we would like to thank dr . ramsey ( pennsylvania state university ) for collaborating in the mcdonald observing run ( 2.1 m telescope , texas , usa ) and the staff of mcdonald observatory for their allocation of observing time and their assistance with our observations . thanks to dr . nadya gorlova and dr . joern rossa for their great help on the english redaction and corrections . this work was supported the spanish `` programa nacional de astronoma y astrofsica '' under grants aya2005 - 02750 and aya2008 - 00695 , and the `` comunidad de madrid '' under pricit project s-0505/esp-0237 ( astrocam ) . glvez , m. c. , montes , d. , ramsey , l. w. , fernndez - figueroa , m. j. , de castro , e. , & cornide , m. 2003 , proceedings of the v scientific meeting of the spanish astronomical society ( sea ) , `` highlights of spanish astrophysics iii '' kluwer academic publishers ( printed in netherlands ) , ed j. gallego , j. zamorano , n. cardiel , 474 ( li i ) . the observed and the synthesized spectrum are plotted at the top and the subtracted spectrum ( dotted line ) at the bottom . [ fig : li],title="fig:",width=264 ] ( li i ) . the observed and the synthesized spectrum are plotted at the top and the subtracted spectrum ( dotted line ) at the bottom . [ fig : li],title="fig:",width=238 ] line region in mcdonald run . the observed spectrum ( solid - line ) and the synthesized spectrum ( dashed - line ) are plotted in the left panel and the subtracted spectrum ( dotted line ) in the right panel . the position of the h@xmath0 line for the primary ( p ) and secondary ( s ) components are marked too and here we again have the two - gaussian fit superposed . [ fig : ha3],title="fig:",width=219 ] line region in mcdonald run . the observed spectrum ( solid - line ) and the synthesized spectrum ( dashed - line ) are plotted in the left panel and the subtracted spectrum ( dotted line ) in the right panel . the position of the h@xmath0 line for the primary ( p ) and secondary ( s ) components are marked too and here we again have the two - gaussian fit superposed . [ fig : ha3],title="fig:",width=219 ] llllcccc date & telescope & instrument & ccd chip & spect . range & orders & dispersion & fwhm + & & & & ( ) & & ( / pixel ) & ( ) + 12 - 21 01/1998 & 2.1 m os@xmath70 & sandiford & 1200x400 ccd & 6400 - 8800 & 32 & 0.06 - 0.08 & 0.13 - 0.20 + 22 - 25 04/2002 & 2.2m@xmath71 & foces & 2048x2048 24@xmath16 m site#1d & 3510 - 10700 & 112 & 0.04 - 0.13 & 0.10 - 0.40 + ll c c c c l l l r r name & @xmath72 & sb & @xmath12 & @xmath13 & @xmath73 & @xmath14 & _ v_sin_i_@xmath74 & @xmath15 & @xmath16@xmath17 cos @xmath49 & @xmath75@xmath19 + & & & & & ( days ) & ( days ) & ( kms@xmath5 ) & ( mas ) & ( mas / yr ) & ( mas / yr ) + v789 mon & k5v / k5v & 2 & 1.03 & 9.33 & 1.4021 & 1.412 & [email protected]/[email protected] & [email protected]@xmath74 & [email protected] & [email protected] + gz leo & k1v / k1v & 2 & 0.87 & 8.92 & 1.5260 & 1.526 & [email protected]/[email protected] & [email protected] & [email protected] & [email protected] + llrrrrrrrrrrrllrrrrrrrr obs . & hjd & @xmath11 & & + & 2400000 + & ( h@xmath0 ) & @xmath25 @xmath24 @xmath76 & @xmath25 @xmath24 @xmath76 + & & & ( kms@xmath5 ) & ( kms@xmath5 ) + jef(95)@xmath74 & 49054.400 & - & 57.8 @xmath24 2.0 & -46.1 @xmath24 2.0 + jef(95)@xmath74 & 49054.444 & - & 71.9 @xmath24 2.0 & -63.6 @xmath24 2.0 + jef(95)@xmath74 & 49054.456 & - & 81.6 @xmath24 2.0 & -69.8 @xmath24 2.0 + jef(95)@xmath74 & 49054.507 & - & 96.3 @xmath24 2.0 & -84.6 @xmath24 2.0 + jef(95)@xmath74 & 49054.555 & - & 103.2 @xmath24 2.0 & -92.2 @xmath24 2.0 + jef(95)@xmath74 & 49055.394 & - & -89.6 @xmath24 2.0 & 109.2 @xmath24 2.0 + jef(95)@xmath74 & 49055.451 & - & -79.5 @xmath24 2.0 & 102.6 @xmath24 2.0 + jef(95)@xmath74 & 49055.500 & - & -69.1 @xmath24 2.0 & 85.4 @xmath24 2.0 + jef(95)@xmath74 & 49056.449 & - & -21.2 @xmath24 2.0 & 32.4 @xmath24 2.0 + jef(95)@xmath74 & 49056.496 & - & -40.3 @xmath24 2.0 & 54.3 @xmath24 2.0 + jef(95)@xmath74 & 49056.572 & - & -68.1 @xmath24 2.0 & 91.1 @xmath24 2.0 + mcd98 & 50826.835 & 152 & 105.40 @xmath24 0.66 & -100.66 @xmath240.88 + mcd98 & 50828.837 & 76 & -80.61 @xmath24 0.66 & 93.65 @xmath240.93 + mcd98 & 50829.854 & 70 & 74.97 @xmath24 0.76 & -63.90 @xmath241.38 + mcd98 & 50830.848 & 59 & 61.17 @xmath24 0.67 & -51.52 @xmath240.86 + mcd98 & 50831.870 & 46 & -86.62 @xmath24 0.62 & 104.17 @xmath240.95 + mcd98 & 50833.846 & 58 & 101.92 @xmath24 0.67 & -100.26 @xmath240.83 + mcd98 & 50834.879 & 56 & -23.90 @xmath24 0.62 & 33.66 @xmath240.89 + mcd98 & 50835.844 & 47 & -72.60 @xmath24 0.66 & 85.18 @xmath241.00 + mcd98 & 50832.843 & 99 & 2.87 @xmath24 4.28 & 6.44 @xmath244.00 + llrrrrrrrrrrrllrrrrrrrr obs . & hjd & @xmath11 & & + & 2400000 + & ( h@xmath0 ) & @xmath25 @xmath24 @xmath76 & @xmath25 @xmath24 @xmath76 + & & & ( kms@xmath5 ) & ( kms@xmath5 ) + jef(95)@xmath77 & 49052.571 & - & 101.9 @xmath24 2.0 & -92.8 @xmath24 2.0 + jef(95)@xmath77 & 49052.664 & - & 78.5 @xmath24 2.0 & -68.5 @xmath24 2.0 + jef(95)@xmath77 & 49052.757 & - & 40.9 @xmath24 2.0 & -32.2 @xmath24 2.0 + jef(95)@xmath77 & 49054.600 & - & -87.2 @xmath24 2.0 & 95.4 @xmath24 2.0 + jef(95)@xmath77 & 49054.644 & - & -95.9 @xmath24 2.0 & 104.2 @xmath24 2.0 + jef(95)@xmath77 & 49055.389 & - & 100.7 @xmath24 2.0 & -93.2 @xmath24 2.0 + jef(95)@xmath77 & 49055.435 & - & 106.8 @xmath24 2.0 & -99.4 @xmath24 2.0 + jef(95)@xmath77 & 49055.611 & - & 106.2 @xmath24 2.0 & -94.5 @xmath24 2.0 + jef(95)@xmath77 & 49055.679 & - & 88.7 @xmath24 2.0 & -86.9 @xmath24 2.0 + jef(95)@xmath77 & 49055.727 & - & 72.8 @xmath24 2.0 & -65.8 @xmath24 2.0 + fekel(00)@xmath78 & 50200.709 & - & -103.4 @xmath24 2.0 & 109.1 @xmath24 2.0 + fekel(00)@xmath78 & 50400.035 & - & 101.0 @xmath24 2.0 & -92.3 @xmath24 2.0 + fekel(00)@xmath78 & 50401.049 & - & 5.4 @xmath24 2.0 & 5.4 @xmath24 2.0 + fekel(00)@xmath78 & 50576.681 & - & 60.9 @xmath24 2.0 & -52.9 @xmath24 2.0 + fekel(00)@xmath78 & 50632.664 & - & -103.9 @xmath24 2.0 & 111.4 @xmath24 2.0 + fekel(00)@xmath78 & 50831.054 & - & -104.0 @xmath24 2.0 & 111.6 @xmath24 2.0 + fekel(00)@xmath78 & 50831.947 & - & 86.4 @xmath24 2.0 & -78.6 @xmath24 2.0 + fekel(00)@xmath78 & 50927.773 & - & 96.4 @xmath24 2.0 & -88.6 @xmath24 2.0 + mcd98 & 50826.992 & 123 & -77.48 @xmath24 0.44 & 78.84 @xmath24 0.16 + mcd98 & 50827.975 & 133 & 112.42 @xmath24 0.75 & -105.11 @xmath24 0.78 + mcd98 & 50829.032 & 123 & -24.19 @xmath24 0.93 & 31.98 @xmath24 0.96 + mcd98 & 50830.019 & 116 & -64.94 @xmath24 1.10 & 72.53 @xmath24 1.02 + mcd98 & 50831.010 & 86 & 112.36 @xmath24 0.77 & -105.72 @xmath24 0.76 + mcd98 & 50832.024 & 92 & -46.47 @xmath24 0.96 & 52.29 @xmath24 0.90 + mcd98 & 50833.027 & 132 & -50.52 @xmath24 1.28 & 56.33 @xmath24 1.05 + mcd98 & 50834.024 & 100 & 110.18 @xmath24 0.92 & -106.47 @xmath24 0.92 + mcd98 & 50835.025 & 80 & -65.73 @xmath24 1.12 & 72.44 @xmath24 1.01 + mcd98 & 50836.017 & 97 & -24.11 @xmath24 1.26 & 26.04 @xmath24 1.20 + fekel(00)@xmath78 & 51240.763 & - & 110.2 @xmath24 2.0 & -103.7 @xmath24 2.0 + stras(00)@xmath79 & 51659.696 & - & -101.6 @xmath24 2.0 & 107.7 @xmath24 2.0 + foces02 & 52387.409 & 77 & -93.69 @xmath24 0.75 & 102.35 @xmath24 0.78 + foces02 & 52388.458 & 82 & 88.72 @xmath24 0.76 & -76.47 @xmath24 0.96 + foces02 & 52389.460 & 97 & 4.76 @xmath24 0.24 & 4.76 @xmath24 0.24 + foces02 & 52390.496 & 96 & -99.96 @xmath24 0.69 & 108.04 @xmath24 0.79 + foces02 & 52390.512 & 68 & -101.09 @xmath24 082 & 109.61 @xmath24 0.68 + + lccc element & value & uncertainty & units + @xmath23 & 1.4021 & 0.0000 & days + @xmath80 & 49054.980 & 0.048 & hjd + @xmath81 & 269.44 & 0.23 & degrees + @xmath82 & 0.0129 & 0.0024 & + @xmath83 & 101.09 & 0.47 & kms@xmath5 + @xmath84 & 106.57 & 0.69 & kms@xmath5 + @xmath43 & 5.29 & 0.17 & kms@xmath5 + @xmath85 & 1.0542 & 0.0047 & + + @xmath86 sin@xmath87 & 1.9490 & 0.0092 & 10@xmath88 km + @xmath89sin@xmath87 & 2.054 & 0.013 & 10@xmath88 km + @xmath90 sen@xmath87 & 4.003 & 0.016 & 10@xmath88 km + `` & 0.027 & & au + '' & 5.752 & & r@xmath91 + + @xmath92 sin@xmath93 & 0.6675 & 0.0085 & @xmath94 + @xmath95 sin@xmath93 & 0.6332 & 0.0081 & @xmath94 + f(@xmath96 ) & 0.1500 & 0.0021 & @xmath94 + lccc element & value & uncertainty & units + @xmath23 & 1.5260 & 0.0000 & days + @xmath80 & 49051.312 & 0.080 & hjd + @xmath81 & 359.33 & 18.98 & degrees + @xmath82 & 0.0073 & 0.0025 & + @xmath83 & 108.92 & 0.48 & kms@xmath5 + @xmath84 & 110.43 & 0.58 & kms@xmath5 + @xmath43 & 2.93 & 0.12 & kms@xmath5 + @xmath85 & 1.0139 & 0.0029 & + + @xmath86 sin@xmath87 & 2.285 & 0.010 & 10@xmath88 km + @xmath89sin@xmath87 & 2.317 & 0.012 & 10@xmath88 km + @xmath90 sen@xmath87 & 4.603 & 0.016 & 10@xmath88 km + `` & 0.031 & & au + '' & 6.613 & & r@xmath91 + + @xmath92 sin@xmath93 & 0.8400 & 0.0091 & @xmath94 + @xmath95 sin@xmath93 & 0.8285 & 0.0090 & @xmath94 + f(@xmath96 ) & 0.2043 & 0.0027 & @xmath94 + llcclccccccc name & @xmath97 & @xmath98 & @xmath95 & @xmath99 & @xmath100 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & @xmath106 + & & ( @xmath94 ) & ( @xmath94 ) & & ( @xmath107 ) & ( @xmath107 ) & ( k ) & ( @xmath108 ) & ( @xmath108 ) & ( @xmath94 ) & ( @xmath94 ) + v789 mon & k5v & 0.67 & 0.63 & k6-k7v & [email protected] & [email protected] & 4350 & 0.20 & 0.16 & 0.56 & 0.52 + gz leo & k0v & 0.79 & 0.78 & k0v & [email protected] & [email protected] & 5250 & 0.42 & 0.54 & 0.73 & 0.75 + lllll name & @xmath109 & @xmath110 & @xmath111 & @xmath112 + & ( kms@xmath5 ) & ( kms@xmath5 ) & ( kms@xmath5 ) & ( kms@xmath5 ) + v789 mon & [email protected] & [email protected] & [email protected] & 36.53 + gz leo & [email protected] & [email protected] & [email protected] & 10.37 + ccccccccccccccccccc @xmath55 & & + & & & & + & h@xmath0 & @xmath28498 & @xmath28542 & @xmath28662 & h@xmath0 & @xmath28498 & @xmath28542 & @xmath28662 + 0.71 & 0.71/0.72 & 0.27/0.31 & 0.42/0.29 & 0.33/0.29 & 6.83/7.02 & 6.27/6.45 & 6.46/6.42 & 6.35/6.42 + 0.14 & 0.62/0.58 & 0.19/0.22 & 0.33/0.53 & 0.13/0.36 & 6.77/6.93 & 6.11/6.23 & 6.35/6.68 & 5.95/6.51 + 0.87 & 0.48/0.58 & - & - & - & 6.66/6.93 & - & - & - + 0.58 & 0.70/0.65 & 0.30/0.25 & 0.34/0.39 & 0.34/0.28 & 6.82/6.98 & 6.31/6.35 & 6.36/6.55 & 6.36/6.40 + 0.31 & 0.66/0.60 & 0.35/0.34 & 0.32/0.33 & 0.38/0.37 & 6.80/6.94 & 6.38/6.49 & 6.34/6.47 & 6.41/6.52 + 0.00 & 1.24@xmath77 & 0.39@xmath77 & 0.63@xmath77 & 0.52@xmath77 & 7.07@xmath77 & 6.42@xmath77 & 6.63@xmath77 & 6.55@xmath77 + 0.71 & 0.64/0.62 & 0.19/0.28 & 0.43/0.36 & 0.28/0.28 & 6.78/6.96 & 6.11/6.40 & 6.47/6.51 & 6.28/6.40 + 0.45 & 1.39@xmath77 & 0.23@xmath78/0.41@xmath78 & 0.28@xmath78/0.30@xmath78 & 0.28@xmath78/0.40@xmath78 & 7.12@xmath77 & 6.20@xmath78/6.60@xmath78 & 6.28@xmath78/6.43@xmath78 & 6.28@xmath78/6.56@xmath78 + 0.14 & 0.72/0.58 & 0.25/0.18 & 0.36/0.34 & 0.24/0.30 & 6.83/6.93 & 6.23/6.21 & 6.39/6.49 & 6.21/6.43 + lccccccccccc & & + o & @xmath55 & & & & & & & + & & k & h & h@xmath48 & h@xmath49 & h@xmath43 & h@xmath50 & h@xmath0 & @xmath28498 & @xmath28542 & @xmath28662 + m & 0.12 & - & - & - & - & - & - & 0.29/0.30 & 0.30/0.20 & 0.22/0.21 & 0.26/0.26 + m & 0.77 & - & - & - & - & - & - & 0.22/0.41 & 0.23/0.33 & 0.41/0.52 & 0.37/0.40 + m & 0.46 & - & - & - & - & - & - & 0.29/0.42 & @xmath79/0.49 & 0.90@xmath77/@xmath79 & 0.54@xmath77/@xmath79 + m & 0.11 & - & - & - & - & - & - & 0.29/0.28 & - & - & - + m & 0.76 & - & - & - & - & - & - & 0.26/0.34 & 0.13/0.39 & 0.33/0.29 + m & 0.42 & - & - & - & - & - & - & 0.31/0.32 & 0.21/0.25 & 0.29/0.25 & 0.27/0.34 + m & 0.08 & - & - & - & - & - & - & 0.34/0.40 & 0.14/0.18 & 0.35/0.30 & @xmath79/@xmath79 + m & 0.73 & - & - & - & - & - & - & 0.33/0.47 & 0.13/0.28 & 0.32/0.37 & 0.24/0.23 + m & 0.39 & - & - & - & - & - & - & 0.44/0.38 & 0.23/0.20 & 0.37/0.38 & 0.16/0.24 + m & 0.04 & - & - & - & - & - & - & 0.41/0.28 & 0.18/0.32 & 0.27/0.38 & 0.22/0.41 + f & 0.68 & 1.06/1.02 & 0.85/0.92@xmath113 & @xmath114/0.50 & 0.09/0.13@xmath78 & @xmath79/0.08@xmath78 & 0.12/0.08@xmath78 & 0.54/0.48 & 0.33/0.35 & 0.33/0.30 & 0.37/0.35 + f & 0.36 & 0.77/0.84 & 0.80@xmath113/0.68 & 0.17/@xmath114 & 0.05/0.09@xmath78 & 0.08/0.10@xmath78 & 0.09/0.15@xmath78 & 0.32/0.52 & 0.39/0.40 & 0.33/0.37 & 0.34/0.42 + f & 0.02 & 1.52@xmath77 & 1.27@xmath77 & 0.68@xmath77 & 0.10@xmath115@xmath78 & 0.10@xmath115@xmath78 & 0.21@xmath115@xmath78 & 0.74@xmath77 & 0.60@xmath77 & 0.50@xmath77 & 0.78@xmath77 + f & 0.70 & 0.92/0.87 & 0.92/0.59 & 0.20/0.35 & 0.07/0.08@xmath78 & 0.12/0.09@xmath78 & 0.18/0.16@xmath78 & 0.49/0.43 & 0.36/0.37 & 0.33/0.32 & 0.41/0.38 + f & 0.71 & 0.76/0.63 & 0.70/0.68 & 0.40/0.44 & 0.08/0.08@xmath78 & 0.08/0.09@xmath78 & 0.13/0.13@xmath78 & 0.50/0.46 & 0.36/0.37 & 0.35/0.32 & 0.40/0.37 + o observing run : m : mcdonald98 , f : foces02 . + @xmath55 orbital phase ( see table [ tab : reduc1 ] ) + @xmath77 data for the primary and secondary components not deblended . + @xmath78 mean value of measured ( from two apertures in our spectra ) or value from higher @xmath11 aperture + @xmath79 data not measured due to very low @xmath11 . + @xmath114 data not measured due to blending . + @xmath113 ca ii h line of one component blended with h@xmath48 of the other component . lccccccccccc & & + o & @xmath55 & & & & & & & + & & k & h & h@xmath48 & h@xmath49 & h@xmath43 & h@xmath50 & h@xmath0 & @xmath28498 & @xmath28542 & @xmath28662 + m & 0.12 & - & - & - & - & - & - & 6.52/6.54 & 6.37/6.19 & 6.23/6.21 & 6.3 1/6.31 + m & 0.77 & - & - & - & - & - & - & 6.40/6.67 & 6.25/6.41 & 6.50/6.61 & 6.4 6/6.49 + m & 0.46 & - & - & - & - & - & - & 6.52/6.68 & @xmath79/6.58 & 6.84@xmath77/@xmath79 & 6.62@xmath77/@xmath79 + m & 0.11 & - & - & - & - & - & - & 6.52/6.51 & - & - & - + m & 0.77 & - & - & - & - & - & - & 6.48/6.59 & 6.00/6.27 & 6.41/6.48 & 6.3 4/6.35 + m & 0.42 & - & - & - & - & - & - & 6.55/6.57 & 6.21/6.29 & 6.35/6.29 & 6.3 2/6.42 + m & 0.08 & - & - & - & - & - & - & 6.59/6.66 & 6.04/6.15 & 6.43/6.37 & @xmath79/@xmath79 + m & 0.73 & - & - & - & - & - & - & 6.58/6.73 & 6.00/6.34 & 6.40/6.46 & 6.2 7/6.25 + m & 0.39 & - & - & - & - & - & - & 6.70/6.64 & 6.25/6.19 & 6.46/6.47 & 6.0 9/6.27 + m & 0.04 & - & - & - & - & - & - & 6.67/6.51 & 6.15/6.40 & 6.32/6.47 & 6.2 3/6.50 + f & 0.68 & 7.11/7.09 & 7.01/7.05@xmath113 & @xmath114/6.78 & 6.04/6.20@xmath78 & @xmath79/5.98@xmath116 & 6.15/5.98@xmath78 & 6.79/6.74 & 6.41/6.43 & 6.41/6.37 & 6.46/6.43 + f & 0.36 & 6.97/7.01 & 6.99@xmath113/6.92 & 6.32/@xmath114 & 5.78/6.04@xmath78 & 5.98/6.08@xmath78 & 6.03/6.25@xmath78 & 6.57/6.78 & 6.48/6.49 & 6.41/6.45 & 6.42/6.51 + f & 0.02 & 7.27@xmath77 & 7.19@xmath77 & 6.92@xmath77 & 6.08@xmath115@xmath78 & 6.08@xmath115@xmath78 & 6.40@xmath115@xmath78 & 6.93@xmath77 & 6.67@xmath77 & 6.59@xmath77 & 6.78@xmath77 + f & 0.70 & 7.05/7.02 & 7.05/6.86 & 6.39/6.63 & 5.93/5.98@xmath78 & 6.16/6.03@xmath78 & 6.33/6.28@xmath78 & 6.75/6.69 & 6.45/6.46 & 6.41/6.40 & 6.50/6.47 + f & 0.71 & 6.97/6.88 & 6.93/6.92 & 6.69/6.73 & 5.98/5.98 & 5.98/6.03 & 6.19/6.19 & 6.76/6.72 & 6.45/6.46 & 6.43/6.40 & 6.49/6.46 +

the combinatorial theory of rotor - routers has connections with problems of statistical mechanics , graph theory , chaos theory , and computer science . a rotor - router network defines a deterministic walk on a digraph in which a particle walks from a source vertex until it reaches one of several target vertices . motivated by recent results due to giacaglia et al . , we study rotor - router networks in which all non - target vertices have the same type . a rotor type is _ universal _ if every hitting sequence can be achieved by a homogeneous rotor - router network consisting entirely of rotors of type . we give a conjecture that completely classifies universal rotor types . then , this problem is simplified by a theorem we call the reduction theorem that allows us to consider only two - state rotors . a rotor - router network called the compressor , because it tends to shorten rotor periods , is introduced along with an associated algorithm that determines the universality of almost all rotors . new rotor classes , including boppy rotors , balanced rotors , and burd rotors , are defined to study this algorithm rigorously . using the compressor the universality of new rotor classes is proved , and empirical computer results are presented to support our conclusions . prior to these results , less than 100 of the roughly 260,000 possible two - state rotor types of length up to 17 were known to be universal , while the compressor algorithm proves the universality of all but 272 of these rotor types .

1111.1459

the rotor - router model gives a deterministic analogue to random walks on directed graphs ( digraphs ) . it is an offshoot of the abelian sandpile , or chip - firing , model introduced by bak , tang , and wiesenfield @xcite to study physical phenomena collectively known as self - organized criticality . " work by dhar @xcite and others on chip - firing gave new insights into phenomena such as the spreading of forest fires , the formation of fractal - like river networks , and other problems in statistical mechanics . their work , notably the discovery of the sandpile group , laid the groundwork for further progress in chip - firing and rotor - routing . more recent combinatorial work on chip - firing can be found in the work of diaconis and fulton @xcite , and that of levine @xcite . rotor - routers , first introduced by priezzhev et al . @xcite to study self - organized criticality , were then independently rediscovered several times in various connections . these include load - balancing , as in rabani , sinclair and wanka @xcite , and the internal diffusion limited aggregation ( idla ) model , as in diaconis and fulton @xcite . rotor - routers have been studied extensively as a quasirandom analogue of random walks and markov chains @xcite , in particular for understanding idla . recent progress shows that rotor - router aggregation satisfies similar , though much stronger , spherical asymptotics as random idla @xcite . holroyd et al . @xcite outlined a more general theory of rotor - routers on arbitrary digraphs , followed by the 2011 article @xcite by giacaglia et al . that proved the periodicity of rotor - router hitting sequences and the conservation of palindromicity . the universality problem , proposed by propp , has been of interest since a solution can be viewed as a significant generalization of the aforementioned results by giacaglia et al . propp asks for the classification of rotor types , the fundamental units of rotor - router networks , that can model all others . these are called _ universal_. any complete theory of rotor - routers requires an understanding of rotor universality , just as number theory requires an understanding of primes and ring theory an understanding of units . our main results towards this end are the reduction theorem of section [ reductsect ] , reducing the problem to the fairly manageable space of two - state rotors , and the introduction of the compressor algorithm in section [ compressor ] , which proves universality for a large number of rotors and significantly simplifies the problem for the rest . the relevant results from @xcite are presented in section [ backmot ] along with a rigorous description of the rotor - router model and the concept of universality . in section [ examples ] , we give two simple examples to elucidate our definitions , the latter of which is also essential to the problem at hand . in section [ reductsect ] , the reduction theorem is stated and proved . section [ compressor ] introduces the compressor , and its basic properties are proven . in section [ empiricalresults ] , we present various empirical results showing that the compressor algorithm almost always terminates in the rotor @xmath2 for unboppy rotors . sections [ morebal ] and [ abba ] focus on the action of the compressor on specific classes of balanced rotors . finally , section [ future ] makes explicit the many possible directions of future work hinted at in the rest of the report , and in section [ acknowledgements ] the people who made this work possible are acknowledged . on a finite directed graph @xmath3 , allowing self - loops and multiple edges , a rotor - router network is imposed as follows . we choose a _ source _ vertex as well as a set of _ target _ vertices with outdegree @xmath4 . every non - target vertex must be able to reach at least one target . at each non - target vertex , we choose a _ rotor pattern _ denoted @xmath5 comprising an infinite periodic sequence of the directed edges emanating from @xmath6 . a particle travels through the network , beginning at the source and traveling until it reaches a target , whereupon it is returned to the source and commences a new walk . after its @xmath7-th visit to non - target vertex @xmath6 , the particle leaves @xmath6 along edge @xmath8 . at this moment , we say that the first @xmath7 terms of the rotor pattern at @xmath6 have been _ used up _ , or _ it is important to note that this scheduling policy is purely local and deterministic , and does not depend on the in - edge by which the particle enters @xmath6 . the _ hitting sequence _ of the network is defined as the infinite sequence of targets successively visited by this particle . hitting sequences are eventually periodic ; in @xcite , it is shown that they are also periodic from the start . each rotor pattern can be described by a partition @xmath9 of @xmath10 ( this set is denoted @xmath11 $ ] ) where @xmath7 is its period , and @xmath12 $ ] are partitioned into the same block of @xmath9 iff @xmath13 . each block of @xmath9 is associated with a distinct natural number , which gives a representation of the rotor that we call its _ rotor type_. there are an infinite number of such representations for every rotor pattern . for instance , if the pattern at a vertex @xmath6 with two out - edges @xmath14 and @xmath15 is the sequence @xmath16 with period 3 , we can write its type as 1 , 2 , 2 , or as 4 , 7 , 7 . typically if all the numbers are single digits , commas are omitted and we write 122 or 477 instead . often , we abuse these definitions and use the terms rotor pattern , " rotor type , " and rotor " interchangeably to avoid cumbersome language . a rotor is said to have @xmath17 states if exactly @xmath17 distinct natural numbers appear in its type . two rotor types @xmath1 , @xmath18 are said to be _ equivalent _ if they can represent the same rotor pattern . this is denoted @xmath19 . thus @xmath20 . given a rotor type @xmath1 , the notation @xmath21 denotes its period length , and @xmath22 represents its @xmath23-th term , where @xmath23 is taken modulo @xmath21 . propp introduced the idea that an entire rotor - router network can be thought of as a single meta - rotor with rotor pattern equivalent to its hitting sequence , since hitting sequences are always periodic . to formalize this concept , we say that a rotor type @xmath1 can _ model _ a rotor type @xmath18 if there exists a _ homogeneous _ rotor - router network , one containing only rotors of type @xmath1 , with a hitting sequence equivalent to @xmath18 . working off this idea , propp defined a _ universal _ rotor as a rotor that can model any other . propp then raised the natural question : which rotor types @xmath1 are universal ? a rotor type is called _ palindromic _ if it reads the same forwards and backwards ( e.g. @xmath24 ) , and _ block - repetitive _ if it is the concatenation of uniform sequences of the same length @xmath17 ( e.g. @xmath25 ) . the length @xmath17 is known as the _ block length _ of such a rotor . we define : by a _ block of length @xmath17 _ of a rotor type @xmath1 , where @xmath17 divides @xmath21 , we mean a subsequence of @xmath1 of the form @xmath26 , @xmath27 , ... , @xmath28 , where @xmath29 . we call this specific block the @xmath30-th block of length @xmath17 of @xmath1 , denoted @xmath31 . a block - repetitive rotor of block length @xmath17 is one such that every block of length @xmath17 is uniform . it is important to distinguish blocks from _ runs _ , where the two are distinct in that @xmath32 is always a run of @xmath1 but only a block if @xmath33 divides @xmath34 . uniform runs are important in section [ morebal ] . it was proven by giacaglia et al . @xcite that a network composed entirely of palindromic rotors has a palindromic hitting sequence , and similarly that a network composed entirely of block - repetitive rotors of block length @xmath35 has a block - repetitive hitting sequence with the same block length @xmath17 . from these results it follows that neither palindromic nor block - repetitive rotor types can be universal . we believe these two classes encompass all nonuniversal rotors , so the following definition is natural : a rotor type is _ boppy _ if it is palindromic or block - repetitive . otherwise , it is _ unboppy_. henceforth , all bold uppercase letters or words will stand for classes of rotors . we let @xmath36 denote the class of all rotor types , @xmath37 the class of all palindromic rotors , @xmath38 the class of all block - repetitive rotors , @xmath39 the class of all universal rotors , and @xmath40 the class of all boppy rotors . also , @xmath41 denotes the complement of a set @xmath42 of rotors ( with respect to @xmath36 ) . in this notation , giacaglia et al . @xcite showed that @xmath43 . we conjecture that more is true : [ cnjuniv ] as classes of rotor types @xmath44 . let the notation @xmath45 , where @xmath46 and @xmath47 , define the class of rotor types that are palindromic iff @xmath48 and block - repetitive with block length @xmath33 ( but no larger block length ) . the conjecture above can then be strengthened to : [ cnjstrong ] every rotor type @xmath49 is universal in @xmath45 . conjecture [ cnjstrong ] is only slightly stronger than conjecture [ cnjuniv ] , since conjecture [ cnjuniv ] implies that any rotor of class @xmath50 is universal in that class . it would only remain to show the same for the palindromic classes @xmath51 . finally , we also present a weakening that is more directly amenable to our methods : [ central ] every rotor is universal in @xmath45 for some choice of @xmath52 and @xmath33 . in this paper , conjectures [ cnjuniv ] through [ central ] will be our guiding focus , although we will make many related inquiries of independent interest . the second half of this report focuses on the compressor configuration in figure [ compfig ] , so named because it tends to shorten the lengths of the rotors upon which it is applied . the compressor can be viewed alternatively as a rotor - router network , as a mapping from rotor types to rotor types , or simply as a mapping from binary strings to binary strings . this last viewpoint is discussed extensively in sections [ morebal ] and [ abba ] . a related issue put into focus by the action of the compressor is the definition of a complexity measure for periodic binary strings . with the similar process known as the differential operator , or ducci map , the complexity of such strings can be defined in terms of the length of the end cycle when the map is applied repeatedly @xcite , or alternatively as the number of applications of the map before a term repeats . since the compressor acts in an analogous way on balanced binary strings , we can also define a complexity measure of binary strings based on the compressor algorithm . this motivates our discussion of burd and ab - ba rotors as the elements of fixed cycles of the compressor map " on binary strings . as a simple example of the rotor - router network , consider the directed graph in figure [ ex1 ] . the underlying digraph , shown in figure [ ex1](a ) , consists of @xmath53 non - target vertices labeled @xmath54 , of which vertex @xmath55 is the source , as well as three targets @xmath56 , @xmath57 and @xmath58 . the thin dark arrows denote the current state of the rotor at that vertex , or the direction by which the last particle left . in the other two diagrams , two walks of the particle to a target are traced by thicker dark arrows . upon entering a new vertex , the particle shifts the rotor to the next possible exit direction ( counterclockwise ) and leaves along this new rotor setting . the rotor types are rather simple in this case , since the rotor patterns proceed counterclockwise ; they are @xmath59 , and @xmath60 , at vertices @xmath54 , respectively . the hitting sequence begins @xmath58 , @xmath56 , as shown . from this example we already see some applications of rotor - routers . first , a rotor walk is the _ deterministic analogue of a random walk _ , since rotor walks hit targets with the same frequencies as their associated random walks do in expectation @xcite . to be precise , rotor - routers are quasirandom in the sense that they are clearly nonrandom ( unlike pseudorandom processes ) and yet model random processes in some ways . perhaps , if a randomized algorithm can be replaced with one involving rotor - routers , deterministic worst - case bounds may be given in place of probabilistic time bounds . it is not difficult to imagine how rotor - routers could be useful in load - balancing for parallel processing @xcite . this derandomization " has also been used successfully to study the idla problem about random walks on the plane lattice @xcite . rotor walks share another property with random walks : the particle is _ not responsible for knowing its destination _ ( thanks to yan zhang of mit for pointing this out ) . in conventional networking , each of the innumerable packets of information traveling through a network has some responsibility for finding its way to its destination . in a rotor - router network , however , particles are blindly shoved around but end up reaching targets in a predictable fashion . thus rotor - routers could provide a robust mechanism for the distribution of information through large networks , especially if some regular pattern of delivery is required and the processing power at individual nodes is limited . we proceed to our second example , which is of direct interest to the problem at hand . we show that @xmath61 . consider the hitting sequence of a rotor network in the form of a complete binary tree of depth @xmath7 with @xmath62 non - target nodes and @xmath63 targets , such that each rotor in the network is of type @xmath2 . this means that particles leave each node in an alternating fashion . the source lies at the root of the tree . figure [ ex2 ] shows such a tree of depth @xmath64 . now , we ask , what is the sequence of targets hit ? let us show by induction that the hitting sequence is some permutation of the @xmath63 targets . in the case @xmath65 , we have a single non - target node , the source , so the two targets are hit in an alternating fashion . now , suppose the statement is true for some @xmath66 . consider the binary tree of depth @xmath67 . it can be split into two parts : its first @xmath7 layers , forming a smaller binary tree , and its deepest layer , consisting of the @xmath63 non - target nodes directly preceding the targets . using the inductive hypothesis , the first set can be replaced by a single source meta - rotor of type @xmath68 , emitting particles to a permutation of the last layer of rotors . now the result follows immediately by induction ; this meta - rotor must emit two entire periods worth of particles to the last layer before any target is hit twice , since the rotors are of type @xmath2 . it follows that all @xmath69 targets are hit once before any one is hit twice . from the argument above , it follows that @xmath2 can make any rotor type of the form @xmath68 for @xmath66 . in the spirit of cauchy s proof of the am - gm inequality , we modify the tree slightly to show that @xmath2 can make any rotor type of the form @xmath70 , for any @xmath71 . for each @xmath72 , choose @xmath7 such that @xmath73 . consider the binary tree network of depth @xmath7 , with @xmath63 targets , as before . now , remove @xmath74 targets , routing their predecessors directly back to the root . this modification essentially hides " the extra targets the operation of the rest of the network is not affected ( in the next section such a modification is called a _ destructive reduction _ ) . the remainder of the hitting sequence has type @xmath70 , as desired . .,scaledwidth=60.0% ] finally , to form an arbitrary rotor type @xmath75 of period @xmath17 from @xmath2 , first construct the modified binary tree hitting @xmath17 targets , as above . then , merge together the targets such that two targets @xmath76 $ ] are combined iff @xmath77 ( in the next section such an operation is called a _ merging reduction _ ) . the hitting sequence is now the rotor type we want , so @xmath61 . henceforth , all of our attacks on the universality problem will be based on the fact that @xmath2 is universal , since any rotor @xmath1 that can model @xmath2 is also universal . the flexibility of the rotor - router model allows room for the generalization of results about two - state rotors to any number of states . in this section , we propose and prove a theorem regarding the reductions of multi - state rotors to two - state rotors . a rotor type can often be thought of as a special case of another type with more states . the rotor type @xmath78 , for instance , is a special case , or _ reduction _ , of the type @xmath79 , since we may direct each edge in a @xmath79-type rotor numbered @xmath80 in the same direction as the one numbered @xmath81 . this reduction is described informally as @xmath82 , and called a _ merging reduction_. along the same lines , @xmath60 is also a special case of the type @xmath79 , as each edge in a @xmath79 rotor numbered @xmath80 can be turned into a self - loop , eliminating all the @xmath80 s . this reduction is denoted @xmath83 informally , and called a _ destructive reduction_. in general we can always reduce a rotor by merging two of its states or deleting one altogether . the following lemma about reductions gets us close to our goal , theorem [ reduction ] . [ reductlem ] if @xmath84 has more than two states , and all of its merging reductions are boppy , then there exist three states of @xmath1 , which we assume to be @xmath85 , such that the reductions @xmath86 and @xmath87 of @xmath1 are block - repetitive with coprime block lengths and @xmath88 is palindromic . because @xmath89 , there exists some integer @xmath23 such that @xmath22 and @xmath90 differ . without loss of generality , call these two states @xmath55 and @xmath81 , respectively . consider another state , which we call @xmath64 , without loss of generality . by our assumptions , if we apply @xmath86 , the rotor becomes boppy . it can not become palindromic , since @xmath22 and @xmath90 still differ , so it must become block - repetitive . let its block size be @xmath34 . every block of @xmath81 s in @xmath1 must then have length divisible by @xmath34 . similarly , if we reduce @xmath91 , the rotor becomes block - repetitive with some block size @xmath33 , so the @xmath55 s must occur in blocks of length @xmath33 . now if @xmath34 and @xmath33 share any common factors @xmath92 , both reductions are block - repetitive with block length @xmath92 . also , each block of @xmath81 s has length divisible by @xmath92 , as does each block of @xmath55 s . therefore , the remaining @xmath64 s must also be in blocks of length @xmath92 , so that @xmath1 is itself block - repetitive with block length @xmath92 . a contradiction has been reached so @xmath34 and @xmath33 must be coprime . suppose these two merging reductions , @xmath91 and @xmath93 are both block - repetitive , and consider the reduction @xmath88 . first , suppose it too is block - repetitive , with block length @xmath94 . it follows that the @xmath64 s occur in blocks of length @xmath94 . also , if @xmath34 or @xmath33 shares factors with @xmath94 , then we have the same contradiction as before , that @xmath1 itself is block - repetitive . we have proven that @xmath34 , @xmath33 , and @xmath94 are pairwise coprime . choose any @xmath23 not a multiple of @xmath95 . then , @xmath22 and @xmath96 lie in the same uniform block for at least two of the three numbers @xmath34 , @xmath33 , @xmath94 . it follows that @xmath97 . we have immediately that @xmath1 is block - repetitive with block length @xmath95 . thus we have reached a contradiction and @xmath88 must give a palindromic reduction instead , as desired . for the rotor type @xmath98 , five of its six possible reductions are boppy ; two are block - repetitive and three are palindromic . therefore , the following theorem may come as a surprise . [ reduction ] every @xmath89 with more than two states has an unboppy reduction . suppose there exists @xmath89 such that all reductions of @xmath1 lie in @xmath99 . find some three states @xmath55 , @xmath81 , @xmath64 satisfying the conditions of lemma [ reductlem ] . both reductions @xmath86 and @xmath91 of @xmath1 are block - repetitive with some coprime block lengths @xmath100 , and so every state not @xmath55 , @xmath81 , or @xmath64 occurs in blocks of length @xmath101 . therefore , we can discount them entirely ; if we can show that the subsequence of @xmath1 containing only @xmath55 , @xmath81 , or @xmath64 has an unboppy reduction , then it is automatic that @xmath1 itself has an unboppy reduction by the same merging or destructive reduction rule . indeed , if the reduction of this subsequence is not palindromic the corresponding reduction of @xmath1 can not be palindromic . similarly , if the reduction of this subsequence is not block - repetitive , then the corresponding reduction of @xmath1 just has some blocks of length @xmath101 inserted , and is still not block - repetitive . thus it suffices to assume that @xmath1 has exactly three states @xmath55 , @xmath81 and @xmath64 . let us prove that , in this case , the destructive reduction @xmath102 is unboppy . the resulting type is not palindromic ; otherwise , @xmath1 must have been palindromic , since @xmath88 is also palindromic . it remains to show that this rotor is not block - repetitive . if it is , the block length must be a multiple of @xmath33 , since all @xmath55 s occur in such blocks , and a multiple of @xmath34 , since all @xmath81 s occur in such blocks . hence , the resulting type must have block length @xmath101 . but then , the @xmath64 s removed must also have been in blocks of @xmath101 in order for the reductions @xmath91 and @xmath86 to have been block - repetitive . hence @xmath1 itself was block - repetitive , which is a contradiction . thus , @xmath102 is unboppy . it follows that one of the reductions of @xmath1 lies in @xmath103 . after the reduction theorem , it suffices to consider rotor types of only two states ; if conjecture [ cnjuniv ] is proven for every two - state rotor , then it follows in general by the reduction theorem . to tackle this reduced problem , we introduce the compressor in the next section . here we introduce a rotor - router network of fundamental importance to the universality problem , the compressor . in one instance a special case of the compressor was used previously by propp to prove the universality of @xmath104 , but no recognition of its general significance was made . the rotor - router network in figure [ compfig ] is the compressor , a network that applies only to two - state rotors . it consists of three rotors of the same type @xmath1 as well as two targets . the source rotor is labeled @xmath55 and the targets are @xmath80 and @xmath53 . by symmetry , there are four essentially different variations of this configuration , based on the order of the out - edges of rotors @xmath81 and @xmath64 ( a type @xmath2 rotor at rotor @xmath81 could represent either @xmath105 or @xmath106 ) . each configuration can be viewed as a mapping from the set of rotors to itself . the four variations of the compressor will be called @xmath107 , @xmath108 , @xmath109 , and @xmath110 . in each case , the first letter represents whether rotor @xmath81 points up to rotor @xmath55 first or down to rotor @xmath80 , and the second letter represents whether rotor @xmath64 points up to rotor @xmath55 or down to @xmath53 first . for each rotor @xmath1 , we define the mappings @xmath111 to take @xmath1 to the hitting sequence of the corresponding compressor variation with three copies of @xmath1 . the compressor is a special case of the binary tree on three non - target vertices ( with four targets ) , with two of the targets hidden , " in the language of section [ backmot ] . in the language of the previous section , @xmath112 , @xmath113 , @xmath114 , and @xmath115 are all destructive reductions of @xmath116 , which will stand for the result of applying the binary tree to a rotor . note that @xmath117 takes two - state rotors to four - state rotors , and that @xmath116 can be formed from @xmath112 and @xmath115 by splitting these into smaller subsequences , and pasting these together ; one of them will be given the labels @xmath118 instead of @xmath119 . the same can be said of @xmath108 and @xmath109 . this idea is helpful in sections [ morebal ] and [ abba ] . to illustrate this idea , consider the rotor @xmath120 , for instance . we find @xmath121 and @xmath122 ( how this can be quickly determined will become clear presently ) , while @xmath123 , an enmeshing " of @xmath107 and @xmath110 . to prepare for the next two results we define : a _ balanced _ rotor type is a two - state rotor type such that its two states occur with equal frequency . the class of all balanced rotors is denoted @xmath124 . the following result shows that @xmath108 maps two - state rotors to balanced rotors . henceforth , we assume without loss of generality that the two states of any two - state rotor are @xmath55 and @xmath81 , and that the first term in the rotor is @xmath55 . [ udbalanced ] if @xmath125 has two states , then @xmath126 . suppose a period of @xmath1 has @xmath34 @xmath55 s and @xmath33 @xmath81 s . when exactly @xmath127 periods of the source rotor @xmath55 have been fired in network @xmath108 , exactly @xmath34 periods of rotor @xmath81 and exactly @xmath33 periods of rotor @xmath64 will have fired , so the graph is back to its original state . thus some number of full periods of the hitting sequence has been reached . since @xmath101 of each target have been hit , @xmath126 . the proof of theorem [ udbalanced ] also yields an @xmath128 upper bound for the period length of @xmath113 , where @xmath129 , better than the trivial @xmath130 bound . if @xmath125 has two states , then @xmath131 and @xmath132 . in general this argument extends to give @xmath133 bounds for the hitting sequence lengths of homogeneous acyclic digraph ( dag ) networks with size @xmath17 and longest chain of length @xmath92 , a significant improvement over the exponential bounds for general rotor networks . let a rotor network be imposed on a digraph @xmath134 with @xmath17 non - target vertices and all rotors of type @xmath135 . suppose that every directed cycle in @xmath0 contains the source of the network , and @xmath92 is the largest number of non - target vertices in a simple directed path in @xmath0 beginning at source . then the hitting sequence of @xmath0 has period length at most @xmath136 . we can assume that @xmath0 has no directed cycles . otherwise , let the set of in - edges of the source @xmath137 be @xmath138 . modify @xmath0 by adding one target vertex @xmath139 . for each @xmath140 , we replace it with the edge @xmath141 . there are no directed cycles in the new graph @xmath142 , and the period of the hitting sequence of @xmath0 is the destructive reduction @xmath143 , in the notation of the last section , of that of @xmath142 . hence it is sufficient to prove the bound for acyclic networks . it is only necessary to show that after the particle has been fired from the source @xmath136 times , it has passed through each non - target vertex a multiple of @xmath21 times . for each non - target vertex @xmath144 , we let @xmath145 denote the length of the maximal path from @xmath146 to @xmath6 . . inducting on @xmath145 , it is easy to show that the number of times the particle passes through @xmath6 is a multiple of @xmath148 . the result follows immediately . as an example of theorem [ udbalanced ] , consider the action of @xmath108 on the rotor type @xmath104 . we write out ( with foresight ) some periods of the rotor patterns : 1 . 223223223 2 . 114114 3 . 551 since we are applying @xmath108 , the first term in rotor @xmath81 is @xmath55 ( up " ) while the first term in rotor @xmath64 is @xmath53 ( down " ) . since the number of @xmath81 s in three periods of the source rotor @xmath55 corresponds to the length of two periods of @xmath81 , and the number of @xmath64 s to the length of one period of @xmath64 , after everything displayed above is used up the rotor network has returned to its original position . therefore the hitting sequence must contain exactly two @xmath80 s and two @xmath53 s . because of theorem [ udbalanced ] , it suffices to study balanced rotors . [ nonincreasing ] if @xmath149 , then @xmath150 and @xmath151 . if two periods of rotor @xmath55 have been fired , then exactly one period of rotor @xmath81 and one period of rotor @xmath64 will have fired , because @xmath1 is balanced . the rotors have all returned to their original states . therefore , since exactly @xmath21 targets have been hit in this sequence , the period of the hitting sequence must divide @xmath21 . for instance , consider how @xmath152 is formed . we write out two periods of the source and one period of rotors @xmath81 and @xmath64 : 1 . 232332232332 2 . 141441 3 . 151551 since @xmath153 , we see that when every term listed above is used up , the network returns to its original state . hence the hitting sequence is of length dividing @xmath56 , just like the original rotor . in this particular instance the hitting sequence is @xmath154 , or @xmath2 , of length @xmath81 , and the compressor has proven that @xmath155 . this example also illustrates why every variation of the compressor preserves balance , corollary [ balclosedcor1 ] , and hints at theorem [ one21 ] in section [ abba ] . repeated compressor applications never increase a balance rotor s period , since the compressor preserves balance . many rotors can be compressed " to rotors of short period , or even to @xmath2 , as above . explicitly , the compressor algorithm runs as follows . starting with an arbitrary two - state rotor @xmath1 , apply @xmath108 to create balance . thereafter , apply an arbitrary compressor variation to the result ( in which we replace @xmath80 s by @xmath55 s and @xmath53 s by @xmath81 s ) and repeat . in the vast majority of cases if @xmath1 is unboppy this algorithm ends in @xmath2 , proving the universality of @xmath1 . this is the first , and crudest , instance of the use of monovariants associated with the compressor . as will be seen in sections [ morebal ] and [ abba ] , more monovariants further restrict the action of the compressor . finally , we sketch an algebraic way to study the compressor . for a given @xmath156 , write it as an infinite periodic sequence @xmath157 of period @xmath158 . define @xmath159 to be the position of the @xmath17-th @xmath55 in this sequence , and @xmath160 to be the position of the @xmath17-th @xmath81 . also , define @xmath161 to be the number of @xmath55 s in the sequence up through @xmath162 , and @xmath163 the number of @xmath81 s up through @xmath162 . let us find the position of the @xmath17-th @xmath80 in @xmath113 in terms of @xmath164 , @xmath165 , @xmath166 , and @xmath0 . [ positionud ] the @xmath17-th @xmath80 in @xmath113 occurs at position @xmath167 . consider the position of the @xmath17-th @xmath80 in @xmath113 . it must come from directly the @xmath17-th @xmath80 in rotor @xmath81 of the compressor configuration . we count the number of @xmath53 s hit before this @xmath80 to find its absolute position in @xmath113 . it came from the @xmath17-th @xmath80 in rotor @xmath81 , at position @xmath160 in that rotor . this was hit by the particle coming from the @xmath160-th @xmath81 in the source rotor , which is at position @xmath168 , since @xmath81 s in rotor @xmath55 correspond to @xmath55 s in @xmath1 . now , exactly @xmath169 of the terms in the source up to this point are @xmath64 s , and they all went to rotor @xmath64 . of these , exactly @xmath170 hit a @xmath53 in that rotor . it follows that the @xmath17-th @xmath80 in @xmath113 occurs at position @xmath171 . theorem [ positionud ] and its analogues for the other compressor variations give a number of results supporting our intuition . we give only one example here . suppose @xmath172 for some @xmath1 . note that since the @xmath17-th and @xmath173-th @xmath80 s in @xmath113 must be symmetric ( i.e. their positions must add to @xmath174 ) , it follows that @xmath175 by the exact same manipulations on the @xmath53 s in @xmath113 , we can show that @xmath176 these equations say that the original rotor @xmath1 was almost palindromic . " for instance : if @xmath156 , then @xmath172 only if @xmath1 starts and ends with the same term . with equations ( [ pal1a ] ) and ( [ pal1b ] ) in mind , suppose for the sake of contradiction that @xmath1 starts with @xmath55 and ends with @xmath81 ( without loss of generality ) . find the smallest @xmath23 such that @xmath177 , so @xmath178 for all @xmath179 . then , letting @xmath180 in ( [ pal1a ] ) , we see that @xmath181 next , note that @xmath182 , since the number of @xmath81 s up to the @xmath183-th @xmath55 is exactly the position of the @xmath183-th @xmath55 minus @xmath183 , the number of @xmath55 s up to that point . thus equation ( [ cons1 ] ) simplifies further to @xmath184 furthermore , since @xmath185 is the position of the @xmath158-th @xmath55 , it is the last @xmath55 in the second period of @xmath1 . because a period of @xmath1 has length exactly @xmath158 , it follows that @xmath186 , the position of the @xmath7-th @xmath55 . also @xmath187 , so now it remains to show that @xmath188 is impossible . note that @xmath189 trivially , and @xmath190 because @xmath177 . therefore , @xmath191 , from which it follows that @xmath192 , since @xmath193 . thus ( [ cons3 ] ) is impossible and a contradiction has been reached . the rotor type @xmath1 must start and end with the same term . similar local results can be deduced from theorem [ positionud ] for block - repetitive rotors . both empirically and heuristically from the type of argument above , the compressor only immediately fails to prove universality for unboppy rotors when they are close to boppy rotors ( a few terms away ) . we turn to the empirical data in the next section . of all nontrivial two - state rotors , it was previously known that the rotor types @xmath194 ( which denotes @xmath23 @xmath55 s followed by a @xmath81 ) were universal . by judicious application of the compressor , the following general classes have been proven universal : those of the form @xmath195 , and those of the form @xmath196 . a short proof of the universality of the second class is given in section [ abba ] . in addition to these classes , we have proven through c programs the universality of the vast majority of all unboppy two - state rotor types with period up through 17 . the rotor - router model , and in particular our compressor algorithm , lends itself to computer simulation . here is a brief sketch of the programs written to generate our data . based on an input parameter @xmath7 , the length of the rotors to consider , all two - state rotors of that length are generated . those that are trivial , such as @xmath197 , or necessarily nonuniversal ( i.e. boppy ) such as @xmath198 , are immediately discarded . for each of the remaining rotors , the program applies @xmath108 to the rotor to make it balanced , after which it repeatedly applies a pseudorandomly chosen compressor variation for up to @xmath199 times . if @xmath2 is reached within this subroutine , the loop terminates and the rotor is universal . otherwise , up to @xmath200 more such random sequences of applications of the compressor occur , restarting each time with the original rotor . if at the end of this loop the rotor still has not been proven universal , it is recorded in an output file as undecided . the program outputs at termination the total number of undecided rotors found . the running time is necessarily ( at least ) exponential and proved prohibitive for values of @xmath7 greater than @xmath201 . .the number of unboppy rotors not proven universal . [ cols="<,<,<",options="header " , ] in table [ resulttab ] , we summarize these results . for each rotor length up through @xmath201 , the table gives the number of unboppy rotor types not proven universal by the compressor , and its fraction out of the total number of unboppy rotors of that length . in total , at most 272 are not universal out of roughly 260000 . it seems plausible from this data that _ almost all _ unboppy rotor types can be proven universal using the compressor alone , in the sense that the fraction of undecided rotors approaches @xmath4 . it should be stressed that these undecided rotors are not necessarily counterexamples to conjecture [ cnjuniv ] ; for instance , the @xmath57 potential counterexamples of length @xmath56 or @xmath58 have been proven universal by first applying an auxilliary rotor - router network and then using the compressor algorithm . although this method times out for longer rotors it shows that the occasional failure of the compressor algorithm is probably not due to the existence of a third class of nonuniversals . from this data we conjecture that all unboppy rotors of prime length can be proven universal by the compressor . it is true for all the primes up through 17 . variations on the program above produced further data supporting our conjectures . every balanced , unboppy rotor with period less than 24 is universal , and at most 9 such rotors with period 24 are not . in addition , every ab - ba rotor of length @xmath202 with period up through 42 is universal . finally , we note that every rotor tested , up to length 17 , has been shown universal in some class @xmath203 , strongly supporting conjecture [ central ] . we extend the theory of balanced rotors and prove some finer claims than those of section [ compressor ] , culminating in the definition of burd rotors . in this section it is possible to view rotors almost entirely as binary strings upon which the compressor family of maps are applied . we start by refining the original notion of rotor balance . an _ @xmath7-balanced _ rotor is a two - state rotor type with fundamental period divisible by @xmath7 such that for each @xmath29 , the block @xmath204 has equal numbers of @xmath55 s and @xmath81 s . the class of @xmath7-balanced rotors is denoted @xmath205-@xmath124 . if a rotor @xmath1 is @xmath81-balanced , we also call it an _ ab - ba rotor _ ; the class @xmath206-@xmath124 of @xmath81-balanced rotors is also called @xmath207 . note that the compressor preserves @xmath7-balance . as an illustration , consider @xmath208 . in this case the compressor looks like : 1 . 2332223323322233 2 . 41144411 3 . 51155511 because @xmath209-@xmath124 , we can see that when @xmath210 is fired in rotor @xmath55 , balanced blocks @xmath211 and @xmath212 are used up in rotors @xmath81 and @xmath64 . the next block of length eight , which is again @xmath210 , corresponds to balanced blocks @xmath213 and @xmath214 in rotors @xmath81 and @xmath64 . after each balanced block of eight is fired from the source , the corresponding two blocks of four are fired in rotors @xmath81 and @xmath64 , producing the desired @xmath80-balanced effect . formally , we have : [ balclosed ] for any @xmath215 , the class @xmath216-@xmath124 is closed under @xmath217 and @xmath110 . for @xmath108 and @xmath109 , theorem [ nequiv ] gives a stronger result . we need only prove the result for @xmath107 and @xmath110 . consider @xmath112 for an @xmath158-balanced rotor @xmath1 . we wish to show that for each @xmath29 , the @xmath23-th block of length @xmath158 in @xmath112 is balanced . when the block @xmath218 is fired from the source , half of the particles enter rotor @xmath81 , and the other half enter @xmath64 . from @xmath81 and @xmath64 , each fires according to the block @xmath219 , so we hit the two targets @xmath7 times each . hence this block of @xmath158 of the hitting sequence is balanced , as is every other by the same argument . the two most important special cases are : [ balclosedcor1 ] the classes @xmath124 and @xmath207 are closed under @xmath217 and @xmath110 . for the following results it is natural to define a coarser type of equivalence , where only the frequencies of labels within blocks are taken into consideration . two @xmath220 are _ n - equivalent _ if there exists @xmath221 equivalent to @xmath1 with the same states as @xmath18 such that for each @xmath29 , as multisets @xmath222 . we write @xmath223 . in particular , if @xmath224 , then @xmath19 . now that we have the requisite definitions , we can state and prove theorem [ nequiv ] . [ nequiv ] if @xmath225-@xmath124 , then @xmath226 . we show @xmath227 . suppose by induction that this has been shown for the first @xmath23 blocks of length @xmath7 in @xmath113 and @xmath1 . then , consider @xmath219 . at the point that the source rotor first enters this block of its cycle , by the balanced property of the previous blocks the two other rotors have just used up their first @xmath23 blocks of length @xmath7 , respectively . now , in the next @xmath158 firings of the source , the next block of @xmath7 of rotor @xmath81 is used up . there are the same number of @xmath80 s in this block as there were @xmath81 s in @xmath204 . the corresponding block of rotor @xmath64 also has the same number of @xmath53 s as there are @xmath55 s in @xmath204 , so a block of length @xmath7 is completed and we are done by induction . in particular , when theorem [ nequiv ] is applied to @xmath81-balanced , or ab - ba rotors , it shows that all ab - ba rotors are fixed points of @xmath108 and @xmath109 . [ abbafixed ] if @xmath228 , then @xmath229 . theorem [ nequiv ] exhibits a possible obstacle of the compressor algorithm . if a rotor @xmath230-@xmath124 but @xmath231-@xmath124 for some @xmath7 , then @xmath108 and @xmath109 will never make it @xmath158-balanced ; some blocks of length @xmath232 become permanently skewed in one direction . for instance , @xmath233 , and the imbalanced number of @xmath55 s and @xmath81 s in the first block of length @xmath80 is not resolved . however , an almost exactly opposite result holds for @xmath107 and @xmath110 , showing that they distribute the states more evenly . recall that a balanced run of a rotor @xmath1 is a subsequence @xmath234 such that the same number of @xmath55 s appear as @xmath81 s . for the purposes of the next theorem , it is simpler to treat the rotor - router particle as multiple reincarnations , where it becomes a new , distinct particle whenever it returns to the source . thus the particle that uses up the @xmath23-th term of rotor @xmath55 in the compressor is the @xmath23-th particle . in @xmath112 and @xmath115 there are two runs that together correspond to the original run in @xmath1 in the sense that the targets in those subsequences were hit by exactly those particles using up that run in the source . [ uudddistribute2 ] if @xmath156 is such that the runs @xmath235 and @xmath236 are balanced , then the two runs in @xmath112 and @xmath115 corresponding to the @xmath237-th through @xmath238-th particles to leave the source are both balanced and have lengths that add to @xmath239 . this is true unless the run @xmath240 is constant , in which case exactly one of the balanced runs exists and has length exactly @xmath239 . consider first @xmath112 . the condition that @xmath241 is balanced simply tells us that the same number of terms , namely @xmath34 , of rotors @xmath81 and @xmath64 have been fired . the second condition shows that exactly @xmath33 particles among the next @xmath239 enter rotor @xmath81 , and the rest enter rotor @xmath64 . consider the number of @xmath80 s in rotor @xmath81 , which is equal to the number of @xmath53 s in rotor @xmath64 , coming from the run @xmath240 ; whatever this number is , it is the number of @xmath80 s and @xmath53 s in the corresponding balanced run of @xmath107 , and @xmath239 minus this number is the corresponding size of the balanced run in @xmath110 , by the same logic . one of these numbers is zero iff @xmath240 is constant , so we are done . unless the exception occurs , we say that the run in @xmath1 is _ split up _ between @xmath112 and @xmath115 . note that two periods of @xmath1 correspond to one period of @xmath112 and one of @xmath115 , so an individual run can be split up twice , independently . for an example of theorem [ uudddistribute2 ] , consider again the rotor @xmath242 , which has @xmath243 and @xmath244 . we say that the first occurrence of the balanced run @xmath25 is split up between the @xmath245 in positions three and four of @xmath107 and the corresponding @xmath245 of @xmath110 . the second occurrence of @xmath25 , meanwhile , is not split up because the run at half the position with half the length is the last @xmath246 in the rotor , and is uniform . instead , this balanced run corresponds to the @xmath247 that ends @xmath107 . these ideas about balanced and uniform runs motivate two more notions . define a _ balanced run decomposition _ ( brd ) of a balanced rotor @xmath1 as a partition of the period of @xmath1 into contiguous balanced runs . define the balanced run type of this brd as the infinite periodic sequence @xmath248 such that @xmath249 is the length of the @xmath183-th balanced run , modulo the size of the partition . let the _ balance coefficient _ of @xmath1 , denoted @xmath250 , be the maximal number of runs in any brd of @xmath1 , divided by @xmath21 . a rotor may have multiple brd s if we choose to merge adjacent balanced runs . thus @xmath251 can have brd @xmath252 or @xmath253 or @xmath251 . note that theorem [ uudddistribute2 ] , when applied to all the balanced runs of a rotor , gives the inequality @xmath254 this can be seen from the fact that the balanced runs in two periods of @xmath1 are split between @xmath112 and @xmath115 . equality occurs iff no balanced run is split up as in theorem [ uudddistribute2 ] . next we define the analogous decomposition of a rotor into uniform runs . define an _ uniform run decomposition _ ( urd ) of a rotor @xmath1 as a partition of the period of @xmath1 into contiguous uniform runs . define the uniform run type of this urd as the infinite periodic sequence @xmath255 such that @xmath256 is the length of the @xmath183-th run in this urd , @xmath183 taken modulo the number of runs in the urd . a rotor can have multiple urd s if we choose to split up larger uniform runs . thus @xmath251 can have urd @xmath257 or @xmath258 or @xmath259 . based on these two definitions and inequality ( [ balancecoeff ] ) , we can conclude the following : [ burd ] if @xmath149 has a brd of type @xmath248 , then one of the following must hold : a. one of @xmath260 and @xmath261 is strictly larger than @xmath250 . b. there is also a urd of @xmath1 of type @xmath248 . the theorem follows from the fact that either a split of the type given in theorem [ uudddistribute2 ] exists for @xmath112 or @xmath115 , in which case @xmath262 , since an extra balanced run was created , or else every balanced run of length @xmath249 corresponds to a uniform run of length @xmath263 at the position half its own . we define a _ burd rotor _ as a rotor type satisfying condition _ ii _ of theorem [ burd ] . that is , it has a brd and a urd of the same type . from any balanced rotor , repeated applications of the compressor must eventually lead to a burd rotor ; otherwise we could increase the balance coefficient arbitrarily often using @xmath107 or @xmath110 , and it can only take on a finite number of rational values . naturally , the class of burd rotors is denoted @xmath264 . one essential property of burd rotors is the following generalization of corollary [ abbafixed ] : [ burdfixed ] if @xmath265 , then @xmath229 . assume we are looking at @xmath113 , without loss of generality . let a uniform run of the urd of @xmath1 that has a brd counterpart be @xmath266 . the corresponding balanced run in the brd is @xmath267 . thus , when the particles numbered @xmath237 through @xmath238 are emitted by the source rotor , exactly half go to rotor 2 of the compressor , hitting the uniform run corresponding to @xmath268 there , and the other half go to the same uniform run in rotor 3 . only one of these runs is routed down , depending on @xmath269 , so we know that this uniform run is copied faithfully onto @xmath113 . each of the uniform runs in the urd is copied in the same way into @xmath113 . it is now evident that @xmath270 , and the proof of @xmath271 is exactly analogous . note that the fixed cycles of the compressor are the only rotors necessary to consider to prove conjecture [ central ] . a _ fixed cycle _ of the compressor is defined formally as a set @xmath272 of rotors , closed under all compressor operations , such that any element can be reached from any other by way of some sequence of compressor operations . from the preceding theorems , a standard extremal argument shows : [ fixedcycle ] for every fixed cycle @xmath272 of the compressor @xmath273 . suppose for the sake of contradiction that some such cycle @xmath272 exists , containing a rotor @xmath274 . we can choose @xmath1 to have maximal @xmath250 of all rotors in @xmath275 . also , @xmath276 is nonempty , so we can choose @xmath277 with maximal balance coefficient in this set as well . clearly , @xmath278 since there is a sequence of @xmath107 s and @xmath110 s strictly increasing @xmath250 until it becomes burd , by theorem [ burd ] . but then , if @xmath18 can reach @xmath1 , it must also be able to reach some rotor @xmath279 with @xmath280 by inequality [ balancecoeff ] , a contradiction . thus @xmath273 always . the obvious class of burd rotors is @xmath207 , and the generalization thereof to block - repetitive rotors ( i.e. aabb - bbaa rotors and the like ) . however , not all burd rotors are of this form . for instance , the rotor type @xmath281 generates a nontrivial fixed cycle of size @xmath81 in the action of the compressor . however , when the burd rotor also happens to be ab - ba , we have some further results . recall that a balanced rotor @xmath1 is called ab - ba if it is @xmath81-balanced , that is , if @xmath282 and @xmath283 are different for all @xmath29 . intuitively , ab - ba rotors are just one step away from the universal rotor @xmath2 . this is because the compressor keeps ab - ba rotors ab - ba , and they are all fixed points of @xmath108 and @xmath109 . also , they ca nt be block - repetitive . hence , to be boppy , they need to be palindromic and thus have length @xmath284 ( consider the middle two terms ) . if our conjectures hold , then all ab - ba rotors of length @xmath202 should be universal . indeed , we verified using computer programs that all ab - ba rotors of length @xmath285 up to length @xmath286 are universal by the compressor algorithm . now we discuss some extra tools . break an ab - ba rotor @xmath1 into its constituent @xmath2 or @xmath287 blocks . if exactly @xmath23 such blocks are @xmath287 , then define the _ ba - frequency _ of @xmath1 as @xmath288 . this function @xmath17 can be made into a monovariant for ab - ba rotors under the compressor , as in the following result . [ bafreq ] for @xmath228 , the inequality @xmath289 holds . we only need the number of @xmath287 s in the first @xmath21 terms of @xmath112 and the number of @xmath287 s in the first @xmath21 terms of @xmath115 to sum to the number of @xmath287 s in the first @xmath290 terms of @xmath1 . each @xmath287 in the first two periods of @xmath1 corresponds to one @xmath287 in the first @xmath21 terms of either @xmath112 or @xmath115 . for instance , suppose that @xmath291 , @xmath292 , and @xmath293 . then , in configuration @xmath107 , @xmath294rotor @xmath295 and @xmath294rotor @xmath296 both go back to the source , while in @xmath115 , they route to targets @xmath80 and @xmath53 . in @xmath112 this block @xmath287 will not appear in the hitting sequence , while in @xmath115 it contributes a @xmath287 . each block @xmath287 in the @xmath290 terms of @xmath1 contributes a @xmath287 to exactly one of @xmath112 and @xmath115 . in particular , this shows that there is some sequence of @xmath107 s and @xmath110 s that never increase @xmath297 . further , the theorem shows that if @xmath297 is changed at all by one of @xmath107 and @xmath110 , then it must be strictly decreased by one of them . we end with two specific results based on this idea . [ one21 ] every @xmath228 consisting of more than one block of @xmath2 and exactly one block of @xmath287 is universal . let @xmath298 denote the position of the @xmath287 block in such an @xmath299 . if either of @xmath300 , @xmath301 is not equal to @xmath297 , then one of them must be strictly less than @xmath297 and be zero , by theorem [ bafreq ] . otherwise , both @xmath112 and @xmath115 also have exactly one @xmath287 block . if @xmath302 , then one of @xmath112 and @xmath115 is @xmath2 unless @xmath303 or @xmath304 . if @xmath303 , then one of @xmath305 , @xmath306 is @xmath307 , and another application of @xmath107 or @xmath110 suffices . if @xmath304 , then one of @xmath305 , @xmath306 is @xmath308 , and again @xmath1 is universal after another application of @xmath107 or @xmath110 . if @xmath309 , then one of @xmath112 and @xmath115 is @xmath2 if @xmath303 or @xmath304 . otherwise , it is easy to prove that one of @xmath305 , @xmath306 is @xmath310 , so some combination of @xmath112 , @xmath115 must move the @xmath287 block to position @xmath55 . the proof above shows that the ab - ba problem can be reinterpreted as follows : given a periodic sequence @xmath311 of @xmath4 s and @xmath55 s , where @xmath4 represents @xmath2 and @xmath55 represents @xmath287 , we are allowed to perform the following two transforms on the sequence . the first takes @xmath312 to @xmath313 , and the second to @xmath314 , where @xmath315 and @xmath316 . the problem is to determine when these two transforms , applied repeatedly , can take a sequence to a constant one . in this spirit , the next theorem is immediate . if @xmath228 satisfies the following conditions : a. @xmath302 . b. for any two @xmath287 blocks of @xmath1 that are the @xmath34-th and @xmath33-th blocks of @xmath1 , @xmath34 and @xmath33 do not satisfy @xmath317 . c. for every pair of @xmath287 blocks of @xmath1 , their positions as blocks have the same parity . then @xmath318 . it is clear from the proof of theorem [ one21 ] that one of @xmath319 is @xmath2 , since each @xmath287 block acts independently of the others . if we are patient we can extend these two results somewhat , but it does not seem fruitful in general to do so . here we collect together some possible directions of future work on the universality problem and the related string manipulations . the algebraic perspective in section [ compressor ] may have further implications for the compressor algorithm . in section [ empiricalresults ] we mentioned several conjectures inspired by empirical data . it was conjectured that the compressor directly proves the universality of all non - palindromic rotors of prime length and all ab - ba rotors of length @xmath285 ; these are the focus of our current work . another direction to take is the investigation of the universality of burd rotors , which are central to the compressor algorithm . the classification of burd strings is interesting in its own right , especially given that nontrivial examples like 1122222111 exist . also , the fascinating self - reference of the burd condition merits more attention . another problem proposed by propp asks for the universality of rotors when _ cyclic shifts of the same rotor are considered equivalent_. our results point to the conjecture that all rotors are universal under this alternative definition . the universality question can also be generalized in another direction , that of determining which finite sets of rotors are universal when taken together . resolution of the universality problem would almost completely solve this as well . we may try to quantify our results on universality ; instead of asking whether or not rotor @xmath1 makes rotor @xmath18 , we can ask _ how efficiently _ @xmath1 can make @xmath18 . such a question provides another notion of the complexity " or information value " of binary strings . the author is grateful to tanya khovanova of mit for her invaluable guidance and insight throughout this project . she provided much - needed advice during the writing of this paper . the author also received significant support from james propp of u. mass . lowell , who proposed the problem and provided expertise on rotor - routers . the present work was supported by the mit - primes program for high school students in the greater boston area . thanks go out to yan zhang of mit for providing a philosophical comment on the responsibility of memory ( to be found in section [ examples ] ) , and david yang of philips exeter academy for proofreading this paper . a. e. holroyd , l. levine , k. mszros , y. peres , j. propp and d. b. wilson , chip - firing and rotor - routing on directed graphs , _ in and out of equilibrium 2 _ , 331364 , progr . probab . * 60 * , birkhuser , 2008 .

near earth objects ( neos ) are small solar system bodies whose orbits bring them close to the earth s orbit . we are carrying out a warm spitzer cycle 11 exploration science program entitled neosurvey a fast and efficient flux - limited survey of 597 known neos in which we derive diameter and albedo for each target . the vast majority of our targets are too faint to be observed by neowise , though a small sample has been or will be observed by both observatories , which allows for a cross - check of our mutual results . our primary goal is to create a large and uniform catalog of neo properties . we present here the first results from this new program : fluxes and derived diameters and albedos for 80 neos , together with a description of the overall program and approach , including several updates to our thermal model . the largest source of error in our diameter and albedo solutions , which derive from our single band thermal emission measurements , is uncertainty in , the beaming parameter used in our thermal modeling ; for albedos , improvements in solar system absolute magnitudes would also help significantly . all data and derived diameters and albedos from this entire program are being posted on a publicly accessible webpage at nearearthobjects.nau.edu .

1608.03673

near earth objects ( neos ) are small solar system bodies whose orbits bring them close to the earth s orbit . neos lie at the intersection of science , space exploration , and civil defense . because neos are constantly being replenished from sources elsewhere in the solar system , they act as compositional and dynamical tracers and allow us to probe environmental conditions throughout the solar system and the history of our planetary system . neos are the parent bodies of meteorites , one of our key sources of detailed knowledge about the solar system s development , and neo studies provide the needed context for this work . the space exploration of neos is carried out both through robotic spacecraft such as near @xcite ; hayabusa @xcite ; change 2 @xcite ; hayabusa 2 @xcite ; and osiris - rex @xcite ; as well as remote sensing observations using iras , spitzer , wise / neowise , akari , and other assets ( as reviewed in @xcite ) . energetically , some neos are easier to reach with spacecraft than the earth s moon @xcite , and neos offer a large number of targets with a range of physical properties and histories . neos present advanced astronautical challenges , and president obama announced in 2010 that missions to asteroids would form the cornerstone of nasa s exploration for the next decade or more ; the new asteroid retrieval mission @xcite is part of those activities . finally , neos are a civil defense matter : the impact threat from neos is real , as demonstrated again recently in chelyabinsk , russia , in february , 2013 @xcite . understanding the number and properties of neos affects our planning strategies , international cooperation , and overall risk assessment . the spitzer space telescope is a powerful neo characterization system . neos typically have daytime temperatures around 300 k. hence , their thermal emission at 4.5 microns is almost always significantly larger than their reflected light at that wavelength . we can therefore employ a thermal model to derive neo properties , with the primary result being derived diameters and albedos @xcite . measuring the size distribution and albedos and compositions for a large fraction of all known neos allows us to understand the scientific , exploration , and civil defense - related properties of the neo population . spitzer s sensitivity at 4.5 microns is unparalleled , allowing us to observe many hundreds of neos over the course of the 20 month cycle 11 period . we present here a new warm spitzer exploration science program entitled neosurvey : an exploration science survey of near earth object properties ( pid11002 ; pi : trilling ) . in this paper we present the design of our program and results for the first 80 neos observed in this program ; results for the complete sample will appear in a forthcoming paper . we also introduce nearearthobjects.nau.edu , a publicly - accessible web page where all data and results from this program are posted , typically within a week of the data being released by the spitzer science center . in the 710 hours assigned to this program we use spitzer to observe 597 neos ; the vast majority of these are not accessible with any other facility . for each observed target , we use a thermal model to derive the diameter and albedo . this sample , when combined with existing data from our previous exploreneos work ( spitzer cycle 6 + 7 exploration science program ) , together with results from neowise and akari , includes nearly 20% of all neos known at the inception of this project . the majority of the infrared measurements of neos in this combined catalog with @xmath1 bodies smaller than around 175 meters diameter will come from this new program . our technical approach has been honed through observations of more than 600 neos with spitzer in our cycles 6 + 7 exploreneos program @xcite , as well as through a number of spitzer programs to study individual neos of interest ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the observation design has been proven to deliver high quality photometry . the sensitivity estimates we use here are empirically derived from our many measurements of neos , and are more appropriate for observations of moving targets than the estimates provided by spitzer s sensitivity estimating tool sens - pet . all of the technical steps are described in detail in our various papers , including our seminal exploreneos paper , @xcite . we selected our sample targets based on spitzer cycle 11 observability from the 24-oct-2014 version of the minor planet center neo list . observability was checked using the nasa horizons system for each day in cycle 11 ( february , 2015 , through september , 2016 ) ; horizons was accessed with the callhorizons python module , which retrieves ephemerides and orbital elements from the horizons system based on user queries . an object was considered a potential target when its solar elongation with respect to spitzer complied with the observatory constraints ( 82.5120 degrees ) , its positional uncertainty was less than @xmath2 arcsec at 3@xmath3 ( so that the target would fall within the detector fov ) , it had a proper motion of less than @xmath4 arcsec / s ( though in practice no objects are excluded by this constraint ) , and its distance from the galactic plane was larger than @xmath5 . these requirements are based on experience from our successful exploreneos program @xcite . from this list of potential observations and targets , we removed all targets that had already been observed by spitzer ( exploreneos : @xcite ) , wise @xcite , or akari @xcite . spitzer s infrared array camera ( irac ; @xcite ) is the only instrument available in spitzer s warm mission , providing imaging at 3.6 and 4.5 microns ( ch1 and ch2 , respectively ) . the field of view is 5.2 arcmin . fluxes from neos at ch2 are always dominated by thermal emission ( figure [ sed ] ) but the ratio of reflected light to thermal emission in ch1 can not be determined if the albedo at 3.6 microns is not known @xcite . in our previous exploreneos program @xcite we found that ch1 observations are generally not useful for thermal modeling . we therefore do not observe in ch1 in this program , saving more than 50% on our total time request . for each potential target we predicted the thermal - infrared flux density in irac ch2 as a function of time . our flux predictions were based on the absolute optical magnitude @xmath6 , a measure of @xmath7 , as reported by horizons ( @xmath8 is diameter and @xmath9 is albedo ) . @xmath6 values for neos are of notoriously low quality and tend to be skewed to brighter magnitudes @xcite . to account for this , we assumed an @xmath6 offset ( @xmath10 ) of [ 0.6 , 0.3 , 0.0 ] mag for [ low , nominal , high ] fluxes , respectively . optical fluxes were calculated from @xmath11 together with the observing geometry . in our observation planning we assumed that asteroids are 1.4 times more reflective at 4.5 microns than in the @xmath12 band @xcite , although in our analysis pipeline we use an updated value that is derived from neowise data , as described below . thermal fluxes also depend on @xmath9 ( @xmath8 is determined from @xmath6 and @xmath9 ) and @xmath0 ( see section [ modeling ] ) . we assumed @xmath9 = [ 0.4 , 0.2 , 0.05 ] for [ low , nominal , high ] thermal fluxes . the nominal @xmath0 value was determined from the solar phase angle @xmath13 using the linear relation given by @xcite , which is generally in agreement with the newer results of @xcite and the approach used in this work ( see [ beaming ] , below ) ; 0.3 was [ added , subtracted ] for [ low , high ] fluxes to capture the scatter in the empirical relationship derived in @xcite . from these parameters , we calculated the predicted thermal fluxes using the near earth asteroid thermal model ( * ? ? ? * discussed further in section [ modeling ] ) . we calculated [ low , nominal , high ] predicted fluxes in the 4.5 @xmath14 m irac band for each day of cycle 11 . the resulting neatm fluxes were convolved with the ch2 passband to yield `` color - corrected '' in - band fluxes . after removing all dates where an neo s _ high _ predicted flux could saturate the detector ( using saturation values from the irac warm mission handbook ) , we identified a five day window centered on the peak brightness during cycle 11 . we used the minimum flux calculated for the _ low _ predicted values in this five day window to calculate our integration times and build our aors . this five day window dictates the timing constraint that we place on the aor . a window of this size allows good scheduling flexibility while allowing us to use the shortest possible integration times . using the _ low _ prediction values here ensures that our snr requirement will be met or exceeded . our experience has shown that the ratio of observed to _ low _ predicted flux is in the range 110 , so our conservative prediction technique is appropriate . our photometric requirement is snr@xmath1515 , at which point , our detailed thermal modeling shows , the flux density uncertainty becomes small compared to overall model uncertainties . we have empirically measured the warm spitzer ch2 sensitivity ( 5@xmath3 ) for faint moving objects to be 2.2 @xmath14jy in 5 hours and 1.8 @xmath14jy in 10 hours , scaling roughly as the square root of integration time for shorter integrations ( mommert et al . this is less sensitive than predicted by sens - pet due to contamination from background sources that can not be completely subtracted out in the moving frame . our threshold is set to integration times of @xmath1610,000 seconds ( 2.7 hours ) , which produces an aor clock time including overheads of 3.2 hours . all sample targets have a 4.5 @xmath14 m thermal - infrared flux density of 7.2 @xmath14jy or greater . in 710.1 hours of total observation time , including all overheads , we will observe 597 neos . our final sample is shown in figures [ comparison ] and [ hfig ] . we use a standard moving single target aor , in which the telescope tracks at the appropriate moving object rates for the object , and a medium cycling dither , with only the 4.5 micron field integrating on the neo ( as we did recently for 2011 md @xcite ) . as described below , the 4.5 micron band provides the most information on the thermal emission from the neo , whereas the 3.6 micron band can have a significant ( and in general unknown ) contribution from reflected light and does not significantly improve the derivation of an albedo and size . not observing in ch1 yields significant time savings as well ; since most neos are fainter in ch1 than in ch2 ( figure [ sed ] ) , observing in ch1 would more than double the needed time . each target was assigned to one of six uniform observing templates with total frame times [ 120 , 480 , 1200 , 2400 , 6000 , 10000 ] seconds . the individual frame times are 12 sec and 30 sec for the first two bins , respectively , and 100 sec for the remainder ; this allows many dithers for each observation . the frame times used for each object were chosen based on our estimate of the source flux in order to obtain the required sensitivity using the minimum total observation time . however , we have seen that occasionally the neo is brighter than expected , possibly due to errors in the previously determined @xmath6 magnitude , or as in the case of don quixote , the result of cometary activity @xcite . we therefore use hdr mode for the 12 and 30 second frame times , which minimizes the chance of the object saturating . however , if saturation does occur , we can use a psf - fitting technique to recover the neo flux , as we did for don quixote . ( using hdr mode for the brightest targets increases the total overhead only a small amount . ) for the 100 second frame times , where the objects are very likely much fainter than the saturation limit , we do not use hdr mode . we have also checked for objects that move only slightly during the minimum observation time required to reach snr@xmath1515 . if an object moves only a small distance during the aor , it is difficult to subtract the background field and perform accurate photometry , especially if it falls in close proximity to a bright source . we have therefore found all objects that would move less than @xmath173 fwhms during the observation , and added additional integrations for these sources so that the object moves sufficiently to be able to measure and subtract the background field . this has been done for approximately 50 sources , which are all relatively bright targets and have short duration aors . the objects in our target pool are distributed all over the sky and have an essentially uniform distribution of ecliptic longitudes ( except at the galactic plane crossings ) . no aor is longer than 3.2 hours , including all overheads , and more than half the targets have aors that are 30 minutes or less . as such , this program maximizes both total neo yield and efficiency . the data processing is performed in a manner similar to our previous reductions for exploreneos @xcite and our other observations of neos @xcite . the bcds are downloaded from the spitzer archive , and we use iracproc / mopex in the moving object mode to mosaic the frames relative to the neo position . for objects that are relatively bright compared to the background , and for low background fields , the outlier rejection process in the mosaicking software is sufficient to remove background objects . for fainter objects or regions of relatively high background , we often must subtract the background field before constructing the neo mosaic . we first mosaic the data in the standard ( non - moving object ) mode to construct an image of the background field ( the neo is masked from the individual frames before making the background mosaic ) . then we subtract the background image from each bcd , and mask any remaining strong residuals from bright background objects . we then construct the neo mosaic from the background - subtracted bcds . once the neo mosaic has been constructed , we perform aperture photometry to measure the neo flux . we use an aperture correction derived from observations of standard stars during the warm mission to determine the flux . each mosaic is examined during the photometry process to ensure that there are no image artifacts present that will affect the photometry , and that the neo is found in the field and not confused with background objects . the ssc produces a data delivery approximately every two weeks that will contain @xmath1714 new neo observations from the campaign that ended two weeks prior . in most cases , we can extract fluxes and perform thermal modeling within a few days of data delivery . in order to interpret the thermal - infrared flux densities measured by spitzer , we utilize the near - earth asteroid thermal model ( neatm , * ? ? ? * ) . the neatm determines the thermal - infrared flux density emitted from the surface of an asteroid by integrating the planck function over the body s surface temperature distribution . by combining thermal - infrared observations and optical brightness measurements ( @xmath6 ) , neatm is used to fit the model flux density to the measurement in order to independently derive @xmath8 and @xmath9 of the target body . the neatm incorporates a variable adjustment to the model surface temperature through the beaming parameter @xmath0 , allowing for a correction for thermal effects of shape , spin state , thermal inertia , and surface roughness , and enabling the model and observed thermal continua to be accurately matched . neatm has been applied in the successful exploreneos @xcite and neowise @xcite programs , as well as for other asteroid observations , to measure the physical properties of a large number of neos . in contrast , more detailed thermophysical modeling @xcite would require knowledge of shape and spin state , which is generally unavailable for our poorly studied targets . even the hybrid nestm @xcite requires assumptions about or default values of thermal inertia and rotation period , neither of which is generally known for our targets . additionally , @xcite show that neatm appears to overestimate diameters at phase angles larger than 45 degrees , but only by a significant amount ( more than 15% ) for phase angles larger than 80 degrees . only a few percent of our targets are observed at phase angles this large . furthemore , ultimately the @xcite approach is irrelevant in our case because they considered only the cases of floating @xmath0 or fixed @xmath0 . our use of @xmath0 as a linear function of @xmath13 will result in different behavior , and @xcite found no evidence in our approach for a significant phase - angle - dependent bias as a result of neglecting night - side thermal emission ( their figures 4 and 5 ) . we therefore use neatm , as it adequately explains our data and does not limit our analysis or interpretation . in our ( previous ) exploreneos program we fit the neatm to @xmath6 and single - band infrared data ( ch2 only , since ch1 data are heavily contaminated with reflected solar light ; * ? ? ? * ) , deriving robust results : @xcite found that this technique leads to diameters and albedos that are accurate to within 20% and 50% , respectively , compared to previously published spacecraft , radar , and radiometric - based solutions . in this program , we apply the same modeling technique , which uses a value of @xmath0 that is not fit independently to the data but rather is based on a linear relation between @xmath0 and the solar phase angle @xmath13 , but with a more sophisticated statistical analysis of the uncertainties . in the following subsections , we provide a brief introduction to the thermal modeling pipeline and elaborate on concepts of the pipeline that differ from the exploreneos approach @xcite . the measured flux density includes both thermal emission from the asteroid s surface and reflected solar light . the latter is generally small , but in some cases may be significant . we subtract this contribution from the measured flux densites using the method described by @xcite , but with an updated reflectance ratio between the optical and ir wavelengths of [email protected] , which is based on measurements by wise @xcite , who assumed , as do we , that neo albedos are equal at 3.6 and 4.5 microns . this albedo ratio can differ throughout the neo population , and our error bar on this ratio captures this variance . we subtract the ( estimated ) reflected light contribution before calculating the color correction because reflected sunlight has a different spectral shape than asteroid thermal emission ; the reflected light spectral shape is very close to the one used in the irac flux calibration , so the corresponding color corrections for the reflected sunlight is negligible . in our previous program , exploreneos , we corrected solar flux densities by converting from irac in - band flux densities to monochromatic flux densities by integrating over the model surface temperature distribution and the product of the model spectral energy distribution and the irac ch2 bandpass @xcite , requiring an iterative approach . by contrast , in the current program , we use a different approach in which we color - correct locally emitted thermal flux density from monochromatic to irac ch2 in - band fluxes using the exact model temperature of that surface element at the time the planck equation is integrated over the surface temperature distribution . in other words , we derive separate color corrections for each spot on the surface , similar to the approach used in @xcite . using this approach , the color correction is properly iterated as part of the fitting process . the derived color corrections are in the range 5% 20% ; the only uncertainty associated comes from the ratio of reflectivity in the infrared compared to the optical @xcite , and our error bars ( described above ) already capture our lack of knowledge of this reflectance ratio @xcite . in the measured 4.5 micron flux , the reflected light fraction is between 0.01% and 29% , with a mean contribution of 1% and a median value of 0.7% . this new method produces color corrections that agree at the 12% level with our previous approach @xcite . we obtain our optical @xmath6 magnitudes from the astdys database and improve them using statistical corrections provided by @xcite over the given range of @xmath6 magnitudes ; uncertainties are also adopted from that work . for the range of @xmath19 , we apply a generic offset of 0.3 mag ( we reduce the brightness relative to the catalog ) and adopt an uncertainty of 0.3 mag , which is supported by @xcite , @xcite , and @xcite . in the future we will replace astdys optical data with more accurate @xmath6 magnitudes from ground - based surveys , where available . because we model single - band irac data , we are unable to use @xmath0 as a free - floating fitting parameter . instead , we have to make use of existing neo thermal - infrared data to derive a robust estimate for @xmath0 . as a result of asteroid surface roughness , thermal emission from the surface can be preferentially directed toward the sun , which is referred to as infrared beaming . energy conservation therefore requires that an observer at a high phase angle observes less thermal radiation than would be received from a smooth object , leading to a trend of increasing @xmath0 with solar phase angle @xmath13 . this trend has been quantified in the literature @xcite using linear relations between the parameters . in general , all of exploreneos , neosurvey , and warm neowise must use such a derived ( not fitted ) @xmath0 , though neowise was able to fit @xmath0 directly for data obtained during their cryogenic mission , where some neos were detected at multiple thermal wavelengths , and both exploreneos and warm neowise can fit both ch1 and ch2 thermal fluxes for very low albedo objects . we also use a linear relation to derive the best - fit @xmath0 based on its phase angle @xmath13 , but here we take a more sophisticated approach to sample the structure and full range in the @xmath0 distribution of neos . based on values of @xmath0 from neowise @xcite , which provides the largest published sample of fitted @xmath0 values , we derive a new relation . phase angles of the observed objects are derived by querying the minor planet center database for wise observations and represent average values . figure [ fig : etarelation ] ( left ) shows the @xmath0 data , the literature @xmath0-@xmath13 relations , and our newly derived relation ( see next paragraph ) . the neowise fractional uncertainties in @xmath0 are uniformly distributed throughout this space . the high degree of scatter in @xmath0 across the whole range of phase angle is an indication of the heterogeneity in the thermal properties of neos and is a reminder that a simple linear relation is not sufficient to describe the variety in @xmath0 . instead , the whole range of @xmath0 values must be taken into account in a statistical approach to properly interpret thermal - infrared data . here we derive a formalism that provides the most likely @xmath0 as a function of the phase angle , as well as an uncertainty distribution that samples the whole range of possible @xmath0 values . from figure [ fig : etarelation ] ( left ) , we observe that there are no neos with @xmath20 or @xmath21 ( 0.6 is the smallest value that @xmath0 can ever take on @xcite ) . furthermore , there is a clear grouping of low-@xmath0 neos but a lack of low-@xmath0 neos at high phase angles . a comparison of the @xmath0 distributions derived by @xcite and @xcite shows that the large population of neos with low @xmath0 is not a result of the thermal - infrared nature of the neowise survey , as it is also present in the optically - selected sample by @xcite . for the @xmath0 relation used in this work , we adopt a slope of 0.01 deg@xmath22 , which is in agreement with previous estimates and the slope of the lowest @xmath0 values ( we show below that a slightly different slope will not affect the results ) . we choose the intercept of our linear relation in such a way that it divides the sample of @xmath0 values in figure [ fig : etarelation ] ( left ) in two equally - sized subsamples ; the probability to overestimate @xmath0 is the same as to underestimate it . our new relation is @xmath23 . we define the residual @xmath24 as the difference between the measured values of @xmath0 and a specific @xmath25 . figure [ fig : etarelation ] ( right ) shows that @xmath24 for the individual @xmath25 relations agree well with one another ; despite the slightly different slopes , the shapes of the @xmath24 histograms align . we fit a log - normal distribution to the residuals using a least - squares method . the fitted distribution has parameters @xmath26 and @xmath27 , using the notation of @xcite . the median of the @xmath24 distribution is 0.675 . the upper and lower 1@xmath3 confidence interval of @xmath0 uncertainties spans the range from -0.3 to + 0.55 , relative to the nominal value of @xmath0 as derived by our @xmath25 relation ( 3@xmath3 interval : -0.55 to + 1.7 ) . overall , the @xmath0-@xmath13 parameter space shown in figure [ fig : etarelation ] is poorly sampled , and we assume that the averaged log - normal distribution shown in the right panel of figure [ fig : etarelation ] applies across all phase angles . the data in hand do not warrant or allow any more detailed treatment of these uncertainties . uncertainties in diameter and albedo are derived in a monte carlo approach very similar to the one described by @xcite . we perform 10,000 simulations with randomized parameters , including the measured flux densities that are varied based on the measured uncertainties ; the absolute magnitude @xmath6 , with gaussian uncertainties from @xcite or , for @xmath28 , as a gaussian with 1@xmath3 uncertainties of 0.3 ; and the optical / ir reflectance ratio , also varied as a gaussian . the beaming parameter @xmath0 is varied within the log - normal distribution described above . note that occasionally values @xmath21 or @xmath20 are drawn from the distribution ; these extreme values are rejected . clipping the @xmath0 distribution affects the final results in a negligible way and is in agreement with the observations obtained by @xcite . we adopt the median of the diameter and albedo distributions as our nominal solutions . we derive 1@xmath3 and 3@xmath3 confidence intervals as those ranges of values that bracket 68.3% and 99.7% of the values around the median , respectively . in figure [ uncertainties ] we show representative cases of the diameter and albedo probability distributions that result from our thermal modeling . the beaming parameter @xmath0 is not well constrained ( fig . [ fig : etarelation ] ) , and drives the overall uncertainties in our model results . for the results presented here , more than 90% of the total uncertainty in diameter results from uncertainty in @xmath0 . for albedo , 80%90% of the total uncertainty is due to @xmath0 uncertainties , with uncertainties in @xmath6 responsible for most of the remaining uncertainty . uncertainties in other aspects ( such as emissivity ) are about ten times smaller and are therefore negligible . finally , we make no attempt to correct for intrinsic neo lightcurves for the targets in this program . the elapsed time of our observations spans the range 10 minutes to 3.2 hours . for the longest exposures , we might integrate over an entire asteroid rotation period , as periods of seconds to a few hours hours are common for small asteroids , in which case we derive the mean diameter from our observations . for short exposures , we could easily sample just one part of the lightcurve , in which case our solution may represent the effective spherical diameter that corresponds to a lightcurve minimum or maximum , thereby under- or overestimating the true mean diameter . we estimate how many of our observations are affected by lightcurve effects with a monte carlo approach . taking into account the measured lightcurve amplitude distribution of known neos @xcite , we randomly sample sine curves with different amplitudes to simulate the offset from the lightcurve - averaged thermal flux of the target in our observations and produce diameter and albedo solutions that correspond to this instantaneous sampling . we find that 36% of all trials have diameter discrepancies greater than our quoted diameter uncertainty of 40% . hence , we expect the same fraction of our targets to have diameters affected by lightcurve - induced discrepancies greater than 1@xmath3 . for our albedo solutions , 59% of trials have solutions greater than our quoted albedo uncertainty of 70% . if we smooth the lightcurve over 20% of its period ( which is a more realistic approximation of the situation ) , we find that fewer than 5% of our diameters and fewer than 9% of our albedos are affected by this lightcurve sampling . derived albedos that are ( much ) greater than around 0.5 are suspicious , as such values are rarely seen for asteroids for which albedos have been reliably measured ( for radar or spacecraft measurements , for example , as opposed to radiometric approaches ) , and are at odds with expected surface compositions . the most likely explanations for such high albedos are @xmath6 magnitudes that are significantly incorrect and/or ( optical or thermal ) lightcurve amplitudes and periods that produce a high albedo solution . in both cases , additional ground - based photometry can allow us to correct our albedo solution . we have initiated such a program using a range of ground - based telescopes , and in a future paper we will report revised optical magnitudes and/or lightcurves for these high albedo objects , along with re - calculated albedos . ( note that poor @xmath6 values and/or lightcurves could also produce anomalously low albedos , which may be harder to identify . we will carry out ground - based observations of the objects for which we find the lowest albedos to determine if a correction is also needed for those objects . ) a small number of our targets will be observed by neowise it is impossible to calculate which targets will overlap in the future , since the neowise spice kernel ( orbit ) is only published a few months ahead and we will use those targets as cross - checks . furthermore , there are more than 100 neos that have been observed by neowise that were also observed in our exploreneos program @xcite . for these common objects , we use our updated thermal model pipeline and used only ch2 data ( as if the exploreneos data were obtained in this neosurvey program ) to rederive the diameter and albedo from our exploreneos measurements , and compare these to the published neowise diameters and albedos @xcite . the mean ratio of exploreneos values to neowise values for [ diameter , albedo ] is [ 0.9 , 1.28 ] . the median value in this ratio of exploreneos values to neowise values for [ diameter , albedo ] is [ 0.9 , 1.02 ] . this overall agreement , based on independent data measurements and modeling , is quite good , although we caution that for any individual object significantly different solutions may exist . for diameter , 70% of the newly modeled exploreneos targets have 1@xmath3 error bars that overlap with neowise values ; 98% of the targets agree at 3@xmath3 . for albedo , 66% of the newly modeled exploreneos targets have 1@xmath3 error bars that overlap with neowise values ; 98% of the targets agree at 3@xmath3 . this suggests that both the errors are largely random , and that our error bar estimations are appropriate . the largest deviation is in albedo , where uncertainties in @xmath6 magnitude typically introduce large errors , and where the two approaches frequently use somewhat different @xmath6 magnitudes for their modeling . in the future we will carry out a more comprehensive comparison , considering targets that overlap with both exploreneos and neosurvey , and using common @xmath6 values for the thermal modeling of the different thermal measurements . we report here results for the first 80 neos observed in this program . these neos were observed in february and march , 2015 . table [ datatable ] gives the geometries , measured fluxes , and derived diameter and albedo results for these targets . as described above , the probability distributions for the results are asymmetric and non - gaussian . we therefore provide 1@xmath3 and 3@xmath3 upper and lower diameter and albedo values for each target . we introduce here nearearthobjects.nau.edu , a publicly accessible webpage where all information from this program is posted . table [ webpage ] lists the properties for each observed object that are available in our online database , where we list complete observing circumstances and modeling results . the observing cadence for this program is roughly one target per day . data is made available to us at a biweekly cadence . it takes us typically 12 days from receiving the data to producing model results . therefore , our online database grows continuously , in roughly two week intervals , over the duration of this program . the latest version of our results will always be available on that webpage . because of the monte carlo nature of our modeling solutions , the values in this online table can very slightly with time , but all such changes are insignificant compared to the size of the error bars . the database ultimately will also host all exploreneos data , with results derived from the latest version of our pipeline . relevant publications will also be posted at that webpage . as our survey proceeds and our catalog grows , a number of science investigations will be enabled , including the following . ( 1 ) our sample is optically selected , so any study of the properties of the entire neo population requires rigorous debiasing , which is a signficant challenge . using debiasing approaches that we have developed @xcite and continue to enhance , we will measure the size distribution of neos down to 100 meters . we will observe nearly 200 neos smaller than 200 meters ( @xmath28 ; figure [ hfig ] ) . this will allow us to calculate an independent measurement of the neo size distribution , one that is derived from a sample size at 100 meters that is more than three times as large as all the previous thermal infrared surveys combined . ( 2 ) by applying those same debiasing approaches to the combination of neosurvey , exploreneos , and neowise data we will derive the distribution of albedos and compositions as a function of size . ( 3 ) we will search for `` dead comets '' bodies that dynamically , and presumably compositionally , are of cometary origin , but typically exhibit no cometary activity . our target sample includes 32 potential dead comets , based on their low , comet - like tisserand parameters , @xmath29 . our albedo measurements for all these objects may provide further evidence for their cometary nature ( cometary albedos are extraordinarily low ) , as it did for the neo don quixote @xcite . ( 4 ) there are 109 targets in our sample that obey the usual `` mission accessible '' definition of @xmath30 km / s ( and six targets that have @xmath31 km / s ) . we will measure diameters and albedos for these neos and , together with historical temperature modeling as in @xcite , identify the most primitive objects and therefore most interesting for mission planning in our sample . ( 5 ) by determining the albedos of neos , we can measure the compositions of small bodies in the various source regions . these results can be compared to models from @xcite and recent work by @xcite to make connections among meteorite , neos , and parent asteroids or asteroid families in the main belt , thereby providing geologic , mineralogic , and solar system context to the objects studied in our laboratories and by our space missions . the preceding list is not meant to be exhaustive , as many other investigations will be possible , both by members of the survey team and other researchers . the ultimate result will be a catalog of some 2000 neo properties , derived from the union of neowise , exploreneos , akari , and neosurvey results . this will represent nearly 20% of all known neos in late 2016 when our neosurvey concludes . we acknowledge the thorough and prompt hard work of the staff at the spitzer science center , without whom the execution of this program would not be possible . we also appreciate the support that nasa s planetary science division and solar system observations program provide for the spitzer mission and its neo observations . we thank two anonymous referees for their detailed comments that improved this paper . we also thank scott gaudi for useful advice . some of the computational analyses were run on northern arizona university s monsoon computing cluster , funded by arizona s technology and research initiative fund . this work is based in part on observations made with the spitzer space telescope , which is operated by jpl / caltech under a contract with nasa . support for this work was provided by nasa through an award issued by jpl / caltech . we make use of data from the astdys database as well as extensive use of the jpl / horizons system . mommert , m. hora , j. l. , harris , a. w. , et al . 2014a , apj , 781 , 25 mommert , m. , hora , j. l. , farnocchia , d. , et al . 2014b , apj , 786 , 148 mommert , m. , farnocchia , d. , hora , j. l. , et al . 2014c , apjl , 789 , 22 mommert , m. , harris , a. w. , mueller , m. , et al . 2015 , aj , submitted magnitude . the dots show all neos accessible with spitzer in cycle 11 . dashed lines indicate the @xmath32 sensitivities for the neowise survey and this program . the sensitivity limit for neowise w2 ( 890 @xmath14jy ) is derived from flux densities tabulated in mainzer et al . ( 2014b ) and the flux calibration information in wright et al . we set our integration cutoff to be 10,000 seconds , and observe all targets whose peak brightnesses are greater than 7.2 @xmath14jy , as indicated : all dots above the dotted line marked `` this program . '' there are many small neos in this sample , with some 200 targets smaller than 200 meters ( @xmath28 ) . the vast majority of our targets are too faint to be observed by neowise . objects above the neowise flux limit as seen by spitzer may not be bright enough to be detected by neowise when in the neowise field of view , and the reverse : some objects that are below the neowise faint limit as seen by spitzer may be brighter when they are observed by neowise . therefore , a small number of neos will be observed by both spitzer and neowise , which allows for a cross - check of our mutual results . [ comparison ] ] magnitudes for targets to be observed in this sample ( thin solid red line and unshaded histogram ) and by neowise during spitzer s cycle 11 ( shaded gray region ) . the 80 targets observed to date in this program are shown as the shaded red histogram . because the neowise spice kernel ( orbit ) is only published a few months ahead , it is impossible to calculate exactly which neos will be observed by neowise during spitzer s cycle 11 , but we can closely estimate the @xmath6 distribution by scaling the neowise yield since reactivation to the duration of spitzer s cycle 11 . a diameter of 100 meters corresponds roughly to @xmath6 of 22.5 . at @xmath28 , this project contains more than three times as many targets as neowise and exploreneos combined . [ hfig ] ] rlllllcccccc 8567 ( 1996 hw1 ) & 2.07 & 1.90 & 29.3 & 15.76 & 0.22 & 91@xmath189 & 1.14 & 2900 ( -900/+2400 ) & 150012100 & 0.10 ( -0.07/+0.13 ) & 0.000.42 + 35396 ( 1997 xf11 ) & 1.36 & 0.59 & 43.7 & 17.11 & 0.22 & 519@xmath1822 & 1.29 & 940 ( -220/+480 ) & 3801930 & 0.29 ( -0.16/+0.21 ) & 0.030.60 + 162117 ( 1998 sd15 ) & 1.25 & 0.46 & 50.2 & 19.41 & 0.22 & 147@xmath1811 & 1.34 & 350 ( -80/+160 ) & 140660 & 0.25 ( -0.14/+0.18 ) & 0.020.49 + 208023 ( 1999 aq10 ) & 1.08 & 0.14 & 59.3 & 20.60 & 0.30 & 365@xmath1818 & 1.44 & 148 ( -26/+52 ) & 52201 & 0.46 ( -0.22/+0.25 ) & 0.060.69 + ( 2004 tp1 ) & 1.03 & 0.17 & 81.8 & 20.90 & 0.30 & 315@xmath1816 & 1.65 & 207 ( -38/+74 ) & 70253 & 0.18 ( -0.09/+0.11 ) & 0.030.36 + 390929 ( 2005 gp21 ) & 1.04 & 0.17 & 76.5 & 20.80 & 0.30 & 238@xmath1814 & 1.60 & 179 ( -32/+62 ) & 62222 & 0.26 ( -0.12/+0.15 ) & 0.040.47 + 434734 ( 2006 fx ) & 1.05 & 0.36 & 75.4 & 20.30 & 0.30 & 66@xmath188 & 1.59 & 198 ( -35/+69 ) & 72242 & 0.34 ( -0.16/+0.19 ) & 0.060.55 + ( 2008 su1 ) & 1.19 & 0.63 & 58.6 & 19.51 & 0.22 & 113@xmath1810 & 1.42 & 410 ( -90/+190 ) & 170720 & 0.16 ( -0.09/+0.11 ) & 0.020.36 + ( 2008 uf7 ) & 1.41 & 0.62 & 40.7 & 19.41 & 0.22 & 54@xmath187 & 1.25 & 330 ( -80/+170 ) & 140740 & 0.28 ( -0.17/+0.22 ) & 0.020.65 + ( 2010 xp51 ) & 1.24 & 0.39 & 48.1 & 18.93 & 0.22 & 2119@xmath1842 & 1.32 & 970 ( -240/+510 ) & 4302070 & 0.05 ( -0.03/+0.04 ) & 0.000.14 + ( 2011 sm68 ) & 1.09 & 0.50 & 68.3 & 19.89 & 0.22 & 57@xmath187 & 1.51 & 241 ( -44/+90 ) & 86338 & 0.33 ( -0.16/+0.18 ) & 0.050.51 + ( 2011 ws2 ) & 1.52 & 1.22 & 41.8 & 17.59 & 0.22 & 149@xmath1811 & 1.26 & 1240 ( -340/+750 ) & 5903360 & 0.10 ( -0.06/+0.10 ) & 0.010.33 + ( 2012 ad3 ) & 1.12 & 0.47 & 64.7 & 19.89 & 0.22 & 96@xmath189 & 1.48 & 290 ( -60/+120 ) & 110450 & 0.24 ( -0.12/+0.14 ) & 0.030.43 + 14827 hypnos ( 1986 jk ) & 1.32 & 0.93 & 50.2 & 18.65 & 0.22 & 65@xmath187 & 1.35 & 520 ( -120/+260 ) & 2201030 & 0.22 ( -0.12/+0.17 ) & 0.020.50 + 4487 pocahontas ( 1987 ua ) & 1.22 & 0.66 & 56.4 & 17.68 & 0.22 & 789@xmath1826 & 1.41 & 1150 ( -270/+570 ) & 4802110 & 0.11 ( -0.06/+0.09 ) & 0.010.29 + 5332 davidaguilar ( 1990 da ) & 1.64 & 1.32 & 38.2 & 15.09 & 0.22 & 634@xmath1824 & 1.22 & 3300 ( -900/+2000 ) & 15009200 & 0.15 ( -0.10/+0.14 ) & 0.010.44 + 8034 akka ( 1992 lr ) & 1.41 & 0.78 & 44.8 & 18.16 & 0.22 & 78@xmath188 & 1.29 & 540 ( -120/+260 ) & 2201170 & 0.33 ( -0.18/+0.23 ) & 0.030.63 + 32906 ( 1994 rh ) & 1.33 & 0.66 & 48.2 & 16.15 & 0.22 & 2266@xmath1843 & 1.32 & 2080 ( -510/+1100 ) & 8804510 & 0.14 ( -0.08/+0.11 ) & 0.010.36 + ( 1998 wp7 ) & 1.07 & 0.35 & 71.6 & 20.18 & 0.22 & 133@xmath1811 & 1.55 & 256 ( -49/+93 ) & 93346 & 0.23 ( -0.11/+0.13 ) & 0.040.38 + 137078 ( 1998 xz4 ) & 1.40 & 0.82 & 46.1 & 16.63 & 0.22 & 274@xmath1816 & 1.30 & 1070 ( -230/+510 ) & 4302140 & 0.34 ( -0.19/+0.23 ) & 0.030.61 + 53409 ( 1999 lu7 ) & 1.19 & 0.54 & 58.5 & 19.03 & 0.22 & 232@xmath1814 & 1.42 & 510 ( -110/+240 ) & 200870 & 0.17 ( -0.09/+0.12 ) & 0.020.35 + 102873 ( 1999 wk11 ) & 1.58 & 1.00 & 38.6 & 17.78 & 0.22 & 32@xmath185 & 1.23 & 570 ( -130/+290 ) & 2401430 & 0.43 ( -0.25/+0.31 ) & 0.030.85 + 178601 ( 2000 cg59 ) & 1.48 & 1.21 & 42.8 & 17.88 & 0.22 & 90@xmath189 & 1.26 & 920 ( -240/+530 ) & 4202340 & 0.15 ( -0.09/+0.13 ) & 0.010.41 + 159495 ( 2000 uv16 ) & 1.37 & 0.67 & 44.9 & 17.40 & 0.22 & 329@xmath1817 & 1.29 & 870 ( -200/+440 ) & 3501920 & 0.26 ( -0.15/+0.19 ) & 0.020.53 + 162723 ( 2000 vm2 ) & 1.21 & 0.35 & 48.2 & 17.68 & 0.22 & 812@xmath1827 & 1.32 & 600 ( -120/+250 ) & 2101040 & 0.42 ( -0.21/+0.25 ) & 0.040.65 + 162741 ( 2000 wg6 ) & 1.27 & 0.44 & 46.1 & 17.78 & 0.22 & 17336@xmath18191 & 1.30 & 3100 ( -800/+1700 ) & 15007100 & 0.01 ( -0.01/+0.01 ) & 0.000.04 + ( 2000 wp19 ) & 1.04 & 0.06 & 68.1 & 22.70 & 0.30 & 2827@xmath1848 & 1.52 & 176 ( -37/+73 ) & 68261 & 0.05 ( -0.02/+0.04 ) & 0.010.13 + 68346 ( 2001 kz66 ) & 1.35 & 0.95 & 48.9 & 17.11 & 0.22 & 219@xmath1814 & 1.32 & 1010 ( -230/+520 ) & 4002120 & 0.25 ( -0.14/+0.17 ) & 0.030.51 + 283460 ( 2001 pd1 ) & 1.21 & 0.75 & 56.5 & 18.55 & 0.22 & 230@xmath1814 & 1.40 & 700 ( -160/+320 ) & 2901300 & 0.14 ( -0.07/+0.10 ) & 0.020.30 + 317685 ( 2003 no4 ) & 1.53 & 0.81 & 37.5 & 18.45 & 0.22 & 74@xmath188 & 1.21 & 570 ( -150/+320 ) & 2601550 & 0.22 ( -0.13/+0.19 ) & 0.010.57 + 143992 ( 2004 af ) & 1.33 & 0.94 & 49.6 & 16.43 & 0.22 & 1064@xmath1830 & 1.33 & 2050 ( -510/+1100 ) & 8804440 & 0.11 ( -0.07/+0.09 ) & 0.010.28 + 267940 ( 2004 em20 ) & 1.15 & 0.33 & 57.4 & 20.50 & 0.30 & 61@xmath187 & 1.42 & 160 ( -30/+62 ) & 58245 & 0.44 ( -0.21/+0.25 ) & 0.060.72 + 154715 ( 2004 lb6 ) & 1.14 & 0.60 & 62.7 & 18.74 & 0.22 & 129@xmath1810 & 1.47 & 430 ( -80/+170 ) & 150650 & 0.30 ( -0.15/+0.18 ) & 0.040.50 + ( 2004 qd14 ) & 1.04 & 0.12 & 73.9 & 20.90 & 0.30 & 352@xmath1817 & 1.57 & 143 ( -24/+49 ) & 47174 & 0.37 ( -0.18/+0.20 ) & 0.050.58 + 318160 ( 2004 qz2 ) & 1.14 & 0.47 & 62.4 & 18.36 & 0.22 & 256@xmath1815 & 1.46 & 490 ( -90/+190 ) & 170730 & 0.33 ( -0.16/+0.19 ) & 0.040.53 + ( 2005 xy ) & 1.35 & 0.76 & 48.5 & 19.13 & 0.22 & 97@xmath189 & 1.33 & 520 ( -130/+270 ) & 2301140 & 0.15 ( -0.08/+0.12 ) & 0.020.38 + ( 2006 vd13 ) & 1.09 & 0.31 & 68.0 & 19.32 & 0.22 & 360@xmath1818 & 1.53 & 370 ( -70/+140 ) & 130500 & 0.24 ( -0.12/+0.14 ) & 0.030.41 + ( 2006 vd2 ) & 1.10 & 0.41 & 67.6 & 19.51 & 0.22 & 314@xmath1816 & 1.51 & 430 ( -90/+180 ) & 170640 & 0.15 ( -0.08/+0.09 ) & 0.020.28 + 375103 ( 2007 td71 ) & 1.30 & 0.70 & 50.7 & 18.74 & 0.22 & 91@xmath189 & 1.34 & 460 ( -100/+220 ) & 190910 & 0.26 ( -0.15/+0.18 ) & 0.020.53 + ( 2008 sq ) & 1.25 & 0.51 & 52.1 & 19.80 & 0.22 & 86@xmath189 & 1.36 & 300 ( -70/+140 ) & 120580 & 0.24 ( -0.13/+0.17 ) & 0.020.49 + ( 2010 ku10 ) & 1.15 & 0.40 & 60.5 & 19.80 & 0.22 & 236@xmath1814 & 1.44 & 360 ( -80/+160 ) & 140620 & 0.16 ( -0.08/+0.11 ) & 0.020.34 + ( 2010 tj7 ) & 1.17 & 0.41 & 58.4 & 19.99 & 0.22 & 185@xmath1813 & 1.42 & 330 ( -70/+150 ) & 130570 & 0.17 ( -0.09/+0.11 ) & 0.020.34 + ( 2010 vn65 ) & 1.11 & 0.40 & 66.1 & 20.40 & 0.30 & 57@xmath187 & 1.50 & 193 ( -37/+72 ) & 70281 & 0.32 ( -0.16/+0.20 ) & 0.040.65 + ( 2011 hr ) & 1.25 & 0.47 & 50.8 & 20.09 & 0.22 & 42@xmath186 & 1.35 & 202 ( -40/+88 ) & 77366 & 0.40 ( -0.21/+0.24 ) & 0.040.70 + ( 2012 as10 ) & 1.56 & 1.20 & 40.5 & 17.59 & 0.22 & 44@xmath186 & 1.25 & 750 ( -180/+420 ) & 3301900 & 0.29 ( -0.17/+0.23 ) & 0.020.64 + ( 2014 mq18 ) & 1.57 & 0.92 & 38.1 & 16.15 & 0.22 & 4023@xmath1860 & 1.22 & 4800 ( -1400/+3200 ) & 240014000 & 0.03 ( -0.02/+0.03 ) & 0.000.10 + 152560 ( 1991 bn ) & 1.33 & 0.55 & 43.6 & 19.32 & 0.22 & 62@xmath187 & 1.29 & 290 ( -60/+140 ) & 120570 & 0.38 ( -0.21/+0.25 ) & 0.040.71 + 337053 ( 1996 xw1 ) & 1.10 & 0.44 & 67.1 & 19.22 & 0.22 & 105@xmath189 & 1.51 & 296 ( -52/+106 ) & 100393 & 0.41 ( -0.20/+0.21 ) & 0.060.59 + 312956 ( 1997 cz3 ) & 1.04 & 0.17 & 74.8 & 19.80 & 0.22 & 3386@xmath1854 & 1.59 & 580 ( -120/+230 ) & 220780 & 0.06 ( -0.03/+0.04 ) & 0.010.13 + 53430 ( 1999 ty16 ) & 1.53 & 1.02 & 41.0 & 17.01 & 0.22 & 348@xmath1817 & 1.25 & 1580 ( -430/+950 ) & 7404220 & 0.11 ( -0.07/+0.10 ) & 0.010.33 + 331509 ( 1999 ya ) & 1.41 & 0.66 & 41.1 & 18.45 & 0.22 & 104@xmath189 & 1.25 & 490 ( -110/+260 ) & 2001170 & 0.31 ( -0.18/+0.22 ) & 0.030.64 + 68063 ( 2000 yj66 ) & 2.18 & 1.51 & 24.1 & 15.86 & 0.22 & 61@xmath187 & 1.10 & 2100 ( -700/+1700 ) & 11009800 & 0.18 ( -0.13/+0.22 ) & 0.010.63 + 159533 ( 2001 hh31 ) & 1.75 & 1.08 & 31.9 & 17.97 & 0.22 & 37@xmath186 & 1.17 & 710 ( -200/+460 ) & 3502390 & 0.22 ( -0.14/+0.22 ) & 0.010.64 + 286079 ( 2001 tw1 ) & 1.32 & 0.52 & 44.5 & 19.41 & 0.22 & 64@xmath187 & 1.28 & 280 ( -60/+130 ) & 110580 & 0.39 ( -0.21/+0.25 ) & 0.040.70 + ( 2002 vo85 ) & 1.11 & 0.19 & 53.0 & 21.80 & 0.30 & 149@xmath1811 & 1.37 & 113 ( -24/+50 ) & 45189 & 0.26 ( -0.14/+0.19 ) & 0.030.59 + ( 2002 vv17 ) & 1.08 & 0.15 & 60.4 & 20.18 & 0.22 & 803@xmath1826 & 1.44 & 229 ( -43/+91 ) & 80332 & 0.28 ( -0.14/+0.16 ) & 0.040.49 + 280244 ( 2002 wp11 ) & 1.58 & 0.82 & 34.4 & 18.45 & 0.22 & 59@xmath187 & 1.19 & 540 ( -140/+320 ) & 2401510 & 0.25 ( -0.16/+0.22 ) & 0.010.64 + ( 2004 jr ) & 1.22 & 0.76 & 56.1 & 19.03 & 0.22 & 234@xmath1814 & 1.40 & 710 ( -170/+340 ) & 3001360 & 0.09 ( -0.05/+0.07 ) & 0.010.22 + 351508 ( 2005 rn33 ) & 1.37 & 0.55 & 40.4 & 19.89 & 0.22 & 47@xmath186 & 1.25 & 250 ( -60/+130 ) & 110580 & 0.30 ( -0.17/+0.22 ) & 0.020.63 + ( 2005 vc2 ) & 1.23 & 0.40 & 49.3 & 18.07 & 0.22 & 29171@xmath18204 & 1.33 & 3600 ( -900/+1900 ) & 16007400 & 0.01 ( -0.00/+0.01 ) & 0.000.03 + ( 2005 xt77 ) & 1.05 & 0.13 & 69.1 & 21.20 & 0.30 & 177@xmath1812 & 1.52 & 112 ( -19/+38 ) & 37137 & 0.46 ( -0.21/+0.23 ) & 0.070.66 + 417949 ( 2007 tb23 ) & 1.40 & 0.62 & 40.3 & 19.03 & 0.22 & 37@xmath186 & 1.24 & 290 ( -60/+130 ) & 110630 & 0.51 ( -0.28/+0.31 ) & 0.040.80 + ( 2007 wb5 ) & 1.14 & 0.49 & 62.8 & 19.41 & 0.22 & 93@xmath189 & 1.46 & 300 ( -60/+120 ) & 110460 & 0.33 ( -0.16/+0.19 ) & 0.050.53 + ( 2009 ff19 ) & 1.03 & 0.05 & 72.9 & 21.70 & 0.30 & 1938@xmath1841 & 1.56 & 136 ( -26/+49 ) & 47176 & 0.20 ( -0.10/+0.13 ) & 0.030.38 + ( 2009 kn4 ) & 1.09 & 0.52 & 67.7 & 18.55 & 0.22 & 365@xmath1818 & 1.51 & 590 ( -120/+240 ) & 210870 & 0.19 ( -0.10/+0.12 ) & 0.030.35 + 407338 ( 2010 rq30 ) & 1.18 & 0.29 & 49.1 & 18.55 & 0.22 & 1228@xmath1832 & 1.33 & 530 ( -120/+250 ) & 210990 & 0.23 ( -0.13/+0.16 ) & 0.020.46 + ( 2010 vf1 ) & 1.14 & 0.28 & 57.0 & 20.70 & 0.30 & 241@xmath1814 & 1.40 & 229 ( -49/+104 ) & 90388 & 0.17 ( -0.09/+0.13 ) & 0.010.42 + 436761 ( 2012 dn ) & 1.31 & 0.85 & 50.8 & 18.55 & 0.22 & 1015@xmath1829 & 1.35 & 1730 ( -450/+950 ) & 7903730 & 0.02 ( -0.01/+0.02 ) & 0.000.07 + ( 2012 rg3 ) & 1.07 & 0.24 & 69.9 & 18.26 & 0.22 & 1332@xmath1833 & 1.54 & 550 ( -100/+200 ) & 180720 & 0.29 ( -0.14/+0.16 ) & 0.050.46 + ( 2012 ua34 ) & 1.09 & 0.38 & 68.4 & 19.89 & 0.22 & 71@xmath188 & 1.53 & 214 ( -37/+74 ) & 71268 & 0.43 ( -0.19/+0.21 ) & 0.070.59 + ( 2013 qk48 ) & 1.22 & 0.52 & 55.2 & 18.65 & 0.22 & 242@xmath1814 & 1.38 & 510 ( -110/+230 ) & 200910 & 0.24 ( -0.12/+0.15 ) & 0.030.44 + ( 2014 nf64 ) & 1.29 & 0.83 & 51.8 & 18.65 & 0.22 & 146@xmath1811 & 1.35 & 660 ( -160/+340 ) & 2801330 & 0.14 ( -0.08/+0.11 ) & 0.010.34 + ( 2014 ol339 ) & 1.05 & 0.12 & 67.7 & 23.00 & 0.30 & 71@xmath188 & 1.51 & 58 ( -11/+22 ) & 2183 & 0.33 ( -0.16/+0.19 ) & 0.040.57 + ( 2014 qw296 ) & 1.55 & 0.81 & 36.0 & 19.13 & 0.22 & 63@xmath187 & 1.20 & 520 ( -150/+320 ) & 2501470 & 0.15 ( -0.09/+0.14 ) & 0.010.48 + 136618 ( 1994 cn2 ) & 1.12 & 0.60 & 63.5 & 17.11 & 0.22 & 339@xmath1818 & 1.48 & 730 ( -130/+270 ) & 240970 & 0.47 ( -0.22/+0.23 ) & 0.070.61 + 137427 ( 1999 tf211 ) & 2.27 & 2.12 & 26.3 & 15.47 & 0.22 & 46@xmath186 & 1.11 & 2900 ( -1000/+2500 ) & 160013500 & 0.14 ( -0.10/+0.18 ) & 0.010.54 + 54509 yorp ( 2000 ph5 ) & 1.04 & 0.05 & 58.5 & 22.90 & 0.30 & 282@xmath1816 & 1.42 & 47 ( -8/+15 ) & 1660 & 0.56 ( -0.25/+0.27 ) & 0.090.71 + 228368 ( 2000 wk10 ) & 1.12 & 0.52 & 64.2 & 18.65 & 0.22 & 206@xmath1813 & 1.48 & 470 ( -90/+180 ) & 170680 & 0.28 ( -0.14/+0.16 ) & 0.040.47 + ( 2002 aq2 ) & 1.09 & 0.42 & 68.5 & 18.93 & 0.22 & 364@xmath1818 & 1.52 & 490 ( -100/+190 ) & 170690 & 0.20 ( -0.10/+0.12 ) & 0.030.37 + ( 2012 gg1 ) & 1.18 & 0.71 & 58.6 & 18.93 & 0.22 & 219@xmath1814 & 1.43 & 620 ( -140/+270 ) & 2501080 & 0.12 ( -0.06/+0.09 ) & 0.010.28 + llll object number & object name & object designation + semi - major axis & eccentricity & inclination + long . node @xmath33 & arg . perihelion @xmath34 & orbital period + observed ra & observed dec & 3@xmath3 ra uncert . & 3@xmath3 dec uncert . + v magnitude & heliocentric distance & spitzer - centric distance + phase angle & solar elong . angle & galactic latitude & galactic longitude + date of observation & jd of observation & aorkey + frame time & total integration time & elapsed time & notes about the observation or data quality + ch2 flux & ch2 error & ch2 snr + @xmath6 magnitude & uncertainty on @xmath6 magnitude & g ( slope parameter ) + @xmath0 & @xmath0 : @xmath181@xmath3 & @xmath0 : @xmath183@xmath3 + diameter & diameter : @xmath181@xmath3 & diameter : @xmath183@xmath3 + albedo & albedo : @xmath181@xmath3 & albedo : @xmath183@xmath3 +

in this paper a novel neuro - fuzzy system is proposed where its learning is based on the creation of fuzzy relations by using new implication method without utilizing any exact mathematical techniques . then , a simple memristor crossbar - based analog circuit is designed to implement this neuro - fuzzy system which offers very interesting properties . in addition to high connectivity between neurons and being fault - tolerant , all synaptic weights in our proposed method are always non - negative and there is no need to precisely adjust them . finally , this structure is hierarchically expandable and can compute operations in real time since it is implemented through analog circuits . simulation results show the efficiency and applicability of our neuro - fuzzy computing system . they also indicate that this system can be a good candidate to be used for creating artificial brain .

1103.1156

in the field of artificial intelligence , neuro - fuzzy refers to combination of artificial neural networks and fuzzy logics trying to use benefits of these two fields . perhaps the most important advantage of neural networks is their adaptivity . adaptivity comes from learning capability of neural networks which allows these networks to perform well even when the environment varies over time like what human brain does . another significant benefit of neural networks relates to their huge connectivity ( again inspired from real brain ) that offers high fault tolerance and parallel processing power which is also assumed to be the reason of high efficiency of biological systems . on the other hand , fuzzy logic performs an inference mechanism under cognitive uncertainty and provides an inference morphology that enables approximate human reasoning capabilities to be applied to knowledge - based systems . however , in fuzzy inference systems , it usually takes a lot of time to design and tune membership functions and rules . to overcome this problem , similar to what is done in neuro - fuzzy systems , learning techniques of neural networks can automated this tuning process to reduce development time . therefore , it can be said that current neuro - fuzzy systems like anfis @xcite , rulenet @xcite and garic @xcite are fuzzy systems that use a learning algorithm derived from neural network theory to determine their parameters ( fuzzy sets and fuzzy rules ) by processing training data . however , eventually all of these neuro - fuzzy systems are trying to approach an ideal soft - computing tool where the nature of its computing or inference be as similar as possible to computation and inference in human brain . although now we can see significant progresses in this area in the software domain , samples of successful hardware which could even approach computing capabilities of real brain are very rare . one of the main obstacle in front of this purpose relates to the ability of efficiently modeling and construction of synapses . actually , highly parallel processing power of biological systems comes from large connectivity between neurons through synapses ( each neuron in human brain is connected to about 10000 other neurons through these synapses ) and it is widely believed that the adaptation of synaptic weights enables the biological systems to learn and function . therefore , efficient construction of synaptic weights is a critical factor in the success of final system . after the first physical realization of memristor @xcite , it becomes clear that this passive element can be a perfect representative of synapse since similar to synapse , its conductance can be precisely modulated by passing charge and flux through it . in the past few years and by the discovery of memristor and improvements achieved in the construction of powerful digital hardware , extensive works are in progress to build an artificial brain that can adaptively interact with the world in real time . among then we can name projects like ifat 4 g at john hopkins university , brainscales in the european union s neuromorphic chip program , brain simulator c2 introduced by ibm , and modular neural exploring traveling agent ( moneta ) . the most recent one between these projects , _ i.e. _ moneta which is the part of the darpa s synapse program , is a software developed by the researchers at boston university which will run on a brain - inspired microprocessor under development at hp labs in california @xcite . in this still under construction system , in spite of internal structure of real brain , all units are implemented in digital with separate memory and computational units . however , it has been argued that by constructing memristive memories ( to store synaptic weights ) and putting them very close to computational circuits that read and write them , signalling losses and power consumption can be minimized @xcite . in this paper , we introduce another approach to construct a simple neuro - fuzzy computing system which differs significantly from other systems in this category . in fact , we believe that structures similar to what is proposed in this paper merit more to be called neuro - fuzzy compared to currently available systems . this is because of the fact that as can be seen in the rest of the paper , in our proposed system neural network and fuzzy logic fields are completely involved with each other and it is hard to distinguish that which part belongs to fuzzy logic and which part relates to neural networks . moreover , there is no use of exact mathematical techniques . as another advantage , in this system memristor crossbar is used to store synaptic weights but by this difference that memory is assimilated with computational units like what we have in human brain . in addition , because of large connectivity between input and output neurons , this proposed structure is completely fault - tolerant . actually we have shown that even if during the fabrication process or execution phase of the system near half of the memristors become faulty , system can still continue working satisfactory . on the other hand , learning in our structure is based on the creation of fuzzy relations which is shown to be equal to primary hebbian learning rule and therefore all synaptic weights will be always non - negative . however , in spite of almost all developed learning methods , there is no need to precisely adjust these weights which are represented by memristors . finally , our neuro - fuzzy system is hierarchically expandable and since it is constructed with analog circuits , its computations are almost in real - time . the paper is organized as follows . inference method in our proposed neuro - fuzzy system is described in section [ neurofuzzy ] . section [ hardware ] is devoted to the explanation of the hardware implementation of proposed inference method based on memristor crossbar structure . the reason that we have considered our proposed system as a neuro - fuzzy one is presented in section [ reason ] . eventually , a few experimental results are presented in section [ simulation ] , before conclusions in section [ conclusion ] . in this paper we propose a new fuzzy structure with biological support which can do some inferential tasks in fuzzy form . in addition , this fuzzy structure has learning capability like human brain but in spite of neural networks , it performs any operation based on concepts of fuzzy logic . this means that in our proposed structure , inputs and outputs of the structure are fuzzy numbers and any calculation in the structure for doing inferential processes is done in fuzzy without using any precise and accurate mathematical techniques . figure [ figa:1 ] shows such a typical single input single output ( siso ) system where in this system , fuzzy input and output variables are considered to be discrete for simplicity . if this system was a non - fuzzy system , its input or output would be a single crisp number . however , since this system is a fuzzy one , its input and output should be fuzzy numbers or set of pairs like what is depicted in fig . [ figa:1 ] . since the construction of this system with this kind of input and output terminals is impossible ( from hardware aspect ) , we modify input and output terminals of the fuzzy system of fig.[figa:1 ] such as the one presented in fig . [ figa:2 ] . in this figure , i_th input and the _ j_th output of the system represent concepts @xmath0 and @xmath1 respectively . note that resolution or domain of input or output variables can be increased by adding input or output terminals to the system where each of these newly added terminals will represent new values of input or output variables . now , if in the system of fig . [ figa:2 ] , the input applied to the _ i_th input terminal which represents concept @xmath0 be @xmath2 ( @xmath3 : membership function of input fuzzy number ) and the output at the _ j_th output terminal which represents concept @xmath1 be @xmath4 ( @xmath5 : membership function of output fuzzy number ) , the system of fig . [ figa:2 ] will be equal to the fuzzy system of fig . [ figa:1 ] but by this difference that the construction of the fuzzy system shown in fig . [ figa:2 ] is much simpler . consequently , input of the system of fig . [ figa:2 ] is a vector of membership degrees , _ i.e. _ @xmath6 $ ] , and the output of this system is also a vector of membership degrees , _ i.e. _ @xmath7 $ ] where combination of these membership grades with those concepts which are assigned to input and output terminals creates input and output fuzzy numbers . as a result , we again have a totally fuzzy system with fuzzy input and fuzzy output variables . for example , consider a typical fuzzy system shown in fig . [ figa:3 ] . in this example , input of the system is a fuzzy number @xmath8 which can be roughly considered as a fuzzy representation of crisp number @xmath9 and the output of this system for this input is a fuzzy number @xmath10 where its deffuzzification may probably result in a crisp number @xmath11 . now , let s see how the fuzzy inference system of fig . [ figa:2 ] which relates the fuzzy input variable to fuzzy output variable can be constructed . it is well known that the compositional rule of inference describes a composition of a fuzzy set and a fuzzy relation . any fuzzy rule in the form of @xmath12 is usually represented by a fuzzy relation @xmath13 @xcite . having given input linguistic value @xmath14 , we can infer an output fuzzy set @xmath15 by the composition of the fuzzy set @xmath14 and the relation @xmath13 . therefore , doing any inference between fuzzy input and output variables requires a fuzzy relation @xmath13 between these variables . each fuzzy relation can be represented by a surface which we call it _ fuzzy relation surface _ or frs . value of this surface at any point is from a kind of membership degree and therefore will be always non - negative . in literature , several implication methods have been proposed to construct a fuzzy relation based on given input and output fuzzy sets @xcite . in this paper , we will show another method to construct fuzzy relation based on available training fuzzy data when input and output fuzzy sets are not known . as can be seen in the rest of this paper , our proposed method has the ability of learning as well . figure [ figa:4 ] shows one typical frs for the fuzzy system of fig . [ figa:3 ] . note that since input and output variables of the fuzzy system of fig . [ figa:3 ] are discrete , this surface has became discrete as well . one reason that we have made input and output of this system and consequently its representing fuzzy relation discrete is that the hardware implementation of continues fuzzy system is very difficult if not impossible . as stated before , it is evident that based on the requirements of the application that this fuzzy system is intended for , resolution and range of its corresponding discrete fuzzy relation ( surface ) can be increased simply by increasing the number of input and(or ) output terminals of the system . in its simplest form , any fuzzy relation can be defined as follows @xcite : let @xmath16 and @xmath17 be two universes of discourse . binary fuzzy relation , denoted by @xmath13 , are fuzzy sets which map each element in the product set @xmath18 to a membership grade @xmath19 , where @xmath20 and @xmath21 . so any fuzzy relation can be expressed as : @xmath22 now the problem is to determine the membership function of the fuzzy relation @xmath13 based on given input and output fuzzy sets . in this section , two different situation will be considered : ( i ) input and output fuzzy sets are available and ( ii ) input and output fuzzy sets are not in hand but instead of them , some fuzzy input - output training data are available . consider that input fuzzy set @xmath23 with membership function @xmath24 and output fuzzy set @xmath25 with membership function @xmath26 are available and the goal is to determine the membership function of the fuzzy relation described by : @xmath27 based on the knowledge of @xmath28 and @xmath29 where @xmath30 is a fuzzy implication scheme . here , we consider a new and simple implication method defined as : @xmath31 where @xmath32 can be any monolithically increasing function . note that since we are dealing with fuzzy systems , function @xmath32 does not need to be defined precisely . equation [ eqfr ] says that the membership value of the constructed fuzzy relation at any point @xmath33 should be directly proportional to the sum of the values of the input s membership function at point @xmath34 and the output s membership function at point @xmath35 . in other words , whatever both values of input and output membership functions at points @xmath34 and @xmath35 respectively be higher , the membership value of fuzzy relation @xmath13 at point @xmath33 should be higher as well . although this implication method is not similar to other common methods , it has this benefit that as can be seen later , its hardware implementation is straightforward . now , consider the case in which input and output fuzzy sets are not available but we want to construct a fuzzy relation based on @xmath36 existing fuzzy input - output training data . each fuzzy input - output training data for any siso system consists of two fuzzy numbers : input fuzzy number denoted by @xmath14 and output fuzzy number denoted by @xmath15 . the _ i_th fuzzy training data can be considered as @xmath37 where : @xmath38 in above equations , @xmath16 and @xmath17 are universes of discourse and @xmath39 and @xmath40 are membership functions of fuzzy numbers @xmath41 and @xmath42 respectively . similar to what is done in previous subsection , we propose to construct a fuzzy relation representing the relation between input and output fuzzy variables as : @xmath43 however , when all training data are not entirely available at the beginning or when they are entering one by one , fuzzy relation can be formed iteratively . in this case , entrance of any new data should update the current fuzzy relation which its updating rule can be written as : @xmath44 where @xmath45 is the newly observed fuzzy input - output training data . here it is worth to discuss a bit more about our proposed method to construct fuzzy relation based on fuzzy training data . actually , this implication method is inspired from the ink drop spreading concept in active learning method @xcite because of three reasons : ( i ) it has biological support , ( ii ) its hardware implementation is very simple and ( iii ) it does not obey from exact mathematical techniques . figure [ figfuzrelcreation ] shows this process graphically . in this figure , fuzzy relation is interpreted as an image where in this image the intensity of the pixel at location @xmath46 is directly proportional to @xmath19 and pixels with higher intensity are depicted darker . when a new training data is being observed , the value of the fuzzy relation at any point like @xmath46 should be increased ( should be made darker ) . however , increasing amount of fuzzy relation at point @xmath46 should be proportional to membership grades of observed input and output fuzzy numbers at points @xmath34 and @xmath35 respectively . since most of naturally created fuzzy numbers have bell like shapes ( gaussian - like membership functions ) , updating the fuzzy relation due to this new fuzzy input - output training data can be equated with the distillation of one ink drop and then spreading it on the image representing this fuzzy relation . this process is clearly demonstrated in fig . [ figfuzrelcreation ] as well . in a similar way , updating the fuzzy relation with the next fuzzy input - output training data will be equal to dropping another ink drop on the fuzzy relation image . as individual ink drop patterns overlap , the overlapping area become increasingly darker . the effective radius of distilling ink drop on the image is to somehow related to the amount of uncertainty in measuring input and output numbers . for example , when we can measure input and output data with infinite precision , they will become simple crisp numbers and therefore the radius of distilling ink drop should approach zero . gradually by the entrance of more and more data ( by progressing in the learning process and spreading more data ) , a pattern will be formed on this image . finally , this pattern will represent the fuzzy relation which connects input and output fuzzy variables . to conclude , any fuzzy inference needs a fuzzy relation which in its simplest form can be represented by a 2-dimensional matrix ( like the one shown in fig . [ figfuzrelcreation ] ) . consequently , any fuzzy inference hardware that one would like to propose should at least have a structure for storing and manipulating matrixes . in this section , we explain how inference is done in our proposed method . for this purpose , assume that fuzzy relation is created and based on this relation and given input fuzzy number , we want to determine corresponding output fuzzy number . in literature , several different approaches have been proposed to design fuzzy inference systems having linguistic descriptions of inputs and outputs @xcite . however , most of these inference systems have this disadvantage that their hardware implementation is not simple . for example , consider _ min _ and _ max _ operators which are frequently in use in fuzzy inference systems . it is well known that their hardware implementation is not efficient compared to some other t - norm or t - conorm operators . therefore , it is evident that the construction of large scale systems like human brain based on these inference methods is very hard if not impossible . actually , this is because of the fact that these operators are not inherently consistent with the physical behavior of circuit elements . to avoid this problem , we propose a new but unusual inference method as described below . suppose that we have a fuzzy relation @xmath13 and input fuzzy number @xmath47 . now , we propose to compute the membership function of output fuzzy number @xmath15 corresponds to the input fuzzy number @xmath14 as : @xmath48 where @xmath17 is the universe of discourse of output fuzzy number @xmath15 . if we refer back to our proposed structure shown in fig . [ figa:2 ] and consider the input of the structure as a vector of membership grades like @xmath49^t$ ] , in this case output of that structure can be written as : @xmath50^t=\mathbf{\mu_r}\times\mathbf{\mu}_{a'}\ ] ] where @xmath51 is a transposition operator and @xmath52 . therefore , our proposed inference method is nothing else than a simple vector to matrix multiplication . this process is depicted in fig . [ figinfab ] graphically in two cases : ( i ) input of the system is a crisp number and ( ii ) input of the system is a fuzzy number . note that to have better understanding , fuzzy relation in this fuzzy system is considered to be continuous . figure [ figinfaa ] shows that when the input of the fuzzy system is a crisp number , like @xmath0 ( @xmath53 is equal to 2 in fig . [ figinfaa ] ) , output of the system which is acquired by our proposed inference method will be the curve ( representing output fuzzy number ) of intersection of the fuzzy relation with the plane @xmath0 multiplied by @xmath54 . figure [ figinfbb ] shows the other case in which the input of the system is a fuzzy number . however , without loos of generality and to better illustrate the concept behind our inference method , input fuzzy number is considered to be discrete with only two non - zero samples at points @xmath55 and @xmath56 with membership grades of @xmath2 and @xmath57 respectively . when this input is applied to the structure , output fuzzy number @xmath15 will be the weighted sum of two fuzzy numbers obtained by intersecting of the fuzzy relation with planes @xmath0 and @xmath58 which can be written as : @xmath59 where @xmath60 this equation means that each sample of input fuzzy number creates a distinct output fuzzy number which is the curve of intersection of the fuzzy relation with the specific plane . then , these fuzzy numbers are aggregated ( summed ) with each other with different gains to create the output fuzzy number ; as the membership grade of one sample in fuzzy input number be higher , contribution of its corresponding fuzzy number in the creation of final output fuzzy number will be more . therefore , our proposed inference method ( eq . [ eqinfmul ] or eq . [ eqcomp ] ) which is only a simple vector to matrix multiplication is not so much irrelevant and illogical . note that since multiplying the left - hand side of eq . [ eqinfexam ] by a constant will not change the output fuzzy number ( its defuzzification before and after this multiplication will result in a same crisp number ) , the height of the output fuzzy number does not need to be equal to 1 . in the following sections , we will show how this proposed inference method can be simply implemented by using memristor crossbar structure . in addition , the reason that made us to consider this system as a neuro - fuzzy one will be explained as well . in this section , at first physical properties of the forth circuit element _ i.e. _ memristor will be discussed and then the working procedure of memristor crossbar and its application will be explained . finally , we will show that how the fuzzy inference system that we developed in previous section can be implemented with this memristor crossbar structure . after the first experimental realization of the forth fundamental circuit element _ i.e. _ memristor @xcite whose existence was previously predicted in 1971 by leon chua @xcite , extraordinary increased researches are in process in variety of fields like neuroscience , neural networks and artificial intelligence . it has become clear that this passive element can have many potential applications such as non - volatile memory construction @xcite , creation of analog neural network and emulation of human learning @xcite , building programmable analog circuits @xcite , constructing hardware for soft computing tools @xcite , implementing digital circuits @xcite and in the field of signal processing @xcite . memristor , different from other electrical elements namely resistor , capacitor and inductor , denotes the relationship between flux(@xmath61 ) and electric charge ( @xmath62 ) as @xcite : @xmath63 by rewriting this equation , memristance of the memristor can be expressed as : @xmath64 which shows that the unit of memristance is ohm . in fact , a memristor can be thought of as a resistive device that its resistance varies in dependence of its current or flux . memristor is an electrically switchable semiconductor thin film sandwiched between two metal contact with a total length of @xmath65 and consists of doped and un - doped regions which its physical structure with its equivalent circuit model is shown in fig . [ fig1 ] @xcite . the internal state variable @xmath66 determines the length of doped region with low resistance against un - doped region with high resistivity . this internal state variable and consequently the total resistivity of the device can be changed by applying external voltage bias @xmath67 @xcite . if the doped region extends to the full length @xmath65 , the total resistance of the device will be at its lowest level denoted as @xmath68 and if the un - doped region extends to the full length @xmath65 , the total resistance of the device will be at its highest level namely @xmath69 . for example , the mathematical model for the total resistance of the memristor reported by hp can be written as @xcite : @xmath70 where @xmath71 is the initial state for state variable @xmath66 , @xmath72 is the average ion mobility and @xmath73 is the amount of electric charge ( integral of current ) that has passed through the device . above equations show that passing current from memristor in one direction will increase the memristance of the memristor while changing the direction of the applied current will decrease its memristance . in addition , it is obvious that in this element , passing current in one direction for longer period of time ( which means @xmath73 has higher absolute value ) will change the memristance of the memristor more . moreover , by setting the passing current into zero , the memristance of the memristor will not change anymore . finally , note that determining the memristance of the memristor at any time can be done by passing a small current through the memristor and measuring the dropped voltage across it ( see eq . [ eq2 ] ) . as a result , memristor is nothing else than the analog variable resistor where its resistance can be adjusted by changing the direction and the duration of the applied voltage . therefore , memristor can be used as a storage device in which analog values can be stored as a memristance instead of voltage or charge . here it should be emphasized that memristor - based storage devices have some advantages compared to those storage devices which use capacitors . at first , the former can be fabricated much denser than the later one through the nano - crossbar technology @xcite . second , memristor can hold the stored data unchanged theoretically for an infinite period of time without refreshing @xcite . finally , unlike capacitors , memristors can be used in the memristor crossbar structure which has so many potential applications ( one of them is demonstrated in this paper ) . a crossbar array basically consists of two sets of conductive parallel wires intersecting each other perpendicularly . the region where a wire in one set crosses over a wire in the other set is called a crosspoint ( or junction ) . crosspoints are usually separated by a thin film material which its properties such as its resistance can be changed by controlling the voltage applied to it . one of such materials is memristor which is used in our proposed crossbar - based circuits in this paper . figure [ fig2 ] shows a typical memristor crossbar . in this circuit , memristors which are formed at crosspoints are depicted explicitly to have better visibility . in this crossbar , memristance of any memristor can simply be changed by applying suitable voltages to those wires that memristor is fabricated between them . for example , consider the memristor located at coordinate ( 1 , 1 ) ( crossing point of the first horizontal and the first vertical wires ) of the crossbar . memristance of this memristor can be decreased by applying a positive voltage to the first vertical wire while grounding the first horizontal one ( or connecting to negative voltage ) . dropping a positive voltage across the memristor will cause the current to pass through it and consequently , the memristance of this passive element will be decreased . in a similar way , memristance of this memristor can be increased by reversing the polarity of applied voltage . as stated before , application of higher voltages for longer period of time will change the memristance of the memristor more . this means that the memristance of any memristor in the crossbar can be adjusted to any predetermined value by the application of suitable voltages to the specific row and column of the crossbar . however , here it should be emphasized that in this paper , there will be no need to adjust the memristance of the memristors accurately and only increasing or decreasing the memristance of the memristor will be sufficient . to summarize , memristor crossbar is a 2-dimensional grid that analog values can be stored in its crosspoints through the memristance of the memristors . consequently , it seems that the memristor crossbar is a perfect structure to construct and store 2-dimensional patterns like fuzzy relations . in this case , vertical and horizontal wires of the crossbar will represent different discrete values of input and output variables respectively and memristors in crosspoints will have the role of entries of 2-dimensional matrixes representing fuzzy relations . against these mentioned advantages , memristor crossbar structure has some drawbacks . first of all , reading the memristance of the memristors at any point of the crossbar is partly difficult . but the second and the most important disadvantage of memristor crossbar structure relates to the inherent physical limitation of memristor . as stated before , memristance of the memristor can be between @xmath74 and @xmath75 . therefore , it seems that storing analog values through the memristance of the memristors needs a linear mapping from input space to interval @xmath76 $ ] . in the next section , we show how these problems can be solved by slightly modifying the structure of fig . [ fig2 ] and how this structure can be used to construct and store fuzzy relations . figure [ fig3 ] shows the memristor crossbar - based analog circuit that we have proposed to construct fuzzy inference system introduced in previous section . as can be seen later , this structure has biological support and can be considered as a single - layer fuzzy neural network which conforms with the hebbian learning rule but with very new interesting properties compared to today s conventional neural networks . figure [ fig3 ] consists of a simple crossbar that each of its rows is connected to the negative input terminal of a simple opamp . through these opamp - based circuits added to rows of the crossbar we will be able to read the pattern stored in the crosspoints of the crossbar . since these opamps have a fixed resistor with resistance @xmath69 ( maximum memristance value of the memristors used in the crossbar ) as a feedback , combination of each of these opamps and those memristors which are connected directly to the same horizontal wire that opamp is connected to it creates a simple opamp - based summing circuit . note that in the uppermost row of the crossbar , instead of memristor , a resistor with resistance @xmath77 is fabricated in each crosspoint . since this row of the crossbar is connected to an opamp with a feedback resistor @xmath77 , it again forms a simple inverting summing circuit that adds inputs of the crossbar with the same gain _ i.e. _ -1 . therefore , at any time output of this summing circuit , _ i.e. _ @xmath78 , can be written as : @xmath79 where @xmath80 is the input signal ( represented by voltage ) connected to the _ j_th vertical wire of the crossbar and @xmath81 is the number of columns of the crossbar . note that the hat sign in eq . [ eq4 ] is used to distinguish inputs and outputs of the structure from those concepts which are assigned to rows and columns of the circuit . the reason of using this summing circuit will become clear in the rest of this section . now , by applying standard opamp circuit analysis techniques , output voltage of the opamp connected to the _ i_th row of the crossbar can be expressed as : @xmath82 where @xmath83 is the current memristance of the memristor at coordinate @xmath84 of memristor crossbar and @xmath85 is the number of rows of the crossbar . note that to write this equation , memristor is treated like a simple resistor . this is true only when the memristance of the memristor be almost constant during the calculation of output voltages . this can be guaranteed by applying input voltages @xmath86 ( for @xmath87 ) to columns of the crossbar for very short period of time during the execution phase . another condition that this structure should satisfy is that the memristance of all memristors in the crosspoints of the crossbar should initially be equal to @xmath69 . in fact , in all of our proposed structures in this paper , storing any value at coordinate @xmath84 of the crossbar is done by decreasing the initial memristance of the memristor located at this coordinate by the given value . in other words , current stored value at coordinate @xmath84 denoted by @xmath88 can be written as : @xmath89 by this trick , there will be no need to use linear mapping as mentioned in previous section . by substituting [ eq6 ] and [ eq4 ] into [ eq5 ] we will get : @xmath90 now , by assuming that the stored value in every crosspoint is always much smaller than @xmath69 , _ i.e. _ @xmath91 , which can be satisfied by scaling data before storing , eq . [ eq6 - 1 ] can be simplified to : @xmath92 or equivalently can be expressed in matrix form as : @xmath93 where @xmath94 , @xmath95 $ ] , @xmath96^t$ ] , and @xmath97 . it is evident that although we will use this structure for performing fuzzy operations , any matrix multiplication in the form of eq . [ eq6 - 3 ] can be done by this structure . by comparing eqs . [ eq6 - 3 ] and [ eqcomp ] ( or equivalently by comparing structures of fig . [ fig3 ] and fig . [ figa:2 ] ) it becomes clear that if somehow we can store fuzzy relation in the crosspoints of the crossbar ( as matrix @xmath98 ) our proposed structure will be a perfect analog circuit to perform our proposed inference method ( note that in eq . [ eqcomp ] , @xmath99 was a column vector while in eq . [ eq6 - 3 ] @xmath100 is a row vector ) . for this purpose , it is sufficient to consider that the _ i_th column and the _ j_th row of the crossbar represent concepts @xmath0 and @xmath1 respectively similar to inputs and outputs of the system shown in fig . [ figa:2 ] and then connect the membership grades of newly observed input fuzzy number to the inputs of this structure . in this case , the membership grades of the corresponding output fuzzy number will emerge at outputs of the structure instantaneously . as another example , consider the case in which the row vector @xmath101 with only one non - zero element at position @xmath102 is being applied as an input to the structure where fuzzy relation @xmath98 is programmed in its crosspoints . in this case , the output of the structure will be @xmath103 where @xmath104 is the _ k_th column of matrix @xmath98 and @xmath105 is the value of the _ k_th entry of vector @xmath101 . note that this example is equivalent to the inferential process depicted in fig . [ figinfaa ] . as stated in that section , multiplication of output fuzzy number ( @xmath104 which is the intersection of stored 2-dimensional fuzzy relation ( @xmath98 ) with the plane @xmath106 ) with a constant , _ i.e. _ @xmath107 , will not degrade the generated output fuzzy number significantly or it can be easily compensated by using some auxiliary circuits . as stated before , domain and resolution of input and output variables can be increased by adding more horizontal and vertical wires to the structure . simplicity of the circuit and computing in realtime are some of the advantages of our proposed hardware . to summarize , proposed hardware of fig . [ fig3 ] can do vector to matrix multiplication and therefore can do any of the inferential processes previously described in sec . [ secdoinf ] . the only question that we should answer is how to program the memristors in the crossbar or how to create and store fuzzy relation in this memristor crossbar - based structure . this will be addressed in the next subsection . in this subsection , we will consider two cases to construct a fuzzy relation on the proposed structure of fig . [ fig3 ] : ( i ) input and output fuzzy sets are available and ( ii ) instead of input and output fuzzy sets some input - output training data are in hand . in the first case , assume that we have input fuzzy set @xmath23 and output fuzzy set @xmath25 with membership functions @xmath28 and @xmath29 respectively and we want to construct a fuzzy relation between these two fuzzy sets based on fuzzy implication of eq . [ eqfr ] . to create a fuzzy relation from @xmath28 and @xmath29 in the memristor crossbar structure of fig . [ fig3 ] , it is sufficient to interpret membership degrees in each fuzzy set as a voltage and apply them to their corresponding rows and columns of the memristor crossbar for @xmath108 second(s ) . to be exact , each element of @xmath28 should be connected to its corresponding column in the crossbar and the negative of each element of @xmath29 should be connected to its corresponding row in the crossbar ( remember that as depicted in fig . [ figa:2 ] , each input and output terminal of the structure corresponds to exclusive value or concept of input and output fuzzy variables respectively ) . for example , if sets @xmath23 and @xmath25 be defined as : @xmath109 then to create a fuzzy relation which is connecting these input and output fuzzy sets on this structure , these following tasks should be done : * interpret membership grades as a voltage signal and connect @xmath110 to the vertical wire of the crossbar representing concept @xmath111 , @xmath112 to the vertical wire of the crossbar representing concept @xmath113 and so on . * at the same time , connect @xmath114 to the horizontal wire of the crossbar representing concept @xmath115 , @xmath116 to the horizontal wire of the crossbar representing concept @xmath117 and so on . * only wait for @xmath108 second(s ) and then remove applied voltages from the crossbar . this process is depicted in fig . [ figfuzrela ] for two typical fuzzy sets . simultaneous application of positive and negative voltages to rows and columns of the crossbar respectively will cause the current to pass through the memristors in crosspoints . amount of current passing through the memristor located at the intersection point of _ j_th column and _ i_th row of the crossbar corresponding to concepts @xmath58 and @xmath118 respectively ( or equivalently through the memristor at coordinate @xmath84 of the crossbar ) is directly proportional to the dropped voltage over this passive element or equivalently to @xmath119 . as this term increases , memristance of the memristor at coordinate @xmath120 will decrease more during this @xmath108 second(s ) . assuming the hp model for the memristors of the crossbar , memristance of this memristor after this @xmath108 second(s ) can be written as @xcite : @xmath121 where @xmath122 is a constant defined as : @xmath123 . for this purpose it is only sufficient to connect samples of input and output fuzzy sets to their corresponding vertical or horizontal wires for @xmath108 second(s).,title="fig:",width=556,height=297 ] note that to obtain eq . [ eq8 - 11 ] , initial memristance of the memristors , @xmath124 , as expected is assumed to be @xmath69 . using eq . [ eq8 - 11 ] and eq . [ eq6 ] , amount of change of @xmath83 during this @xmath108 second(s ) which is equal to the stored value at coordinate @xmath84 of the crossbar will become : @xmath125 comparing eqs . [ eq8 - 12 ] and [ eqfr ] shows that the implication function @xmath126 considered in eq . [ eqfr ] for this specific memristor crossbar structure is : @xmath127 figure [ figfunfimp ] shows a plot of function @xmath126 for different values of @xmath108 which shows that as expected it is a monolithically increasing function . for plotting this figure , values of other parameters are set as following : @xmath128 , @xmath129 m , @xmath130 and @xmath131 . although here we tried to obtain mathematical expression for the implication method of our proposed hardware , it is clear that while working with this structure , there is no need to be involved with these exact mathematics . the only thing that we should do is to connect voltages to rows and columns of the crossbar and wait for @xmath108 second(s ) and then fuzzy relation will automatically be created and stored on the crossbar . finally note that to satisfy the mentioned condition for properly working of the memristor crossbar structure , _ i.e. _ @xmath132 , it is necessary to choose a small value for @xmath108 ( for example , in the above test , @xmath133 will be a good choice ) . defined in eq . [ eq8 - 122 ] for different values of @xmath108 which shows that this is a monolithically increasing function.,title="fig:",width=384,height=211 ] it is well known that one of the major applications of memristor is in the construction of non - volatile memories . this is because of the fact that memristance of the memristor will remain fixed theoretically for an infinite period of time when there is no applied voltage or current . this can be seen in eq . [ eq3 ] by setting @xmath73 to zero . this means that the created fuzzy relation on the crossbar will remain unchanged without refreshing ( which is required for most of currently available capacitor - based memories ) . this property of memristor says that final memristance of the memristor after the application of specific voltage ( or current ) will act as a initial memristance of the memristor during the application of next voltage ( or current ) . this means that effects of these sequentially applied voltages on the memristance of the memristor will be added to each other . by this explanation , it would be logical to think that one simple way for the construction of fuzzy relation on the proposed hardware based on available input - output fuzzy training data can be the repetition of the process described in previous subsection for each of these training data . to illustrate this process a bit more , consider that @xmath36 input - output fuzzy training data like @xmath134 for @xmath135 are available and we want to construct a pattern on our memristor crossbar circuit corresponding to the fuzzy relation between input and output fuzzy variables based on these training data . for this purpose , it is sufficient to behave as follows : for @xmath135 repeat these steps : * connect membership grades of fuzzy number @xmath41 and the negative of membership grades of fuzzy number @xmath42 to their corresponding columns and rows of the crossbar respectively . * wait for @xmath108 . presentation of each of these training data will decrease the memristance of the memristors in some areas of the crossbar . therefore , the effect of observing any new training data will be added to the currently stored pattern in the crossbar . in other words , entrance of any new training data simply updates the stored pattern . hence , by presenting more training data to the structure , gradually a pattern ( fuzzy relation ) will begin to form on the crossbar . here , three important aspects of this described procedure should be emphasized . first , @xmath108 should be decreased by the increase of @xmath36 to satisfy the condition @xmath136 . second , since a crisp number is a special case of fuzzy number , it is evident that the described procedure will be also applicable to crisp training data . creating fuzzy relation based on crisp data is possible either by using large number of training data to form a pattern or by converting crisp training data to their corresponding fuzzy numbers before applying them to the structure ( for example by using gaussian membership function ) . the third and the most important note pertains to how to create voltages corresponding to membership grades of input and output fuzzy numbers and how to apply(remove ) them to(from ) inputs and outputs of the structure . as will be demonstrated in the section of simulation results as well , there is no need to have any other auxiliary circuit to perform these tasks . this is because of the fact that our proposed circuit has this capability that it can be directly connected to other similar circuits . in this case , output of one circuit will be the input of the next structure and by this way , signals will propagate easily in the entire system . at the same time , based on the current values of input and output of each of these proposed memristor crossbar - based circuits in the whole system , stored pattern in their crossbar will automatically be updated . the neuromorphic paradigm is attractive for nanoscale computation because of its massive parallelism , potential scalability , and inherent defect- and fault - tolerance @xcite . in biological systems , the synaptic weights between neurons can be precisely adjusted by the ionic flow through them and it is widely believed that the adaptation of synaptic weights enables biological systems to learn and function . however , before the first physical realization of memristor , experimental construction of these neuromorphic systems which consist of neurons and synapses , especially in the electronic domain , has remained somewhat difficult . the primary problem was the lack of a small and efficient circuit that can emulate essential properties of synapses namely : having low power consumption , ability to be fabricated in high density ( human brain has about @xmath137 synapses ) and plasticity . after the first experimental realization of memristor @xcite , it became widely accepted that memristor is a good candidate to emulate synapse . this is because of the fact that memristors can be fabricated in high density through the crossbar technology . in addition , similar to biological synapse , the conductance of the memristor can be changed by passing current from it or applying voltage to it @xcite . based of these evidences , several authors have tried to use memristor as a synapse and showed that this passive element can facilitate hardware implementation of artificial neural networks and their corresponding learning rules such as spike time dependent plasticity ( stdp ) @xcite . figure [ figconvneur : a ] shows the memristor crossbar - based structure that has been proposed in almost all of these works . in this figure , each fabricated memristor at each crosspoint represents one synapse which connects one presynaptic to one postsynaptic neuron . in this case , every neuron ( shown by a triangle ) in the input layer is directly connected to every neuron in output layer with unique synaptic weights . if in fig . [ figconvneur : a ] , we denote output voltages of presynaptic neurons by a row vector @xmath138 , output voltages at postsynaptic neurons by a column vector @xmath139 and weights of synapses by matrix @xmath140 , the structure of fig . [ figconvneur : a ] implements the following mathematical operation : @xmath141 where to obtain this equation , we have assumed identity activation function for neurons . note that eq . [ eqneuro1 ] is a very familiar equation we have in conventional neural networks like the typical one depicted in fig . [ figconvneur : b ] @xcite . therefore , fig . [ figconvneur : a ] does not present anything new but only shows how memristor crossbar can be used as an electrical representation of synapses weights . learning can be accomplished in this structure based on learning methods mostly inspired from hebbian learning rule ( or equivalently stdp in spiky neural networks ) which in its simplest form says : _ neurons that fire together , wire together _ @xcite . this means that when one input and one output neuron fire simultaneously , the weight of that synapse which is connecting these neurons should be increased ( strengthened ) . for this purpose , in fig . [ figconvneur : a ] memristance of the memristor connecting these neurons should be modified ( actually it should be decreased ) . however , unfortunately it is well known that hebbian learning rule ( although is very simple ) is not enough to train neural networks and other complex methods like backpropagation learning algorithm are needed @xcite . moreover , hardware implementation of currently in use learning methods like stdp on this structure is difficult and needs some auxiliary circuits @xcite . in addition , in these kinds of learning rules synaptic weights may be either increased or decreased . these mentioned drawbacks have caused the efficient hardware implementation of neural networks very difficult . now , compare the structure of fig . [ figconvneur : a ] and eq . [ eqneuro1 ] with fig . [ fig3 ] and eq . [ eq6 - 3 ] respectively . in addition to their structural similarities , both of these structures perform the same task ( vector to matrix multiplication ) during the execution phase . however , in learning phase , we can see some differences . in the structure representing hardware implementation of artificial neural networks such as the one depicted in fig . [ figconvneur : a ] , learning is usually carried out based on methods like back propagation or stdp . on the other hand , learning in our proposed method is based on the creation of fuzzy relation between input and output variables or concepts . however , here we will show that _ learning based on the creation of fuzzy relation is the same as fuzzy learning in artificial neural networks but_. to illustrate this theorem better , consider the creation of fuzzy relation on the circuit of fig . [ figa ] based on available fuzzy training data described in section [ fuzreltrdat ] . in this process , simultaneous application of input and output fuzzy numbers to inputs and outputs of the structure is equivalent with the simultaneous firing of input and output neurons and the value of applied membership grades to each of input or output terminals actually determines the firing strength of each neuron . by this statement we mean that each simultaneous firing of input and output neurons generates one fuzzy input - output training data . therefore , in our belief , all of neurons are active at the same time during learning phase but with different confidence degrees where output of each neuron determines the confidence degree of its activation . in fact , when we connect @xmath54 to the column of the crossbar representing concept @xmath0 it means that the neuron in input layer which represents concept @xmath0 is firing with the confidence degree of @xmath54 . by interpreting applied membership grades as a firing strength of neurons , the concept of ink drop spread introduced in section [ inkdrop ] to create fuzzy relation on the crossbar ( see fig . [ figfuzrelcreation ] ) becomes equal to primary hebbian learning rule in neural networks . for example , fig . [ figfuzrelcreation ] from another point of view says that weights of those connections which are connecting simultaneously firing input and output neurons should be strengthened . however , the amount of increase of the weight of each connection should be proportional to the firing strength of those input and output neurons . when confidence degrees of firing of both input and output neurons are high , the weight of the connection connecting these input and output neurons should be increased more ( see eq . [ eqfr ] or eq . [ updateink ] ) . note that if we have crisp input - output training data instead of fuzzy one , ink drop spread method in fig . [ figfuzrelcreation ] will become completely the same as hebbian learning rule in conventional neural networks . therefore , it seems that our proposed fuzzy inference system implements the same function as neural networks and also learns in a similar way . in this case , a simple question may arise : _ what is the benefit of our proposed structure and inference method compared to conventional neural networks?_. to answer this question , the reader should note that in reality , by these explanations we tried to show that neural networks and fuzzy are not two disjoint fields . in fact , we strongly believe that the reason which causes hebbian learning not to be able to learn neural networks properly is the misinterpretation of the nature of input and output signals in neural networks . let s look at neural networks from another point of view for a short time . suppose that each neuron represents one and only one concept like @xmath0 where @xmath34 can be a numerical or linguistic variable . note that in this situation , spatial location of each neuron in the entire system determines its corresponding concept . in addition , assume that output ( firing strength ) of each neuron in the network specifies the confidence degree of the activation of that neuron or its corresponding concept . in this case , combination of those neurons ( or concepts ) lied in one layer and their corresponding confidence degrees ( outputs ) simply creates a fuzzy number . now , application of primary hebbian learning method to this described neural network with this kind of input and output signals will result in a creation of fuzzy relation on the matrix representing connection weights . note that changing the functionality of neural networks to work with confidence or membership degrees is not in contrast with biological findings but in return it offers some interesting properties . first of all , as we will illustrate in the section of simulation result , in spite of conventional neural networks a simple primary hebbian learning method ( or equivalently creating fuzzy relation ) is completely enough to learn these kinds of networks . consequently , as we showed in this paper their hardware implementation becomes more simple than other learning methods in artificial neural networks such as stdp or backpropagation . second , since in hebbian learning the connection weights are only strengthened , they will always be non - negative . note that we have the same case in our proposed fuzzy inference system since the created fuzzy relation on the crossbar is always non - negative . it is clear that working with non - negative weights is much simpler than working with weights which can have both positive and negative values . third , if after the training of the conventional neural networks , we plot the matrix which is holding connection weights as a surface it will have no meaningful shape and no information can be obtained from it . however , corresponding surface in a neural network which works with confidence or membership degrees is a fuzzy relation where its shape describes the overall behavior of input and output variables versus each other . remember that human brain remembers most concepts and relations through the images ( surfaces ) . fourth , in the field of artificial neural networks and based on biological findings , it is common to put a threshold on the outputs of neurons . however , putting threshold on confidence or membership degrees ( output of our proposed structure ) seems to be more logical than putting threshold on meaningless output values in conventional neural networks . finally , it is interesting to note that that in sec . [ fuzreltrdat ] we told that by increasing the number of training data , @xmath108 should approach zero . in this case , signals that propagate in the network will become similar to spike and therefore we would have spiking neural networks . to summarize , in this section we tried to show that it is possible to look at the working procedure of neural networks from another aspect . if we accept that biological neurons transmit information in the form of confidence degrees , then the computational task which is done in conventional neural network will become equal to our proposed fuzzy inference method with extra advantages . in this section , we want to verify the efficiency of our proposed hardware and fuzzy inference method by conducting several simulations . in the first simulation , we show how mathematical functions can be constructed in an unprecise manner and how function composition can be done in our proposed structure . for this purpose , consider the creation of two fuzzy relations representing two different functions @xmath142 and @xmath143 on two distinct samples of the circuit shown in fig . [ fig3 ] . to construct each of these functions on the circuit of fig . [ fig3 ] near 500 input - output fuzzy training data are used . these training data are created artificially by firstly generating 500 crisp input - output training data uniformly distributed on input domain and then converting them to their corresponding fuzzy numbers by using gaussian membership function . although in this paper we have used gaussian membership function for the fuzzification of crisp training data , repeating this test showed that any other membership function can also be used without degrading output results significantly . figures [ figfunfafb : a ] and [ figfunfafb : b ] show the constructed fuzzy relations on the memristor crossbar of the circuit of fig . [ fig3 ] based on these training data by following the procedure described in section [ fuzreltrdat ] . note that these figures indicate final stored values ( or created fuzzy relations ) on the crossbars , _ i.e. _ matrix @xmath144 , after the accomplishment of training process where they are plotted as a continues surface to have better visibility . in this test , it is assumed that input variable of both functions @xmath145 and @xmath146 , _ i.e. _ @xmath34 , is bounded between 0 and 1 and memristor crossbar of the circuit of fig . [ fig3 ] has 100 rows and 100 columns ( 100 vertical and 100 horizontal wires ) . other parameters are set as follows : @xmath128 , @xmath129 m , @xmath130 , @xmath133 and @xmath131 . these crossbars are simulated in hspice software by utilizing the spice model proposed in @xcite for memristors . figure [ figfunfafb : a ] and [ figfunfafb : b ] demonstrate that application of hebbian learning to our proposed structure which we showed that it is equal to the creation of fuzzy relation creates a meaningful surface on the memristor crossbar since it is easy to recognize shapes of functions @xmath147 and @xmath148 in these figures . therefore , it is clear that in our structure , relations between input and output variables are stored in the system based on their shapes and not through their exact mathematical formulas . now , let s see how these two programmed circuits where each of them is a representative of one function can be combined with each other to create new functions . without loss of generality , consider the creation of function @xmath149 . to build this function , it is only sufficient to directly connect outputs of the circuit on which function @xmath146 is constructed to their corresponding inputs in the circuit which represents function @xmath145 . this process is depicted schematically in fig . [ figcomb : a ] . note that since the resolution of output variable in function @xmath146 may not be the same as the resolution of input variable in function @xmath145 , in fig . [ figcomb : a ] output terminals of the first circuit is not connected sequentially to input terminals of the second circuit . connecting two circuits representing functions @xmath145 and @xmath146 with fuzzy relations of figs . [ figfunfafb : a ] and [ figfunfafb : b ] respectively in this way will create a new function @xmath149 . since @xmath150 is a identity function its input and output should always be the same . figure [ figcomb : b ] shows the result of testing this circuit which is obtained by applying randomly generated crisp number to the input of the system and then plotting the result of the defuzzification of the output fuzzy number versus original input . this figure shows that the newly constructed function by merging two functions @xmath147 and @xmath148 is a good approximation of function @xmath149 . from the results of this simulation , it can be concluded that by sequentially connecting several circuits of fig . [ fig3 ] where each of them represents different but simple function , any complex function can be reproduced . here it is worth to mention that since in combinational circuits like the one shown in fig . [ figcomb : a ] output of one circuit directly connects to inputs of other circuit , input signals of one circuit will come from previous circuits . actually , these are fuzzy numbers that propagate in the structure from one block to others and therefore no defuzzification process will be required in the entire system . moreover , in this case there will be no need to any other auxiliary circuits to generate these signals since they are coming from other blocks . functions with more than one input variable play an important roles in the construction of complex systems and inference methods . note that functions like @xmath151 , @xmath152 , @xmath153 and @xmath154 belong to this category . until now , we have deal only with 1-input functions in the form of @xmath155 and our proposed structure in fig . [ fig3 ] seems to be able to implement only these kinds of functions ( because each binary fuzzy relation has one input and one output variable ) . in this part , we want to show how efficiently our proposed method and circuit can model multi - input functions . to use the structure of fig . [ fig3 ] for modeling multi - input functions , there is no need to apply any changes to it . this is because of the fact that actually this structure in its current form is inherently implementing multi - input functions . to demonstrate this concept , note that if we assume that the concepts which are assigned to different input terminals of the circuit of fig . [ fig3 ] ( its vertical wires ) be completely independent from each other , then each of these inputs can be considered as a distinct input variable . for example , assume that we want to implement a 2-input function @xmath156 on the structure of fig . [ fig3 ] . for this purpose , we can split the vertical wires of this circuit into two sections ; wires in one section represent different values of one input variable and wires in other section represent different values of other input variable . this process is depicted in fig . [ figmultinpfun ] for better illustration . note that by adding more vertical wire to the structure of fig . [ fig3 ] , resolution and domain of each of these input variables can be simply increased . it is obvious that generalization of this method for the implementation of multi - input functions with any number of input variables is straightforward . since we have not changed the circuit of fig . [ fig3 ] , learning of this circuit ( creation of fuzzy relation ) can be done the same as before . for this purpose , it is sufficient to apply fuzzy numbers of independent input variables to their corresponding columns of the crossbar and the fuzzy number of output variable to rows of the crossbar for @xmath108 second(s ) . repeating this process for other fuzzy input - output training data will cause the fuzzy relation to be formed on the memristor crossbar of the circuit . however , in spite of previous tests , when implementing multi - input functions in this way instead of single fuzzy relation , several fuzzy relations will be formed on the crossbar . actually , in this case one fuzzy relation per each input variable will be created on the crossbar as shown in fig . [ figmultinpfun ] as well . each of these fuzzy relations specifies the overall behavior of output variable versus one of input variables . in this situation , structure s output fuzzy number for one given input sample will be computed from several intermediate output fuzzy numbers where each of them is obtained through one of these fuzzy relations based on the same method described in section [ secdoinf ] . to be precise , in fact as shown in fig . [ figinfbb ] as well , these fuzzy numbers are summed to each other to generate final output fuzzy number corresponding to given input . to illustrate this procedure better , we do several simulations . in the first simulation , ability of our proposed structure in modeling 2-input functions is investigated . the following nonlinear function is considered for this purpose : @xmath157 can be used to implement multi - input functions . for this purpose , one can split input terminals into several sections one for each of input variables . training of this structure will be the same az before but it should be noted that in spite of the structure of fig . [ fig3 ] , in this structure several fuzzy relation will be formed on the crossbar.,title="fig:",width=652,height=336 ] where its graph is shown in fig . [ fig161:a ] . this modeling test is conducted on the structure similar to the circuit depicted in fig . [ figmultinpfun ] with 180 vertical wires ( 90 wires for each of input variables ) and 100 horizontal wires ( for output variable ) . since input variables are bounded between 1 and 10 and there are 90 wires to cover this interval , minimum achievable resolution for input variables by assuming that these wires are spread uniformly over this interval will be : @xmath158 in a similar way , minimum achievable resolution for output variable @xmath159 can be computed which becomes equal to @xmath160 . to create fuzzy relations on the structure of fig . [ figmultinpfun ] , 800 fuzzy input - output training data are used . since no fuzzy input - output data are available , we have to create them artificially . for this purpose , at first 800 crisp input - output training data which are uniformly distributed over input space are generated and then they are converted to their corresponding fuzzy input - output training data by using gaussian membership function . application of these training data to rows and columns of the crossbar of fig . [ figmultinpfun ] will cause fuzzy relations to be created on the crossbar . figures [ fuzxyzrel : a ] and [ fuzxyzrel : b ] show these formed fuzzy relations on the crossbar separately after this training process . now , similar to what is done in previous simulation , _ i.e. _ by application of randomly generated crisp data and defuzzification of produced output fuzzy number , the model of function defined in eq . [ eq161 ] can be reconstructed . the result of this modeling test is presented in fig . [ fig161:b ] . the mean square error ( mse ) between the target function and constructed model is 0.021 . this simulation clearly demonstrates the efficiency of our proposed inference method in modeling multi - input functions . nowadays , nano - scale devices and molecular electronics promise to overcome the fundamental physical limitation of lithography - based silicon vlsi technology @xcite . furthermore , it has been demonstrated that nano devices such as nanoscale crossbars can be fabricated efficiently by using bottom - up self - assembly techniques without relying on lithography to define the smallest feature size @xcite . however , non - determinism in bottom - up self - assembly chemical processes at molecular scale results in more defects compared to highly controlled lithography - based manufacturing processes currently used in cmos technologies @xcite . therefore , defect tolerance is necessary for circuits which are realized through the usage of nano - scale devices @xcite . in the next simulation , we will illustrate the excellent fault tolerance capability of our proposed hardware . figure [ fig161:c ] shows the result of modeling test obtained by repeating the above simulation but this time with the hardware which has some defects . in this test , we have assumed that approximately half of the memristors in the memristor crossbar of the circuit of fig . [ figmultinpfun ] is not working . this means that these randomly chosen memristors have the memristance @xmath69 and their memristance can not be changed during the training process . in other words , we have assumed that these faulty memristors behave completely the same as a simple resistor with the resistance of @xmath69 and nothing can be stored at these points . the training set and the value of other parameters in this simulation are the same as the previous test . the mean square error ( mse ) between the target function and constructed model is 0.0281 . figure [ fig161:c ] indicates that although 50 percent of all of the memristors in the crossbar are working incorrectly , our proposed hardware still can model functions with acceptable accuracy . in this paper we proposed a new fuzzy inference system and a simple method to create fuzzy relation based on input and output fuzzy sets . in addition our proposed scheme has this benefit that fuzzy relation can be formed in it based on fuzzy training data and therefore offering learning capability . since one of the main problems of fuzzy systems relates to their efficient hardware implementation , we have also designed simple memristor crossbar - based circuit as a hardware implementation of our proposed inference method . simulation results show that this structure can effectively be used to construct multi - input functions . moreover , we illustrated that samples of this memristor crossbar - based analog circuit can be sequentially connected to each other to create a cellular structure on which complex functions can be constructed . finally , according to the similarities between our proposed fuzzy inference method and recently suggested memristor crossbar - based circuits for constructing artificial neural networks , we showed that if input and output signals of neural networks be of a kind of confidence degree , then the computational task which is done in conventional neural network will become equal to our proposed fuzzy inference method with extra advantages . shouraki , `` a novel fuzzy approach to modeling and control and its hardware implementation based on brain functionality and specifications , '' ph.d . dissertation , the univ . of electro - communications , chofu , japan , march 2000 . m. murakami and n. honda , `` a study on the modeling ability of the ids method : a soft computing technique using pattern - based information processing , '' international journal of approximate reasoning , vol . 45 , pp . 470487 , 2007 . y. v. pershin , and m.d . ventra , `` practical approach to programmable analog circuits with memristors , '' ieee transactions on circuits and systems i : regular paper , vol . 57 , no . 8 , pp . 18571864 , aug . 2010 . f. merrikh - bayat , and s. b. shouraki , `` memristor crossbar - based hardware implementation of sign - sign lms adaptive filter , '' analog integrated circuits and signal processing , doi : 10.1007/s10470 - 010 - 9523 - 3 . n. gergel - hackett , b. hamadani , b. dunlap , j. suehle , c. richter , c. hacker , and d. gundlach , `` a flexible solution - processed memristor , '' ieee electron device letters , vol . 7 , pp . 706708 , july 2009 . k. cantley , a. subramaniam , h. stiegler , r. chapman and e. vogel , `` hebbian learning in spiking neural networks with nano - crystalline silicon tfts and memristive synapses , '' accepted in ieee transactions on nanotechnology , 2011 . a. afifi , a. ayatollahi and f. raissi , `` implementation of biologically plausible spiking neural network models on the memristor crossbar - based cmos / nano circuits , '' european conference on circuit theory and design , pp . 563566 , 2009 . j. a. carrasco , c. zamarreno - ramos , t. serrano - gotarredona , and b. linares - barranco , `` on neuromorphic spiking architectures for asynchronous stdp memristive systems , '' ieee international symposium on circuits and systems , pp . 16591662 , 2010 . d. biolek , z. biolek and v. biolkova , `` spice modeling of memristive , memcapacitative and meminductive systems , '' european conference on circuit theory and design ( ecctd2009 ) , pp . 249252 , antalya , 2327 augost 2009 . j. huang , m. b. tahoori , and f. lombardi , `` on the defect tolerance of nano - scale two - dimensional crossbars , '' 19th ieee international symposium on defect and fault tolerance in vlsi systems , pp . 96104 , oct . 2004 . c. p. collier , e. w. wong , m. belohradsky , f. m. raymo , j. f. stoddart , p. j. kuekes , r. s. williams , and j. r. heath , `` electronically configurable molecular - based logic gates , '' ieee trans . on nanotechnology , science , vol . 391394 , 1999 .

multifractal properties of the distribution of topological invariants for a model of trajectories randomly entangled with a nonsymmetric lattice of obstacles are investigated . using the equivalence of the model to random walks on a locally nonsymmetric tree , statistical properties of topological invariants , such as drift and return probabilities , have been studied by means of a renormalization group ( rg ) technique . the comparison of the analytical rg results with numerical simulations as well as with the rigorous results of p.gerl and w.woess demonstrates clearly the validity of our approach . it is shown explicitly by direct counting for the discrete version of the model and by conformal methods for the continuous version that multifractality occurs when local uniformity of the phase space ( which has an exponentially large number of states ) has been broken . = -0.25 in = 6.7 in =

cond-mat0001138

the phenomenon of multifractality consists in a scale dependence of critical exponents . it has been widely discussed in the literature for a wide range of issues , such as statistics of strange sets @xcite , diffusion limited aggregation @xcite , wavelet transforms @xcite , conformal invariance @xcite or statistical properties of critical wave functions of massless dirac fermions in a random magnetic field @xcite . the aim of our work is not only to describe a new model possessing multiscaling dependence , but also to show that the phenomenon of multifractality is related to local nonuniformity of the exponentially growing ( `` hyperbolic '' ) underlying `` target '' phase space , through an example of entangled random walks distribution in homotopy classes . indeed , to our knowledge , almost all examples of multifractal behavior for physical @xcite or more abstract @xcite systems share one common feature all target phase spaces have a noncommutative structure and are locally nonuniform . we believe that multiscaling is a much more generic physical phenomenon compared with uniform scaling , appearing when the phase space of a system possesses a hyperbolic structure with local symmetry breaking . such perturbation of local symmetry could be either regular or random from our point of view the details of the origin of local nonuniformity play a less significant role . we discuss below the basic features of multifractality in a locally nonuniform regular hyperbolic phase space . we show in particular that a multifractal behavior is encountered in statistical topology in the case of entangled ( or knotted ) random walks distribution in topological classes . the paper is organized as follows . in section [ sect:2 ] we consider a 2d @xmath0step random walk in a nonsymmetric array of topological obstacles and investigate the multiscaling properties of the `` target '' phase space for a set of specific topological invariants the `` primitive paths '' . the renormalization group computations of mean length of the primitive path , as well as return probabilities to the unentangled topological state are developed in section [ sect:3 ] . section [ sect:4 ] is devoted to the application of conformal methods to a geometrical analysis of multifractality in locally nonuniform hyperbolic spaces . the concept of multifractality has been formulated and clearly explained in the paper @xcite . we begin by recalling the basic definitions of rnyi spectrum , which will be used in the following . let @xmath1 be an abstract invariant distribution characterizing the probability of a dynamical system to stay in a basin of attraction of some stable configuration @xmath2 @xmath3 . taking a uniform grid parameterized by `` balls '' of size @xmath4 , we define the family of fractal dimensions @xmath5 : [ 1:dq ] d_q=_l0 as @xmath6 is varied , different subsets of @xmath7 associated with different values of @xmath6 become dominant . let us define the scaling exponent @xmath8 as follows @xmath9 where @xmath8 can take different values corresponding to different regions of the measure which become dominant in eq.([1:dq ] ) . in particular , it is natural to suggest that @xmath10 can be rewritten as follows : @xmath11 \right|_{l\to 0}\ ] ] where @xmath12 is the probability to have the value @xmath8 lying in a small `` window '' @xmath13 $ ] and @xmath14 is a continuous function which has sense of fractal dimension of the subset characterized by the value @xmath8 . supposing @xmath15 one can approximately evaluate the last expression via the saddle point method . thus , one gets ( see , for example , @xcite ) : @xmath16 what together with ( [ 1:dq ] ) leads to the following equations [ 1:dq2 ] l ( q)=q(q)-f[(q ) ] + ( q)=(q ) where @xmath17 . hence , the exponents @xmath18 and @xmath19 $ ] are related via legendre transform . for further details and more advanced mathematical analysis , the reader is refered to @xcite . topological constraints essentially modify physical properties of the broad class of statistical systems composed of chain like objects . it should be stressed that topological problems are widely investigated in connection with quantum field and string theories , 2d gravitation , statistics of vortices in superconductors , quantum hall effect , thermodynamic properties of entangled polymers etc . modern methods of theoretical physics allow us to describe rather comprehensively the effects of nonabelian statistics on the physical behavior of some systems . however the following question remains still obscure : what are the fractal ( and as it is shown below , multifractal ) properties of the distribution function of topological invariants , characterizing the homotopy states of a statistical system with topological constraints ? we investigate this problem in the framework of the model `` random walk in an array of obstacles '' ( rwao ) . the rwao model can be regarded as physically clear and as a very representative image for systems of fluctuating chain like objects with a full range of nonabelian topological properties . this model is formulated as follows : suppose that a random walk of @xmath0 steps of length @xmath20 takes place on a plane between obstacles which form a simple 2d rectangular lattice with unit cell of size @xmath21 . we assume that the random walk can not cross ( `` pass through '' ) any obstacles . it is convenient to begin with the lattice realization of the rwao model . in this case the random path can be represented as a @xmath0step random walk in a square lattice of size @xmath22 ( @xmath23)see fig.[1:fig:1 ] . it had been shown previously ( see , for example @xcite ) that for @xmath24 a lattice random walk in the presence of a regular array of obstacles ( punctures ) on the dual lattice @xmath25 is topologicaly equivalent to a free random walk on a graph a cayley tree with branching number @xmath26 ( see fig.[1:fig:2 ] ) . an outline of the derivation of this result is as follows . the different topological states of our problem coincide with the elements of the homotopy group of the multi - punctured plane , which is the free group @xmath27 generated by a countable set of elements . the translational invariance allows to consider a local basis and therefore to study the factored group @xmath28 , where @xmath29 is a free group with @xmath30 generators whose cayley graph is precisely a @xmath31branching tree . the relation between cayley trees and hyperbolic geometry is discussed in details in section [ sect:3 ] . intuitively such a relation could be understood as follows . the cayley tree can be isometrically embedded in the hyperbolic plane @xmath32 ( surface of constant negative curvature ) . the group @xmath29 is one of the discrete subgroups of the group of motion of the hyperbolic plane @xmath33 , therefore the cayley tree can be considered as a particular discrete realization of the hyperbolic plane . returning to the rwao model , we conclude that each trajectory in the lattice of obstacles can be lifted to a path in the `` universal covering space '' i.e to a path on the @xmath31branching cayley tree . the geodesic on the cayley graph , i.e the shortest trajectory along the graph which connects ends of the path , plays the role of a complete topological invariant for the original trajectory in the lattice of obstacles . for example , the random walk in the lattice of obstacles is closed and contractible to a point ( i.e. is not entangled with the array of obstacles ) if and only if the geodesic length between the ends of the trajectory on the cayley graph is zero . hence , this geodesic length can be regarded as a topological invariant , which preserves the main nonabelian features of the considered problem . we would like to stress two facts concerning our model : ( i ) the exact configuration of a geodesic is a complete topological invariant , while its length @xmath34 is only a partial topological invariant ( except the case @xmath35 ) ; ( ii ) geodesics have a clear geometrical interpretation , having sense of a bar ( or `` primitive '' ) path which remains after deleting all even times folded parts of a random trajectory in the lattice of obstacles . the concept of `` primitive path '' has been repeatedly used in statistical physics of polymers , leading to a successful classification of the topological states of chain like molecules in various topological problems @xcite . even if many aspects of statistics of random walks in fixed lattices of obstacles have been well understood ( see , for example @xcite and references therein ) , the set of problems dealing with the investigation of fractal properties of the distribution of topological invariants in the rwao model are practically out of discussion . thus we devote the next section to the study of fractal and multifractal structures of the measure on the set of primitive paths in the rwao model for @xmath36 . the classification of different topological states of a @xmath0step random walk in a rectangular lattice of obstacles in the case @xmath36 turns out to be a more difficult and more rich problem than in the case @xmath37 discussed above . however , after a proper rescaling , the mapping of a random walk in the rectangular array of obstacles to a random walk on a cayley tree can be explored again . to proceed we should solve two auxiliary problems . first of all we consider a random walk inside the elementary rectangular cell of the lattice of obstacles . let us compute : \(i ) the `` waiting time '' , i.e the average number of steps @xmath38 which a @xmath39step random walk spends within the rectangle of size @xmath40 ; \(ii ) the ratio of the `` escape probabilities '' @xmath41 and @xmath42 through the corresponding sides @xmath43 and @xmath44 for a random walk staying till time @xmath39 within the elementary cell . the desired quantities can be easily computed from the distribution function @xmath45 which gives the probability to find the @xmath39step random walk with initial @xmath46 and final @xmath47 points within the rectangle of size @xmath40 . the function @xmath48 in the continuous approximation ( @xmath49 ) is the solution of the following boundary problem [ 1:2 ] \ { l p(x , y , t ) = ( + ) p(x , y , t ) + p(0,y , t)=p(c_x , y , t)=p(x,0,t)=p(x , c_y , t)=0 + p(x , y,0)=(x_0,y_0 ) . where @xmath20 is the length of the effective step of the random walk and the value @xmath50 has sense of a diffusion constant . the solution of eqs.([1:2 ] ) reads [ 1:3 ] p(x_0,y_0,x , y , t)= _ m_x=1^_m_y=1^ e^-(+ ) t the `` waiting time '' @xmath38 can be written now as follows [ 1:3a ] = _ 0^c_x dx_0 _ 0^c_y dy_0 _ 0^c_x dx _ 0^c_y dy _ 0^dt p(x_0,y_0,x , y , t ) while the ratio @xmath51 can be computed straightforwardly via the relation : [ 1:3b ] = in the `` ground state dominance '' approximation we truncate the sum ( [ 1:3 ] ) at @xmath52 and get the following approximate expressions : [ 1:4a ] = ; = in the symmetric case ( @xmath53 ) eq.([1:4a ] ) gives @xmath54 and @xmath55 , as it should be for the square lattice of obstacles . now the distribution function of the primitive paths for the rwao model can be obtained via lifting this topological problem to the problem of _ directed _ random walks on the 4branching cayley tree , where the random walk on the cayley tree is defined as follows : \(a ) the total number of steps @xmath56 on the cayley tree is @xmath57 ( the value @xmath38 has been computed in ( [ 1:4a ] ) ) . \(b ) the distance ( or `` level '' @xmath34 ) on the cayley tree is defined as the number of steps of the shortest path between two points on the tree . each vertex of the cayley tree has 4 branches ; the steps along two of them carry a boltzmann weight @xmath58 , while the steps along the two remaining ones carry a boltzmann weight @xmath59 as it is shown in fig.[fig:4_cayley ] . the value of @xmath59 is fixed by eq.([1:4a ] ) , which yields [ 1:4c ] = = the ultrametric structure of the topological phase space , i.e. of the cayley tree @xmath60 , allows us to use the results of paper @xcite for investigating multicritical properties of the measure of all primitive ( directed ) paths of @xmath34 steps along the graph @xmath60 with nonsymmetric weights @xmath58 and @xmath59 ( see fig.[fig:4_cayley ] ) . a rigorous mathematical description of such weighted paths on trees ( called cascades ) can be found in @xcite , where the authors derive multifractal spectra , but for different distributions of weights . = 1.00 mm ( 153.33,55.00 ) ( 80.00,55.00)(-1,-1)20.00 ( 80.00,55.00)(1,-1)20.00 ( 80.00,55.00)(3,-1)60.00 ( 80.00,55.00)(-3,-1)60.00 ( 20.00,35.00)(-2,-3)10.00 ( 20.00,35.00)(0,-1)15.00 ( 20.00,35.00)(2,-3)10.00 ( 60.00,35.00)(-2,-3)10.00 ( 60.00,35.00)(0,-1)15.00 ( 60.00,35.00)(2,-3)10.00 ( 100.00,35.00)(-2,-3)10.00 ( 100.00,35.00)(0,-1)15.00 ( 100.00,35.00)(2,-3)10.00 ( 140.00,35.00)(-2,-3)10.00 ( 140.00,35.00)(0,-1)15.00 ( 140.00,35.00)(2,-3)10.00 ( 50.00,20.00)(-1,-3)3.33 ( 50.00,20.00)(0,-1)10.00 ( 50.00,20.00)(1,-3)3.33 ( 60.00,20.00)(-1,-3)3.33 ( 60.00,20.00)(0,-1)10.00 ( 60.00,20.00)(1,-3)3.33 ( 70.00,20.00)(-1,-3)3.33 ( 70.00,20.00)(0,-1)10.00 ( 70.00,20.00)(1,-3)3.33 ( 10.00,20.00)(-1,-3)3.33 ( 10.00,20.00)(0,-1)10.00 ( 10.00,20.00)(1,-3)3.33 ( 20.00,20.00)(-1,-3)3.33 ( 20.00,20.00)(0,-1)10.00 ( 20.00,20.00)(1,-3)3.33 ( 30.00,20.00)(-1,-3)3.33 ( 30.00,20.00)(0,-1)10.00 ( 30.00,20.00)(1,-3)3.33 ( 130.00,20.00)(-1,-3)3.33 ( 130.00,20.00)(0,-1)10.00 ( 130.00,20.00)(1,-3)3.33 ( 140.00,20.00)(-1,-3)3.33 ( 140.00,20.00)(0,-1)10.00 ( 140.00,20.00)(1,-3)3.33 ( 150.00,20.00)(-1,-3)3.33 ( 150.00,20.00)(0,-1)10.00 ( 150.00,20.00)(1,-3)3.33 ( 90.00,20.00)(-1,-3)3.33 ( 90.00,20.00)(0,-1)10.00 ( 90.00,20.00)(1,-3)3.33 ( 100.00,20.00)(-1,-3)3.33 ( 100.00,20.00)(0,-1)10.00 ( 100.00,20.00)(1,-3)3.33 ( 110.00,20.00)(-1,-3)3.33 ( 110.00,20.00)(0,-1)10.00 ( 110.00,20.00)(1,-3)3.33 ( 38.00,44.00)(0,0)[cc]@xmath58 ( 74.00,44.00)(0,0)[cc]@xmath59 ( 86.00,44.00)(0,0)[cc]@xmath58 ( 123.00,44.00)(0,0)[cc]@xmath59 ( 22.00,24.00)(0,0)[cc]@xmath58 ( 10.00,24.00)(0,0)[cc]@xmath59 ( 31.00,24.00)(0,0)[cc]@xmath59 ( 102.00,24.00)(0,0)[cc]@xmath58 ( 90.00,24.00)(0,0)[cc]@xmath59 ( 111.00,24.00)(0,0)[cc]@xmath59 ( 62.00,24.00)(0,0)[cc]@xmath59 ( 50.00,24.00)(0,0)[cc]@xmath58 ( 71.00,24.00)(0,0)[cc]@xmath58 ( 142.00,24.00)(0,0)[cc]@xmath59 ( 130.00,24.00)(0,0)[cc]@xmath58 ( 151.00,24.00)(0,0)[cc]@xmath58 ( 7.00,7.00)(0,0)[cc]@xmath58 ( 10.00,7.00)(0,0)[cc]@xmath59 ( 13.00,7.00)(0,0)[cc]@xmath58 ( 27.00,7.00)(0,0)[cc]@xmath58 ( 30.00,7.00)(0,0)[cc]@xmath59 ( 33.00,7.00)(0,0)[cc]@xmath58 ( 17.00,7.00)(0,0)[cc]@xmath59 ( 20.00,7.00)(0,0)[cc]@xmath58 ( 23.00,7.00)(0,0)[cc]@xmath59 ( 57.00,7.00)(0,0)[cc]@xmath58 ( 60.00,7.00)(0,0)[cc]@xmath59 ( 63.00,7.00)(0,0)[cc]@xmath58 ( 47.00,7.00)(0,0)[cc]@xmath59 ( 50.00,7.00)(0,0)[cc]@xmath58 ( 53.00,7.00)(0,0)[cc]@xmath59 ( 67.00,7.00)(0,0)[cc]@xmath59 ( 70.00,7.00)(0,0)[cc]@xmath58 ( 73.00,7.00)(0,0)[cc]@xmath59 ( 87.00,7.00)(0,0)[cc]@xmath58 ( 90.00,7.00)(0,0)[cc]@xmath59 ( 93.00,7.00)(0,0)[cc]@xmath58 ( 107.00,7.00)(0,0)[cc]@xmath58 ( 110.00,7.00)(0,0)[cc]@xmath59 ( 113.00,7.00)(0,0)[cc]@xmath58 ( 97.00,7.00)(0,0)[cc]@xmath59 ( 100.00,7.00)(0,0)[cc]@xmath58 ( 103.00,7.00)(0,0)[cc]@xmath59 ( 137.00,7.00)(0,0)[cc]@xmath58 ( 140.00,7.00)(0,0)[cc]@xmath59 ( 143.00,7.00)(0,0)[cc]@xmath58 ( 127.00,7.00)(0,0)[cc]@xmath59 ( 130.00,7.00)(0,0)[cc]@xmath58 ( 133.00,7.00)(0,0)[cc]@xmath59 ( 147.00,7.00)(0,0)[cc]@xmath59 ( 150.00,7.00)(0,0)[cc]@xmath58 ( 153.00,7.00)(0,0)[cc]@xmath59 we construct the partition function @xmath61 which counts properly the weighted number of all @xmath62 different @xmath34step primitive paths on the graph @xmath60 . define two partition functions @xmath63 and @xmath64 of @xmath34step paths , whose last steps carry the weights @xmath58 and @xmath59 correspondingly . these functions satisfy the recursion relations for @xmath65 : [ 1:6 ] \ { l a_k+1=a_k+2b_k + b_k+1=2 a_k+ b_k . ( k1 ) with the following initial conditions at @xmath66 : [ 1:6a ] \ { l a_1=2 + b_1=2 . combining ( [ 1:6 ] ) and ( [ 1:6a ] ) we arrive at the following 2step recursion relation for the function @xmath63 : [ 1:6b ] \ { ll a_k+2=(1 + ) a_k+1 + 3 a_k & ( k1 ) + a_1=2 & ( k=1 ) + a_2=2 + 4 & ( k=2 ) . whose solution is [ 1:6c ] a_k= _ 1^k-1 + _ 2^k-1 where [ 1:6d ] _ 1,2=(1 + ) taking into account that @xmath64 is given by the same recursion relation as @xmath63 but with the initial values @xmath67 and @xmath68 , we get the following expression for the partition function @xmath69 : [ 1:6sum ] ( , k)= _ 1^k-1 + _ 2^k-1 the partition function @xmath61 contains all necessary information about the multifractal behavior . following eqs.([1:dq])([1:dq2 ] ) , we associate the set of stable configurations @xmath70 with the set of @xmath71 vertices of level @xmath34 . hence , we define [ 1:set ] _ i=1^n ^q(c_i)= + taking into account that the uniform grid has resolution @xmath72 for @xmath65 and using eq.([1:dq ] ) , we obtain [ 1:7 ] ( q)=_k which allows to determine the generalized hausdorff dimension @xmath5 via the relation [ 1:7a ] d_q=(q)/(q-1 ) the corresponding plots of the functions @xmath73 for different values of @xmath74 are shown in fig.[1:fig:4 ] ( the numerical computations of eqs.([1:7])([1:7a ] ) are carried out for @xmath75 ) . the fact that @xmath73 depends on @xmath6 clearly demonstrate the multifractal behavior . consider a random walk on a 4-branching cayley tree and investigate the distribution @xmath76 giving the probability for a @xmath56step random walk starting at the origin of the tree to have a primitive ( shortest ) path between ends of length @xmath34 . the random walk is defined as follows : at each vertex of the cayley tree the probability of a step along two of the branches is @xmath77 , and is @xmath78 along the two others ; @xmath77 and @xmath78 satisfy the conservation condition @xmath79 . using eq.([1:4c ] ) , the following expressions hold : [ 2:pxpy ] \ { l p_x= + p_y= . the symmetric case @xmath80 ( which gives @xmath81 ) has already been studied and an exact expression for @xmath76 has been derived in @xcite . the rigorous mathematical description of random walks on graphs can be found in @xcite . importance of spherical symmetry ( i - e the fact that all vertices of a given level are strictly equivalent ) is discussed in @xcite . another example of nonsymmetric model on a tree ( case of randomly distributed transition probabilities , the so - called rwre model ) is described in @xcite . to our knowledge , the solution for the nonsymmetric random walk which we defined above is known only for @xmath34 fixed and @xmath82 @xcite . here we consider the case @xmath83 , and in particular we study the distribution in the neighborhood of the maximum . breaking the symmetry by taking @xmath84 affects strongly the structure of the problem , since then the phase space becomes locally nonuniform : we have now vertices of two different kinds , @xmath85 and @xmath86 , depending on whether the step toward the root of the cayley tree occurs with probability @xmath77 or @xmath78 . in order to obtain a master equation for @xmath76 , we introduce the new variables @xmath87 and @xmath88 , which define the probabilities to be at the level @xmath34 in a vertex @xmath85 or @xmath86 after @xmath56 steps . we recursively define the same way the probabilities @xmath89 , ( @xmath90 ) to be at level @xmath34 in a vertex such that the sequence of vertices toward the root of the tree is @xmath91 . one can see that the recursion depends on the total `` history '' till the root point , what makes the problem nonlocal . the master equation for the distribution function @xmath76 [ 2:p11 ] lll p(k,+1 ) & = & ( 2p_x+p_y)l_y(k-1,)+(2p_y+p_x)l_x(k-1,)+ + & & p_yl_y(k+1,)+p_xl_x(k+1 , ) is coupled to the hierarchical set of functions @xmath92 which satisfy the following recursion relation [ 2:p12 ] lll l_a_1 ... a_n(k,+1 ) & = & ( 2-_a_1,a_2 ) p_a_1l_a_2 ... a_n(k-1,)+ + & & p_xl_xa_1 ... a_n(k+1,)+p_yl_ya_1 ... a_n(k+1 , ) where @xmath91 cover all sequences of any lengths ( @xmath93 ) in @xmath85 and @xmath86 . in order to close this infinite system at an arbitrary order @xmath94 we make the following assumption : for any @xmath95 we have [ 2:ap ] .|_kn_0n_0 _ a_1 ... a_n with @xmath96 constant . using the approximation ( [ 2:ap ] ) we rewrite ( [ 2:p11])([2:p12 ] ) for large @xmath34 and @xmath56 in terms of the function @xmath76 and constants @xmath96 ( @xmath97 ) . taking into account that @xmath98 we arrive at @xmath99 independent recursion relations for one and the same function @xmath76 , with @xmath100 independent unknown constants @xmath101 . in order to make this system self consistent , one has to identify coefficients entering in different equations , what yields @xmath102 compatibility relations for the constants @xmath101 , and the system is still open . this fact means that all scales are involved and the evolution of @xmath103 depends on @xmath104 , the evolution of @xmath104 depends on @xmath105 and so on . at each scale we need informations about larger scales . this kind of scaling problem naturally suggests to use a renormalization group approach , which is developed in the next section . to begin with the renormalization procedure , we need to estimate the values of the constants @xmath101 for the first ( i.e. the smallest ) scale . let us denote @xmath106 and define @xmath107 as follows : @xmath108 now we set [ 2:ap2 ] p_x_xa_1 ... a_n+p_y_ya_1 ... a_n= ( p_x+p_y(1-))_a_1 ... a_n what means that we neglect the correlations between the constants @xmath96 and @xmath109 at different scales . as it is shown in the next section , the renormalization group approach allows us to get rid of the approximation ( [ 2:ap2 ] ) . with ( [ 2:ap2 ] ) one can obtain the following generic master equation [ 2:sys1 ] p(k,+1 ) = ( 2-_a_1,a_2)p(k-1,)+(p_x+(1-)p_y)p(k+1 , ) where @xmath91 again cover all possible sequences in @xmath85 and @xmath86 . we have now @xmath100 unknown quantities with @xmath100 compatibility relations ( [ 2:sys1 ] ) , what makes the system ( [ 2:sys1 ] ) closed . for illustration , we derive the solution for @xmath110 : [ 2:2step ] \ { l p(k,+1)=p(k-1,)+ ( p_x+(1-)p_y)p(k+1 , ) + p(k,+1)=p(k-1,)+ ( p_x+(1-)p_y)p(k+1 , ) + p(k,+1)=p(k-1,)+ ( p_x+(1-)p_y)p(k+1 , ) + p(k,+1)=p(k-1,)+ ( p_x+(1-)p_y)p(k+1 , ) . note that ( [ 2:2step ] ) displays clearly a @xmath111 symmetry : @xmath112 . compatibility conditions for system ( [ 2:2step ] ) read : [ 2:values ] = = = which finally gives [ 2:sol ] \ { l = + v_x= + v_y= . as it has been said above , without ( [ 2:ap2 ] ) the system ( [ 2:p11])([2:p12 ] ) is open , giving a single equation for the unknown function @xmath76 depending on the unknown parameter @xmath113 : [ 2:p2 ] lll p(k,+1 ) & = & ( ( 2p_x+p_y)(1-)+ ( 2p_y+p_x))p(k-1,)+ + & & ( p_y(1-)+p_x)p(k+1 , ) eq.([2:p2 ] ) describes a 1d diffusion process with a drift [ drift ] = 2p_y+2(1-)p_x and a dispersion [ dispersion ] = = 1 - 4(p_y+(1-)p_x)^2 which provides for @xmath114 and @xmath115 the usual gaussian distribution with nonzero mean ( see @xcite ) . the value of @xmath8 obtained in ( [ 2:sol ] ) using the approximation ( [ 2:ap2 ] ) gives a fair estimate of the drift compared with the numerical simulations , as it is shown in fig.[2:num ] . in order to improve the results obtained above , we recover the information lost in the approximation ( [ 2:ap2 ] ) and take into account `` interactions '' between different scales . namely , we follow the renormalization flow of the parameter @xmath116 at a scale @xmath4 supposing that a new effective step is a composition of @xmath117 initial lattice steps . let us define : * the probability @xmath118 of going forth ( with respect to the location of the root point of the cayley tree ) from a vertex of kind @xmath20 ; * the probability @xmath119 of going back ( towards the root point of the cayley tree ) from a vertex of kind @xmath20 ; * the probability @xmath116 of being at a vertex of kind @xmath85 ; * the conditional probability @xmath120 to reach a vertex of kind @xmath20 starting from a vertex of kind @xmath20 under the condition that the step is forth ; * the conditional probability @xmath121 to reach a vertex of kind @xmath20 starting from a vertex of kind @xmath20 under the condition that the step is back ; * the effective length @xmath122 of a composite step . then the drift @xmath123 at scale @xmath4 is given by ( compare with [ drift ] ) : [ 2:kl ] ( l)=d(l ) we say that the problem is scale independent if the flow @xmath123 is invariant under the decimation procedure , i.e. with respect to the renormalization group . we compute the flow counting the appropriate combinations of two steps , depending on the variable considered : [ 2:renorm ] lll w_a(l+1 ) & = & ( 1-w_a(l))(1-w_|a(l))+w_a^2(l ) + v_a(l+1 ) & = & ( 1-v_a(l))(1-v_|a(l))+v_a^2(l ) + f_a(l+1 ) & = & + b_a(l+1 ) & = & + d(l+1 ) & = & d(l ) + ( l+1 ) & = & ( l)+ + & & ( 1-(l ) ) where @xmath124 when @xmath125 ( and @xmath126 when @xmath127 ) and the value @xmath128 ensures the conservation condition @xmath129 because we do not consider the combinations of two successive steps in opposite directions . the transformation of @xmath113 in ( [ 2:renorm ] ) needs some explanations . we consider the drift @xmath123 as a probability to make a ( composite ) step forward . the equation for @xmath113 is given by counting the different ways of getting to a vertex of kind @xmath85 . one can check that @xmath123 given by ( [ 2:kl ] ) remains invariant under such transformation , what is considered as a verification of the scale independence ( i.e. of renormalizability ) . following the standard procedure , we find the fixed points for the flow of @xmath130 . first of all we realize that the recursion equations for @xmath120 and @xmath121 can be solved independently , providing a continuous set of fixed points : @xmath131 and @xmath132 . using the initial conditions ( [ 2:sol ] ) for @xmath121 and deriving straightforwardly the absent initial conditions for @xmath120 , we get [ 2:ic ] \ { l v_x(1)=v_x + v_y(1)=v_y + w_x(1)=w_x= + w_y(1)=w_y= . ( we recall that these values are obtained by taking into account the elementary correlations for two successive steps ) . with the initial conditions ( [ 2:ic ] ) we find the following renormalized values @xmath133 and @xmath134 at the fixed point [ 2:limv ] \ { l v^0=v^0()=_lv_x(l)=1-_lv_y(l)= + w^0=w^0()=_lw_x(l)=1-_lw_y(l)= . where @xmath135 is the @xmath136 iteration of the function @xmath137 we then obtain successively all renormalized values at the fixed point [ 2:fp ] \ { l f_a^0=1 + b_a^0=0 + d^0== + ^0= . where the invariance of the drift @xmath138 is taken into account : @xmath139 in figure [ 2:num ] we compare the theoretical results with numerical simulations . it is worth mentioning the efficiency of the renormalization group method , which yields a solution in very good agreement with numerical simulations in a broad interval of values @xmath59 . in addition we compare our results with the exact expression obtained by p.gerl and w.woess in @xcite for the probability @xmath140 to return to the origin after @xmath56 random steps on the nonsymmetric cayley tree . this distribution function @xmath140 reads [ woess1 ] p(0,)^^-3/2 with ( ) = \{-t++ |t>0 } let us assert now without justification that eq.([2:p2 ] ) ( which is actually written for @xmath114 and @xmath115 ) is valid for any values of @xmath34 and @xmath56 . the initial conditions for the recursion relation ( [ 2:p2 ] ) are as follows [ 2:init ] \ { l p(0,+1)=(2p_x+2(1- ) p_y)p(1 , ) + p(k,0)=_k,0 . one can notice that eq.([2:p2 ] ) completed with the conditions ( [ 2:init ] ) can be viewed as a master equation for a _ symmetric _ random walk on a cayley tree _ with effective branching _ @xmath31 continuously depending on @xmath59 : z()= hence , we conclude that our problem becomes equivalent to a symmetric random walk on a @xmath141-branching tree . for @xmath35 the solution , given in @xcite is [ our ] p(0,)^^-3/2 this provides the same form as the exact solution ( [ woess1 ] ) . it has been checked numerically that for @xmath142 the discrepancy between ( [ woess1 ] ) and ( [ our ] ) is as follows @xmath143 thus , we believe that our self consistent rg approach to statistics of random walks on nonsymmetric trees can be extended with sufficient accuracy to all values of @xmath34 . we have claimed in sections [ sect:1][sect:2 ] that local nonuniformity and the exponentially growing structure of the phase space of statistical systems generates a multiscaling behavior of the corresponding partition functions . the aim of the present section is to bring geometric arguments to support our claim by introducing a different approach of the rwao model . the differences between the approach considered in this section and the one discussed in section [ sect:2 ] are as follows : * we consider a _ continuous _ model of random walk topologically entangled with either a symmetric or a nonsymmetric _ triangular _ lattice of obstacles on the plane . * we pursue the goal to construct _ explicitly _ the metric structure of the topological phase space via conformal methods and _ to relate directly the nonuniform fractal relief _ of the topological phase space _ to the multifractal properties _ of the distribution function of topological invariants for the given model . consider a random walk in a regular array of topological obstacles on the plane . as in the discrete case we can split the distribution function of all @xmath0step paths with fixed positions of end points into different topological ( homotopy ) classes . we characterize each topological class by a topological invariant similar to the `` primitive path '' defined in section [ sect:2 ] . introducing complex coordinates @xmath144 on the plane , we use conformal methods which provide an efficient tool for investigating multifractal properties of the distribution function of random trajectories in homotopy classes . let us stress that explicit expressions are constructed so far for triangular lattices of obstacles only . that is why we replace the investigation of the rectangular lattices discussed in sections [ sect:1][sect:2 ] by the consideration of the triangular ones . moreover , for triangular lattices a continuous symmetry parameter ( such as @xmath145 in case of rectangular lattices ) does not exist and only the triangles with angles @xmath146 , @xmath147 , @xmath148 are available only such triangles tessellate the whole plane @xmath31 . in spite of the mentioned restrictions , the study of these cases enables us to figure out an origin of multifractality coming from the metric structure of the topological phase space . suppose that the topological obstacles form a periodic lattice in the @xmath31plane . let the fundamental domain of this lattice be the triangle @xmath149 with angles either @xmath146 ( symmetric case ) or @xmath148 ( nonsymmetric case ) . the conformal mapping @xmath150 establishes a one - to - one correspondence between a given fundamental domain @xmath149 of the lattice of obstacles in the @xmath31plane with a zero angled triangle @xmath151 lying in the upper half plane @xmath152 of the plane @xmath153 , and having corners on the real axis @xmath154 . to avoid possible misunderstandings let us point out that such transform is conformal everywhere except at corner ( branching ) points see , for example @xcite . consider now the tessellation of the @xmath31plane by means of consecutive reflections of the domain @xmath149 with respect to its sides , and the corresponding reflections ( inversions ) of the domain @xmath151 in the @xmath155plane . few first generations are shown in fig.[fig : lacun ] . the obtained upper half plane @xmath156 has a `` lacunary '' structure and represents the topological phase space of the trajectories entangled with the lattice of obstacles . the details of such a construction as well as a discussion of the topological features of the conformal mapping @xmath150 in the symmetric case can be found in @xcite . we recall the basic properties of the transform @xmath150 related to our investigation of multifractality . the topological state of a trajectory @xmath157 in the lattice of obstacles can be characterized as follows . * perform the conformal mappings @xmath158 ( or @xmath159 ) of the plane @xmath31 with symmetric ( or nonsymmetric ) triangular lattice of obstacles to the upper half - plane @xmath160 , playing the role of the topological phase space of the given model . * connect by nodes the centers of neighboring curvilinear triangles in the upper half plane @xmath160 and raise a graph @xmath161 ( or @xmath162 ) ( which is , as shown below an isometric cayley tree embedded in the poincar plane ) . * find the image of the path @xmath157 in the `` covering space '' @xmath160 and define the shortest ( primitive ) path connecting the centers of the curvilinear triangles where the ends of the path @xmath157 are located . the configuration of this primitive path projected to the cayley tree @xmath161 ( or @xmath162 ) plays the role of topological invariant for the model under consideration . the cayley trees @xmath163 have the same topological content as the one described in the section [ sect:2 ] , but here we determine the boltzmann weights @xmath164 associated with passages between neighboring vertices ( see fig.[fig:3_cayley ] ) directly from the metric properties of the topological phase space obtained via the conformal mappings @xmath165 . = 1.00 mm ( 113.33,55.00 ) ( 20.00,35.00)(-2,-3)10.00 ( 20.00,35.00)(2,-3)10.00 ( 60.00,35.00)(-2,-3)10.00 ( 60.00,35.00)(2,-3)10.00 ( 100.00,35.00)(-2,-3)10.00 ( 100.00,35.00)(2,-3)10.00 ( 10.00,20.00)(-1,-3)3.33 ( 10.00,20.00)(1,-3)3.33 ( 30.00,20.00)(-1,-3)3.33 ( 30.00,20.00)(1,-3)3.33 ( 90.00,20.00)(-1,-3)3.33 ( 90.00,20.00)(1,-3)3.33 ( 110.00,20.00)(-1,-3)3.33 ( 110.00,20.00)(1,-3)3.33 ( 50.00,20.00)(-1,-3)3.33 ( 50.00,20.00)(1,-3)3.33 ( 70.00,20.00)(-1,-3)3.33 ( 70.00,20.00)(1,-3)3.33 ( 30.00,44.00)(0,0)[cc]@xmath166 ( 64.00,44.00)(0,0)[cc]@xmath167 ( 90.00,44.00)(0,0)[cc]@xmath58 ( 8.00,24.00)(0,0)[cc]@xmath58 ( 33.00,24.00)(0,0)[cc]@xmath167 ( 48.00,24.00)(0,0)[cc]@xmath166 ( 73.00,24.00)(0,0)[cc]@xmath58 ( 88.00,24.00)(0,0)[cc]@xmath167 ( 113.00,24.00)(0,0)[cc]@xmath166 ( 7.00,7.00)(0,0)[cc]@xmath166 ( 13.00,7.00)(0,0)[cc]@xmath167 ( 27.00,7.00)(0,0)[cc]@xmath58 ( 33.00,7.00)(0,0)[cc]@xmath166 ( 47.00,7.00)(0,0)[cc]@xmath167 ( 53.00,7.00)(0,0)[cc]@xmath58 ( 67.00,7.00)(0,0)[cc]@xmath166 ( 73.00,7.00)(0,0)[cc]@xmath167 ( 87.00,7.00)(0,0)[cc]@xmath58 ( 93.00,7.00)(0,0)[cc]@xmath166 ( 107.00,7.00)(0,0)[cc]@xmath167 ( 113.00,7.00)(0,0)[cc]@xmath58 ( 60.00,55.00)(-2,-1)40.00 ( 60.00,55.00)(2,-1)40.00 ( 60.00,55.00)(0,-1)20.00 it is well known that random walks are conformally invariant ; in other words the diffusion equation on the plane @xmath31 preserves its structure under a conformal transform , but the diffusion coefficient can become space dependent @xcite . namely , under the conformal transform @xmath150 the laplace operator @xmath168 is transformed in the following way [ laplace ] = before discussing the properties of the jacobians @xmath169 for the symmetric and nonsymmetric transforms , it is more convenient to set up the following geometrical context . the connection between cayley trees and surfaces of constant negative curvature has already been pointed out @xcite , mostly through volume growth considerations . therefore it becomes more natural to regard the upper half plane @xmath160 as the standard realization of the hyperbolic 2space ( surface of constant negative curvature @xmath170 , with here arbitrarily @xmath171 ) , that is to consider the following metric : [ met ] ds^2=(d^2+d^2 ) let us rewrite the laplace operator ( [ laplace ] ) in the form [ laplace2 ] = d(,)^2 ( + ) where the value @xmath172 can be interpreted as the normalized space dependent diffusion coefficient on the poincar upper half plane : [ diffusion ] d()= the methods providing the conformal transform @xmath158 for the symmetric triangle with angles @xmath146 have been discussed in details in @xcite . the generalization of these results to the conformal transform @xmath159 for the nonsymmetric triangle with angles @xmath148 is very straightforward . we here expose the jacobians of those conformal mappings without derivation : [ jac ] l |z_s()|^2=|_1(0,e^i)|^8/3 + where @xmath173 and @xmath174 @xmath175 are the standard definitions of jacobi elliptic functions @xcite . combining ( [ diffusion ] ) and ( [ jac ] ) we define the effective inverse diffusion coefficients in symmetric ( @xmath176 ) and in nonsymmetric ( @xmath177 ) cases : [ diff_s_ns ] l d^-1_s()=^2|z_s()|^2 + d^-1_ns()=^2|z_ns()|^2 the corresponding 3d plots of the reliefs @xmath178 and @xmath178 are shown in fig.[prof ] . the functions @xmath179 and @xmath180 are considered as quantitative indicators of the topological structure of the phase spaces ; in particular a cayley tree can be isometrically embedded in the surface @xmath179 . it can be shown that the images of the centers of the triangles of the symmetric lattice in the @xmath31plane correspond to the local maxima of the surface @xmath181 in the @xmath155plane . we define the vertices of the embedded tree as those maxima . the links connecting neighboring vertices are defined in the next paragraph . let us define the _ horocycles _ which correspond to repeating sequences of weights in fig.[fig:3_cayley ] with minimal periods . there are only three such sequences : @xmath182 , @xmath183 and @xmath184 . the horocycles are images ( analytically known ) of certain circles of the @xmath31plane . they proved to be a convenient tool for a constructive description of the trajectories in the @xmath31plane starting from the trajectories in the covering space @xmath185 . the first generation of horocycles ( closest to the root point of the cayley tree ) is shown in fig.[prof ] . let us consider the symmetric case . following a given horocycle we follow a ridge of the surface , and we pass through certain maxima of this surface ( that is through certain vertices of the tree ) . we therefore define _ locally _ the links of the tree as the set of ridges connecting neighboring maxima of @xmath178 . we recall that the ridge of the surface can be defined as the set of points where the gradient of the function @xmath178 is minimal along its isoline . even if this gives a proper definition of the tree , extracting a direct parametrization is difficult , that is why henceforth we will approximate the tree by arcs of horocycles . to give a quantitative formulation of the local definition of the embedded cayley tree , we consider the path integral formulation of the problem on the @xmath185plane . define the lagrangian @xmath186 of a free particle moving with the diffusion coefficient @xmath187 in the space @xmath185 . following the canonical procedure and minimizing the corresponding action @xcite , we get the equations of motion in the effective potential @xmath188 : [ euler ] q_i=(q_j_ju)q_i- _ iu where @xmath189 and @xmath190 . even if eq.([euler ] ) is nonlinear with a friction term , one can show that the trajectory of extremal action between the centers of two neighboring triangles follow the ridge of the surface @xmath187 . it is noteworthy that obtaining an analytical support of cayley graphs is of great importance , since those graphs clearly display ultrametric properties and have connections to @xmath191adic surfaces @xcite . the detailed study of metric properties of the functions @xmath192 and @xmath193 is left for separate publication . while the self similar properties of the jacobians of those conformal mappings appear clearly in fig.[prof ] , one could wonder how the local symmetry breaking affects the continuous problem . we can see that if @xmath192 is univalued along the embedded tree , @xmath193 does vary , what makes the tree locally nonuniform and leads to a multifractal behavior . in other words , different paths of same length along the tree have the same weights in the symmetric case , but have different ones in the nonsymmetric case . the probability of a random path @xmath157 of length @xmath194 can be written in terms of a path integral with a wiener measure [ weight ] p_c = d\{s}\{-_0^l ( ) ^2dt } where @xmath195 is a parametric representation of the path @xmath157 . the first horocycles in fig.[prof ] can be parameterized as follows [ para ] \ { = ( - ) + = ( 1- ) . with @xmath196 running in the interval @xmath197 the condition ensuring the constant velocity @xmath198 along the horocycles gives with ( [ met ] ) @xmath199 hence [ t ] ( t)= ( ) with proper choice of the time unit . this parameterization is used to check that the embedded tree is isometric . indeed , the horocycles shown in fig.[prof ] correspond to a periodic sequence of steps like @xmath182 , @xmath200 or @xmath201 . it is natural to assert that a step carries a boltzmann weight characterized by the corresponding local values of @xmath202 . therefore the period of the plot shown in fig.[iso ] is directly linked to the spacing of the tree embedded in the profile @xmath202 . coming back to the probability of different paths covered at constant velocity , one can write [ heights ] -p_c_t_1^t_2 the figure [ slo ] shows the value @xmath203 in symmetric and nonsymmetric cases for different paths starting at @xmath204 and ending at @xmath39 . in the symmetric case all plots are the same ( solid line ) , whereas in the nonsymmetric case they are different : dashed and dot dashed curves display the corresponding plots for the sequences @xmath205 and @xmath206 . following the outline of construction of the fractal dimensions @xmath5 in section [ sect:2 ] , we can describe multifractality in the continuous case by [ haus ] d_q=-_l where @xmath207 is the area of the surface covered by the trajectories of length @xmath194 . this form is consistent with definitions [ 1:dq ] and [ 1:7 ] . indeed , if instead of the usual wiener measure one chooses a discrete measure @xmath208 , which is nonzero only for trajectories along the cayley tree , we recover the following description . define the distribution function @xmath209 , which has sense of the weighted number of directed paths of @xmath34 steps on the nonsymmetric 3branching cayley tree shown in fig.[fig:3_cayley ] . the values of the effective boltzmann weights @xmath166 and @xmath167 are defined in terms of the local heights of the surface @xmath210 along the corresponding branches of the embedded tree . we set l = 1.07 ; + = 1.19 where @xmath211 are adjusted so that @xmath212 represents a step weighted with @xmath213 for @xmath214 and a step weighted with @xmath215 for @xmath216 for right hand side horocycles while @xmath217 represents a step weighted with @xmath213 for @xmath214 and a step weighted with @xmath218 for @xmath216 for left hand side horocycles . the partition function @xmath219 can be computed via straightforward generalization of eq.([1:6 ] ) ; it can be written in the form : ( , , k)= a_0 _ 1^k-1 + b_0 _ 2^k-1 + c_0 _ 3^k-1 ( k1 ) where @xmath220 , @xmath221 and @xmath222 are the roots of the cubic equation @xmath223 and @xmath224,@xmath225 and @xmath226 are the solutions of the following system of linear equations @xmath227 knowing the distribution function @xmath219 , eq.([haus ] ) with the discrete measure @xmath208 reads now ( compare to ( [ 1:7])([1:7a ] ) ) [ haus2 ] d_q=-_k the plot of the function @xmath73 is shown in fig.[fig : haus2 ] ( the plot is drawn for @xmath75 ) . the results presented in sections [ sect:2][sect:4 ] are summarized ; they underline several problems still unsolved related to our work , and raise the issue of their possible applications to real physical systems . the basic concepts of multifractality have been clearly formulated mainly for abstract systems in @xcite . in the present work , we have tried to remain as close as possible to these classical formulations , while adding to abstract models of ref.@xcite the new physical content of topological properties of random walks entangled with an array of obstacles . our results point out two conditions which generate multifractality for any physical system : ( i ) an exponentially growing number of states , i.e. `` hyperbolicity '' of the phase space , and ( ii ) the breaking of a local symmetry of the phase space ( while on large scales the phase space could remain isotropic ) . in section [ sect:2 ] we have considered the topological properties of the discrete `` random walk in a rectangular lattice of obstacles '' model . generalizing an approach developed earlier ( see for example @xcite and references therein ) we have shown that the topological phase space of the model is a cayley tree whose associated transition probabilities are nonsymmetric . transition probabilities have been computed from the basic characteristics of a free random walk within the elementary cell of the lattice of obstacles . the family of generalized hausdorff dimensions @xmath73 for the partition function @xmath228 ( where @xmath34 is the distance on the cayley graph which parameterizes the topological state of the trajectory ) exhibits nontrivial dependence on @xmath6 , what means that different moments of the partition function @xmath228 scale in different ways , e.g. that @xmath228 is multifractal . the main topologically probabilistic issues concerning the distribution of random walks in a rectangular lattice of obstacles have been considered in section [ sect:3 ] . in particular we have computed the average `` degree of entanglement '' of a @xmath56step random walk and the probability for a @xmath56step random walk to be closed and unentangled . results have been achieved through a renormalization group technique on a nonsymmetric cayley tree . the renormalization procedure has allowed us to overcome one major difficulty : in spite of a locally broken spherical symmetry , we have mapped our problem to a symmetric random walk on a tree of effective branching number @xmath31 depending on the lattice parameters . to validate our procedure , we have compared the return probabilities obtained via our rg approach with the exact result of p.gerl and w.woess @xcite and found a very good numerical agreement . the problem tackled in section [ sect:4 ] is closely related to the one discussed in section [ sect:2 ] . we believe that the approach developed in section [ sect:4 ] could be very important and informative as it explicitly shows that multifractality is not attached to particular properties of a statistical system ( like random walks in our case ) but deals directly with metric properties of the topological phase space . as we have already pointed out , the required conformal transforms are known only for triangular lattices , what restricts our study . however we explicitly showed that the transform @xmath159 maps the multi - punctured complex plane @xmath31 onto the so - called `` topological phase space '' , which is the complex plane @xmath155 free of topological obstacles ( all obstacles are mapped onto the real axis ) . we have connected multifractality to the multi valley structure of the properly normalized jacobian @xmath229 of the nonsymmetric conformal mapping @xmath230 . the conformal mapping obtained has deep relations with number theory , which we are going to discuss in a forthcoming publication . the `` random walk in an array of obstacles''model can be considered as a basis of a mean field like approach to the problem of entropy calculations in sets of strongly entangled fluctuating polymer chains . namely , we choose a test chain , specify its topological state and assume that the lattice of obstacles models the effect of entanglements with the surrounding chains ( the `` background '' ) . changing @xmath43 and @xmath44 one can mimic the affine deformation of the background . investigating the free energy of the test chain entangled with the deformed media is an important step towards understanding high - elasticity of polymeric rubbers @xcite . neglecting the fluctuations of the background as well as the topological constraints which the test chain produces by itself , leads to information losses about the correlations between the test chain and the background . yet , even in this simplest case we obtain nontrivial statistical results concerning the test chain topologically interacting with the background . the first attempts to go beyond the mean field approximation of rwao model and to develop a microscopic approach to statistics of mutually entangled chain like objects have been undertaken recently in @xcite . we believe that investigating multifractality of such systems is worth attention . 99 t.halsey , m.jensen , l.kadanoff , i.procaccia , b.shraiman , phys.rev . a * 33 * , 1141 ( 1986 ) h.stanley , p.meakin , nature * 335 * , 405 ( 1988 ) a.arneodo,g.grasseau,m.holschneider , phys.rev.lett . * 61 * , 2281 ( 1988 ) b.duplantier , cond - mat/9901008 i.kogan , c.mudry , a.tsvelik , phys.rev.lett * 77 * , 707 ( 1996 ) ; j.caux , i.kogan , a.tsvelik , nucl.phys . b * 466 * , 444 ( 1996 ) c.chamon , c.mudry , x.wen , phys.rev.lett * 77 * , 4196 ( 1996 ) h.castillo , c.chamon , e.fradkin , p.goldbart , c.mudry , phys.rev . b * 56 * , 10668 ( 1997 ) b.derrida , h.spohn , j.stat.phys . * 51 * , 817 ( 1988 ) ya.pesin , h.weiss , chaos , * 7 * , 89 ( 1997 ) s.nechaev , a.r.khokhlov , phys.lett.a * 112 * , 156 ( 1985 ) ; s.nechaev , a.n.semenov , m.k.koleva , physica a , * 140 * , 506 ( 1987 ) a.r.khokhlov , f.f.ternovskii , sov jetp , * 63 * , 728 ( 1986 ) ; a.r.khokhlov , f.f.ternovskii , e.a.zheligovskaya , physica a , * 163 * , 747 ( 1990 ) e.helfand , d.pearson , j.chem.phys . , * 79 * , 2054 ( 1983 ) ; e.helfand , m.rubinstein , j.chem.phys . , * 82 * , 2477 ( 1985 ) s.nechaev , sov.phys . uspekhi , * 41 * , 313 ( 1998 ) ; s.nechaev , lect . at les houches 1998 summer school `` topological aspects of low dimensional systems '' , cond - mat/9812205 h.kesten , trans.am.math.soc . , * 92 * , 336 ( 1959 ) p.gerl , w.woess , prob.theor.rel.fields , * 71 * , 341 ( 1986 ) w.woess , bull.london math.soc . , * 26 * , ( 1994 ) w.koppenfels , f.stalman _ praxis der konformen abbildung _ ( berlin : springer , 1959 ) ; h.kober , _ dictionnary of conformal representation _ : dover , 1957 ) k.ito , h.p.mckean , _ diffusion processes and their sample paths _ ( berlin : springer , 1965 ) m.abramowitz , i.stegun , _ handbook on mathematical functions _ ( nat.bur.stand . , 1964 ) r.feynman , a.r.hibbs , _ quantum mechanics and path integrals _ ( mcgraw - hill , new york 1965 ) l.brekke , p.freund , phys.rep . * 233 * , 1 ( 1993 ) j.desbois , s.nechaev , j.phys . a : math.gen . , * 31 * , 2767 ( 1998 ) ; a.vershik , s.nechaev , r.bikbov , preprint ihes / m/99/45 ; a.vershik , russ.math.surv . , to appear r.lyons , ann.prob . , * 20 * , 125 ( 1992 ) r.lyons , r.pemantle , ann.prob . , * 18 * , 931 ( 1992 ) r. holley , e.waymire , ann.app.prob . , * 2 * , 819 ( 1992 )

we prove that for the two - dimensional steady complete compressible euler system , with given uniform upcoming supersonic flows , the following three fundamental flow patterns ( special solutions ) in gas dynamics involving transonic shocks are all unique in the class of piecewise smooth functions , under appropriate conditions on the downstream subsonic flows : the normal transonic shocks in a straight duct with finite or infinite length , after fixing a point the shock - front passing through ; the oblique transonic shocks attached to an infinite wedge ; a flat mach configuration containing one supersonic shock , two transonic shocks , and a contact discontinuity , after fixing the point the four discontinuities intersect . these special solutions are constructed traditionally under the assumption that they are piecewise constant , and they have played important roles in the studies of mathematical gas dynamics . our results show that the assumption of piecewise constant can be replaced by some more weaker assumptions on the downstream subsonic flows , which are sufficient to uniquely determine these special solutions . mathematically , these are uniqueness results on solutions of free boundary problems of a quasi - linear system of elliptic - hyperbolic composite - mixed type in bounded or unbounded planar domains , without any assumptions on smallness . the proof relies on an elliptic system of pressure and the tangent of the flow angle obtained by decomposition of the euler system in lagrangian coordinates , and a newly developed method for the estimate that is independent of the free boundaries , by combining the maximum principles of elliptic equations , and careful analysis of shock polar applied on the ( maybe curved ) shock - fronts .

1004.2002

in discussing steady flows which are supersonic in the entrance section of a duct and then become subsonic , r. courant and k. o. friedrichs wrote in their classical monograph _ supersonic flow and shock waves _ @xcite that _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ @xmath7 we know in general that this assumption ( the flow is continuous ) is not tenable and that we must consider flows involving shocks . it is a question of great importance to know under what circumstances a steady flow involving shocks is uniquely determined by the boundary conditions at the entrance , and when further conditions at the exit are appropriate . " _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this paper is exactly devoted to showing that , for the two - dimensional steady complete compressible euler system , with given uniform upstream supersonic flows , under certain reasonable conditions of the downstream of the subsonic flows , in the class of piecewise @xmath0 functions , the uniqueness of the following three important flow patterns in gas dynamics : * the transonic normal shock in a two dimensional straight duct with finite or infinite length , after fixing a point the shock - front passing through ( see figure 1 ) ; + * the transonic oblique shocks attached to an infinite wedge against the uniform supersonic flow ( see figure 2 ) ; + * a flat mach configuration involving three shocks and a contact discontinuity , where one shock is supersonic , and another two are transonic , with the contact line separating two subsonic regions , after fixing a point @xmath8 where the four discontinuities intersect ( see figure 3 ) . these flow patterns have been constructed in ( * ? ? ? * , pp.332 - 333 ) by using shock polar , under the assumptions that the shock - fronts ( contact line ) are straight lines , and the subsonic flows behind the shock - fronts are also uniform i.e. , the solution is piecewise constant ( see 2 below ) . the uniqueness results we obtain in this work indicate that under certain conditions of the downstream subsonic flows , these assumptions may be relaxed to that the flow fields are only piecewise continuously differentiable then for given uniform upcoming supersonic flows , the transonic shock - fronts ( and the contact line appeared in the mach configuration ) must be straight lines and the subsonic flows behind them must be uniform . from mathematical point of view , since the steady euler system for subsonic flows is of elliptic - hyperbolic composite - mixed type , these are all results on uniqueness of solutions of free boundary value problems of a nonlinear system of mixed type , with the transonic shock - fronts and contact line being the free boundaries . ( 300 , 200)(0,0 ) ( 0,0 ) and zero attacking angle in uniform supersonic flows.,title="fig : " ] ( 133,92)@xmath9 ( 133,83)@xmath9 ( 300 , 200)(0,0 ) ( 0,0 ) is a supersonic shock lying in the quadrant @xmath10 , @xmath11 are transonic shocks lying in quadrant @xmath12 respectively , and @xmath13 is a contact line . they are all straight lines . the flow field is piecewise constant.,title="fig : " ] ( 77,65)@xmath14 ( 40,75)@xmath15 ( -10,100 ) ( 100,145)@xmath16 ( 70,155 ) ( 133,75)@xmath17 ( 150,100 ) ( 133,30)@xmath18 ( 150,0 ) ( 85,180)@xmath19 ( 235,75)@xmath20 ( 25,165)@xmath21 ( 90,-5)@xmath22 ( 190,170)@xmath23 ( 220,20)@xmath24 these special while important flow patterns have been the focus of research for many years , since the pioneering work on transonic shocks in ducts by g .- chen , m. feldman @xcite for the potential flow equation , and s. chen @xcite for the euler system . in a series of papers @xcite , g .- q . chen and m. feldman established the stability of transonic shocks in finitely long or infinitely long duct with square sections or arbitrary sections , under various conditions at the exit of the duct . z. xin and h. yin studied similar problems but employed different methods @xcite . for the euler system , h. yuan showed in @xcite various stability and instability results by using lagrangian coordinates introduced by s. chen in @xcite and characteristic decomposition technique @xcite . chen , j. chen and m. feldman also introduced the method of stream function in lagrangian coordinates by which the euler system can be written as a single second order equation to study the stability , local uniqueness and asymptotic behavior of transonic shocks in infinite ducts for the euler system . z. xin , h. yin and their collaborators also studied the problem of stability of transonic shock in nozzles for the steady euler system , see @xcite . for the study of stability of oblique transonic shocks attached to a wedge in supersonic flow , there are papers of s. chen , b. fang @xcite , e. h. kim @xcite , and h. yin , c. zhou @xcite . both the potential flow equation and the complete euler system were used as models and the authors showed stability under suitable downstream conditions . for the mach configurations , one has to use the euler system since there is a contact discontinuity . s. chen firstly studied the stability of a mach configuration under perturbations of upcoming supersonic flow , for the steady euler flow and pesudo - steady euler flow @xcite . later on he and b. fang also showed the stability of a wave pattern of regular reflection - refraction of shocks upon an interface @xcite , which is similar to a mach configuration . comparing to stability , the progress on the study of uniqueness is rather slow . for potential flows , g .- q . chen and h. yuan @xcite proved the transonic shock in a two - dimensional or three - dimensional straight duct is unique modulo a translation . l. liu showed uniqueness of subsonic potential flow in various domains @xcite , see also @xcite . by proposition 3.1 in @xcite , in two - dimensional case , many of the uniqueness results on subsonic flows also hold for the euler system , since under appropriate boundary conditions , the flow is in itself irrotational , i.e. , governed by the potential flow equation . however , to our knowledge , there is no any result on global uniqueness of transonic shocks for the complete euler system before . in this paper , we are going to establish the global uniqueness , for the complete euler system , of the three flow patterns we previously mentioned . it turns out that the method of maximum principles employed in @xcite for potential flow also works for euler flow , but the point is that one should consider now the extreme values of pressure @xmath4 and slope of the flow angle @xmath5 , with @xmath25 the velocity component of the flow along @xmath26 axis , respectively . in @xcite , h. yuan has shown that @xmath4 and @xmath27 satisfy a first order elliptic system for subsonic flow . then as in @xcite , one needs to show if there are nontrivial extremes of @xmath4 ( or @xmath27 ) on the shock - front @xmath28 , there would be contradictions by a hopf - type boundary point lemma . it is here we need to utilize some nice properties of the rankine - hugoniot conditions particularly , a rather simple curve called @xmath29 shock polar ( see figure 4 ) , which shows the relation between @xmath4 and @xmath27 behind the shock - front . we recommend @xcite for detailed computations of these relations . this paper is organized as follows . in 2 , we formulate the above mentioned physical problems ( flow patterns ) and state rigorously our results , theorem [ thm1]theorem [ thm6 ] . we also introduce briefly the @xmath29 shock polar for the comfort of those readers not familiar with this subject . then we prove in 3 the uniqueness of transonic shocks in straight ducts ( theorem [ thm1]theorem [ thm2 ] ) , and in 4 , the uniqueness of oblique transonic shocks attached to an infinite wedge ( theorem [ thm3]theorem [ thm5 ] ) . theorem [ thm6 ] , on the uniqueness of the mach configuration studied by s. chen @xcite , is established in 5 . the complete compressible euler system governing the steady motion of polytropic gas flow in two - dimensional space @xmath30 expresses the conservation of mass , momentum and total energy ( cf . @xcite ) : @xmath31 the last equation is also called as bernoulli law , with @xmath32 the bernoulli constant which is invariant on each flow trajectory , even across a shock - front . here @xmath4 , @xmath33 , @xmath34 , @xmath35 @xmath36 are the pressure , mass density , sonic speed , velocity component along @xmath37 and @xmath19 axis respectively , of the flow , and @xmath38 is the adiabatic exponent . for polytropic gas , the state function is @xmath39 , with @xmath40 the entropy , we then have @xmath41 the mach number of the flow is defined by @xmath42 . when @xmath43 ( resp . @xmath44 ) , the flow is called _ subsonic flow _ ( resp . _ supersonic flow _ ) . it is well known that the system is symmetric hyperbolic for supersonic flow , while of elliptic - hyperbolic composite - mixed type for subsonic flow . as in @xcite , we introduce @xmath5 to denote the slope of the flow angle is the angle between the velocity @xmath45 and the @xmath37 axis . hence @xmath46 in case @xmath47 , which , together with @xmath4 , plays a key role in our analysis later . let @xmath48 be a shock - front that is , a curve where the state of the gas is discontinuous and physical entropy condition holds ( cf . definition [ def101 ] in 2.2 below ) . it is well known that the following rankine - hugoniot conditions should hold across @xmath28 @xcite : @xmath49-f'(y)[\rho v]=0 , \label{rh1}\\ & & [ \rho u^2+p]-f'(y)[\rho uv]=0 , \label{rh2}\\ & & [ \rho uv]-f'(y)[\rho v^2+p]=0 . \label{rh3}\end{aligned}\ ] ] here @xmath50 $ ] denotes the jump of a quantity across @xmath28 . for example , if the pressure on the left hand side of @xmath28 ( upstream ) is @xmath51 , on the right hand side of @xmath28 ( downstream ) is @xmath4 , then we set @xmath52=p - p_0.$ ] note that @xmath32 in is constant along a flow trajectory even across the shock - front , which is the rankine - hugoniot condition corresponding to conservation of energy . ( 300 , 200)(40,0 ) ( 0,0 ) shock polar . @xmath53 represents the upcoming supersonic flow ; @xmath54 represents the strongest normal transonic shock ; @xmath27 reaches its maximum at @xmath55 , and @xmath56 represents a shock that the state behind the shock - front is exactly sonic . @xmath57 and @xmath58 are reflections of @xmath59 with respect to the @xmath4 axis . the open arc @xmath60 represents all possible transonic shocks . the open arc @xmath61 and @xmath62 represent all possible supersonic shocks . the dashed line below @xmath63 corresponds to those states that do not satisfy physical entropy condition . @xmath64 represents the weak shock ; @xmath65 represents the strong shock.,title="fig : " ] ( 200,20)@xmath14 ( 205,180)@xmath66 ( 200,155)@xmath67 ( 197,164)@xmath68 ( 202,48)@xmath69 ( 197,58)@xmath68 ( 314,130)@xmath70 ( 305,132)@xmath68 ( 75,130)@xmath71 ( 88,132)@xmath68 ( 280,90)@xmath72 ( 274.5,94.5)@xmath68 ( 106,90)@xmath73 ( 117,94.5)@xmath68 ( 350,20)@xmath74 ( 240,20)(0,1)180 ( 240,70)@xmath75 ( 240,40)@xmath76 ( 240,155)@xmath77 for a fixed point @xmath78 on the shock - front , supposing the uniform state @xmath79 of the gas ahead of the shock - front is given , in is @xmath80 we consider the four algebraic equations , with five independent variables @xmath81 , which are the state of the flow on the right hand side of @xmath28 , and @xmath82 in @xcite , it has been shown that , the flow angle @xmath83 , horizonal velocity @xmath84 , and density @xmath85 , as well as the slope of the shock - front @xmath86 , can all be expressed by @xmath87 . particularly we have the following important relation on @xmath4 and @xmath27 @xcite : @xmath88 here @xmath89 is the mach number of the flow ahead of @xmath28 , and @xmath90 figure 4 shows the curve represented by this expression in @xmath91 plane , which is traditionally called a _ @xmath29 shock polar_. it is easy to see that this curve is symmetric with respect to the line @xmath92 due to @xmath93 " in . there are several critical points on the @xmath29 shock polar . @xmath94 is the point where @xmath4 reaches its maximum ( denoted by @xmath95 ) ; @xmath96 is the point where @xmath27 reaches its maximum @xmath97 . the point @xmath98 has the special property that the state of the gas behind the shock - front is exactly sonic , i.e. , @xmath99 for all points on the shock polar with @xmath100 ( i.e. , on the open arc @xmath60 ) , the state behind the shock - front they represent are all subsonic . @xmath101 are the mirror images of @xmath102 with respect to the line @xmath103 for convenience of later reference , we also write out the following relations on the state of the flow behind the shock - front ( * ? ? ? * , p.48 ) : @xmath104 here @xmath105 is the angle between the tangent of the shock - front and the @xmath37 axis . now we formulate the physical problems we are concerned with , construct the special solutions ( flow patterns ) by using the shock polar , and then state the results on their uniqueness . let @xmath106 be a straight duct with finite length . we set @xmath107 and @xmath108 as the entry and exit of the duct , @xmath109 and @xmath110 as the lower and upper wall of the duct , respectively . on the walls , it is natural to impose the impenetrability or slip condition @xmath111 on the entry @xmath112 , we assume the flow is supersonic and moves in @xmath13 , so the euler system is hyperbolic in the positive @xmath37 direction . therefore we impose the initial condition @xmath113 , with @xmath114 all being constants and satisfying @xmath115 on the exit , one dirichlet condition on @xmath4 is given : @xmath116 it is indicated in ( * ? ? ? * , p.385 ) that this restriction on pressure is more physically interesting for nozzle flows . therefore we get a boundary value problem of in @xmath13 . we call this _ problem a. _ we are interested in those solutions of problems like this with the following structure , i.e. , transonic shock solutions : [ def101 ] suppose @xmath13 is a domain in @xmath117 for a @xmath118 function @xmath119 defined on an interval @xmath120 of @xmath121 , let @xmath122 then @xmath123 is a _ transonic shock solution _ of if it is supersonic in @xmath124 and subsonic in @xmath125 , satisfies equations in @xmath126 and the boundary conditions pointwise in the classical sense , the rankine - hugoniot jump conditions , as well as the physical entropy condition across @xmath28 : @xmath127>0.\ ] ] by and entropy condition , there holds @xmath128)$ ] and the denominator never vanishes . so if @xmath129 , we naturally have @xmath130 in definition [ def101 ] . for nozzle flows , we know the transonic shock - front will meet the walls perpendicularly . so as explained in @xcite , at the corner points where they intersect , the solution will in general be singular . however , for uniform upcoming supersonic flow and straight duct , it can be shown that the first order compatibility condition holds ( cf . lemma 3.3 in @xcite ) , that guarantees the solution @xmath131 to be @xmath0 at the corners , and hence @xmath0 up to boundary in @xmath13 . for oblique shocks attached to a wedge , in our proof we do not require the solution @xmath131 to be @xmath0 up to the vertex , where the shock - front meet the wedge ( continuous is enough , see remark [ rem41 ] ) . hence to prove uniqueness , our assumptions on regularity in definition [ def101 ] are reasonable . as a simple application of the shock polar , we show existence of special normal transonic shock solutions to problem a. for given uniform upcoming supersonic flow @xmath132 , a @xmath29 shock polar is determined ( see figure 4 ) , and for @xmath133 we have a unique @xmath134 ( i.e. , the point @xmath94 in figure 4 ) representing a subsonic state , and hence @xmath135 , and @xmath136 may be calculated by with @xmath4 there replaced by @xmath95 . suppose the shock - front passes the point @xmath137 with @xmath138 , then we may take the shock - front as @xmath139 , and the state behind it is @xmath140 which is uniform and subsonic . it is easy to see that what we get in this way is a transonic shock solution to problem a , with @xmath141 in . we denote this solution by @xmath142 . the following theorem is one of the main results of this paper . [ thm1 ] for given supersonic initial data @xmath143 on @xmath112 , supposing the flow behind the shock - front satisfies @xmath144 then problem a has a transonic shock solution ( in the sense of definition [ def101 ] ) if and only if @xmath141 in . if this holds , let @xmath145 be a transonic shock solution to problem a with shock - front passing through the point @xmath137 , then @xmath145 must be the special solution @xmath142 constructed above . this is to consider transonic shocks in an infinitely long straight duct @xmath147 . the boundary conditions are the same as in problem a , except that is dropped . replaced by @xmath148we call this boundary value problem of euler system in the unbounded domain @xmath149 as _ problem @xmath146 . _ of course @xmath142 is a solution to this problem . results like theorem [ thm1 ] also hold : [ thm2 ] for given supersonic initial data @xmath150 on @xmath112 , supposing the flow behind the shock - front satisfies @xmath151 and @xmath152 with fixed numbers @xmath153 and @xmath154 then any transonic shock solution to problem @xmath146 must be the special solution @xmath142 , provided that the shock - front passes the point @xmath155 . like theorem 1.3 in @xcite , here we do not need any asymptotic condition of the flow at infinity , under the rather strong assumptions that the flow is uniformly subsonic behind the shock - front , i.e. , @xmath156 , and moves uniformly to the exit at infinity . we also note that theorem [ thm1 ] and theorem [ thm2 ] imply particularly the instability of the special normal transonic shocks under any perturbations of the downstream pressure , as have been shown in @xcite . next we consider the transonic oblique shocks attached to an infinite wedge against a uniform supersonic flow . by symmetry with respect to the @xmath37 axis ( see figure 2 ) , we may consider only the half - wedge " ( a ramp ) with an open angle @xmath157 and zero attacking angle against the uniform supersonic flow . let @xmath158 , and set @xmath159 be the entry " where the uniform supersonic flow @xmath160 . ] flows from left to right to the ramp @xmath161 with wall @xmath162 . on the wall we have the slip condition @xmath163 we would like to find transonic shock in the flow field . we call this _ problem b. _ assuming the transonic shock solution is piecewise constant that is , the shock - front is a straight line issuing from the origin @xmath8 , which is the vertex of the ramp , and the subsonic flow behind the shock - front is also uniform , we may obtain special transonic shock solution to problem b by the shock polar . indeed , for @xmath164 ( see figure 4 ) , the line @xmath165 meets the @xmath29 shock polar at three different points @xmath166 . the point @xmath167 represents a state that does not satisfy the physical entropy condition , so we ignore it . the point @xmath168 corresponds to a weaker supersonic shock , that is , the flow behind the shock - front is also supersonic . this supersonic shock has been intensively studied ( see , for example , @xcite and references therein ) , and is not of our concern . the point @xmath169 represents a transonic shock which we are interested in , and we may obtain the slope of the shock - front , the density and velocity behind the shock - front by . we call this solution the _ strong transonic shock @xmath170 . _ it always corresponds to the point @xmath171 , p = p_{strong})$ ] on the upper part of the shock polar ( i.e. , on the closed arc @xmath172 ) . for @xmath173 both the weak shock and strong shock are transonic shocks , and for @xmath174 , they coincide . we will denote the weak transonic shock by @xmath175 ; it corresponds to the point @xmath176 on the lower part of the shock polar ( i.e. , on the open arc @xmath177 ) . for @xmath178 , detached bow shock will appear , and for @xmath179 , the weak shock is sonic . we will not consider these two cumbersome cases in this paper . for the special strong transonic shock , we have the following uniqueness result . [ thm3 ] for given supersonic initial data @xmath143 on @xmath180 , supposing the flow behind the shock - front satisfies @xmath181 and the following asymptotic condition : @xmath182 here @xmath183,$ ] then any transonic shock solution to problem b with the restriction at the vertex of the ramp @xmath184 must be the strong transonic shock solution @xmath185 under a further rather strong asymptotic condition on the shock - front at infinity , namely , @xmath186 with @xmath187 a constant less than @xmath188 we also have uniqueness of the weak transonic shock solution . [ thm4 ] for given supersonic initial data @xmath143 on @xmath180 , supposing the flow behind the shock - front satisfies @xmath144 and the asymptotic conditions , , then any transonic shock solution to problem b with the restriction at the vertex of the ramp @xmath189 must be the weak transonic shock solution @xmath190 here @xmath191 here we note that by shock polar , one can always determine that at @xmath8 either or holds . therefore our requirements in the theorems are natural . we may replace in theorem [ thm3 ] by @xmath192 to guarantee uniqueness of @xmath170 , and replace in theorem [ thm4 ] by @xmath193 as well as dropping to show uniqueness of @xmath175 . [ thm5 ] for given supersonic initial data @xmath143 on @xmath180 , supposing the flow behind the shock - front satisfies @xmath144 and the asymptotic condition resp . , then any transonic shock solution to problem b with the restriction resp . at the vertex of the ramp must be the strong ( resp . weak ) transonic shock solution @xmath170 ( resp . @xmath175 ) . we recall that , in discussing the shocks attached to a wedge , r. courant and k. o. friedrichs had written @xcite:quite aside from the question of stability , the problem of determining which of the possible shocks occurs can not be formulated and answered without taking the boundary conditions at infinity into account . @xmath7 if the pressure prescribed there is below an appropriate limit , the weak shock occurs in the corner . if , however , the pressure at the downstream end is sufficiently high , a strong shock may be needed for adjustment . under appropriate circumstances this strong shock may begin just in the corner and thus , of the two possibilities mentioned , the one giving a strong shock may actually occur . @xmath7 all statements made here are conjectures so far . while there is little doubt that they are in general correct , they should be supported , if possible , by detailed theoretical investigation . " out results above , may be considered as a part of such investigation . mach configuration is one of the fundamental flow patterns in gas dynamics which is closely connected to shock reflection - diffraction phenomena . as explained by r. courant and k. o. friedrichs in @xcite : three shocks separating three zones of different continuous states are impossible . @xmath7 configurations involving three shock fronts must therefore involve an additional discontinuity . the simplest assumption , in agreement with many observations , is that there occurs in addition to the three shock fronts a single contact discontinuity line . " such a wave structure is called a _ mach configuration _ figure 3 shows a possible mach configuration , where @xmath194 is an incident shock separating the two supersonic regions @xmath195 and @xmath196 ; @xmath197 is called the reflected shock from the contact discontinuity line @xmath13 ; @xmath198 is called a mach front . the flow in @xmath199 and @xmath200 are both subsonic . so @xmath194 is a supersonic shock and @xmath11 are transonic shocks . following terminologies in @xcite , we call @xmath201 the supersonic part of the mach configuration , and @xmath202 its subsonic part . for the case that all the discontinuities are straight lines and the flow field is piece - wise constant , the existence of such a flat " mach configuration showed in figure 3 can be proved by using shock polar . in figure 5 , the point @xmath63 represents the upstream supersonic flow in @xmath195 . @xmath203 corresponds to the flow state in @xmath196 . by the shock polar @xmath204 roots at @xmath203 , we can determine all the possible states in @xmath200 ; and by the polar roots at @xmath63 , we may determine all the possible states in @xmath199 . since @xmath27 and @xmath4 should be continuous across the contact line @xmath13 , the point @xmath205 where the two loops intersect represents the @xmath27 and @xmath4 in @xmath206 . hence by , the slope of the lines @xmath194 , @xmath198 , @xmath197 , @xmath13 and @xmath207 , as well as @xmath33 , in @xmath208 can all be uniquely determined . we recommend @xcite or @xcite for a detailed discussion . ( 300 , 200)(40,0 ) ( 0,0 ) shock polar.,title="fig : " ] ( 200,20)@xmath14 ( 205,180)@xmath66 ( 200,155)@xmath67 ( 197,164)@xmath68 ( 202,48)@xmath69 ( 197,58)@xmath68 ( 314,130)@xmath70 ( 305,132)@xmath68 ( 97,130)@xmath71 ( 88,132)@xmath68 ( 280,90)@xmath72 ( 274.5,94.5)@xmath68 ( 126,100)@xmath73 ( 122,94.5)@xmath68 ( 350,20)@xmath74 ( 132,81)@xmath209 ( 170,155)@xmath210 ( 65,155)@xmath211 it arises naturally the question of uniqueness : under what conditions , can we show the flat mach configuration constructed above is the only possible one in the class of piecewise @xmath0 functions ? we have the follows . [ thm6 ] suppose the supersonic part , i.e. , the uniform upcoming supersonic flow in @xmath195 and a straight supersonic shock - front @xmath194 , are given ( see figure 6 ) , and @xmath181 , as well as the following asymptotic condition in the subsonic region @xmath212 : @xmath213 where @xmath214 is determined in . for definiteness , we also fix the point @xmath215 where it meets the possibly curved discontinuities @xmath216 then any weak entropy solution @xmath217 of the euler system must be the flat mach configuration . here @xmath218 is the set of those functions continuous in @xmath219 and continuously differentiable in any compact subset of @xmath220 . may be replaced by @xmath221 and the same conclusion holds . ( 300 , 200)(0,0 ) ( 0,0).,title="fig : " ] ( 77,65)@xmath14 ( 40,75)@xmath15 ( 100,145)@xmath16 ( 150,90)@xmath17 ( 150,40)@xmath18 ( 85,180)@xmath19 ( 235,75)@xmath37 ( 25,165)@xmath21 ( 150,-5)@xmath22 ( 180,160)@xmath23 ( 220,100)@xmath24 ( -10,100 ) ( 70,155 ) ( 160,30 ) ( 150,100 ) for the four problems stated above , by the theory of classical solutions of hyperbolic systems @xcite , it is easy to show that the supersonic flow ahead of the transonic shock - fronts must be uniform ( cf . @xcite for a sketch ) . particularly , we may prescribe supersonic initial data on @xmath222 in problem c ( see figure 6 ) . to prove these uniqueness results , one then only needs to show the state behind the transonic shock must be uniform , and hence by shock polar , the shock - front must be straight . we will show how these hold and prove theorem [ thm1 ] , theorem [ thm2 ] in 3 , theorem [ thm3] theorem [ thm5 ] in 4 , and theorem [ thm6 ] in 5 , by using a rather unified and newly developed approach that combining various maximum principles and analysis of the shock polar . supposing @xmath223 the euler system may be written in the lagrangian coordinates @xmath224 as ( see @xcite ) @xmath225 together with the invariance of entropy along flow trajectories @xmath226 and the bernoulli law @xmath227 here @xmath228 , and @xmath229 measures the mass flux along two flow trajectories . the flow trajectory in lagrangian coordinates is the line @xmath230 . the coefficients in and are @xmath231 these may be obtained from ( 4.8)(4.19 ) in @xcite , by taking @xmath232 there . note our assumption @xmath233 in implies @xmath234 , which is required in ( 4.21 ) in @xcite . by ( 3.11 ) or ( 3.12 ) in @xcite and our assumptions in definition [ def101 ] , the transformation @xmath235 is @xmath118 and invertible , provided that @xmath236 . it is easy to check that for subsonic flows , and constitute an elliptic system , which is the elliptic part of the euler system . equations are the hyperbolic part of the euler system . the equations may also be expressed as @xmath237 with @xmath238 the gradient of a function @xmath27 , and @xmath239 the matrix @xmath240 note that @xmath241 so @xmath239 is invertible . by acting @xmath242 from left to , we get a second order elliptic equation of @xmath4 in divergence form : @xmath243 by our assumptions in definition [ def101 ] , the coefficients , @xmath244 . ] are all @xmath0 functions , and the equation is uniformly elliptic in any fixed compact domain . therefore the strong maximum principle for weak solutions of elliptic equations in divergence form is available , see theorem 8.19 in @xcite . we also need the following boundary point lemma due to r. finn and d. gilbarg ( see lemma 7 in @xcite ) . [ hopf ] let the coefficients of the system @xmath245 be hlder continuous in the closure of a simply connected domain @xmath246 with @xmath247 boundary ( @xmath248 ) , and @xmath249 is a solution to . suppose there is a @xmath250 satisfying @xmath251 ( @xmath252 ) for all @xmath253 then there holds @xmath254 where @xmath255 is the inward drawn normal of the curve @xmath256 at @xmath257 . it is clear that the inner normal @xmath255 in may be replaced by any vector @xmath258 at @xmath257 with @xmath259 since along the tangent @xmath260 of @xmath256 we have @xmath261 . in this paper we always use @xmath262 " to denote the scalar product of vectors . under the assumptions in definition [ def101 ] and theorem [ thm1 ] , we show holds . we directly have @xmath181 and @xmath43 . by , and , the entropy @xmath263 has the estimate @xmath264 with two numbers @xmath265 depending only on the upcoming supersonic flow @xmath114 and the adiabatic exponent @xmath266 hence from @xmath181 and @xmath43 we get @xmath267 , therefore @xmath268 in @xmath269 one then checks that for , @xmath270 is bounded away from zero in a compact subset of @xmath271 . so we may employ the results in 3.1 for the following analysis . the duct , in the lagrangian coordinates , is now the rectangle @xmath272 with @xmath273 ( see figure 7 ) . at the exit @xmath274 we still have the dirichlet condition for pressure : @xmath275 with @xmath276 a constant ( cf . ) . on the upper and lower walls @xmath277 there should hold @xmath133 hence @xmath278 and by , this is a neumann condition of @xmath4 : @xmath279 since now we have @xmath280 and @xmath281 ( 250 , 130)(0,0 ) ( 0,0 ) coordinates.,title="fig : " ] ( 105,120)@xmath229 ( 90,101)h ( 230,20)@xmath282 ( 150,60)@xmath283 without loss of generality , we suppose the shock - front @xmath28 , with the equation @xmath284 $ ] in lagrangian coordinates , passing through the origin @xmath8 ( i.e. , @xmath285 or @xmath286 ) . we do not need write out the rankine - hugoniot conditions in lagrangian coordinates here because the relations do not depend on coordinate system ( these rankine - hugoniot conditions are available in @xcite ) . since @xmath92 on the walls , the shock - front always meets the walls perpendicularly ( i.e. , @xmath287 , even in the @xmath224 coordinates./[p].$ ] ] let @xmath283 be the domain bounded by the shock - front , walls and exit , where the flow is subsonic . now if @xmath4 is a constant in @xmath283 , that is , @xmath288 according to , then by , @xmath27 is a constant , and by boundary conditions on walls , we have @xmath289 . so from the shock polar , we can infer that @xmath290 if @xmath291 , then we get a contradiction and either the solution is not constant or @xmath292 solution does not exist . we further may solve on @xmath28 that @xmath293 and @xmath294 by , and by the hyperbolic part , obtain that @xmath295 in @xmath283 . hence if @xmath4 is a constant in @xmath296 then @xmath141 and the solution is exactly the special solution @xmath142 . the uniqueness is therefore proved . now we show that if @xmath4 is not a constant in @xmath283 , then there will occur contradictions by using maximum principles . in the following , we always assume that @xmath4 is not a constant in @xmath283 . first , by and the strong maximum principle ( theorem 8.19 in @xcite ) , @xmath297 , @xmath298 , the maximum and minimum of @xmath4 in @xmath219 , respectively , can only be attained on the boundary @xmath299 since @xmath276 is a constant , @xmath297 and @xmath298 can not be attained simultaneously on @xmath300 by lemma [ hopf ] and , none of them can be attained on the walls . ( @xmath301 ) excluding their extreme points , hence they are open . note that , however , by our definition , @xmath302 and @xmath28 are closed . ] so at least one of them should be attained on @xmath303 . now we suppose that @xmath297 is attained on @xmath303 . by shock polar , we can see the maximum is @xmath95 ( corresponding to @xmath92 , that is the point @xmath94 in figure 4 ) and is certainly attained at @xmath304 and @xmath305 . let @xmath306 and @xmath307 be respectively the tangential and inner normal vector of @xmath28 at @xmath308 , where @xmath4 attains its maximum . by , we have @xmath309 but @xmath310 by lemma [ hopf ] we get @xmath311 note that at the corner points @xmath312 we can not employ lemma [ hopf ] directly . however , since both the corner angles are @xmath313 , we may use a reflection argument to overcome this difficulty . taking @xmath8 as an example . we first reflect @xmath314 with respect to @xmath315 and obtain its image @xmath316 since the corner angle is @xmath317 the curve @xmath318 is still @xmath319 let @xmath320 be an open disk centered at @xmath8 with a very small radius @xmath321 , we set @xmath322 . then we use even reflection to extend @xmath4 ( both the solution @xmath4 and those appeared in the coefficients through ) to @xmath323 , and by neumann condition , @xmath324 similarly we use even reflection to extend @xmath325 , odd reflection to extend @xmath326 to @xmath327 hence in @xmath323 equation still holds and the coefficients are at least lipschtiz continuous . we then use lemma [ hopf ] at @xmath328 to obtain . hence whenever @xmath4 reaches its maximum , we have @xmath329 by and . thus from @xmath330 we may infer that @xmath27 is always negative on @xmath331 . this is impossible since @xmath332 step 3 . next we consider the case that only the minimum of @xmath4 is attained on @xmath28 . if @xmath333 then by shock polar , we see that on @xmath28 there should always hold @xmath334 and @xmath133 hence @xmath335 we then get a contradiction from and : the left hand side of is zero , while the right hand side should be positive due to lemma [ hopf ] . we then consider the case that @xmath336 first we show @xmath337 . indeed , if @xmath338 , by shock polar there are only two possibilities : the points corresponding to @xmath96 or @xmath339 in figure 4 , where @xmath340 , and @xmath27 attains its maximum @xmath97(at @xmath96 ) , or minimum @xmath341 ( at @xmath339 ) . by , we still see the left hand side of is zero , while the right hand side is positive , a contradiction ! step 5 . then we conclude that @xmath342 we remark that @xmath298 should be larger than @xmath343 by definition [ def101 ] , since for those points on shock polar below @xmath344 and @xmath345 , the flow is supersonic , and sonic at @xmath346 . considering @xmath4 as a function of @xmath27 for @xmath347 and @xmath348 respectively , we see that @xmath349 now suppose @xmath4 attains its minimum @xmath298 at a point @xmath350 on @xmath28 , then at @xmath350 we have @xmath351 due to . therefore @xmath352 a contradiction to lemma [ hopf ] by , which yields @xmath353 . this finishes the proof of theorem [ thm1 ] . we may replace or , the dirichlet condition of @xmath4 , by @xmath92 on the exit @xmath300 the proof is similar to those presented below in 4.1 , and we omit the details . for infinitely long duct , we set @xmath354 as the subsonic region behind the shock - front @xmath28 : @xmath355 we first show that under the assumptions in theorem [ thm2 ] , there holds @xmath356 depending only on @xmath357 and @xmath358 . indeed , by bernoulli law , the bounds of @xmath25 and @xmath34 is @xmath359 but @xmath360 , by , @xmath33 is bounded . hence due to @xmath361 , @xmath4 is also bounded . we claim the coefficients in equation is bounded , with a bounds determined by @xmath362 and @xmath363 . to prove this , from , we have a lower bound on @xmath33 : @xmath364 then by , it is easy to check that @xmath365 and the claim is proved . finally , since @xmath366 it is straightforward to check that the equation is uniformly elliptic in @xmath354 , and holds in @xmath354 . we now state an extreme principle of bounded solutions to general linear elliptic equations in the unbounded domain @xmath354 with neumann conditions on the lateral boundaries . [ lem2 ] let @xmath367 be a linear uniformly elliptic equation with bounded coefficients @xmath368 is the set of those functions which are @xmath0 in any compact subset of @xmath369 . ] suppose @xmath370 is a bounded weak solution function @xmath371 is a weak solution to a mixed boundary value problem if it satisfies the dirichlet condition on @xmath28 in the sense of trace , the neumann condition on @xmath372 as stated in lemma [ lem3 ] below . ] to this equation that satisfies the homogeneous neumann conditions on the lateral boundaries @xmath373 and @xmath374 then either @xmath375 or @xmath376 is attained on @xmath377 by this lemma , we see at least one of the extremes , @xmath378 and @xmath379 , which are respectively the infimum and supremum of @xmath4 in @xmath369 , is attained on @xmath303 . the analysis in 3.2 then shows the uniqueness of the transonic shock solution , hence theorem [ thm2 ] is proved . in the proof of theorem [ thm2 ] . the assumption @xmath380 on regularity is enough . ] to prove lemma [ lem2 ] , we need an up to boundary harnack inequality for elliptic equations with neumann boundary conditions : [ lem3 ] let @xmath381 be a domain with part of its boundary @xmath382 lying in a plane , and @xmath383 a linear equation with coefficients @xmath384 satisfying @xmath385 for two fixed positive constants @xmath386 suppose @xmath387 is a continuous nonnegative weak solution satisfying the homogeneous neumann condition on @xmath388 , that is , for any @xmath389 whose support @xmath390 is disjoint with @xmath391 , there holds @xmath392 then for any compact subset @xmath393 of @xmath394 , there holds @xmath395 with the positive constant @xmath63 relying only on @xmath396 let @xmath397 be the open ball centered at @xmath398 with radius @xmath399 for @xmath400 is the classical interior harnack inequality ( theorem 8.20 in @xcite ) . we then consider the case that @xmath401 without loss of generality , we may suppose @xmath388 lying in the hyperplane @xmath402 by even reflections of @xmath384(@xmath403 or @xmath404 ) and @xmath371 with respect to @xmath405 , and odd reflections of @xmath384(@xmath406 or @xmath407 but @xmath408 ) with respect to @xmath405 , they are all extended from @xmath409 to the whole ball @xmath410 for a small @xmath321 . we also get an elliptic equation satisfying in @xmath410 . it is straightforward to check that @xmath371 is still a nonnegative weak solution ( in the sense of ( 8.2 ) in @xcite ) to this equation . hence by the interior harnack inequality we also have @xmath411 and follows with @xmath412 for @xmath393 any compact subset of @xmath394 , we may cover it by finite balls ( or half balls with center on @xmath388 ) , and holds by an argument as that in the proof of theorem 2.5 in @xcite . _ proof of lemma [ lem2 ] . _ if @xmath371 is a constant , then obviously lemma [ lem2 ] holds . in the following we always assume that @xmath371 is not a constant . we prove by contradiction , that is , we assume that @xmath413 the argument below is similar to that of proof of theorem 1.2 in @xcite . set @xmath414 and @xmath415 by , and applying lemma [ hopf ] , as well as the strong maximum principle in @xmath416 , we infer that there is an increasing sequence @xmath417 tends to infinity , and there are points @xmath418 such that @xmath419 as @xmath420 we may choose @xmath421 rather large . then define @xmath422 and @xmath423 by translation along @xmath282 : @xmath424 we may translate @xmath425 to @xmath426 , and @xmath427 to @xmath428 and the equation @xmath429 in @xmath425 are transformed to @xmath430 in @xmath431 , with @xmath432 now let @xmath433 which is positive and also a weak solution of @xmath434 in @xmath431 . applying lemma [ lem3 ] to @xmath435 , by the strong maximum principle , we have @xmath436 it is crucial to note that by our assumptions in lemma [ lem2 ] , the constant @xmath63 here actually does not depend on @xmath437 . then taking @xmath438 we have @xmath439 a contradiction desired . this finishes the proof of lemma [ lem2 ] . now we consider the oblique transonic shocks attached to an infinite wedge against uniform supersonic flow ( see figure 2 ) . ( 400 , 180)(0,0 ) ( 0,0 ) ( 285,15)@xmath282 ( 85,100)@xmath229 ( 220,70)@xmath283 in lagrangian coordinates , the wall of the wedge is @xmath440 see figure 8 . on @xmath441 the slip condition is still @xmath442 as in 3.2 , let the equation of the shock - front @xmath28 be @xmath443 with @xmath444 and @xmath445 and @xmath446 . condition is now @xmath447 which holds at the origin @xmath8 , and is replaced by @xmath448 by the latter requirement and shock polar , we infer that on @xmath28 , there holds @xmath449 we derive a second order elliptic equation of @xmath27 . indeed , by , we have @xmath450 then @xmath27 satisfies @xmath451 and is now @xmath452 one may check that lemma [ hopf ] holds for , and the strong maximum principle holds for @xmath453 if @xmath27 is a constant , i.e. , @xmath454 in @xmath283 , then by and shock polar , we infer that there should hold @xmath455 on @xmath28 and @xmath28 is the special strong transonic shock - front by . from and , @xmath371 and @xmath33 can also be calculated on @xmath28 . then using the hyperbolic part , we show the solution must be @xmath170 , and theorem [ thm3 ] is proved . in the following we show there are contradictions if @xmath27 is not a constant . suppose now that @xmath27 is not identical to @xmath456 in @xmath283 . then by lemma [ hopf ] and the strong maximum principle , either the supremum of @xmath27 in @xmath219 , @xmath457 , or the infimum of @xmath27 in @xmath458 @xmath459 , must be attained on @xmath28 . ( by and , @xmath457 is indeed the maximum of @xmath27 , while @xmath459 is the minimum of @xmath27 . ) we consider firstly the case that @xmath460 step 1 . suppose @xmath457 is attained at a point @xmath350 on @xmath28 . if @xmath461 , then @xmath462 note that by shock polar , we may regard @xmath27 as a function of @xmath4 , @xmath463 , which shares @xmath464 for @xmath465 now considering the directional derivative of @xmath27 along @xmath28 at @xmath350 , @xmath466 , which vanishes , we have @xmath467 hence @xmath468 however , by and lemma [ hopf ] , one gets @xmath469 , a contradiction ! secondly consider the case @xmath470 whence @xmath471 . then by and lemma [ hopf ] , we have @xmath472 this means @xmath4 is strictly increasing once it attains @xmath473 . however , let @xmath350 be the first point on @xmath28 where @xmath27 attains its maximum as @xmath229 varies from @xmath474 to @xmath475 . note that @xmath476 , so @xmath4 decreases near the left side of @xmath350 , we should have @xmath477 a contradiction to the assumption of @xmath478 . we then conclude that @xmath479 which is attained at @xmath480 so @xmath481 and definitely @xmath459 is attained on @xmath482 . if @xmath483 and @xmath484 , then as in step 1 above we can show a contradiction since @xmath485 . now for @xmath486 as that of , we may get @xmath487 therefore , whenever @xmath27 attains its minimum @xmath341 , @xmath4 is strictly decreasing ( hence less than @xmath473 ) and @xmath27 should run into @xmath488 ( the arc @xmath489 of shock polar ) , this contradicts to our asymptotic condition : @xmath490 step 5 . for the case @xmath491 where @xmath4 attained its maximum on @xmath482 , we get @xmath492 , a contradiction . secondly , for the case that @xmath493 , we have @xmath494 by and . this fact , together with , and strong maximum principle , implies @xmath495 should be attained on @xmath496 we then repeat step 3 step 5 above to reach contradictions . this finishes proof of theorem [ thm3 ] . next we consider the uniqueness of weak transonic shock . now at the vertex of the ramp @xmath8 , we have @xmath497 by . the condition at infinity is plus @xmath498 with @xmath187 a constant less than @xmath499 as in 4.1 , to prove uniqueness , we only need to show that there are contradictions if @xmath27 is not a constant , and then either its supremum @xmath457 or infimum @xmath459 is attained on @xmath28 . step 1 . now consider the case @xmath457 is attained on @xmath28 . if @xmath500 , the proof is the same as in step 1 in 4.1 . for @xmath501 , by , once @xmath27 reaches its maximum , @xmath4 should increase ( larger than @xmath473 ) . this would be a contradiction to our assumption and . step 2 . for the case that @xmath502 is attained on @xmath28 , if @xmath459 lies on the arc @xmath503 , by continuity , @xmath27 should first attain its maximum @xmath97 ( note now @xmath27 runs counterclockwise on the shock polar from @xmath504 to @xmath339 on the arc @xmath503 ) . there is a contradiction as we showed above in step 1 ; if @xmath459 lies on the arc @xmath505 , just repeat step 1 in 4.1 for a contradiction . theorem [ thm4 ] is also proved . finally , we show theorem [ thm5 ] holds . the arguments are very similar to that in 3.2 , so we just outline the main ideas . by and , one gets an oblique derivative condition of @xmath4 on @xmath239 : @xmath506 the asymptotic conditions are respectively @xmath507 to show uniqueness , as before , it is sufficient to show contradictions if @xmath4 is not constant . in the following we always assume @xmath4 is not identical to @xmath508 ( resp . @xmath509 ) . then by and strong maximum principles applied to equation , either @xmath510 or @xmath511 is attained on @xmath377 step 1 . consider the case that @xmath512 is attained at a point @xmath513 . if @xmath514 , then as of step 2 in 3.2 , @xmath27 should be negative on @xmath28 for @xmath515 . note that implies particularly @xmath516 so we get @xmath517 hence @xmath518$]>0 $ ] and @xmath181 are both bounded away from zero on the shock - front which can be checked by using shock polar . ] has a negative upper bound for large @xmath229 , which means @xmath28 will eventually intersect with the boundary @xmath519 where hyperbolic initial data is prescribed , that is not allowed . if @xmath520 , then @xmath379 lies on the arc @xmath172 of @xmath29 shock polar , and by there is a contradiction . since @xmath4 is not constant , we infer @xmath522 and is attained on @xmath482 . just repeat step 4step 5 in 3.2 for contradictions . this finishes the proof of theorem [ thm5 ] for the case of strong transonic shocks . note that the arguments here also work for the case @xmath523 consider as well the case that @xmath524 is attained at a point @xmath513 . if @xmath514 , then the argument is the same as in step 1 in 4.3.1 . if @xmath525 , refer step 4 in 3.2 for a contradiction . if @xmath526 refer step 5 there . hence we conclude that @xmath527 and @xmath528 is attained on @xmath496 therefore @xmath529 and refer also step 5 in 3.2 for a contradiction . this finishes proof of theorem [ thm5 ] for the case of weak transonic shock . [ rem41 ] in the above proofs we do not need to take care of the corner points @xmath8 , so we may only require that @xmath530 is the set of those functions that are continuous in any compact subset of @xmath219 . ] since the contact line @xmath13 is also a trajectory of the flow , by virtue of lagrangian coordinates , it may be transformed to the @xmath282 axis , see figure 9 . in the following we use superscript @xmath531 , such as @xmath532 to denote quantities in @xmath533 , with @xmath534 . @xmath13 is a contact line means @xmath535 that is , @xmath27 and @xmath4 are continuous in @xmath536 , although others , such as , @xmath537 , are not . the supersonic initial data is prescribed on the @xmath229 axis . ( 300 , 200)(0,0 ) ( 0,0).,title="fig : " ] ( 77,65)@xmath14 ( 40,75)@xmath15 ( 90,145)@xmath16 ( 150,120)@xmath17 ( 150,40)@xmath18 ( 85,180)@xmath229 ( 235,75)@xmath282 ( 25,165)@xmath21 ( 150,0)@xmath22 ( 160,180)@xmath23 ( 190,80)@xmath24 ( -10,100 ) ( 50,160 ) ( 160,30 ) ( 170,120 ) now we begin to prove theorem [ thm6 ] . it is divided into several steps . we consider only the case of the condition . the case of may be proved similarly by adopting arguments in 4.1 and then combining the following reasoning , so we omit it . if both @xmath544 and @xmath545 are constants in @xmath199 , @xmath200 respectively , then we see the subsonic flow in @xmath206 are uniform , and by shock polar , the transonic shock - fronts are straight lines , and then according to analysis in 2.2.4 , the mach configuration must be the flat one constructed there . theorem [ thm6 ] is then proved . now we show there will be contradictions if either @xmath544 or @xmath545 is not constant . there are two cases : @xmath546 one is constant while the other is not ; @xmath547 both are not constant . let @xmath552 and @xmath553 be respectively the supremum and infimum of @xmath545 in @xmath554 if @xmath555 or @xmath556 , by the strong maximum principle , they can only be attained on the shock - front @xmath557 . as shown in 4.3 , there will be contradictions . we firstly show @xmath561 implies contradictions . suppose now @xmath562 then by the strong maximum principle , the extreme @xmath558 can only be attained on the shock - front @xmath563 or on the contact line @xmath564 as shown in 4.3 , there will be contradictions if this happens on @xmath198 . so it is only possible to take place on @xmath564 by , we infer @xmath565 as before , @xmath552 also can only be attained on @xmath564 hence we get @xmath566 and they are attained at the same point @xmath257 ( @xmath567 ) on the contact line . therefore , by lemma [ hopf ] , we have @xmath568 now let us look at the equations . we may solve ( cf . ) @xmath569 then by and , we may get , at @xmath257 , @xmath570 on the other hand , note that we have @xmath571 by , therefore at @xmath257 , @xmath572 we then get @xmath573 however , by , we should have @xmath574 a contradiction ! we have now @xmath575 since @xmath544 is not constant , one has @xmath576 arguments as in step 4.1 show that we must have @xmath577 and they are attained at the same point @xmath567 on the contact line , and a contradiction similar to then follows . b. fang is supported in part by nnsf of china under grant no . l. liu is supported in part by nnsf of china under grant no . h. yuan is supported in part by nnsf of china under grants no . 10901052 , no . 10871071 , and chenguang program ( 09cg20 ) sponsored by shanghai municipal education commission and shanghai educational development foundation .

we present an efficient method to generate large simulations of the epoch of reionization ( eor ) without the need for a full 3-dimensional radiative transfer code . large dark - matter - only simulations are post - processed to produce maps of the redshifted 21 cm emission from neutral hydrogen . dark matter haloes are embedded with sources of radiation whose properties are either based on semi - analytical prescriptions or derived from hydrodynamical simulations . these sources could either be stars or power - law sources with varying spectral indices . assuming spherical symmetry , ionized bubbles are created around these sources , whose radial ionized fraction and temperature profiles are derived from a catalogue of 1-d radiative transfer experiments . in case of overlap of these spheres , photons are conserved by redistributing them around the connected ionized regions corresponding to the spheres . the efficiency with which these maps are created allows us to span the large parameter space typically encountered in reionization simulations . we compare our results with other , more accurate , 3-d radiative transfer simulations and find excellent agreement for the redshifts and the spatial scales of interest to upcoming 21 cm experiments . we generate a contiguous observational cube spanning redshift 6 to 12 and use these simulations to study the differences in the reionization histories between stars and quasars . finally , the signal is convolved with the lofar beam response and its effects are analyzed and quantified . statistics performed on this mock data set shed light on possible observational strategies for lofar . [ firstpage ] quasars : general cosmology : theory observation diffuse radiation radio lines : general .

0809.1326

the history of our universe is largely unknown between the surface of last scattering ( @xmath0 ) down to a redshift of about 6 . because of the dearth of radiating sources and the fact that we know very little about this epoch , it is often referred to as the `` dark ages '' . theoretical models suggest that around redshifts 10 20 , the first sources of radiation appeared that subsequently reionized the universe . two different experiments provide the bounds for this epoch of reionization ( eor ) ; the high polarization component at large spatial scales of the temperature - electric field ( te ) cross - polarization mode of the cosmic microwave background ( cmb ) providing the upper limit for the redshift at @xmath1 @xcite and the rapid increase in the lyman-@xmath2 optical depth towards redshift 6 , observed in the spectrum of high redshift quasars @xcite , the lower limit . although the redshifted 21 cm hyperfine transition of hydrogen was proposed as a probe to study this epoch decades ago @xcite , the technological challenges to make these observations possible are only now being realised . in the meantime , theoretical understanding of the eor has improved greatly @xcite . over the past few years there have been considerable efforts in simulating the 21 cm signal from the epoch of reionization . almost all of the methods employed in simulating the 21 cm involve computer intensive full 3-d radiative transfer calculations @xcite . theories predict that the process of reionization is complex and sensitively dependent on many not - so - well - known parameters . although stars may be the most favoured of reionization sources , the role of mini - quasars ( miniqsos ) , with the central black hole mass less than a few million solar masses , are debated @xcite . even if the nature of the sources of radiation would be relatively well constrained , there are a number of `` tunable '' parameters like the photon escape fraction , masses of these first sources , and so on , that are not well constrained . in a couple of years , next generation radio telescopes like lofar and mwa will be tuned to detect the 21 cm radiation from the eor . although the designs of these telescopes are unprecedented , the prospects for successfully detecting and mapping neutral hydrogen at the eor critically depends on our understanding of the behaviour and response of the instrument , the effect of diffuse polarized galactic & extra - galactic emission , point source contamination , ionospheric scintillations , radio frequency interference ( rfi ) and , not least , the characteristics of the desired signal . a good knowhow of the above phenomena would enable us to develop advanced signal processing / extraction algorithms , that can be efficiently and reliably implemented to extract the signal . in order to test and confirm the stability and reliability of these algorithms , it is imperative that we simulate , along with all the effects mentioned above , a large range of reionization scenarios . [ fig : bigpic ] shows the basic building blocks of the simulation pipeline being built for the lofar - eor experiment . this paper basically constitutes the first block , i.e. , simulation of the cosmological 21 cm eor signal . this then passes through a sequence of blocks like the foreground simulation @xcite , instrument response and extraction ( lambropoulous et al . , _ in prep _ ) . the extracted signal is then compared with the original signal to quantify the performance of the extraction scheme . this process needs to be repeated for various reionization scenarios to avoid any bias the extraction scheme would have if only a subsample of all possible signal characteristics were used . simulating observing windows as large as the field of view ( fov ) of lofar ( @xmath3 ) and for frequencies corresponding to redshift 6 to 12 is a daunting task for conventional radiative transfer codes because of multiple reasons such as a requirement for high dynamic range in mass for the sources of reionization , their large number towards the end of reionization and the size of the box which strains the memory of even the largest computer cluster . in order to facilitate the simulation of such large mock data sets for diverse reionization scenarios , we need to implement an approximation to these radiative transfer methods that mimic the `` standard '' simulations to good accuracy . it was clear from the onset that the details of the ionization fronts like its complex non - spherical nature will not be reproduced by the semi - analytical approach that we propose here . but the argument towards overlooking this discrepancy is that when the outputs of our semi - analytical approach and that of a radiative transfer code are passed through the machinery of the lofar - eor pipeline , they are experimentally indistinguishable . the reason being the filtering nature of the telescope s point spread function ( psf ) across the sky and the substantial bandwidth averaging along the frequency / redshift direction that is needed to recover the signal , smoothes out the structural details captured by codes . recently , several authors @xcite have proposed schemes to reduce the computational burden of generating relatively accurate 21 cm maps . these methods do fairly well , although there are some caveats , like for example the intergalactic medium ( igm ) ionization being treated as binary , i.e. , the igm is either ionized or neutral @xcite . although this might be the case for stellar - like sources , others with a power - law component could exhibit an effect on the igm wherein the ionizing front is extended and hence this assumption need not hold @xcite . added to this , the schemes presented in order to compute the 21 cm maps , make the assumption that the spin temperature of hydrogen , @xmath4 is much larger than the cmb temperature . towards the end of reionization ( @xmath5 ) this might very well be valid . but the dawn of reionization would see a complex spatial correlation of igm temperatures with the sources of radiation , its clustering and spectral energy distributions @xcite . in the current paper we have assumed that the spin temperature is coupled to the kinetic temperature and that they are much higher than the cmb temperature . at higher redshifts this need not be a valid assumption . the effects of heating by different types of radiative sources on the igm and the coupling ( both ly-@xmath2 and collisional ) between the spin and kinetic temperature will be simulated accurately in a follow - up paper _ thomas et al . , in prep _ , using the same scheme , but now applied to heating . in this paper we propose a method of post processing numerical simulations in order to rapidly generate realistic 21 cm maps . briefly , the algorithm consists of simulating the ionization fronts created by the `` first '' radiative sources for a range of parameters which include the power spectrum , source mass function and clustering . we then identify haloes in the outputs of n - body simulations and convert them to a photon count using semi - analytical prescriptions , or using the photon count derived from a smoothed particle hydrodynamics ( sph ) simulation . depending on the photon count and the spectrum , we embed a sphere around the centre - of - mass ( com ) of the halo whose radial profile matches that of a profile from the table created by the 1-d radiative transfer code of @xcite . appropriate operations are carried out to conserve photon number . since , the basic idea is to expand bubbles around locations of the sources of radiation , we call this method bears ( bubble expansion around radiative sources ) . for the sake of brevity and comparison with full radiative transfer codes , we restrict ourselves to monochromatic radiative transfer with a fixed temperature . a following paper will include a full spectrum along with the temperature evolution . the results of our semi - analytic scheme will be compared to those obtained with the full monte carlo radiative transfer code crash @xcite . in [ sec : simulations ] we describe the various steps involved in implementing the bears algorithm on the outputs of n - body simulations . we describe the specifications of the n - body simulations , the 1-d radiative transfer code used to produce the catalogue of ionization profiles , the algorithm employed to embed the sources and finally an illustrative example of the procedure to correct for the overlap of ionized bubbles . in [ sec : comparison ] the fully cosmological radiative transfer code crash is summarized and the results of its qualitative and statistical comparison with bears are discussed . [ sec : datacube ] describes the method of generating the cube with maps of the brightness temperature ( @xmath6 ) at all the frequencies that will be observed by an eor experiment . in [ sec : starsvquasars ] we use the simulation of the cube , with maps of the sky at different frequencies , to study the difference between two popular sources of reionization , i.e. , stars and quasars . these maps in the cube are then filtered through the lofar antenna response to output the final data cube in [ sec : instrumental ] . finally in [ sec : conclusions ] we summarize our results and outline further improvements that need to be made in our approach in order to start exploring the large parameter space involved in reionization studies . in this section the various components of the simulation that lead towards 21 cm brightness temperature ( @xmath7 ) maps are explained . the very first step of the entire process consists of generating a catalogue of 1-d ionization profiles for different masses / luminosities of the source , their spectra and the density profiles that surround them at different redshifts . although we assume the density around each source to be constant , we do vary its value as detailed below . given the locations of the centres of mass of haloes , the photon rate emanating from that region is calculated based on a semi - analytical description ( discussed in section [ sec : starsvquasars ] ) . given the spectrum , luminosity and the overdensity around the source , a spherical bubble is embedded around that pixel whose radial profile is selected from the table . the justification for this simple - minded approach is that the universe is relatively homogeneous in density at these high redshifts on the scales probed by upcoming eor experiments . therefore , we can assume spherical symmetry in our construction of the ionized regions . in the following subsections we summarize the n - body simulations employed and the 1-d radiative transfer code that was used to generate the catalogue , and we describe in detail the algorithm used to embed the spherical bubbles and the method used to conserve photons in case of overlap . we used a modified version of the n - body / treepm / sph code gadget-2(@xcite ) to perform a dark matter ( dm ) cosmological simulation containing @xmath8 particles in a box of size @xmath9 and a dm+sph cosmological simulation containing @xmath10 dm and @xmath10 gas particles in a box of size @xmath11 . the dm particle masses were @xmath12 and @xmath13 , respectively . initial particle positions and velocities were obtained from glass - like initial conditions using cmbfast ( version 4.1 ; @xcite ) and employing the zeldovich approximation to linearly evolve the particles down to redshift z = 127 . we assumed a flat @xmath14cdm universe and employed the set of cosmological parameters @xmath15 , @xmath16 , @xmath17 , @xmath18 , @xmath19 and @xmath20 , in agreement with the wmap 3-year observations ( @xcite ) . data was generated at @xmath21 equally spaced redshifts between @xmath22 and @xmath23 . halos were identified using the friends - of - friends algorithm ( @xcite ) , with linking length @xmath24 . the gas in the dm+sph simulation is of primordial composition , with a hydrogen mass fraction @xmath25 and a helium mass fraction @xmath26 . radiative cooling and heating are included assuming ionisation equilibrium , using tables generated with the publicly available package cloudy ( version 05.07 of the code last described by @xcite ) . the gas is allowed to cool by collisional ionisation and excitation , emission of free - free and recombination radiation and compton cooling off the cosmic microwave background . molecular cooling ( by hydrogen and deuterium ) is prevented by the inclusion of a soft ( i.e. cut off at the lyman - limit ) uv background . we employed the star formation recipe of @xcite , using a @xcite initial mass function ( imf ) with mass range @xmath27{~\mbox{m}_{\odot}}$ ] . for further analysis , sph particle masses were assigned to uniform meshes of size @xmath28 and @xmath10 cells using tsc ( @xcite ) and the gas densities were calculated . the density field was smoothed on the mesh with a gaussian kernel with standard deviation @xmath29 comoving @xmath30 . each cell was further assigned a hydrogen - ionizing luminosity according to the stellar mass it contained . star particles were treated as simple stellar populations and their luminosity was calculated with the population synthesis code of @xcite . we used stellar masses and ages as determined by the simulation , a chabrier imf consistent with the star formation recipe and assumed a fiducial metal mass fraction @xmath31 . the catalogue of ionization profiles for different redshifts ( which also translates to different densities at a given redshift ) , spectral energy distributions ( sed ) , times of evolution and masses / luminosities , was created using the 1-d radiative transfer code developed by @xcite . following @xcite , a set of rate equations are solved at every cell as a function of time . the equations follow the time - evolution of h@xmath32 , h@xmath33 , he@xmath32 , he@xmath33 , he@xmath34 and temperature at every grid cell . the ionization rates are integrals over the spectrum and the cross sections of the various species . these were pre - computed and stored in a table to facilitate faster execution . case - b recombination coefficients are used to calculate the rate at which hydrogen recombines . the 1-d radiative transfer code starts the simulation at @xmath35 , typically @xmath36 physical @xmath37 from the location of the source . all hydrogen and helium is assumed to be completely ionized inside this radius , @xmath38 . each cell is then updated for time @xmath39 . this @xmath39 is not the intrinsic time - step used to solve the differential equation itself because that is adaptive in nature and varies according to the tolerance limit set in the ordinary differential equation ( ode ) solver . on the other hand , the @xmath39 here decides for how long a particular cell should evolve before moving on to the next . the code is causal in the sense that cell @xmath40 + 1 is updated after cell @xmath40 . the light travel time is not taken into consideration explicitly since the ionization front ( i - front ) is typically very subluminal . hence , all cells are updated to time @xmath41 at the @xmath42 time - step . after all cells except the last cell @xmath43 have been updated to time @xmath39 , the resulting values are stored and then passed on as initial conditions for the evolution of the cell in the next interval of time . the update of all @xmath44 cells is repeated @xmath45 times such that @xmath46 , where @xmath47 is the life time of the radiating source . the various quantities of interest can be stored in a file at intervals of choice . a radial coverage of @xmath48 is chosen _ a priori _ which , depending on the problem , can be set to any value . typically we do not need to solve the radiative transfer beyond ten comoving mega - parsecs . we use an equally spaced grid in radius with a resolution of @xmath49 which , like the time resolution , is decreased to half its value until it meets a given convergence criterion , which here is that the final position of the i - front converges to within @xmath50 . the implementation of the code is modular . hence it is straightforward to include different spectra corresponding to different ionizing sources . our 1-d code can handle x - ray photons and the secondary ionization and heating it causes and therefore performs well for both high ( quasars ) and low energy ( stars ) photons . further details of the code can be found in @xcite . in the following section we discuss the algorithm employed to expand reionization bubbles around the locations of radiative sources . the numerical simulation provides us with the density field at every grid point , the centres of mass of the haloes identified by the fof algorithm , the velocities of these particles and , if the simulation also contains gas , the ionizing luminosity associated with the halo . equipped with this information about the simulation box at a given redshift , we follow the steps enumerated below to embed the ionized bubbles around the source locations : 1 . given the redshift , ionizing luminosity and the time for which the ionization front should evolve ( this depends on the type of source ) , select a corresponding file from the catalogue of ionization profiles generated earlier . 2 . the sources are usually in an overdense region and the density around the source follows a profile . since the profiles vary from source to source we use the following approximation : we calculate the overdensity around the source for a radius @xmath51 ( where the subscript `` od '' stands for overdensity ) . we then assume that the source is embedded in a uniform density whose value is the average overdensity within the radius @xmath52 . this naturally translates in selecting the same ionization profile but now from the table at a higher redshift . this radius is estimated as described in [ sec : avgrad ] . 3 . at lower redshifts there is considerable overlap between bubbles . thus , in order to conserve the number of photons , we need to redistribute the photons that ionize the overlapped regions to other regions which are still neutral . the details of this correction process are given below ( see [ sec : corr_overlap ] ) . 4 . when computing the reionization history , ionized regions are mapped one - to - one from the current simulation snapshot onto the next . note that ionization due to recombination radiation has not been included . the radius @xmath51 is calculated as follows : * for any given source ( we choose the largest ) in the box we perform the radiative transfer and compute the ionization profile using the ( exact ) spherically averaged radial density profile of the surrounding gas ; * the density within a radius @xmath51 , which we initially choose to be very small , is spherically averaged around the source and the ionization profile corresponding to this mean density is selected from the catalogue . this step is repeated for increasing values of @xmath51 until the extent of the ionization profile matches that of the `` exact '' radiative transfer calculation done in the previous step . we will refer to the resulting radius as the calibrated @xmath53 ; * @xmath51 for all the other sources are calculated by scaling them with the luminosity of the source according to @xmath54 , where @xmath55 and @xmath56 are the luminosities of the source under consideration and the calibration source respectively . a comment to be made here is that because the dynamic range in the halo masses derived from the simulation is not very high the value of @xmath51 does not vary considerably between sources . at lower redshifts the sources become more massive and more numerous . this causes considerable overlap between spheres of close - by sources . the regions of overlap correspond to some number of unused photons ( depending on the density ) . the relative simplicity of the implementation of the algorithm arises due to the fact that the ionization fronts created by most sources are `` sharp '' , at least for the resolutions we are able to simulate . if the ionization front is extended , we can modify the procedure below to accommodate it . the procedure for correction is illustrated with the help of a simple example shown in fig . [ fig : incidence ] : * create an analogue to an incidence / admittance matrix * a * ( ref equation [ eq : incidence ] for the connection between ionized regions of sources ) : the matrix * a * is a square matrix of size @xmath57 , whose elements are set to 1 if two sources overlap and 0 otherwise . here @xmath58 is the total number of sources in the box . + @xmath59 * segment the simulation box into regions that are connected . this is done by identifying rows that are identical in the matrix * a*. thus , for the example shown in fig . [ fig : incidence ] there are three different `` connected regions '' , reg1:(1,6,5 ) , reg2:(2,3 ) and reg3:(4 ) . * calculate the total volume of the overlap zones in each of the connected regions . for example for reg1 , we have overlap(1,6)+overlap(5,6 ) . * the sizes of all the bubbles in a particular connected region is increased so as to entail a volume that corresponds to the average overlapped region , i.e. , total overlapped region divided by the number of bubbles in that connected region . * the sizes of the bubbles are iteratively increased until the volume of the regions that are `` newly ionized '' , i.e. , excluding the regions ionized by the sources before the correction , equals the total overlapped volume . this ensures a homogeneous redistribution of unused photons all around the region . the unshaded bubbles in reg1:(1,6,5 ) and reg2:(2,3 ) in fig . [ fig : incidence ] depict the expansion of the bubbles . the bears algorithm is used to generate a series of cubes of ionized fractions at different redshifts and resolutions for the case in which the radiative sources are stars . in this section these cubes are compared with those obtained from the full radiative transfer code crash using the same source list i.e. , source locations , luminosities and spectra . in all comparisons made below we make use of ionizing luminosities derived from the n - body / sph simulation ( refer [ sec : nbody ] ) . we have assumed the spectra to be monochromatic at 13.6 @xmath60 . we first summarize the essentials of the radiative transfer code crash and then discuss a set of visual and statistical comparisons between the results generated using these two approaches . all comparisons except for the case of @xmath11 , @xmath61 box , in this section is done on boxes where we have not included the reionization history . one of the major reasons for this is that a full 3-d radiative transfer code like crash would require enormous computational resources to achieve this task . and on the other hand , it is easier to judge the performance of bears when there a large number of sources overlapping at lower redshifts , whereas if the reionization history is included , the lower redshifts would be completely ionized and most of the interesting characteristics of the comparison would be wiped out . crash is a ray - tracing radiative transfer code based on monte carlo ( mc ) techniques that are used to sample the probability distribution functions ( pdfs ) of several quantities involved in the calculation , e.g. spectrum of the sources , emission direction and optical depth . the mc approach and the code architecture enables applicability over a wide range of astrophysical problems and allows for additional physics to be incorporated with minimum effort . the propagation of the ionizing radiation can be followed through any given h / he static density field sampled on a uniform mesh . at every grid point and time step the algorithm computes the variations in temperature and ionization state of the gas . the code allows for the possibility of the addition of multiple point sources at any specified point in the box , and also diffuse radiation ( e.g. the ultraviolet background or the radiation produced by h / he recombinations ) can be self - consistently incorporated . the energy emitted by point sources in ionizing radiation is discretized into photon packets , beams of ionizing photons , emitted at regularly spaced time intervals . more specifically , the total energy radiated by a single source of luminosity @xmath62 , during the total simulation time , @xmath63 , is @xmath64 . for each source , @xmath65 is distributed in @xmath66 photon packets , emitted at the source location at regularly spaced time intervals , @xmath67 . the time resolution of a given run is thus fixed by @xmath66 and the time evolution is marked by the packets emission : the j - th packet is emitted at time @xmath68 , with @xmath69 . thus , the total number of emissions of continuum photon packets is @xmath70 . the emission direction of each photon packet is assigned by mc sampling the angular pdf characteristic of the source . the propagation of the packet through the given density field is then followed and the impact of radiation - matter interaction on the gas properties is computed on the fly . each time the packet pierces a cell i , the cell optical depth for ionizing continuum radiation,@xmath71 , is estimated summing up the contribution of the different absorbers ( @xmath72 ) . the probability for a single photon to be absorbed in the @xmath40th cell is : @xmath73 the trajectory of the packet is followed until its photon content is extinguished or , if open boundary conditions are assumed , until it exits the simulation volume . the time evolution of the gas physical properties ( ionization fractions and temperature ) is computed solving in each cell the appropriate discretized differential equations each time the cell is crossed by a packet.the reader is referred to @xcite for details of the code . figure [ fig : comparered6 ] shows a view of the ionized fraction of hydrogen calculated using crash ( left panel ) and bears ( right panel ) , with isosurfaces shaded dark . the box shown here is a @xmath10-simulation at a redshift of six for a box with a comoving length of @xmath74 on each side . globally the two boxes do look very similar . statistics on these cases and others are presented later . also bear in mind that this case , i.e. , redshift six , is the one for which we should expect maximum discrepancy between the two methods . reasons being : one , the universe is much less homogeneous at redshift six than at higher redshifts which is contrary to the basic assumption in bears that the igm is predominantly uniform , two , the bubbles from these ionizing sources become larger ( because the sources grow more massive and because the gas density decreases ) and they invariably overlap with several others which in our case is dealt with in an approximate manner as explained before . in figs . [ fig : overlaps256red6 ] & [ fig : overlaps256red9 ] we compare the ionization isocontours from bears ( blue contours ) and crash ( red contours ) for redshifts 6 and 9 , respectively . the images refer to the simulations in a @xmath10 box with side of length @xmath75 . as expected , redshift 6 is structurally the most complex with many more sources and multiple overlaps . in all these images we see a distinct feature of the bears algorithm , i.e. , that the ionized regions are perfectly spherical when there is no overlap . also , in case of overlap , the shapes of ionized regions are much more regular than those of crash . the reason for this discrepancy is the local inhomogeneity of the underlying density field . as explained in [ sec : bears ] , the bears algorithm averages over a radius of @xmath52 and uses the same density in all directions , whereas crash follows the local density separately in each direction . with higher resolution the agreement between the two simulations is expected to decrease because the density field is not as isotropic , and in situations like this the codes are better at following the non - spherical nature of the ionizing front . ) , randomly selected along a direction in the @xmath74 , @xmath10 box , are plotted that displays the contours ( three levels [ 0 , 0.5 , 1 ] ) of the neutral fraction of crash ( red ) and bears ( cyan ) at @xmath76 . the underlying `` light gray '' contours represent the dark matter overdensities . ] but for redshift @xmath77 . ] in figs.[fig : overlapsred6 ] and [ fig : overlapsred9 ] , we compare bears and crash for a lower resolution simulation , i.e. , @xmath78 in a @xmath74 box for redshifts 6 to 9 . these figures indeed show a much better agreement because the detailed structure of the ionization front traced by crash at a higher resolution has been smoothed out . ) , randomly selected along a direction in the @xmath74 , @xmath78 box , are plotted that displays the contours ( three levels [ 0 , 0.5 , 1 ] ) of the neutral fraction of crash ( red ) and bears ( cyan ) at @xmath76 . the underlying `` light gray '' contours represent the dark matter overdensities . ] but for redshift @xmath77 . ] all the previous figures of comparison of crash and bears were performed without taking into account the history of reionization , i.e. , radiative transfer was performed on each snapshot assuming that it had not been previously ionized . in figure [ fig : overlap6withmem ] and the top panel of figure [ fig : overlap9withmem ] we plot four slices of the @xmath79 , @xmath61 box at redshifts of 6.2 and 9 respectively , which do include the memory of ionization from previous redshifts . we observe that the agreement in this case is not as good as in the case without the history of reionization , but given that this comparison is made at the lowest redshift of interest and for a high resolution ( @xmath74 ) simulation , the results are acceptable . the bottom panel of figure [ fig : overlap6withmem ] shows the mean ( solid line ) and variance ( dashed line ) of the mass - weighted ionized fraction as a function of redshift for a @xmath11 , @xmath61 box . the reason why the statistics of mass - weighted ionized fraction agree so well whereas the contour plots have some descrepancy is that when the underlying density is high , bears estimates the extent of the ionization correctly , whereas in case of under - dense regions , bears overestimates the size of the bubble . ) from a @xmath11 , @xmath61 box are randomly selected and contours ( three levels [ 0 , 0.5 , 1 ] ) of the neutral fraction of crash ( red ) and bears ( blue ) at @xmath76 are plotted . the underlying `` light gray '' contours represent the dark matter overdensities . in this figure we have taken into account the ionization history , i.e. , the radiative transfer is performed on the box which had been partially ionized at earlier redshifts . ] but for redshift @xmath77 . the bottom panel shows the percentage difference between crash and bears in the mean ( solid line ) and variance ( dashed line ) of the mass - weighted ionized fraction ( ionized fraction @xmath80 0.95 ) in the @xmath61 box as a function of redshift when the history of reionization is included.,title="fig : " ] but for redshift @xmath77 . the bottom panel shows the percentage difference between crash and bears in the mean ( solid line ) and variance ( dashed line ) of the mass - weighted ionized fraction ( ionized fraction @xmath80 0.95 ) in the @xmath61 box as a function of redshift when the history of reionization is included.,title="fig : " ] although there are still differences in the images at lower redshifts due to the overlap , we expect that convolving the image with the beam response of the antenna will result in very similar images . this is indeed what we see in fig . [ fig : compconv ] . the slice used in the figure was obtained from a @xmath10-box at redshift 6 . the beam response ( see [ sec : instrumental ] ) eliminates the detailed structures of the ionizing front tracked by the radiative transfer code . box at redshift six . the corresponding figures below show the images after being smoothed by the beam response of the antenna . details of the instrument are given in [ sec : instrumental ] . as expected , the images look almost identical after the convolution operation . ] in this section we describe several statistical comparisons between the results obtained by crash and bears . the bottom panel of fig . [ fig : overlap9withmem ] shows the difference between the mass - weighted mean and variance of the neutral fraction in the @xmath11 , @xmath61 box as a function of redshift including the reionization history . the difference is within 2% for high redshifts ( @xmath81 ) , and even at redshift six it is around 5% . as a second probe , fig . [ fig : comphist ] shows a histogram of the fractional volume in a @xmath11 , @xmath61 box occupied by different neutral fractions . this plot reveals details of the discrepancy in the two approaches . at higher redshifts the volume is predominantly neutral , whereas at lower redshifts , as reionization proceeds , radiation ( partially ) ionizes parts of the volume . we see that the black solid line in the figure , which corresponds to bears agrees very well with the histogram of crash ( red - dashed ) at very low ( @xmath82 ) and at very high ionization levels ( @xmath83 ) , but during the intermediate ionization levels they do not compare very well . the reason for this is that the ionized bubbles in bears have sharp transitions from neutral to a fixed ionized fraction at the ionization front whereas in crash , depending on the density distribution , the ionized fraction across the ionization front falls off slowly . , @xmath61 box at four different redshifts as indicated . the black solid line corresponds to bears and the dashed red line to crash . ] a final diagnostic is presented in fig . [ fig : densvionfrac ] . the neutral fraction in each voxel is plotted against the overdensity in that pixel . as in the previous diagnostic there is overlap between the two methods at high and very low neutral fractions . but we see that crash ( red ) spans the entire range of neutral fractions whereas the points corresponding to bears ( black ) are clustered . this again can be attributed to the sudden transition in the neutral fraction around radiative sources used by bears . also , most higher density environments are ionized by bears because in most cases a source is centred on the high density pixels of the box . another aspect of the implementation of the bears scheme is apparent in this plot , i.e. , at higher densities ( @xmath84 ) the agreement between crash and bears is extremely good . this is because sources are usually located at overdense location and bears uses the density around the source location to estimate the radius of ionized sphere . thus , around the source and consequently in high dense regions , bears correctly estimates the radius of the ionized spheres but fails to do so in the low density enviroment . , @xmath61 box at redshift 6 , including the entire reionization history . outputs from different frequency channels of an interferometre / lofar measure the radiation ( in our case the 21 cm emission ) from varying redshifts . thus , the spectral resolution of the telescope dictates the scales over which structures in the universe are smoothed or averaged along the redshift direction . the shape of the power spectrum and the characteristics of the line of sight signal are influenced by this operation . in order to study its effect on the `` true '' underlying signal it is therefore imperative to create , from outputs of different individual redshifts , a contiguous data set of maps on the sky at different frequencies . this section explains how we create just such a frequency cube ( fc ) . for the purpose of the lofar - eor experiment we create a cube spanning the observing window of the experiment , i.e. , from redshift 6 ( @xmath85 200 @xmath86 ) to 11.5 ( @xmath85 115 @xmath86 ) . the following steps were employed to obtain maps of the 21 cm eor signal at different redshifts , given a number of snapshots from a cosmological simulation between the redshifts of interest : * inputs : start and end frequencies , @xmath87 & @xmath88 ( corresponding to redshifts bounded by 6 & 12 ) ; the resolution in frequency @xmath89 , at which the maps are created . we typically choose a value @xmath89 smaller than the bandwidth of the experiment , because once this data set has been created , we can always smooth along frequency to any desired bandwidth . * create an array of frequencies and redshifts at which the maps are to be created , @xmath90 where @xmath91 , and convert each of these frequencies into a redshift array , @xmath92 where @xmath93}$ ] , the rest frequency of the 21 cm hyperfine hydrogen line . * now , for each of these redshifts @xmath94 , we identify a pair of simulation snapshots whose redshifts bracket @xmath94 . let us call these snapshots @xmath95 and @xmath96 , such that @xmath97 . * although all voxels in the output of a simulation are at the same redshift ( @xmath95 in this case ) , we assume that the voxels of the central slice ( along the redshift or frequency direction ) correspond to the redshift of the simulation box . we then move to a slice , @xmath98 ( we wrap around the box if necessary , since the simulations are performed with periodic boundary conditions ) at a distance corresponding to the radial comoving distance between @xmath95 and @xmath94 given by ; @xmath99 with @xmath100 defined as ; @xmath101 we use parameters obtained from _ wmap3 _ ( @xcite ) . * with slice @xmath98 as the centre , we average all the slices within @xmath102 , where @xmath103 is the redshift interval corresponding to a frequency width of @xmath104 at redshift @xmath95 . lets call this average @xmath105 . * all the steps performed above do not account for the time - evolution of the slice between the two outputs and thus it is important to interpolate between two corresponding slices ( to preserve the phase information of the underlying density field ) . therefore , we identify the corresponding slice , @xmath106 ( in position w.r.t the central slice ) in the snapshot @xmath96 . again , we average slices within @xmath102 with slice @xmath106 as the centre and call it @xmath107 . note , however , that the number of slices in the previous step need not be the same as in this step , owing to the fact that the @xmath103 corresponding to @xmath104 is a function of @xmath108 . * the final step in the generation of the slice @xmath109 , at @xmath94 , involves the interpolation of @xmath105 and @xmath107 : @xmath110 fig . [ fig : alongfreqion ] shows an example of a slice through the box for the the ionization history due to stars constructed using the algorithm above . because the size of the box used is @xmath111 in comoving coordinates and the comoving radial distance between redshifts of 6 and 12 is @xmath112 , we would expect there to be repetition of structures along the frequency / redshift direction . in plotting the above figure this has been reduced by a factor @xmath113 since the slice has been extracted along the diagonal of the fc . in the following sections we will do all our analysis and comparisons on the box generated in the above manner . although the radiative transfer is performed on each box with the underlying density in real space , we produce the fc for the brightness temperature with densities estimated in redshift space to account for the redshift distortions introduced by the peculiar velocities . although the general consensus is that the dominant sources of reionization are stellar in nature , there is also evidence to support quasars as additional sources of reionization @xcite . one of the main differences between stars and quasars , apart from the extent of reionization caused by individual sources , is the heating due to quasars . we simulate only the reionization history for the cases involving either stars or quasars and leave simulating the effect of heating for future work . we use a @xmath8 particle , @xmath114 mpc ( @xmath115 ) dark - matter - only simulation for this purpose . for details on the n - body simulation see section [ sec : nbody ] . about 75 simulation snapshots between redshifts 12 and 6 were used to create the fc . the entire operation of doing the radiative transfer on each of these snapshots and converting them into the fc takes approximately 12 hours on an 8-dual - core amd processor machine with 32 gb of memory . in the sections below we describe the prescriptions used to embed the dark matter haloes with quasars and stars . following that we discuss some of the statistical differences in the ionization fractions and contrast the statistical nature of the @xmath7 for these different scenarios , as a function of frequency . we caution that the scenarios described here are mostly meant to illustrate the potential of the techniques being developed here . for example , although the quasar model described below seems to ionize the universe effectively , care has not been taken to constrain the quasar population based on the soft x - ray background excess between 0.5 and 2 kev @xcite . from the output of the n - body simulation , dark - matter haloes were identified using the friends - of - friends algorithm and a linking length @xmath116 . black holes embedded in them were assigned a mass according to @xcite , @xmath117 where the factor @xmath118 reflects the magorrian relation between the halo mass ( @xmath119 ) and black hole mass ( @xmath120 ) , and @xmath121 gives the baryon ratio @xcite . the template spectrum we assumed for the quasars is a power law of the form , @xmath122 where @xmath2 is the power - law index which is set to unity . a quasar of mass @xmath123 shines at @xmath124 times the eddington luminosity , @xmath125.\ ] ] therefore @xmath126 is given by : @xmath127,\ ] ] where @xmath128 . in the above model , we have assumed that the spectral energy distribution ( sed ) of the quasar extends well above the ionization energy of hydrogen . this ensures that copious amounts of photons are available to ionize the hydrogen in the igm while at the same time reducing the number of hard x - ray photons that could potentially heat the igm . because we concentrate on the ionization history in this paper , we postpone a detailed study of the effect of cutoff in the sed at the higher and lower end of the energies . we associate stellar spectra with dark matter halos using the following procedure . the global star formation rate was calculated using ; @xmath129}{\beta - \alpha + \alpha\exp\left[\beta(z - z_m)\right]}~~[{\rm m}_\odot~{\rm yr}^{-1}~{\rm mpc}^{-3 } ] , \label{eq : starformrate}\ ] ] where @xmath130 , @xmath131 , @xmath132 marks a break redshift , and @xmath133 fixes the overall normalisation @xcite . now , if @xmath134 is the time interval between two outputs in years , the total mass density of stars formed is @xmath135.\ ] ] notice that this approximation is valid only if the typical lifetime of the star is much smaller than @xmath136 , which is the case in our model because we assume 100@xmath137 stars as the source that have a lifetime of about few myrs @xcite . therefore , the total mass in stars in the box is @xmath138 $ ] . this mass in stars is then distributed among the halos according to the mass of the halo as , @xmath139 where @xmath140 is the mass of stars in `` halo '' , @xmath141 the mass of the halo and @xmath142 the total mass of halos in the box . we then assume that all of the mass in stars is distributed in stars of 100 solar masses , which implies that the number of stars in the halo is @xmath143 . the number of ionizing photons from a 100 @xmath144 star is taken from table 3 of @xcite and multiplied by @xmath145 to get the total number of ionizing photons emanating from the `` halo '' and the radiative transfer is done assuming these photons are at @xmath146 . the escape fractions of ionizing photons from early galaxies is assumed to be 10% . using the models of ionizing sources and the algorithm to generate the fc described above , we plot a slice along the frequency direction of the ionization histories in fig . [ fig : alongfreqion ] . we see from the figure that the stars start ionizing much earlier than the quasars , but the rate at which they ionize is low . therefore , even though the effect of quasars on the ionization is seen later , they manage to ionize the igm earlier than stars . this is mainly due to the efficiency ( per unit baryon ) of the quasar to ionize the medium and to the fact that in the model we have constructed the miniqso population is not constrained based on the excess of soft x - ray background radiation and therefore the miniqso number density is higher than it would be in reality . it should be noted that if the escape fractions are allowed to change the ionization histories due to stars can be altered significantly . this can be seen more clearly and quantitatively in fig . [ fig : meanboxes ] , which shows the volume averaged mean neutral fraction as a function of redshift ( top panel ) and the volume fraction that is ionized ( @xmath147 ) in the box ( bottom panel ) . [ fig : alongfreqdtb ] shows a slice through the fc of @xmath7s for the case of quasars ( first panel from the top ) and stars ( second panel ) which was calculated following @xcite : @xmath148^{1/2 } , \label{eq : dtb}\end{aligned}\ ] ] where @xmath149 is the hubble constant in units of @xmath150 , @xmath151 is the mass density contrast , and @xmath152 and @xmath153 are the mass and baryon densities in units of the critical density . @xmath154 is the temperature of the cosmic microwave background at redshift @xmath108 and @xmath155 the spin temperature . in all our calculations we have assumed @xmath156 . the third panel in fig.[fig : alongfreqdtb ] shows the variance of @xmath7 as a function of frequency for stars ( red dashed line ) and quasars ( black solid lines ) . since quasars start ionizing late and completely ionize the universe by redshift 6 , the variance stays below that for stars at the beginning and end of reionization but peaks at around 160 @xmath86 corresponding to a redshift of 7.8 . interestingly , for stars the peak of the variance occurs around the same frequency . this is a welcome coincidence from an observation point of view . lofar will initially observe using instantaneous bandwidths of 32 @xmath86 ( lambropoulous et al . , _ in prep _ ) . thus , it becomes important to make an educated guess of the frequency around which the observation has to be made , to be sure that we capture reionization activity at its peak . and because the variance of these two very different scenarios peak around the same redshift , this frequency ( @xmath157}$ ] ) is a reasonable guess . it is worth emphasizing that these are predictions based on the simple - minded models prescribed above . nevertheless this exercise illustrates the capability of our algorithm to implement diverse scenarios and help guide the observational strategies of lofar . finally , in the fourth panel we plot the image ( slice ) entropy as a function of redshift . image entropy is a measure of randomness in the image and in our case it reflects the inhomogeneity of reionization as a function of redshift . the entropy of an image is calculated as follows : first , a histogram of the intensities in the image , which in our case correspond to @xmath7 values , is made and normalized . if there are @xmath158 bins and each bin @xmath40 has a normalized value @xmath159 , then the entropy is given by , @xmath160 the meaning of image entropy is illustrated with an example in section [ sec : instrumental ] , aided with fig . [ fig : mockentrop ] . in the context of our study the image entropy has the following meaning : 1 . _ before convolution with the lofar beam ( ref . [ fig : alongfreqdtb ] ) _ : + at low frequencies ( @xmath161 ) the universe is still mostly neutral and homogeneous . thus a histogram of @xmath7 at a given frequency less than 145 @xmath86 would be very narrow and hence have lower entropy ( ref . eq [ eq : entropy ] ) . but as the sources start ionizing the igm , it introduces fluctuations in @xmath7 and the histogram spreads out thus increasing the entropy . similarly , at much higher frequencies ( @xmath162 ) the entropy drops down because the universe is largely ionized and the histogram of @xmath7 is narrow but now centred at 0 . 2 . _ after convolution with the lofar beam ( ref . [ fig : postconvstats ] ) _ : + in this case , although the overall behaviour is similar to that of the previous case before convolution , the effect is enhanced . this can be understood with the aide of fig . [ fig : mockentrop ] . lower frequencies ( @xmath163 ) is analogoues to case ( a ) & ( c ) and its corresponding histogram in ( ha ) & ( hc ) . because , at these frequencies the universe is still largely neutral with just a few sources and the effect of smearing @xmath7 with the observing beam spreads the histrogram thus increasing the entropy . at intermediate frequencies ( @xmath164 ) , the effect of decreasing entropy is analogoues to case ( b ) & ( d ) with its corresponding histogram in ( hb ) & ( hd ) . although the entropy increased from ( b ) to ( d ) and also in case of @xmath7 after the convolution , the increase is relatively less . this is because the smearing action was on scales larger than the fluctuations in @xmath7 hence narrowing the histogram to a few values . at higher frequencies of course @xmath7 is zero and hence the entropy drops to 0 as well . a final statistic that was calculated was the probability distribution function ( pdf ) of the @xmath7 at four different redshifts for stars and quasars . if the igm is not ionized and if the neutral hydrogen is measurable ( i.e. the signal - to - noise ratio is high enough ) , the distribution of @xmath7 reflects the underlying density field ( eq . [ eq : dtb ] ) . but as reionization proceeds , larger and larger regions become ionized and the pdf of @xmath7 skews because of the excess of @xmath7 close to zero . this fact can be made use of as a tool for the statistical detection of the eor signal as will be explored in an upcoming paper by harker et al . , _ in prep_. the features of the pdf seem to vary dramatically between the two scenarios . the rationale behind producing these simulations is to test the effect of the interferometric response on the signal and other contributors with the same frequency band . this section gives an overview of simulations of the lofar response and shows how the signal will be seen by lofar in the absence of noise or other calibration errors . a more detailed discussion of the lofar response and the data model for the lofar - eor experiment will be provided in lambropoulous et al . , _ in prep_. for the eor experiment we plan to use the lofar core which will consist of about 24 dual stations of tiles . each dual station will have a diameter of approximately 35 meters and the maximum baseline between stations will be about 2 km . for the simulations in this paper we make some simplifying assumptions regarding the telescope response . we assume the narrow bandwidth condition and we also assume that the image - plane - effects have been calibrated to a satisfactory level . this includes a stable primary beam and adequate compensation for the ionospheric effects such as the ionospheric phase introduced during the propagation of electromagnetic waves in the ionosphere and the ionospheric faraday rotation . in an interferometric observation , the measured correlation of the electric fields between two sensors @xmath40 and @xmath165 is called visibility and is given by @xcite : @xmath166 where @xmath167 is the primary beam , @xmath168 the sky brightness , @xmath169 are the direction cosines and @xmath170 are the coordinates of the baseline ( in units of observing wavelength ) as seen from the source . we further assume that simulated maps are a collection of point sources : each pixel corresponds to a point source with the relevant intensity . note that the equation above takes into account the sky curvature . the visibilities are sampled for all station pairs , but also at different pair positions as the earth rotates . for every baseline and frequency the _ _ uv _ _ track points sample different scales of the fourier transform of the sky at that frequency . we compute the _ uv _ tracks for each interferometer pair for 6 hours of synthesis with an averaging interval of 100 sec and we grid them onto a regular grid in the _ uv _ plane . using the gridded tracks as a sampling function , we sample the fourier transform of our model sky , which in our case is the 21 cm eor signal , at the corresponding grid cells . this gives us the time series of the visibilities for every station pair . the fourier ( or inverse fourier ) transform of the sampled visibilities is called the `` dirty '' map . it is actually the sky map convolved with the fourier transform of the sampling function , which is called the `` dirty '' beam or the psf . this is a simple - minded approach in estimating the sky brightness as it uses linear operations . with the lofar core we expect to reach a sensitivity of 350 mk in one night of observations at 150 @xmath86 . after 100 nights the final sensitivity will drop to 35mk . for total intensity observations we expect gaussian noise with zero average and the above mentioned rms to drop to 35 mk . we assume that calibration errors , after 100 nights of integration accumulate in such a way that they follow a gaussian distribution obeying the central limit theorem . the effect of the `` original '' image passing through the response of the lofar antenna at a frequency of 130 @xmath86 is shown in fig . [ fig : dirty_clean ] . the figure shows the brightness temperature in both cases normalized to the largest value in the corresponding figure . note that the simulation is @xmath9 across which is only about @xmath171 of the total ( @xmath172 ) field of view of lofar . therefore we are missing the very large scale fourier modes . we are planning to run larger simulations to overcome this problem . along with the effect of the spatial resolution of lofar on the sky , we also studied the effect of its spectral resolution . the maps produced by lofar will have a bandwidth of about 1 @xmath86 . this is essential to beat the noise level down . [ fig : losdtb_stars ] and [ fig : losdtb_qsos ] show @xmath7 along five different lines of sight . the black solid lines indicate the signal before convolution and spectral smoothing , whereas the over - plotted blue dashed lines include the effect of lofar s spatial and spectral response . the entire fc was convolved with the antenna response of lofar calculated separately for each frequency . this `` convolved cube '' is then used to plot fig . [ fig : postconvstats ] . this figure is identical in structure to fig . [ fig : alongfreqdtb ] . as expected , due to the smoothing action of the antenna beam pattern , the variance of the signal in the `` convolved cube '' drops by approximately 30% relative to that of the fc while conserving the overall behaviour across all frequencies . the entropy in each of these slices on the other hand shows a dramatically different behaviour . the rise and fall of the entropy in the case of quasars is much more pronounced for the `` convolved cube '' and the entropy for stars shows a steady increase . it is interesting that the slice - entropies for stars and quasars are very different . this could be used as a discriminant for different scenarios of reionization . [ fig : mockentrop ] is a simple example to illustrate the fact that smoothing increases the entropy in the figure . in essence , the averaging along the frequency direction introduces correlations between pixels of adjacent slices . for the simulations considered here it corresponds to 10 adjacent slices ( since the resolution of the simulation is 0.1 @xmath86 and that of lofar is 1 @xmath86 ) . this effect is enhanced in the case of quasars because of the rapid rate of ionization and we see that the entropy peaks around 150 @xmath86 where the transition from a neutral to an ionized universe accelerates . . images ( c ) and ( d ) are the convolutions of ( a ) and ( b ) with a square top hat function with side of length @xmath173 of the side of the image . the entropies of these images are 4.1 and 2.2 , which is larger than the entropies we started with . therefore , when we introduced correlations between pixels by smoothing them , their entropies increased . images ( ha ) to ( hd ) are the histograms corresponding to images ( a ) to ( d ) respectively . as we can see , the histograms of images ( a ) and ( b ) have the same distribution , whereas image ( hc ) has a histogram that has a large spread compared to ( hd ) . from an information theory point of view this large spread corresponds to higher information content and hences higher entropy . similarly , when we smooth / average along the frequency direction in the fc , we introduce correlations between pixels , hence increasing the entropies considerably when the scale of smoothing is smaller than the typical correlation length of the pixels in the original image . , title="fig:",scaledwidth=50.0% ] . images ( c ) and ( d ) are the convolutions of ( a ) and ( b ) with a square top hat function with side of length @xmath173 of the side of the image . the entropies of these images are 4.1 and 2.2 , which is larger than the entropies we started with . therefore , when we introduced correlations between pixels by smoothing them , their entropies increased . images ( ha ) to ( hd ) are the histograms corresponding to images ( a ) to ( d ) respectively . as we can see , the histograms of images ( a ) and ( b ) have the same distribution , whereas image ( hc ) has a histogram that has a large spread compared to ( hd ) . from an information theory point of view this large spread corresponds to higher information content and hences higher entropy . similarly , when we smooth / average along the frequency direction in the fc , we introduce correlations between pixels , hence increasing the entropies considerably when the scale of smoothing is smaller than the typical correlation length of the pixels in the original image . , title="fig:",scaledwidth=50.0% ] we emphasized the need to expand the scenarios that need to be explored for reionization and their role in building a reliable and robust pipeline , thus necessitating fast realizations of different scenarios . we built a scheme in which a range of 1-d ionization profiles were catalogued for a number of luminosities , redshifts , densities and source spectra and which were subsequently coupled to an @xmath174-body simulation to obtain an approximation of a `` standard '' radiative transfer code . the results obtained were validated using crash , a full radiative transfer code with ray tracing . the agreement between the two methods was excellent for early redshifts ( @xmath175 ) , but as expected , the discrepancy grew towards lower redshifts . several visual comparisons of the slices of different boxes were made along with three different statistical measures of the similarity between the simulations . many snapshots ( @xmath176 ) were used between a redshift of 15 and 6 and the radiative transfer done on them as described in the preceding sections . these snapshots were then used to make a contiguous cube running from redshift 6 to 12 . in terms of the observing frequency , the cube spans 115 @xmath86 to 200 @xmath86 with a frequency resolution matching that of lofar , about one @xmath86 . cubes were generated for scenarios involving only stars or quasars and some diagnostics are provided to quantitatively differentiate between them . for both models the variance in @xmath7 peaked at around 160 @xmath86 . although neither may reflect reality , these scenarios demonstrate the use of the techniques we have developed to span large parameter spaces of variables . the pdf of @xmath7 ( see fig . [ fig : dtb_pdf ] ) provides a statistical discriminant between the different source scenarios and could be used in the future to look for the statistical detection of the signal . the cubes generated provide @xmath7 as a function of frequency . the cube was then averaged over a bandwidth of 1 @xmath86 and convolved with the beam pattern of lofar to understand the distortions caused by incomplete sampling by an interferometer . even if the images were blurred by the operation , the overall characteristics of the signal remain detectable . although the behaviour of the variance of the signal before and after the convolution remained the same ( except that the latter had lower values on average ) , the image / slice entropy showed a very different behaviour . in the former case , i.e. , before the convolution , the image entropy remained almost flat throughout the frequency range whereas in the latter the entropy steeply rises at around 160 @xmath86 . as a note , it is important to mention that although in principle the ionized bubbles do move with a peculiar velocity @xmath177 , where @xmath178 is the typical peculiar velocity of galaxies at redshift zero , assuming @xmath179km / s , a redshift 10 object would have a peculiar velocity of 200 km / s . for a typical lifetime of the source considered , i.e. , 10 myr , this corresponds to motion of about a couple of kpc . this is an order of magnitude less than the resolution of the simulation box at that redshift . therefore we ignore this effect . on the other hand we have taken into account the effect of redshift distortions whose effects are relatively more important . the simulations and comparisons in this paper have focused on purely stellar or quasars sources , but it is plausible that the early sources of reionization were a mixture of stars and quasars or other yet unknown sources . it is therefore important to simulate reionization by a mixture of these sources , taking into account their clustering properties . the simulations presented in this paper did not take into account the contraints on the population of ionizing sources imposed by various measurements like the infrared excess in the case of stars and the soft x - ray excess for quasars . apart from the ionization patterns induced by these sources , the @xmath7 maps will also depend on the kinetic temperature which is coupled to the spin temperature via collisions or ly-@xmath2 pumping . hence , it is imperative that we include these temperature effects on the igm . we will incorporate the mixture of sources and the effect of the temperature in an upcoming paper ( thomas et al . , _ in prep _ ) . one of the main astrophysical hurdles for the detection of the eor signal is the existence of prominent galactic and extragalactic foregrounds . typically , the difference between the mean amplitude of the eor signal and the foregrounds is expected to be 4 to 5 orders of magnitude , but an interferometer like lofar measures only the fluctuations which in this case are expected to be different by ` only ' three orders of magnitude @xcite . in subsequent papers we will explore the effects of foregrounds and their removal strategies . lofar is being partially funded by the european union , european regional development fund , and by `` samenwerkingsverband noord - nederland '' , ez / kompas . the author rmt would also like to thank the max planck institute of astrophysics , garching , for the support provided during a work visit to the institute . 99 bruzual g. , charlot s. , 2003 , mnras , 344 , 1000 ciardi b. , ferrara a. , white s. d. m. , 2003 , mnras , 344 , l7 ciardi b. , madau p. , 2003 , apj , 596 , 1 ciardi b. , ferrara a. , marri s. , raimondo g. , 2001 , mnras , 324 , 381 chabrier g. , 2003 , pasp , 115 , 763 davis m. , efstathiou g. , frenk c. s. , white s. d. m. , 1985 , apj , 292 , 371 dijkstra m. , haiman z. , loeb a. , 2004 , apj , 613 , 646 fan x. , et al . , 2006 , aj , 131 , 1203 ferland g. j. , korista k. t. , verner d. a. , ferguson j. w. , kingdon j. b. , verner e. m. , 1998 , pasp , 110 , 761 ferrarese l. , 2002 , apj , 578 , 90 fukugita m. , kawasaki m. , 1994 , mnras , 269 , 563 gnedin n. y. , abel t. , 2001 , newa , 6 , 437 hockney r. w. , eastwood j. w. , 1988 , computer simulations using particles , taylor & francis pawlik a. h. , schaye j. , 2008 , arxiv , 802 , arxiv:0802.1715 perley r. a. , schwab f. r. , bridle a. h. , 1989 , aspc , 6 , hogan c. j. , rees m. j. , 1979 , mnras , 188 , 791 jeli v. , et al . , 2008 , mnras , 891 kuhlen m. , madau p. , 2005 , mnras , 363 , 1069 lambropoulou p. , et al . , in prep madau p. , meiksin a. , rees m. j. , 1997 , apj , 475 , 429 maselli a. , ferrara a. , ciardi b. , 2003 , mnras , 345 , 379 mellema g. , iliev i. t. , alvarez m. a. , shapiro p. r. , 2006 , newa , 11 , 374 mesinger a. , furlanetto s. , 2007 , apj , 669 , 663 nakamoto t. , umemura m. , susa h. , 2001 , mnras , 321 , 593 nusser a. , 2005 , mnras , 359 , 183 page l. , et al . , 2007 , apjs , 170 , 335 pritchard j. r. , furlanetto s. r. , 2007 , mnras , 376 , 1680 razoumov a. o. , cardall c. y. , 2005 , mnras , 362 , 1413 rijkhorst e .- j . , plewa t. , dubey a. , mellema g. , 2006 , a&a , 452 , 907 ritzerveld j. , icke v. , rijkhorst e .- j . , 2003 , astro , arxiv : astro - ph/0312301 schaerer d. , 2002 , a&a , 382 , 28 schaye j. , dalla vecchia c. , 2008 , mnras , 383 , 1210 scott d. , rees m. j. , 1990 , mnras , 247 , 510 seljak u. , zaldarriaga m. , 1996 , apj , 469 , 437 spergel d. n. , et al . , 2007 , apjs , 170 , 377 springel v. , 2005 , mnras , 364 , 1105 springel v. , hernquist l. , 2003 , mnras , 339 , 312 sunyaev r. a. , zeldovich i. b. , 1975 , mnras , 171 , 375 susa h. , 2006 , pasj , 58 , 445 thomas r. m. , zaroubi s. , 2008 , mnras , 384 , 1080 venkatesan a. , giroux m. l. , shull j. m. , 2001 , apj , 563 , 1 whalen d. , norman m. l. , 2006 , apjs , 162 , 281 wiersma r. p. c. , schaye j. , smith b. d. , 2008 , mnras , submitted , preprint ( arxiv:0807.3748 ) zahn o. , lidz a. , mcquinn m. , dutta s. , hernquist l. , zaldarriaga m. , furlanetto s. r. , 2007 , apj , 654 , 12 zaroubi s. , thomas r. m. , sugiyama n. , silk j. , 2007 , mnras , 375 , 1269 zaroubi s. , silk j. , 2005 , mnras , 360 , l64

we solve a truncated system of coupled dyson schwinger equations for the gluon and ghost propagators in su( ) yang mills theories in faddeev popov quantization on a four - torus . this compact space - time manifold provides an efficient mean to solve the gluon and ghost dyson schwinger equations without any angular approximations . we verify that analytically two power - like solutions in the very far infrared seem possible . however , only one of these solutions can be matched to a numerical solution for non - vanishing momenta . for a bare ghost - gluon vertex this implies that the gluon propagator is only weakly infrared vanishing , , , and the ghost propagator is infrared singular , . for non - vanishing momenta our solutions are in agreement with the results of recent su(2 ) monte - carlo lattice calculations . the running coupling possesses an infrared fixed point . we obtain for all gauge groups su( ) . above one gev the running coupling rapidly approaches its perturbative form . + + _ keywords : _ confinement ; non perturbative qcd ; running coupling constant ; + gluon propagator ; dyson schwinger equations ; infrared behavior . + + _ pacs : _ 12.38.aw 14.70.dj 12.38.lg 11.15.tk 02.30.rz

hep-ph0202195

it is generally accepted that the theory of strong interactions , qcd , should describe the observed phenomenon of confinement : colored objects like quarks and gluons occur only in hadrons . a possible route for gaining more understanding of this phenomenon is the study of the infrared behavior of qcd green s functions , for a recent review see @xcite . in addition to shed light on the fundamental properties of qcd the knowledge of these green s functions provides the basis for a successful description of hadronic physics @xcite . based on the idea of infrared slavery older works on this subject assumed a strongly infrared singular gluon propagator . recent studies based either on dyson schwinger equations @xcite or lattice calculations @xcite in landau gauge indicate quite the opposite : an infrared finite or even infrared vanishing gluon propagator . these two techniques are complementary in the following sense : on the one hand , monte - carlo lattice calculations include all non - perturbative physics of yang mills theories but can not make definite statements about the very far infrared due to the finite lattice volume . on the other hand , dyson schwinger equations allow to extract the leading infrared behavior analytically and the general non - perturbative behavior with moderate numerical effort but these equations , consisting of an infinite tower of coupled non - linear integral equations , have to be truncated in order to be manageable . as we will also see in the course of this article , the propagators of su(2 ) and su(3 ) landau gauge yang mills theory in faddeev popov quantization coincide for these two different approaches reasonably well . thus we are confident that our results for the qualitative features of these propagators are trustable . especially , these recent results on the landau gauge propagators imply that the kugo ojima confinement criterion @xcite is fulfilled ( for a short summary on this topic see _ e.g. _ ref . it is gratifying to note that no truncation to the dyson schwinger equations has to be applied to arrive at this conclusion if one assumes that the involved green s functions can be represented in the infrared by asymptotic expansions @xcite . in landau gauge , a sufficient condition for the kugo ojima confinement criterion is that the nonperturbative ghost propagator is more singular than a massless pole in the infrared @xcite : @xmath5 this behavior is also correlated to other aspects of yang mills theories . first , the oehme zimmermann superconvergence relations @xcite can be derived from ward takahashi identities assuming the kugo ojima confinement criterion @xcite . these superconvergence relations formalize a long known contradiction between asymptotic freedom and the positivity of the spectral density for transverse gluons in the covariant gauge . second , eq . ( [ ghprop ] ) agrees with zwanziger s horizon condition @xcite . this amounts to gribov s prescription to cut off the functional integral at the first gribov horizon @xcite . noting that this horizon is a convex hypersurface in @xmath6-space that surrounds the origin @xcite allows one to conclude that the dyson are not changed @xcite . however , one has to note that this treatment of the functional integral is related to the resolution of an ambiguity in the solution of these equations @xcite . nevertheless , supplementing the faddeev popov quantization with this additional constraint might be not sufficient to provide an exact solution of the problem because there exist gribov copies within the first gribov horizon @xcite . as already stated , the obtained values for the infrared exponents of the gluon and ghost propagators depend on the employed approximation for the dyson schwinger equations . beyond the necessary truncation of this set of integral equations in numerical calculations also some approximations for the angular integrals have been used so far @xcite . on the other hand , employing infrared expansions ( without using any angular approximation ) which are strictly valid only in the limit of vanishing momentum , @xmath7 , @xcite yield also some quite different values for the infrared critical exponent depending on the truncation scheme . with respect to these analytical calculations the question arises whether for every extracted value of the infrared exponent the corresponding numerical solution exists also for finite values of momenta . as will be detailed in this paper a tool to overcome angular approximations is the treatment of the dyson schwinger equations on a compact euclidean four - manifold , in our case a four - torus . this allows us to answer the above question : we will see that not every analytically extracted infrared exponent can be matched to a numerical solution for non - vanishing momenta . to make this paper self - contained we will shortly summarize truncation schemes for the gluon and ghost dyson schwinger equations in flat ( infinite ) euclidean space - time , which have been solved recently , in the first two sections of sect.ii . in the following section we introduce a novel truncation scheme . in sect . iii we present the dyson schwinger equations formulated on the momentum grid which is the dual space to the compact four - torus . in sect . iv we will present solutions done in otherwise exactly the same approximation scheme as previous solutions for flat space - time . the comparison to these previous solutions allows to chose a suitable regularization and renormalization procedure . ( as in previous work we adopt a modified mom scheme . ) in sect . iv also the solutions without any angular approximations will be discussed . the central result of this paper is : only one of two solutions allowed by the infrared analysis can be matched to a numerical solution for non - vanishing momenta . e.g. _ for bare vertices this implies that the gluon propagator is only weakly infrared vanishing , @xmath1 , @xmath8 , and the ghost propagator is infrared singular , @xmath3 . in sect . v we present our conclusions . furthermore , in appendix a we present our approximation to impose one - loop scaling of the gluon and ghost propagators in the ultraviolet . in appendix b we discuss the influence of gluon and ghost zero modes on the solutions . the numerical methods to solve the gluon and ghost dyson schwinger equations in flat euclidean space - time will be described in appendix c. in this section a short summary of previously employed truncation and approximation schemes for the coupled gluon and ghost dyson schwinger equations @xcite will be given first . we will also provide the underlying formula for a new truncation scheme . all these schemes include all diagrams in the ghost equation and neglect contributions from the two - loop diagrams in the gluon equation , see fig . [ gluon ] , where the full gluon dyson schwinger equation of qcd is represented diagrammatically . in addition , as we will be only concerned with pure yang mills theory in this paper the quark loop will be neglected . the tadpole term provides in landau gauge only an ( ultraviolet divergent ) constant and will drop out during renormalization anyhow . thus , we will effectively study the coupled system of equations as depicted in fig . [ gluonghost ] . the necessary trunction of the gluon dyson schwinger equation immediately leads to a problem : the gluon polarization , which due to gauge symmetry would be transverse to the gluon momentum in an exact calculation , acquires spurious longitudinal terms . these terms are in general quadratically ultraviolet divergent and thus highly ambiguous because they depend on the momentum routing in the loop integral . in addition , a gauge invariant regularization scheme is required to avoid these unphysical longitudinal terms . such schemes are , however , hard to implement in dyson schwinger studies , for the corresponding use of dimensional regularization see _ e.g. _ refs.@xcite . an alternative unambiguous procedure is to isolate the part free of quadratic ultraviolet divergences by contracting with the projector @xmath9 which is constructed such that @xmath10 , and therefore the ambiguous term proportional to @xmath11 is projected out @xcite . note that the use of this projector also removes the tadpole term . as has become obvious recently @xcite ( see also ref . @xcite for a corresponding discussion in a much simpler truncation scheme ) the use of the projector ( [ rproj ] ) interferes with the infrared analysis of the coupled gluon - ghost system . for technical reasons we will employ a one - parameter family of projectors @xmath12 which allows us to interpolate continuously from the projector ( [ rproj ] ) to the transversal one ( with @xmath13 ) furthermore , we will also use the general form of the quadratically ultraviolet divergent tadpole to remove these unwanted ultraviolet divergencies . here we employ the conventions and notations of ref.@xcite . as usual for dyson schwinger studies all integrals are formulated in euclidean space - time . @xmath14 is the gluon renormalization function defined via the gluon propagator in landau gauge @xmath15 the deviation of @xmath14 from its tree - level value @xmath16 provides a measure for renormalization of the gluon field due to the considered interactions . the ghost renormalization function @xmath17 is defined analogously via the ghost propagator , see eq . ( [ ghprop ] ) . in landau gauge the ghost - gluon vertex does not attribute an independent ultraviolet divergence , _ i.e. _ one has @xmath18 , @xcite . therefore a truncation based on the tree - level form for the ghost - gluon vertex function , @xmath19 is compatible with the desired short distance behaviour of the solutions . this will be exploited in the following . to proceed we will first consider the truncation schemes of refs.@xcite in the next two subsections . both these truncation schemes employ only the projector @xmath20 ( [ rproj ] ) in the gluon equation . the main difference between these truncation schemes consists in the treatment of the three - point functions . whereas in ref . @xcite the form of ghost - gluon and three - gluon vertex function has been related to the gluon and ghost renormalization functions using slavnov taylor identities ( and then the resulting system has been solved self - consistently ) , in ref.@xcite bare three - point functions have been used . amazingly , though , both schemes provide results with identical qualitative infrared behavior : the gluon propagator vanishes in the infrared , the ghost propagator is highly singluar there , and the strong running coupling ( which can be related to the gluon and ghost renormalization functions using the specific form of the ghost - gluon vertex in landau gauge @xcite ) has an infrared fix point . because this infrared behavior is determined by the interplay between the ghost loop in the gluon equation ( the gluon loop being subleading in the infrared ) and the ghost equation such a scenario is also found in the ghost - loop only approximation @xcite . we will exploit these two recent truncation schemes in the course of this article for various tests of our method . in ref . @xcite an approximation scheme for the longitudinal parts of ghost - gluon and 3-gluon vertex functions has been employed which ensures consistency at the level of one - particle greens functions , _ i.e. _ propagators . the detailed form of the vertex functions can be found in ref . @xcite . here we provide directly the ghost , @xmath21 @xmath22 and the gluon @xmath23 equations , respectively . the functions @xmath24 and @xmath25 are given in appendix c of ref . @xcite . in eq . ( [ ghdse1 ] ) we have already exploited the identity @xmath26 . this leaves the gluon and ghost field renormalization constants @xmath27 and @xmath28 as well as the gluon vertex renormalization constant @xmath29 to be determined correspondingly to the employed truncation . note that these constants depend on the ultraviolet cutoff @xmath30 and the renormalization scale @xmath31 . in ref . @xcite different angular approximations for @xmath32 and for @xmath33 have been employed . in the latter case @xmath34 and @xmath35 have been set which obviously preserves the limit @xmath36 of the integrand . with this approximation one obtains from eq . ( [ ghdse1 ] ) upon angular integration @xmath37 where we introduced an @xmath38invariant momentum cutoff @xmath30 to account for the logarithmic ultraviolet divergence , which will have to be absorbed by the renormalization constants . with some further assumptions the angular approximation for the gluon equation ( [ zdse ] ) yields @xmath39 in the infrared the solutions @xmath40 and @xmath41 behave power - like , @xmath42 in this truncation scheme one obtains @xmath43 @xcite . to solve the coupled system for all momenta the power laws , eq . ( [ power ] ) , are used to perform the integrals from @xmath44 to an infrared point @xmath45 analytically , while the remaining part of the integrals is done with the help of numerical routines , see _ e.g. _ ref . @xcite . the non - perturbative subtraction scheme of ref . @xcite implies a strong running coupling with infrared fixed point . starting again from the non - renormalization of the ghost - gluon vertex @xmath46 one can readily show that the product @xmath47 is renormalization group invariant . therefore , in absence of any dimensionful parameter , this ( dimensionless ) product is a function of the running coupling @xmath48 only . analysing this renormalization group invariant product more closely one concludes that it is identical to the running coupling @xmath49 . as the infrared powers in the product @xmath50 cancel exactly the running coupling is a finite constant for @xmath51 . substituting the tree - level ghost - gluon vertex for the dressed one and neglecting the gluon loop the coupled system of equations ( [ ghdse1],[zdse ] ) reads @xmath52 to simplify notation we introduce the abbreviations @xmath53 , @xmath54 , @xmath55 and @xmath56 . in the angular approximation of ref.@xcite ( ymax approximation ) where @xmath57 and @xmath58 are substituted for @xmath59 and @xmath60 eqs.([ghost],[gluon1 ] ) are simplified to @xmath61 imposing as renormalization conditions @xmath62 to determine @xmath27 and @xmath28 these equations may be solved numerically for a given ( o(4 ) invariant ) cutoff . as a matter of fact , we use subtracted finite equations in our numerical procedure , see appendix c for details . in the infrared , also the solutions of eqs . ( [ ghab],[glab ] ) behave power - like , _ _ eq . ( [ power ] ) , with @xmath63 @xcite . in the same truncation but with no angle approximation employed a solution @xmath64 has been extracted for the infrared behaviour . however , it will be explained below that we could not find a numerical solution for non - vanishing momenta connected to the @xmath64 infrared behaviour . in this subsection we will detail a novel truncation scheme which employs bare vertices . nevertheless it will be constructed such that it reproduces the correct perturbative limit for large momenta . to analyse the gluon loop we will use the class of projectors @xmath65 ( [ paproj ] ) . a smooth interpolation between the brown pennington projector ( @xmath66 ) and the transverse one ( @xmath67 ) will be helpful in the analysis of unphysical quadratic ultraviolet divergencies for @xmath68 in the gluon equation . their careful removal is essential for a numerical solution retaining the infrared behaviour of the solutions . the coupled equations for the ghost and gluon dressing functions using bare vertices read as follows ( the ghost equation is , of course , identical to eq . ( [ ghost ] ) , it is only repeated for a coherent representation . ) : @xmath69 the kernels ordered with respect to powers of @xmath70 have the form : @xmath71 it is straigtforward to verify that for @xmath66 the kernel @xmath72 is identical to the kernel in eq . ( [ gluon1 ] ) . the quadratic ultraviolet divergencies of the integrals are most easily discovered approximating the angular integrals as done in the previous subsections ( note that we will use these approximations only for an analysis of the ultraviolet behaviour ) and introducing an euclidean sharp cutoff . displaying only the ultraviolet divergent integrals @xmath73 one sees that for @xmath66 the terms independent of the integration momentum @xmath74 vanish . thus these integrals are then only logarithmically ultraviolet divergent as could be expected on the basis of the results summarized in the previous subsections . of course , the quadratic ultraviolet divergencies are artefacts of the employed truncation . due to gauge invariance they would cancel against the tadpole and similar divergencies in the two loop terms . the calculation of the latter being beyond the scope of this paper we will simply subtract these divergent terms . this can not be done straightforwardly at the level of integrands : such a procedure would disturb the infrared properties of the dyson schwinger equations . as we anticipate from previous studies and analytic work @xcite that the ghost loop is the leading contribution in the infrared the natural place to subtract the quadratically ultraviolet divergent constant is the gluon loop . we do this by employing the substitution @xmath75 in eq . ( [ gluonbare ] ) . at the first sight , due to the presence of the prefactor @xmath29 in the gluon loop , this seems not to be sufficient to remove also the quadratic ultraviolet divergence of this ghost loop . however , note that in the next step we will enforce consistency of the logarithmic divergencies which entails then cancelation of the quadratic divergencies if @xmath76 is employed in the gluon loop , see below . to achieve the correct one - loop scaling in the ultraviolet we will adopt a similar treatment to the one of ref . please note that within the presented class of truncation schemes it is impossible to fulfill both , correct one - loop scaling and the slavnov taylor identity @xmath77 . for large euclidean momenta and to one loop the behavior of the propagator functions can be described as @xmath78^\gamma \ ; , \label{gluon_uv}\\ g(x ) & = & g(s ) \left [ \omega \log\left(\frac{x}{s}\right)+1 \right]^\delta \ ; . \label{ghost_uv}\end{aligned}\ ] ] @xmath79 and @xmath80 denote the value of the dressing functions at some renormalization point @xmath55 , @xmath81 and @xmath82 are the respective anomalous dimensions . to one loop one has @xmath83 and @xmath84 for arbitrary number of colors @xmath0 and no quarks , @xmath85 @xcite . furthermore , @xmath86 . employing these expressions in the ghost equation ( [ ghostbare ] ) and approximating the angular integrals as done previously one obtains to the order of the leading logarithms @xmath87^{-\delta } & = & \tilde{z}_3 - \frac{9 g^2 z(s ) g(s)}{64 \pi^2 \omega(\gamma+\delta+1 ) } \times \nonumber\\ & & \left\ { \left [ \omega \log\left(\frac{l}{s}\right)+1 \right]^{\gamma+\delta+1 } - \left [ \omega \log\left(\frac{x}{s}\right)+1 \right]^{\gamma+\delta+1 } \right\}\end{aligned}\ ] ] where the abbreviation @xmath56 has been used again . the renormalization constant @xmath88 cancels the cutoff dependence of the first term in the bracket . thus , the power and the prefactor of the second term have to match with the left hand side of this equation . this leads to two conditions : @xmath89 eq . ( [ gd1 ] ) is of course nothing else but consistency of the ghost equation with one - loop scaling . plugging the ultraviolet behavior ( [ gluon_uv ] ) and ( [ ghost_uv ] ) into the gluon equation ( [ gluonbare ] ) and keeping the leading order logarithms leads to @xmath90^{-\gamma } & = & z_3 - \frac{g^2 g^2(s)}{32 \pi^2 \omega \gamma } \times \nonumber \\ & & \left\ { \left [ \omega \log\left(\frac{l}{s}\right)+1 \right]^{-\gamma } - \left [ \omega \log\left(\frac{x}{s}\right)+1 \right]^{-\gamma } \right\ } \nonumber \\ & & - \frac{7g^2z(s)g(s)}{16 \pi^2 } \int_x^l dy z_1 \left[\omega \log\left(\frac{y}{s}\right)+1 \right]^{\gamma+\delta } \ ; . \label{gluoneq_uv}\end{aligned}\ ] ] here we see that due to the employed truncation we have to give up either one - loop scaling or the slavnov taylor identity @xmath77 . note , however , that @xmath91 has the correct cutoff dependence to make this equation finite with finite @xmath92 . nevertheless , in order to get the correct leading log contribution from the gluon loop the integrand in the last line should be proportional to @xmath93^{-\gamma-1}$ ] . this not only contradicts the identity @xmath77 but also requires the renormalization _ constant _ @xmath92 to acquire a momentum dependence if correct one - loop scaling is to be enforced . following case c ) described in chapter 6 of ref . @xcite we demand therefore @xmath94^{-\delta-2\gamma-1 } \ ; .\ ] ] generalizing the approach of ref . @xcite we note that the following two - parameter ansatz @xmath95 with arbitrary @xmath96 and @xmath97 satisfies the required proportionality . this can be straightforwardly verified with the help of expression ( [ gd1 ] ) obtained from the ghost equation . in addition ansatz ( [ z1_ansatz ] ) ensures the cancelation of quadratic ultraviolet divergencies as discussed above also if @xmath98 . carrying out the remaining integral in ( [ gluoneq_uv ] ) and matching powers one obtains @xmath99 first of all , we reproduce correct one - loop scaling also from the gluon equation as can be seen from the equivalence of eqs . ( [ gd2 ] ) and ( [ gd1 ] ) . second , with the perturbative renormalization conditions @xmath62 which are possible and appropriate at a large euclidean renormalization point @xmath31 one obtains together with eq . ( [ go1 ] ) the correct anomalous dimensions @xmath100 and @xmath101 . a reasonable choice of parameters is of course one which keeps @xmath102 as weakly dependent as possible on the momenta @xmath74 and @xmath103 , _ c.f . _ [ z1.dat ] in appendix a. note that the choice @xmath104 corresponds to the truncation scheme of @xcite whereas @xmath105 together with the appropriate vertex dressings reproduces case c ) of ref . the infrared behavior of the gluon loop in the gluon equation depends strongly on @xmath96 and @xmath97 . _ e.g. _ for @xmath106 one can distinguish three cases : for @xmath107 the gluon loop is subleading in the infrared , for @xmath108 as in ref . @xcite the gluon loop produces the same power as the ghost loop , for @xmath109 the gluon loop becomes the leading term in the infrared . in the latter case we did not find a solution to the coupled gluon - ghost system . in appendix a we will demonstrate that @xmath110 minimizes the momentum dependence of @xmath102 . thus we will use these values except stated otherwise explicitely . the infrared behavior of the propagator functions in this truncation scheme and for the transverse projector ( @xmath67 ) has been determined very recently @xcite . it can be shown straightforwardly that the relation between ghost and gluon infrared behavior is again as in eq.([power ] ) . this is expected also on general grounds @xcite . due to this one can immediately use the anstze @xmath111 as is explained in detail in ref.@xcite the renormalization constants @xmath27 and @xmath112 can be dropped for very small momenta @xmath113 : they are either subleading in the infrared ( gluon equation ) or have to be zero when the renormalization takes place at @xmath114 ( ghost equation ) . the remaining infrared integrals can be evaluated using the formula @xcite @xmath115 where again @xmath116 , @xmath117 and @xmath118 . from the resulting two conditions @xmath119 one obtains the determining relation for @xmath120 by equating both left hand sides , see fig . [ kappa.dat ] . for the brown pennington projector , _ i.e. _ @xmath66 , one then finds the known solution @xmath121 @xcite . however , as can be seen immediately the left hand side of the second equation possesses a zero for @xmath122 which is canceled by a pole only for @xmath66 . lowering @xmath123 only slightly a further solution with @xmath120 slightly larger than 0.5 exists . for the transverse projector , _ i.e. _ @xmath67 , this latter solution becomes @xmath124 in accordance with refs . also the solution @xmath121 changes continuously when lowering @xmath123 . the corresponding @xmath120 are then all larger than 1 and contradict the masslessness condition , see chapter 5 of ref . @xcite for a discussion of this condition . the main result of this paper is that the infrared behavior @xmath125 matches to a numerical solution whereas no numerical solution could be found with the infrared behavior @xmath121 . the renormalization group analysis for the running coupling given in ref . @xcite certainly applies here as well . the renormalization group invariant expression for the running coupling is therefore given by @xmath126 as can be seen directly from equations ( [ kappa1],[kappa2 ] ) , in the here presented truncations the product @xmath127 is constant for a given @xmath120 . with @xmath128 one concludes immediately that @xmath129 is proportional to @xmath130 . furthermore , the ghost and gluon dressing functions @xmath40 and @xmath41 are independent of the number of colors : @xmath0 enters the dyson - schwinger equations only in the combination @xmath131 at our level of truncation . from the solution @xmath132 of the infrared analysis with the transverse projector @xmath67 one determines the infrared fixed point of the running coupling to be @xmath133 for @xmath134 . there are three central aims connected to the investigation of the dyson schwinger equations on a four - torus . the first one is purely technical : this allows to study finite volume effects also in dyson schwinger approach . monte - carlo simulations on a lattice necessarily have to be done in a finite volume . therefore in the latter kind of approach infrared properties are only accessible by extrapolations to an infinite volume where the avalaible data are gained on several different volumes which , due to limitations in computer time , do only differ at best by one order of magnitude . we will see in the present dyson schwinger approach that available volumes cover several orders of magnitude . and more importantly , in several truncations and approximations one can compare to the results obtained in an infinite volume . the second issue is the solution of the ( truncated ) dyson schwinger equations without any angular approximations . we will detail below why and how this can be achieved on the momentum grid dual to the four - torus . this will also lead us to the main result of this paper discussed in the following section . the third aim relates to the possibility of topological obstructions on a compact manifold . it is well known _ e.g. _ that a four - torus allows for a non - vanishing pontryagin index . we hope that we will be able to describe the solutions of dyson schwinger equations with twisted boundary conditions @xcite in a subsequent publication . and choosing an asymmetric four - torus might allow the introduction of a non - vanishing temperature in a relatively simple way . from a technical point of view using as underlying manifold a four - torus or choosing periodic boundary conditions on a hypercube is identical . note that the definition of the faddeev popov operator necessitates periodic boundary conditions for ghosts instead of their grassmannian nature , see _ e.g. _ ref . @xcite . with @xmath135 being the length in every direction of the hypercube the four - dimensional momentum integral has to be substituted by a sum over four indices , @xmath136 on the other hand , the quantities of interest , the gluon and ghost renormalization function , @xmath14 and @xmath17 , resp . , do only depend on the o(4 ) invariant squared momenta as long as all directions are treated on an equal footing on the torus . this suggests to relabel the points on the momentum grid not according to a cartesian but a hypersherical coordinate system , @xmath137 where the index @xmath138 numbers the hyperspheres @xmath139 . the index @xmath140 , which numbers the grid points on each hypersphere respectively , will be dropped in the following . in the integrals to be discretized there appear three momenta , the external momentum , labeled @xmath141 , the loop momentum @xmath142 and for the second propagator in the loop @xmath143 . we will use the following notation : @xmath144 on the hypercubic momentum grid dual to the four - torus the momentum @xmath143 is located on the grid for every pair of grid momenta @xmath141 and @xmath142 as can be seen from elementary vector operations ( or from the analogue of cubic lattices in solid state physics ) . note that we have introduced an o(4 ) invariant cutoff @xmath30 into eqs.([odgdse],[odzdse ] ) . as we will see below in more detail a corresponding regularization of the sums over grid momenta is required . in a first step we set up a momentum grid with @xmath145 . the hypercube property of this lattice seem to suggest in the first place to cut off the sums such that over complete hypercubes is summed . such a method , however , breaks o(4 ) invariance and introduces sizeable numerical errors . ( note that in lattice monte - carlo simulations the analysis of the resulting data for o(4 ) invariance necessitates special kinds of cuts through the lattice @xcite . ) as can be seen from fig . [ latt ] an o(4 ) invariant cutoff of the sums necessitates to neglect the `` edges '' : the sum extends only over the fully drawn hypersheres , and we omit the summation over the dashed ones . to be able to compare to already known flat space solutions we will first solve the torus analogue of the equations with angle approximations , _ i.e. _ eqs.([odgdse],[odzdse ] ) or eqs . ( [ ghab],[glab ] ) . to this end we start from eqs . ( [ ghdse1],[zdse ] ) or eqs . ( [ ghost],[gluon1 ] ) , respectively . after formulating the corresponding four - dimensional integrals as sums over lattice momenta the functional dependence of the propagator functions is approximated accordingly to the rules described in the subsections ii.a and ii.b , respectively . we give the explicit expressions for the bare vertex ghost - loop only truncation . the corresponding ones for dressed vertex truncation can be derived analogously in a straightforward manner , however , as these are quite lengthy we do not give their explicit form . the dyson schwinger equations in bare vertex ghost - loop only truncation with angle approximation read on the torus @xmath146 as shown in the previous subsection the momentum @xmath143 which fulfills @xmath147 lies on the momentum grid , and furthermore , @xmath143 is determined uniquely from @xmath141 and @xmath142 for which @xmath148 and @xmath149 by using elementary vector operations . note , however , that @xmath150 might be larger than the ultraviolet cutoff even if @xmath151 and @xmath152 are not . the actual treatment of the sums in eqs . ( [ disgdse],[diszdse ] ) can nevertheless be done straightforwardly as the exact algebraic expressions for the kernels are known . the main difference between the equations on the torus and the flat space equations is the effective treatment in the infrared . the finite volume in coordinate space leads to a finite value of squared momentum for the first hypersphere @xmath153 . thus one has not to worry about possible infrared singularities . on the other hand , there exists zero modes on the four - torus . in all the calculations presented here they are neglected , an estimate of their possible contribution is given in appendix b. note furthermore that for the numerical solution as described in refs.@xcite the disadvantage of dealing with infrared singularities has been turned into a benefit : the infrared behavior of the propagator functions has been determined via an asymptotice series calculating the corresponding coefficients and powers analytically . matching these series at some finite momentum to the functions as obtained from standard numerical techniques finally enabled to numerically solve the coupled integral equations . such a complicated process is not necessary employing a torus . however , one anticipates already at this level some deviations in the infrared between the solutions obtained in these different ways . in the next section the corresponding numerical results will be dicussed . they demonstrate that using a torus as infrared cutoff works surprisingly well . the intricacy of the infrared analysis has prevented so far a solution of the coupled gluon - ghost system without angular approximations . using a torus as an infrared regulator opens up the possiblity to solve these equations directly without any recourse to an infrared asymptotic expansion . as recently arguments have been provided that the use of a bare ghost - gluon vertex is not inferior to employing a dressed one @xcite we will solve the system in the truncation described in sect . with the above discussed replacement of @xmath154 the equations ( [ ghostbare],[gluonbare ] ) read on the torus : @xmath155 as already stated @xmath156 might be larger than the ultraviolet cutoff even if @xmath157 and @xmath158 are not . again the kernels can be nevertheless evaluated straightforwardly , however , if @xmath103 with @xmath159 is the argument of a propagator function we set @xmath160 , _ i.e. _ we approximate @xmath161 or @xmath162 then by @xmath163 or @xmath164 , respectively . another more elaborate treatment consists of matching the corresponding perturbative ultraviolet tail to the function under consideration . this has been applied in some cases to test the viability of the method . the solutions of the dyson schwinger equations on the torus can be compared in two different ways to the ones obtained in flat space - time . first , one can use the solutions of the `` continuum '' equations for a certain cutoff @xmath30 and certain renormalization scale @xmath31 to provide the values for @xmath165 and @xmath166 as input for the equations on the torus which are then solved , _ c.f . _ the discussion in appendix c where the numerical methods employed to obtain the solutions of the `` continuum '' equations are summarized . second , one can subtract both equations at the squared momenta @xmath167 and @xmath168 and therefore trade the two renormalization constants for the values of the dressing functions at these momenta , namely @xmath169 and @xmath170 . for the sake of comparison one can read of these values from the continuum solution to be used in the torus equations . if @xmath168 and @xmath167 are taken to be sufficently far in the ultraviolet region of momentum , where finite volume effects play a minor role , the two procedures lead to the same results . we have verified that this is indeed the case within the limits of numerical accuracy . our results for the ghost dressing function , the gluon dressing function and the running coupling in the bare - vertex ghost - only and the dressed vertex truncation can be seen in figs . [ bare.dat ] and [ dressed.dat ] , respectively . in both truncation schemes we solved for three different momentum spacings corresponding to different volumes in coordinate space . to keep the cutoff identical for all the spacings within each truncation scheme we have chosen three different grid sizes respectively . for the bare vertex truncation they are @xmath171 and for the dressed vertex truncation they are @xmath172 . a physical momentum scale can not be determined in truncation schemes which do not provide the correct perturbative running of the coupling in the ultraviolet . with the missing gluon loop this is the case in the bare vertex only truncation scheme . we therefore have to stick to an internal momentum scale without physical units in this case . the situation is different , however , in the dressed vertex truncation scheme . here we have fixed the momentum scale by calculating the running coupling for the colour group su(3 ) and using the experimental value @xmath173 at @xmath174 to fix a physical scale . in the bare vertex truncation scheme we have chosen @xmath175 as renormalization condition where we have taken @xmath176 , the ( squared ) ultraviolet cutoff . of course , this choice is by no means special and one is completely free to choose the renormalization point wherever one likes . for the numerical calculation in flat space we have chosen two different subtraction points for the ghost and the gluon equations as described in appendix c. for good convergence of the iteration process the ghost equation is most conveniently subtracted at zero momentum , whereas the gluon equation can be subtracted at any value of squared momentum in the region where the equations are solved numerically . in our calculation we chose the cutoff @xmath177 in internal units as subtraction point for the gluon equation . this allows us to use the renormalization condition @xmath178 directly as input in the calculation . the second input is provided by the coefficient @xmath6 of the leading order infrared expansion of the gluon dressing function , @xmath179 . the condition @xmath180 then leads to @xmath181 . the value of the coupling at the renormalization point , @xmath182 , is taken to be 0.97 . again one is completely free to chose this number arbitrarily up to the maximum value of the running coupling which is reached in the very infrared . for the torus calculations the unsubtracted equations have been used . the same coupling @xmath183 has been taken and instead of @xmath184 and @xmath6 the values @xmath185 and @xmath186 for the renormalization constants determined from the `` continuum '' solution have been employed . as has already been mentioned , in the dressed vertex truncation scheme we determine a `` physical '' momentum scale according to experimental results . in the numerical treatment again the ghost equation has been subtracted at zero momentum whereas the gluon equation is subtracted at finite momentum . we solved both equations similar to the method described in ref . @xcite , especially we introduced also the auxiliary functions @xmath187 and @xmath188 as defined in ref . @xcite . as input values serve the infrared expansion of @xmath188 , @xmath189 , and the value @xmath190 at the gluon subtraction point @xmath191gev@xmath192 . for the calculations on the torus we use the unsubtracted equations with the values @xmath193 and @xmath194 for the renormalization constants which have been determined from the `` continuum '' solution . our numerical results , depicted in figs . [ bare.dat ] and [ dressed.dat ] , show similar properties for both truncation schemes . compared with the respective `` continuum '' solutions the ones obtained on a torus show deviations for the first few lattice points in the infrared . for higher momenta all functions on the torus approach the continuum ones . deviations are there only visible for the curves with the largest spacings . the biggest effect can be seen for the running coupling @xmath195 , which is the `` observable '' of the system . as @xmath195 is proportional to the product @xmath196 the deviations of the dressing functions from the `` continuum '' curve amplify in the infrared in a somewhat erratic way , so that the points in the very infrared can not be connected by a smooth line . comparing larger and smaller spacings of momentum grids one clearly sees that the effect is always one of the first spheres on the respective lattices and therefore moves to the infrared for smaller spacings . the most important properties of the solutions in flat space can still be found in the torus solutions despite some deviations in the infrared . going from larger to smaller spacings a powerlike behaviour of the dressing functions in the infrared with the correct exponents can still be infered . for the truncation scheme with dressed vertices the gluon dressing function on the torus has the same shape and the same hight of the bump in the bending region of the curve . as all curves with sufficently small momentum spacing follow the correct power behaviour within numerical errors we conclude that one surely can adress the question of the value of the power @xmath120 in the infrared on a manifold with finite volume as has been used here or as one uses in lattice calculations . this is the central result of the present section : employing a torus as infrared regularization is possible . in the following section we will use this to go beyond the angular approximation . in this section we discuss our numerical results for the bare vertex truncation scheme with various projectors in the gluon equation . we show results for the gluon and ghost dressing function and the running coupling calculated on a torus without any angular approximations . note that up to now in the gluon - ghost dyson schwinger equations solutions for non - vanishing momenta have been obtained only employing some sort of angular approximation . going beyond these approximations we will use our new truncation scheme introduced in sect.2.3 . the reason for this is , that the dressed vertex truncation scheme is quite complicated , and in its derivation one has dismissed terms which seem to be crucial for the infrared behavior beyond the angle approximation . as the survey of gluon and ghosts infrared behaviors related to different anstze for vertex functions @xcite does not provide any evidence for crucial differences between bare and dressed vertex functions the use of the latter seems to be an unnecessary complication . in the case of the bare - vertex ghost - loop only truncation with brown pennington projector no solutions for finite momenta without angular approximation have been reported yet . in fact the results presented in this section suggest that these solutions might not exist . this time the values for the renormalization constants @xmath27 and @xmath112 can not be taken from a `` continuum '' solution . thus we use the subtracted equations on the torus in the course of the numerical solution by iteration . both equations are subtracted at the renormalization point @xmath197 . similar to the case of the dressed vertex truncation from the last section the physical units are gained from experimental input : @xmath173 at @xmath174 . the input values for the dressing functions at the renormalization point are @xmath198 . requiring @xmath199 then fixes the overall scale for @xmath41 . for the value of the coupling at the renormalization point we chose again @xmath183 similar to the two truncation schemes in the last section . for the presented solution on three different volumes lattice sizes of @xmath200 have been used . as mentioned in the last section we did check cutoff effects by extrapolating the propagator functions at @xmath201 with a logarithmic tail with the correct anomalous dimensions . the results as compared to the ones obtained by simply setting @xmath202 and @xmath203 for all @xmath204 did change by less than @xmath205 . our results for the dressing functions obtained with a transverse projector as shown in fig . [ new.dat ] are in agreement with the expected power behavior . the infrared critical exponent as calculated in refs . @xcite , @xmath206 , thus has been verified : a corresponding ( numerical ) solution for non - vanishing momenta exists . the gluon dressing function is remarkably stable against changes of the volume and approaches more and more the expected power solution for small momenta . for the ghost dressing function one observes again deviations of the first points in the infrared : an extraction of the correct infrared critical exponent from the numerical solution for the ghost function is hardly possible . only some points come close to the analytical value of the continuum before the curve starts bending down again in the very infrared . for the extracted value of the running coupling in the infrared this leads to a distinct mismatch to what one is to expect on the basis of analytical results . at first sight the fact that the power solution for the ghost dressing function could be not reproduced numerically to a reasonable precision may seem disappointing . nevertheless these numerical results themselves show that the ghost dressing function is highly infrared singular . this reflects the long - range correlation of ghosts in landau gauge . therefore one should expect the ghost dressing function to be the one affected most by a finite volume . on the contrary the gluon dressing function vanishes in the infrared and consequently it is much less affected by a finite volume . we expect the ghost dressing function together with the running coupling to approach more and more the correct power solution in the infrared as lattice spacings are decreased and lattice sizes are increased . furthermore , we add a remark on the transversality of the gluon propagator as obtained in the bare vertex truncation scheme . of course , the full gluon propagator calculated from the complete gluon equation with fully dressed vertices in landau gauge is transversal due to slavnov taylor identities . thus such a hypothetical solution would be independent of the form of the projector , _ i.e. _ in our notation with the projector @xmath207 independent of the parameter @xmath123 . in practice , using bare vertices as in our truncation scheme this is certainly not the case . our numerical results for different values of the parameter @xmath123 can be seen in fig . [ proj.dat ] . although our solutions show the expected dependence on the form of the projector this dependence is not too drastic and in general the behaviour of these different solutions is very similar . for the gluon dressing function one observes that the more @xmath123 grows the greater is the deviation from the pure power behaviour . the points in the very infrared can not be connected by a smooth line any more . correspondingly this happens for the running coupling . based on the infrared analysis one might anticipate that @xmath120 should approach the value @xmath208 more and more as @xmath123 grows until there is a jump to the solution from @xmath209 to @xmath210 as the brown pennington limit @xmath66 is reached . we do not observe such a qualitative jump in our solutions on a torus . the solution shown for @xmath66 is approached smoothly when @xmath123 approaches this limit . this clearly indicates that the solution @xmath210 might not exist at all if one removes the torus as a regulator . finally we compare our results to recent su(2 ) lattice calculations @xcite . as has already been stated above the ghost and gluon dressing functions from dyson - schwinger equations are independent from the numbers of colors at least to our level of truncation . the only caveat in comparing our results with the lattice ones is the adjustment of the momentum scale which is certainly different from the case of su(3 ) above . to get an appropriate momentum scale we therefore used the lattice result @xmath211 at @xmath212 as input . the two graphs in figure ( [ lattice.dat ] ) show that the main qualitative features , the vanishing of the gluon and the divergence of the ghost dressing functions in the infrared , are common properties of both , the lattice solutions and the one from dyson - schwinger equations . even the power @xmath213 of the gluon dressing function on the torus is very close to the one that can be extracted from the lattice fit to be @xmath125 . the main difference of both approaches lies in the medium energy region around one gev , where the dyson - schwinger solutions suffer from the missing two loop contributions that are certainly present in lattice monte - carlo simulations . in this paper we have presented numerical solutions of truncated systems of dyson schwinger equations for the gluon and ghost propagators in landau gauge su(n ) yang mills theories . we have employed a four - torus , _ i.e. _ a compact space - time manifold , as an infrared regulator . this enabled us to overcome angular approximations used so far in previous studies @xcite . the basis for our numerical calculations have been provided by an analytic determination @xcite of the exponents governing the infrared power - like behavior of the gluon and ghost propagators . typically the infrared analysis provide two possible values each for the gluon and the ghost , respectively . the central result of this paper is : only one of these two possible infrared behaviors in a given truncation scheme can be matched to a numerical solution for non - vanishing momenta . in a truncation scheme with a bare ghost - gluon vertex and a transverse projection of the ( truncated ) gluon equation this implies that the gluon propagator is only weakly infrared vanishing , @xmath214 , @xmath215 , and the ghost propagator is highly infrared singular , @xmath3 . the running coupling possesses an infrared fixed point whose value is given by @xmath216 ( or , for a general number @xmath0 of colors , @xmath217 ) . our results compare very favorably with results of recent lattice calculations performed for two colors . due to the finite lattice volume the lattice results can not , of course , be extended into the far infrared . in this respect our results are complementary to the lattice ones : we do obtain the infrared behavior analytically . on the other hand , lattice calculations include , at least in principle , all non - perturbative effects whereas we had to rely on truncations . _ e.g. _ the deviations for the gluon renormalization functions at intermediate momenta depicted in fig . [ lattice.dat ] might be due to the neglect of the four - gluon vertex function in our calculations . as an outlook we would like to mention that employing a four - torus as underlying manifold might serve for a number of interesting studies of dyson schwinger equations . amongst these , the use of twisted boundary conditions on the torus might enable one to shed some light on the importance of topologically non - trivial gauge field configurations for the infrared behavior of qcd green s functions . we would like to note that recent lattice calculations @xcite indicate a relation between the existence of center vortices in the maximal center gauge , the area law of the wilson loop and the infrared behavior of the gluon propagator in landau gauge . furthermore , we want to point out that choosing an asymmetric four - torus might provide an efficient mean to extend these calculations to non - vanishing temperatures . here the main qualitative question arises about the fate of the kugo ojima confinement criterion at the deconfinement transition . finally , we want to remark that qcd green s functions are an important input in many calculations in hadron physics @xcite . the next necessary step towards such phenemenological applications is a study of the quark propagator . such an investigation might hopefully also provide some insight into the mechanism of quark confinement in covariant gauges . we thank jacques bloch for many useful hints in the early stages of this work . we are especially grateful to lorenz von smekal for many valuable discussions and for making ref . @xcite available to us . we are indebted to kurt langfeld for communicating and elucidating his lattice results , partly prior to publication . furthermore , we thank sebastian schmidt , peter watson , andreas wipf and daniel zwanziger for helpful discussions . this work has been supported by the dfg under contract al 279/3 - 3 and by the european graduate school tbingen basel . in the framework of the truncation scheme presented in sect . iic we have shown that the approximation @xmath218 for the gluon vertex renormalization constant yields the correct one loop scaling of the gluon loop in the gluon dyson - schwinger equation . this is true for any values @xmath96 and @xmath97 . of course , in a full treatment of the coupled ghost gluon system @xmath219 would be independent of momentum . therefore a choice of @xmath96 and @xmath97 which keeps @xmath102 as weakly varying as possible seems the most reasonable one . this choice can be infered using the scaling of the dressing functions extracted from the renormalization group equation , see ref . @xcite for details . the dressing functions can be expressed as @xmath220 where the running coupling provides the correct one loop scaling in the ultraviolet . consequently the function @xmath188 approaches unity for high momenta . furthermore , from the known infrared behavior of @xmath40 , @xmath41 and @xmath129 one infers that @xmath188 is proportional to @xmath221 in the infrared . writing @xmath102 in terms of @xmath129 and @xmath188 yields @xmath222 in the perturbative region @xmath223 and the function @xmath102 is therefore slowly varying for any @xmath96 and @xmath97 according to the logarithmic behaviour of the running coupling @xmath195 . in the infrared , however , @xmath195 approaches its fixed point while the functions @xmath224 behaves like a power . consequently the choice @xmath110 guarantees the weakest momentum dependence of @xmath102 which is illustrated in fig . [ z1.dat ] . shown is the @xmath74-dependence of the function @xmath225 . ( note also that due the symmetry @xmath226 and the absence of an explicit @xmath113-dependence this is sufficient to demonstrate its momentum dependence . ) in the perturbative momentum regime the function @xmath102 does not vary with the parameter @xmath96 . so all three choices give the same logarithmic running in momentum as required to give the correct one loop scaling behaviour of the integral . in the infrared , however , a change in @xmath96 gives rise to substantial changes in the behaviour of @xmath102 , with only the choice @xmath227 leading to a constant . an important point when formulating the dyson schwinger equations on the torus could be the treatment of the zero modes . in addition , on the torus an infrared analysis like the one in flat euclidean space - time is not possible , and one is left with the problem how the dressing functions behave at vanishing momenta . guided by the intuition that especially the long ranged modes should be affected by the finite volume we assume in the following @xmath228 just like in the continuum and @xmath229 if zero modes are neglected . phrased otherwise we assume that the zero modes are the missing ingredient to ensure the correct infinite volume limit for the torus results . therefore , if on tori of different volumes @xmath230 shows no sign of becoming divergent , the infrared enhancement seen in @xmath231 or in the flat space - time results has to be due to the torus zero modes of gluons and ghosts . therefore , in this appendix , we will show that the assumption @xmath232 does not lead to a contradiction in the equations on the torus if zero modes are neglected . to this end we focus on the truncation scheme without angular approximations . first we rewrite equations ( [ ghostbare],[gluonbare ] ) as @xmath233 according to appendix a we have chosen @xmath110 , where @xmath101 , the anomalous dimension of the ghost . the kernels have the form : @xmath234 we first analyse the behavior of the integrands in the limit @xmath235 for finite momenta @xmath113 . then @xmath236 and @xmath237 and the kernels times the respective dressing functions are to appropriate order in momentum @xmath74 : @xmath238 furthermore , @xmath239 has been used and terms proportional to @xmath240 have been dropped , as they either integrate to zero in the continuum or cancel each other in the sums on the torus . each of the expressions ( [ kernel_y1],[kernel_y2],[kernel_y3 ] ) is then the appropriate term for @xmath241 on the right hand side of the dyson - schwinger equations on the torus . clearly one observes , that only a finite ghost mode @xmath242 avoids trouble with divergencies . this is especially true for the kernel @xmath243 of the gluon loop , as @xmath244 is more singular than the simple pole , so this kernel vanishes for small momenta @xmath74 . the other two expressions ( [ kernel_y1 ] ) and ( [ kernel_y2 ] ) are finite . one is then left with the ambiguous quantities @xmath245 and @xmath246 which will be replaced by their integrals from zero to @xmath247 in the calculation at the end of this section . the arbitrarines of this procedure is considerably milded by the observation that any number plugged in for the trigonometric functions yields the same qualitative result at the end of this section . second , we take the limit @xmath248 , which on the torus is identical to @xmath249 . the ghost kernel @xmath250 alone would certainly diverge as @xmath249 , but taking into account the power law behaviour @xmath251 for the gluon dressing function the integrand is zero in this limit . this is valid for @xmath252 , which is in agreement with the infrared analysis in the continuum . we therefore may omit the points @xmath253 in the ghost equation . the situation is different in the gluon equation where the kernel of the ghost loop has a finite limit @xmath248 : @xmath254 . therefore with a finite ghost dressing function @xmath242 the points @xmath253 in the ghost loop contribute but no divergencies occur . in the gluon loop the kernel @xmath243 multiplied by the dressing functions approaches zero for vanishing momentum @xmath103 due to the power law behaviour of @xmath255 . to obtain a definite value for @xmath242 we now investigate the behaviour of the equations ( [ ghostbare_app],[gluonbare_app ] ) in the limit @xmath256 . the integrands are then given by @xmath257 where the kernel @xmath243 is of no interest , as we know the gluon loop to be subleading in the infrared . clearly , the kernel @xmath258 of the ghost loop in the gluon equation is now singular for @xmath259 , corresponding to a vanishing gluon dressing function in the infrared . this result confirms our working hypothesis that the gluon mode @xmath260 is not affected by the finite volume of the torus . the integrand of the ghost equation is finite up to the point @xmath44 . there the pole in the kernel is canceled by the behavior of the gluon dressing function @xmath261 resulting in a zero for vanishing momentum @xmath113 and @xmath74 . we therefore arrive at a consistent set of equations for a finite ghost mode @xmath242 and a vanishing gluon mode @xmath260 . in fig . [ null.dat ] we show the results for the ghost dressing function gained on two different volumes on the torus . within numerical accuracy the values of @xmath242 are the same for the two volumes . obviously terms with high loop momentum @xmath74 contribute most to the right hand side of the ghost equation for vanishing momentum @xmath113 . furthermore , one observes that the actual value of @xmath242 is not in accordance with an extrapolation of the ghost curves to the infrared . there is also no sizeable change of the gluon and the ghost dressing function when @xmath242 is set to zero by hand . this has been done in all calculations in the main body of this paper . having shown that @xmath262 on the torus even in the infinite - volume limit and assuming that the torus should provide a reasonable infrared regularization of physics in flat space - time we conclude that the divergence of @xmath242 is very probably due to the torus zero modes of gluons and ghosts . noting furthermore that a diverging @xmath242 is related to the zwanziger gribov horizon condition and the kugo ojima confinement criterion this indicates a direct relation between zero modes , the gribov horizon and confinement . the numerical methods to solve the coupled gluon - ghost system in flat euclidean space - time after applying angular approximations has been described in great detail in refs . note that different numerical techniques have been employed in refs.@xcite and ref.@xcite . we have used all these techniques to verify the independence of our solutions from details of the numerical treatment . first , we discuss the numerical renormalization process . as we know the characteristics of the solutions _ a priori _ we may exploit the possibility of independent subtractions of the gluon and the ghost equation . the procedure followed in ref . @xcite was to eliminate renormalization constants @xmath27 and @xmath112 by subtracting the equation at the renormalization scale @xmath263 . the results presented in sect . iv.a are generated by solving the equations @xmath264 for the truncation scheme of sect . ii.b and analogous equations for the scheme of sect . ii.a thereby introducing subtraction points @xmath168 and @xmath167 . the parts of the integrals from @xmath44 to an infrared matching point @xmath265 are carried out analytically , with the dressing functions in the integral beeing replaced by their leading infrared behaviour ( [ zabg ] ) @xmath266 the infrared analysis leads to fixed values @xmath120 and @xmath267 , so @xmath6 is left as a free parameter in the infrared expansion . for numerical reasons it is convenient to choose the subtraction points @xmath268 and @xmath168 to lie above the maximum of @xmath40 . the two input parameters that determine the renormalization point where we require the normalization conditions @xmath269 are then @xmath6 and the value the gluon dressing function at its subtraction point , _ i.e. _ @xmath169 . the equations are solved using the newton iteration method thereby generating values for the propagator functions . these are then plugged into the unsubtracted equations to obtain the respective values of the renormalization constants @xmath27 and @xmath112 for a given cutoff @xmath30 and a given renormalization scale @xmath31 which allows finally to determine the numerical solutions respecting the normalization conditions @xmath269 . r. alkofer and l. von smekal , phys . rept . * 353 * , 281 ( 2001 ) [ arxiv : hep - ph/0007355 ] . c. d. roberts and s. m. schmidt , prog . * 45 * , s1 ( 2000 ) . l. von smekal , r. alkofer and a. hauck , phys . rev . lett . * 79 * , 3591 ( 1997 ) ; annals phys . * 267 * , 1 ( 1998 ) [ arxiv : hep - ph/9707327 ] . d. atkinson and j. c. bloch , phys . d * 58 * , 094036 ( 1998 ) [ arxiv : hep - ph/9712459 ] ; mod . lett . * a13 * , 1055 ( 1998 ) [ arxiv : hep - ph/9802239 ] . p. watson and r. alkofer , phys . lett . * 86 * , 5239 ( 2001 ) [ arxiv : hep - ph/0102332 ] ; r. alkofer , l. von smekal and p. watson , proceedings of the ect * collaboration meeting on dynamical aspects of the qcd phase transition , trento , italy , march 12 - 15 , 2001 , arxiv : hep - ph/0105142 . d. zwanziger , arxiv : hep - th/0109224 . c. lerche and l. von smekal , arxiv : hep - ph/0202194 ; + c. lerche , diploma thesis , erlangen university , june 2001 ( in german ) . j. e. mandula , phys . rept . * 315 * , 273 ( 1999 ) . f. d. bonnet , p. o. bowman , d. b. leinweber and a. g. williams , phys . d * 62 * , 051501 ( 2000 ) [ arxiv : hep - lat/0002020 ] . f. d. bonnet , p. o. bowman , d. b. leinweber , a. g. williams and j. m. zanotti , phys . d * 64 * , 034501 ( 2001 ) [ arxiv : hep - lat/0101013 ] . k. langfeld , h. reinhardt and j. gattnar , nucl . phys . b * 621 * , 131 ( 2002 ) [ arxiv : hep - ph/0107141 ] ; arxiv : hep - lat/0110025 . t. kugo and i. ojima , prog . * 66 * , 1 ( 1979 ) . n. nakanishi and i. ojima , `` covariant operator formalism of gauge theories and quantum gravity , '' _ world scientific lecture notes in physics 27 ( 1990)_. l. von smekal and r. alkofer , proceedings of the 4th international conference on quark confinement and the hadron spectrum , vienna , austria , 3 - 8 jul 2000 ; arxiv : hep - ph/0009219 . r. oehme and w. zimmermann , phys . d * 21 * , 471 ; 1661 ( 1980 ) . k. nishijima , czech . j. phys . * 46 * ( 1996 ) 1 ; m. chaichian and k. nishijima , arxiv : hep - th/9909158 ; and references therein . d. zwanziger , nucl . b * 364 * , 127 ( 1991 ) ; nucl . b * 399 * , 477 ( 1993 ) ; nucl . b * 412 * , 657 ( 1994 ) . v. n. gribov , nucl . b * 139 * , 1 ( 1978 ) . d. zwanziger , nucl . b * 209 * , 336 ( 1982 ) . p. van baal , nucl . b * 369 * , 259 ( 1992 ) ; talk given at nato advanced study institute on confinement , duality and nonperturbative aspects of qcd , cambridge , england , 23 jun - 4 jul 1997 , arxiv : hep - th/9711070 . v. p. gusynin , a. w. schreiber , t. sizer and a. g. williams , phys . d * 60 * , 065007 ( 1999 ) [ arxiv : hep - th/9811184 ] . a. w. schreiber , t. sizer and a. g. williams , phys . rev . d * 58 * , 125014 ( 1998 ) [ arxiv : hep - ph/9804385 ] . l. von smekal , p. a. amundsen and r. alkofer , nucl . a * 529 * , 633 ( 1991 ) . n. brown and m. r. pennington , phys . d * 38 * , 2266 ( 1988 ) . a. hauck , l. von smekal and r. alkofer , comput . commun . * 112 * , 149 ( 1998 ) [ arxiv : hep - ph/9604430 ] ; comput . phys . commun . * 112 * , 166 ( 1998 ) [ arxiv : hep - ph/9804376 ] . j. c. taylor , nucl . b * 33 * , 436 ( 1971 ) . t. muta , `` foundations of quantum chromodynamics , '' _ singapore , world scientific ( 1998 ) 409 p. _ j. c. bloch , phys . d * 64 * , 116011 ( 2001 ) [ arxiv : hep - ph/0106031 ] . g. t hooft , nucl . b * 153 * , 141 ( 1979 ) ; p. van baal , `` twisted boundary conditions : a nonperturbative probe for pure nonabelian gauge theories , '' _ phd thesis , utrecht , 1984 _ , inis - mf-9631 .

we investigate the generalized kronecker algebra with arrows . given a regular component of the auslander - reiten quiver of , we show that the quasi - rank can be described almost exactly as the distance between two non - intersecting cones in , given by modules with the equal images and the equal kernels property ; more precisley , we show that the two numbers are linked by the inequality utilizing covering theory , we construct for each a bijection between the field and . as a consequence , we get new results about the number of regular components of a fixed quasi - rank .

1702.04206

let @xmath9 be an algebraically closed field of arbitrary characteristic . the finite - dimensional algebra @xmath3 is defined as the path algebra of the quiver @xmath11 with vertices @xmath12 , @xmath13 arrows @xmath14 and called the generalized kronecker algebra . we denote by @xmath15 the category of finite - dimensional left - modules of @xmath3 . + it is well known that for @xmath1 the hereditary algebra @xmath3 is of wild representation type @xcite , every regular component in the auslander - reiten quiver of @xmath3 is of type @xmath16 @xcite and there is a bijection between the regular components and the ground field @xmath9 ( * ? ? ? * xviii 1.8 ) . therefore , the problem of completely understanding the category @xmath15 or all regular components is considered hopeless and it is desirable to find invariants which give more specific information about the regular components . + one important invariant @xmath17for any wild hereditary algebra@xmath18 , introduced in @xcite , is the quasi - rank @xmath19 of a regular component @xmath2 . for a quasi - simple module @xmath20 in @xmath2 , @xmath21 is defined as @xmath22 where @xmath23 is the space of non - invertible homomorphisms from @xmath20 to @xmath24 . + another interesting invariant @xmath5 was recently introduced in @xcite . motivated by the representation theory of group algebras of @xmath25-elementary abelian groups of characteristic @xmath26 , the author defines the category @xmath27 of modules with the equal kernels property and the category @xmath28 of modules with the equal images property in the framework of @xmath3 . she shows the existence of uniquely determined quasi - simple modules @xmath29 and @xmath30 in @xmath2 such that the cone @xmath31 of all successors of @xmath29 satisfies @xmath32 and the cone @xmath33 of all predecessors of @xmath30 satisfies @xmath34 . the width @xmath35 of @xmath2 is defined as the unique number @xmath5 such that @xmath36 , i.e. the distance between the two cones . + utilizing homological descriptions of @xmath27 and @xmath28 from @xcite involving a family of elementary modules , we show that the two invariants @xmath21 and @xmath35 are linked by the inequality @xmath37 motivated by this connection , we construct for each @xmath38 a regular component @xmath2 with @xmath39 . in order to do so , we consider representations over the universal covering @xmath40 of @xmath11 . we define classes @xmath41,@xmath42 of representations over @xmath40 such that @xmath43 is in @xmath41 @xmath17resp . @xmath44 if and only if for each arrow @xmath45 of @xmath40 the linear map @xmath46 is injective @xmath17resp . + let @xmath47 be the push - down functor @xcite and @xmath43 be indecomposable . we prove that @xmath48 is in @xmath41 @xmath17resp . @xmath44 if and only if @xmath49 is in @xmath27 @xmath17resp . @xmath50 . since a component @xmath51 of the auslander - reiten quiver of @xmath40 which is taken to a regular component @xmath52 is also of type @xmath16 , we can lift the definition of @xmath35 to @xmath51 . we define @xmath53 as the distance between the cones @xmath54 and @xmath55 and show that @xmath56 . + for @xmath57 , we denote by @xmath58 its quasi - length . if @xmath20 has certain properties , we show how to construct a short exact sequence @xmath59 with indecomposable middle term @xmath60 in a component @xmath61 such that @xmath62 the construction of @xmath63 relies on the fact that @xmath40 is an infinite @xmath64-regular tree with bipartite orientation . using @xmath65 , we construct for each @xmath38 a component @xmath66 with @xmath67 . + in conjunction with a natural action of @xmath68 on @xmath69 , we arrive at our main theorem : let @xmath7 . there is a bijection @xmath70 . as an immediate consequence we get the following statements , which are generalizations of results by kerner and lukas @xcite , @xcite for the kronecker algebra . let @xmath1 , then for each @xmath38 there are exactly @xmath71 regular components with quasi - rank in @xmath72 . assume that @xmath9 is uncountable and @xmath73 . the set of components of quasi - rank @xmath74 is uncountable . throughout this article let @xmath9 be an algebraically closed field of arbitrary characteristic . for a quiver @xmath75 , @xmath76 let @xmath77 moreover let @xmath78 . if @xmath79 , then by definition @xmath80 and @xmath81 . we say that @xmath82 starts in @xmath83 and ends in @xmath84 . a quiver @xmath85 is called 1 . locally finite if @xmath86 is finite for each @xmath76 , 2 . of bounded length if for each @xmath76 there is @xmath87 such that each path in @xmath85 which starts or ends in @xmath88 is of length @xmath89 , 3 . locally bounded if @xmath85 is locally finite and of bounded length . from now on we assume that @xmath85 is locally bounded . note that this implies that @xmath85 is acyclic . moreover , we assume that @xmath85 is connected . a finite dimensional representation @xmath90 over @xmath85 consists of vector spaces @xmath91 and linear maps @xmath92 such that @xmath93 is finite . a morphism @xmath94 between representations is a collection of linear maps @xmath95 such for each arrow @xmath79 there is a commutative diagram @xmath96 \matrix [ matrix of math nodes , row sep=3em , column sep=3.0em , text height=2.0ex , text depth=0.25ex ] ( s ) { & |[name = a_1]| m_x & |[name = b_1]| m_y \\ & |[name = a_2]| n_x & |[name = b_2]| n_y . \\ } ; \draw[- > ] ( a_1 ) edge node[auto ] { $ m(\alpha)$ } ( b_1 ) ( a_2 ) edge node[auto ] { $ n(\alpha)$ } ( b_2 ) ( a_1 ) edge node[auto ] { $ f_x$ } ( a_2 ) ( b_1 ) edge node[auto ] { $ f_y$ } ( b_2 ) ; \end{tikzpicture}\ ] ] the category of finite dimensional representations over @xmath85 is denoted by @xmath97 . the path category @xmath98 has @xmath99 as set of objects and @xmath100 is the vector space with basis given by the paths from @xmath88 to @xmath101 . the trivial arrow in @xmath88 is denoted by @xmath102 . since @xmath85 is locally bounded , the category @xmath98 is locally bounded in the sense of @xcite . a finite - dimensional module over a locally bounded category @xmath103 is a functor @xmath104 such that @xmath105 is finite . the category @xmath106 has auslander - reiten sequences @xmath17see @xcite@xmath18 . since a finite dimensional module over @xmath98 is the same as a representation of @xmath85 , the category @xmath97 has auslander - reiten sequences . + if moreover @xmath99 is a finite set , we denote with @xmath107 the path algebra of @xmath85 with idempotents @xmath108 . in this case @xmath107 is a finite dimensional , associative , basic and connected @xmath9-algebra . we denote by @xmath109 the class of finite - dimensional @xmath107 left modules . given @xmath110 we let @xmath111 . the categories @xmath109 and @xmath97 are equivalent @xmath17see for example ( * ? ? ? * iii 1.6)@xmath18 . we will therefore switch freely between representations of @xmath85 and modules of @xmath107 , if one of the approaches seems more convenient for us . we assume that the reader is familiar with auslander - reiten theory and basic results on wild hereditary algebras . for a well written survey on the subjects we refer to @xcite , @xcite and @xcite . recall the definition of the dimension function @xmath112 if @xmath113 is an exact sequence , then @xmath114 . a finite quiver @xmath85 defines a @xmath17non - symmetric ) bilinear form @xmath115 given by @xmath116 , which coincides with the euler - ringel form @xcite on the grothendieck group @xmath117 , i.e. for @xmath118 we have @xmath119 we always assume that @xmath1 . denote by @xmath11 the @xmath64-kronecker quiver , which is given by two vertices @xmath12 and arrows @xmath14 . = [ ] = [ ] \(1 ) @xmath120 ; ( 2 ) [ right of=1 ] @xmath120 ; at ( 1.4,0.7 ) @xmath121 ; at ( 1.4,-0.7 ) @xmath122 ; at ( 1.4,-0.2 ) @xmath123 ; at ( 1.4,0.4 ) @xmath123 ; at ( 0,-0.3 ) @xmath124 ; at ( 3,-0.3 ) @xmath125 ; \(1 ) edge[bend right ] node [ left ] ( 2 ) ( 1 ) edge[bend left ] node [ left ] ( 2 ) ( 1 ) edge [ ] node [ left ] ( 2 ) ; we set @xmath126 and @xmath127 , @xmath128 . the modules @xmath129 and @xmath130 are the indecomposable projective modules of @xmath15 , @xmath131 and @xmath132 . as figure @xmath133 suggests , we write @xmath134 . for example @xmath135 and @xmath136 . , scaledwidth=70.0%,height=50 ] figure @xmath137 shows the notation we use for the components @xmath138 in the auslander - reiten quiver of @xmath3 which contain the indecomposable projective modules @xmath139 and indecomposable injective modules @xmath140 . the set of all other components is denoted by @xmath141 . ringel has proven @xcite that every component in @xmath141 is of type @xmath16 . a module in such a component is called regular and the class of all regular indecomposable modules is denoted by @xmath142 . an irreducible morphism in a component of type @xmath16 @xmath17for any algebra@xmath18 is injective if the corresponding arrow is uprising and surjective otherwise . a representation @xmath48 in a @xmath16 component is called quasi - simple if the ar sequence terminating in m has an indecomposable middle term . these modules are the modules in the bottom layer of the component . if @xmath48 is quasi - simple in a component @xmath2 of type @xmath16 , then there is an infinite chain @xmath17ray@xmath18 of irreducible monomorphisms @xmath17resp . epimorphisms@xmath18 @xmath143 \to m[2 ] \to m[3 ] \to \cdots \to m[l ] \to \cdots\ ] ] @xmath144 and for each indecomposable module @xmath20 in @xmath2 there are unique quasi - simple modules @xmath145 and @xmath146 with @xmath147 $ ] . the number @xmath148 is called the quasi - length of @xmath20 . + the indecomposable modules in @xmath149 are called preprojective modules and the modules in @xmath150 are called preinjective modules . moreover we call an arbitrary module preprojective @xmath17resp . preinjective , regular@xmath18 if all its indecomposable direct summands are preprojective @xmath17res . preinjective , regular@xmath18 . we have @xmath151 in @xmath149 @xmath152 in @xmath153 if and only if there is @xmath154 with @xmath155 @xmath156 for @xmath157 . let @xmath158 be the subcategory of all modules without non - zero projective direct summands and @xmath159 the subcategory of all modules without non - zero injective summands . since @xmath3 is a hereditary algebra , the auslander - reiten translation @xmath160 induces an equivalence from @xmath158 to @xmath159 , that will be often used later on without further notice . let us start by recalling the definitions of @xmath21 and @xmath35 . let @xmath161 be a wild hereditary algebra and @xmath2 be a regular component with a quasi - simple module @xmath20 in @xmath2 , then @xmath162 it was shown in @xcite , that @xmath163 is finite and @xmath164 . in other words , there are lots of morphisms in @xmath165-direction . also in @xmath166-direction one finds a lot of morphisms since @xmath167 @xmath17see @xcite@xmath18 . another invariant that can be attached to a regular component @xmath2 of the kronecker algebra is defined in @xcite and denoted by @xmath35 . in order to define @xmath35 , we first recall the definitions of @xmath27 and @xmath28 . for @xmath168 and @xmath169 we consider the @xmath9-linear map @xmath170 and let @xmath171 be the cokernel of the embedding @xmath172 where @xmath173 , @xmath174 . @xcite we define the classes of representations with the equal kernels property and with the equal images property as 1 . @xmath175 and 2 . @xmath176 given a regular component @xmath2 , there exist uniquely determined quasi - simple representations @xmath29 and @xmath30 in @xmath2 such that @xmath177 and @xmath178 @xmath17see @xcite@xmath18 . now @xmath35 is defined as the unique integer with @xmath179 . since @xmath180 it follows @xmath181 . let @xmath2 be a regular component , then @xmath182 . on the one hand let @xmath183 , then there is @xmath184 with @xmath185 @xcite . the auslander - reiten formula @xcite yields @xmath186 . on the other hand let @xmath187 , then there is @xmath188 with @xmath189 @xcite . by the euler - ringel form we have @xmath190 hence @xmath191 . since @xmath192 and @xmath193 are elementary @xcite , we get a non - zero morphism by @xcite @xmath194 @xmath195 hence @xmath196 , since @xcite together with @xmath197 implies that @xmath30 is not a brick . by @xcite it follows that @xmath198 . the second inequality follows from the proof of @xcite . in @xcite , the inequality @xmath199 in conjunction with @xmath200 was used to prove that @xmath201 . we choose a different approach and study the number @xmath35 to draw conclusions for @xmath21 . there are regular components with @xmath202 and @xmath203 for @xmath204 @xcite and a component @xmath51 with @xmath205 and @xmath206 @xcite . we follow @xcite and @xcite and consider the universal cover @xmath40 of the quiver @xmath11 . the underlying graph of @xmath40 is an @xmath64-regular tree and @xmath40 has bipartite orientation . that means each vertex @xmath207 is a sink or a source and @xmath208 . in the following we recall the construction of @xmath40 . + for a quiver @xmath75 with arrow set @xmath209 we write @xmath210 for the formal inverses of @xmath209 . moreover we extend the functions @xmath211 and @xmath212 to @xmath213 by defining @xmath214 and @xmath215 . a walk @xmath216 in @xmath209 is a formal sequence @xmath217 with @xmath218 , @xmath219 such that @xmath220 for all @xmath221 , where @xmath222 for all @xmath223 . we set @xmath224 and @xmath225 . let @xmath226 be the equivalence relation on the set of walks @xmath227 of @xmath11 generated by @xmath228 let @xmath229 be the involution on @xmath227 given by @xmath230 . now consider the fundamental group @xmath231 of @xmath11 in the point @xmath124 , i.e. the elements of @xmath231 are the equivalence classes of unoriented paths starting and ending in @xmath124 , with multiplication given by concatenation of paths , inverse elements @xmath232^{-1 } : = [ w^{-1}]$ ] and identity element @xmath233 $ ] . note that @xmath231 is a free group in the @xmath234 generators @xmath235 \mid 2 \leq j \leq r\}$ ] and in particular torsionfree . + the quiver @xmath40 is given by the following data : 1 . @xmath236 is the set of equivalence classes of paths starting in @xmath124 . 2 . there is an arrow from @xmath232 $ ] to @xmath237 $ ] whenever @xmath238 for some @xmath239 . = [ sibling angle=90 , level distance=2.6 cm ] = [ sibling angle=90 , level distance=1.2 cm ] = [ sibling angle=90 , level distance=0.9 cm ] = [ ] = [ ] child [ color = black ] foreach in @xmath120,@xmath120,@xmath120,@xmath120 node edge from parent[- > ] child [ color = black ] foreach in @xmath120,@xmath120,@xmath120 node edge from parent[<- ] child [ color = black ] foreach in @xmath120,@xmath120,@xmath120 node edge from parent[- > ] ; let @xmath240 be the quiver morphism given by @xmath232 \mapsto t(w)$ ] and @xmath241 \to [ \gamma_i w ] ) \mapsto \gamma_i$ ] . the morphisms @xmath242 is a @xmath243-galois cover for @xmath244 , where the action of @xmath243 on @xmath40 is given by concatenation of paths : if @xmath245 \in \pi(\gamma_r)$ ] and @xmath246,[u ] \in ( c_r)_0 $ ] with arrow @xmath247 \to [ \gamma_i u]$ ] then @xmath248 = [ v w^{-1 } ] \ \text{and}\ ] ] @xmath249 \to [ \gamma_i u ] ) = ( [ u w^{-1 } ] \to [ \gamma_i u w^{-1}]).\ ] ] we define @xmath250 , @xmath251 and get an induced action on @xmath252 by shifting the support of representations via @xmath244 : given @xmath43 and @xmath253 we define @xmath254 , where @xmath255 by identifying the orbit quiver @xmath256 with @xmath11 we define the push - down functor @xmath47 on the objects via @xmath257 , where @xmath258 @xmath259 if @xmath260 is a morphism in @xmath252 then @xmath261 with @xmath262 by @xcite @xmath263 is an exact functor . @xcite , @xcite[theoremringelgabriel ] the following statements hold . 1 . @xmath263 sends indecomposable representations in @xmath252 to indecomposable representations in @xmath69 . if @xmath43 is indecomposable then @xmath264 if and only if @xmath265 for some @xmath253 . 3 . @xmath263 sends @xmath266 sequences to @xmath266 sequences and @xmath263 commutes with the auslander - reiten translates , i.e. @xmath267 . if @xmath43 is indecomposable in a component @xmath51 with @xmath49 in a component @xmath2 , then @xmath263 induces a covering @xmath268 of translation quivers . let @xmath43 be indecomposable . @xmath48 is called regular if @xmath49 is regular . a component @xmath51 of @xmath252 is called regular if it contains a regular representation @xmath48 . in this case we denote by @xmath269 the component containing @xmath49 . moreover we let @xmath270 be the set of all regular components of the auslander - reiten quiver of @xmath252 . let @xmath51 be regular component , then the covering @xmath271 is an isomorphism of translation quivers . in particular , @xmath51 is of type @xmath16 . moreover a component @xmath61 is regular if and only if @xmath61 is of type @xmath16 . by @xcite @xmath272 is a simply connected translation quiver . by @xcite,@xcite the quiver morphism @xmath273 is an isomorphism . if @xmath274 then there exists a vertex @xmath88 with @xmath1 successors @xmath17see figure @xmath275 , since a covering is surjective on arrows . hence @xmath61 is not of type @xmath16 . recall @xcite that the duality @xmath276 is defined by setting @xmath277 and @xmath278 , where @xmath279 is the involution with @xmath280 . + we define an involution @xmath281 via @xmath232 \mapsto [ \overline{w } \gamma_1]$ ] , where @xmath282 , @xmath283 and @xmath284 . this induces a quiver anti - morphism @xmath285 in the following way . + if @xmath232 \to [ \gamma_i w]$ ] is an arrow of @xmath40 , then by definition there is a unique arrow @xmath286 \to [ \gamma_i w])$ ] starting in @xmath287 ) = [ \gamma^{-1}_i \overline{w } \gamma_1]$ ] and ending in @xmath288 ) = [ \overline{w } \gamma_1]$ ] , since @xmath289 . note that @xmath290 = @xmath291 , @xmath292 and @xmath293 . + we define a duality @xmath294 by setting @xmath295 where @xmath296 and @xmath297 . by construction we have @xmath298 . in the following we give a characterization of the equal images and equal kernels property for indecomposable representations of the form @xmath49 . let @xmath299 be the unique map with @xmath300 for all @xmath301 . note that if @xmath302or @xmath303 then the restriction of @xmath304 to @xmath305 @xmath17resp . @xmath306 is a bijective map to @xmath307 . let @xmath308 be a set of vertices and @xmath309 be a tree . 1 . the unique minimal tree containing @xmath20 is denoted by @xmath310 . a vertex @xmath311 is called a leaf of @xmath312 , if @xmath313 . let @xmath207 and @xmath48 be a representation of @xmath40 . 1 . the set @xmath314 is called the support of @xmath48 . 2 . for @xmath315 we let @xmath316 be the induced representation with @xmath317 . 3 . the vertex @xmath88 is a leaf of @xmath48 if @xmath88 is a leaf of @xmath318 . @xmath48 is called balanced provided that @xmath48 is indecomposable and @xmath48 has leaves in @xmath319 and @xmath291 . observe that if @xmath48 is indecomposable we have @xmath320 . we define @xmath321 @xmath322 [ propositiongeneral ] let @xmath43 be an indecomposable representation . the following statements are equivalent : 1 . @xmath323 has the equal kernels property . @xmath324 is injective for all @xmath239 . 3 . @xmath325 . @xmath326 : clear from the definition of @xmath27 . + @xmath327 : let @xmath79 and @xmath328 . denote by @xmath329 the natural embedding and let @xmath330 . we conclude @xmath331 \circ \iota_x(m_x ) = ( n(\gamma_{\overline{\alpha } } ) \circ \iota_x)(m_x).\ ] ] hence @xmath332 and @xmath333 . + @xmath334 : let @xmath335 , @xmath336 and @xmath337 . we assume w.l.o.g that @xmath338 . write @xmath339 . we have to show that @xmath340 for all @xmath341 . + let @xmath342 and suppose that @xmath343 . let @xmath344 be the minimal tree that contains @xmath345 . since @xmath346 every leaf belongs to @xmath347 , by the minimality of @xmath344 . fix a leaf @xmath88 in @xmath344 and let @xmath348 be the unique arrow with @xmath349 . then @xmath350 by the injectivity of @xmath351 there is @xmath352 with @xmath353 and @xmath354 . it follows @xmath355 and @xmath356 . since @xmath40 is a tree we get @xmath357 . + since @xmath358 , we assume without loss of generality that @xmath359 , so that @xmath360 . let @xmath361 be the unique arrow with @xmath362 . then @xmath363 hence there is @xmath364 with @xmath365 and @xmath366 are in @xmath367 . we have shown that @xmath368 and @xmath101 are in @xmath367 . this is a contradiction since @xmath88 is a leaf . [ corollarychareip ] let @xmath43 be an indecomposable representation . the following statements are equivalent . 1 . @xmath323 has the equal images property . @xmath324 is surjective for all @xmath239 . 3 . @xmath369 . @xmath334 : let @xmath370 be surjective for each @xmath371 , then @xmath372 is injective for each @xmath371 . by @xmath373 , the representation @xmath374 has the equal kernels property . therefore @xmath375 has the equal images property , since @xmath376 . let @xmath377 and @xmath52 . then there exist uniquely determined quasi - simple representations @xmath378 and @xmath379 in @xmath51 such that 1 . @xmath380 and @xmath381 . 2 . @xmath382 and @xmath383 . the unique integer @xmath384 with @xmath385 is given by @xmath386 . assume that @xmath43 is balanced , then @xmath48 is regular . if an indecomposable representation @xmath387 is in @xmath41 @xmath17respectively @xmath44 , then all leaves of @xmath20 are in @xmath319 @xmath17resp . @xmath388 . hence @xmath389 by @xmath373 . by @xcite the representations of the components @xmath149 and @xmath150 are contained in @xmath390 . from now on we write @xmath391 \in ( c_r)_0 $ ] for the vertex in @xmath236 given by the trivial walk starting in the vertex @xmath124 . for @xmath239 , we let @xmath392 be the unique arrow with @xmath393 and @xmath394 . moreover let @xmath395 . we define an indecomposable representation @xmath396 in @xmath252 via : @xmath397 and @xmath398 for all @xmath399 . by definition we have @xmath400 and @xmath401 . in view of @xcite , @xcite ) @xmath402 and @xmath373 we conclude the following . [ corollaryeipekp ] let @xmath43 be an indecomposable representation . the following statements are equivalent : 1 . @xmath323 has the equal kernels property . @xmath403 for all @xmath239 . @xmath404 for all @xmath239 and all @xmath253 . the following lemma will be needed later on . [ lemmainjective ] let @xmath38 , @xmath48 be regular indecomposable , @xmath239 and @xmath405 . 1 . for @xmath406 , the linear map @xmath407 is surjective for all @xmath371 . if @xmath408 and @xmath409 is a non - zero morphism , then each @xmath410 is injective . 3 . if @xmath406 , @xmath411 and @xmath412 , then @xmath413 . this means @xmath414 . @xmath415 for @xmath406 , we have @xmath416 @xmath17see @xcite@xmath18 . + @xmath417 for @xmath408 , it is known that each proper factor of @xmath418 is preinjective , see for example @xcite . let @xmath419 , then @xmath420 is injective , where @xmath421 so each @xmath410 is injective , since @xmath422 . + @xmath423 let @xmath411 and @xmath424 . since @xmath88 is a sink , there is @xmath425 and by @xmath415 , @xmath426 is surjective . hence @xmath427 . by @xmath417 , we get @xmath428 and @xmath429 . let us recall what we have shown so far . given a component @xmath51 in @xmath430 , the natural number @xmath384 is the distance between the two non - empty , non - intersecting cones @xmath55 and @xmath54 . moreover we know that @xmath431 . + let now @xmath57 be indecomposable . then there exists an integer @xmath432 such that @xmath433 , since @xmath54 is non - empty . we also find @xmath408 with @xmath434 . since @xmath54 is closed under @xmath435 we conclude @xmath436 . therefore the following minima exist . let @xmath387 be a regular indecomposable representation . we define @xmath437 note that @xmath438 is the distance of @xmath20 to the border of the cone @xmath55 . let @xmath387 be indecomposable in a regular component @xmath51 . then @xmath439 since the equality is obvious for @xmath20 quasi - simple we assume @xmath440 . let @xmath441 be the unique quasi - simple representation with @xmath442 $ ] . by induction we get @xmath443 ) + d^-(z[l-1 ] ) - ( l-1)$ ] . now observe that @xmath444 ) + 1 = d^+(z[l])$ ] and @xmath445 ) = d^-(z[l])$ ] . hence @xmath446 ) - 1 + d^-(z[l ] ) - ( l-1 ) = d^+(x ) + d^-(x ) - \operatorname{ql}(x)$ ] . we show in the following how to modify a regular representation @xmath447 such that the obtained representation @xmath448 is regular , @xmath449 , @xmath450 and @xmath451 . in this section , we show how to construct non - split exact sequences with indecomposable middle term in @xmath252 . let @xmath452 be indecomposable . the pair @xmath453 is called leaf - connected if there is @xmath454 s.t . 1 . @xmath88 is a leaf of @xmath48 , @xmath101 is a leaf of @xmath455 and 2 . @xmath456 . note that the assumption @xmath48 and @xmath455 being indecomposable together with properties @xmath415 and @xmath417 already implies the uniqueness of @xmath82 . if @xmath453 is leaf - connected , @xmath82 is called the connecting arrow and @xmath457 is not leaf - connected . let @xmath453 be leaf - connected with connecting arrow @xmath79 and @xmath458 a non - zero linear map . we define a representation @xmath459 by setting @xmath460 with @xmath461 for @xmath462 and @xmath463 . moreover we denote by @xmath464 and @xmath465 the natural morphisms of quiver representations . the @xmath9-linear map along the connecting arrow is called a connecting map for @xmath453 . note that @xmath466 is just an extension of @xmath48 by @xmath455 with corresponding exact sequence @xmath467 . the next lemma shows that @xmath468 is indecomposable . in particular , @xmath469 does not split . one can also show that the map @xmath470,\ ] ] is an isomorphism of vector spaces , where @xmath471 $ ] is the neutral element in the abelian group @xmath472 with respect to the baer sum . [ lemmaproperties ] let @xmath453 be leaf - connected . 1 . the representation @xmath466 is indecomposable . 2 . if @xmath48 and @xmath455 are regular , then @xmath466 is regular . @xmath415 let @xmath473 with @xmath474 . hence we get @xmath475 since @xmath43 is indecomposable , there is a unique @xmath157 with @xmath476 . we assume @xmath477 . by the same token there is a unique @xmath478 with @xmath479 . since @xmath480 , it follows that @xmath481 . hence @xmath482 and @xmath483 is indecomposable . + @xmath417 by construction @xmath455 is a subrepresentation of @xmath468 and @xmath48 a factor representation . since @xmath484 is indecomposable with regular factor representation @xmath49 , it is regular itself or preprojective . by the same token @xmath484 is regular or preinjective . [ lemmaconnected ] let @xmath485 be indecomposable and assume @xmath486 is a leaf of @xmath48 and @xmath487 is a leaf of @xmath455 . then there exists @xmath253 such that @xmath488 is leaf - connected with connecting arrow @xmath489 . since @xmath490 are leaves , we find at most one arrow @xmath491 and at most one arrow @xmath492 with @xmath493 and @xmath494 . since @xmath1 we find an arrow @xmath495 with @xmath496 . in particular @xmath497 and @xmath498 . + now let @xmath499 be the unique element with @xmath500 . then @xmath424 is a leaf of @xmath501 . by construction , we have @xmath502 and conclude @xmath503 . it follows that @xmath504 . hence @xmath488 is leaf - connected with connecting arrow @xmath82 . let @xmath406 and @xmath505 be indecomposable . the tuple @xmath506 is called leaf - connected , provided that @xmath507 is leaf - connected for all @xmath508 . a tuple @xmath509 of @xmath9-linear maps is called a connecting map for @xmath506 if @xmath510 is a connecting map for @xmath511 for all @xmath508 . let @xmath512 and @xmath513 be leaf - connected with connecting maps @xmath514 and @xmath515 , then @xmath456 , since @xmath40 is a tree . hence @xmath516 and @xmath517 is connected . therefore the next definition is well - defined . let @xmath406 , @xmath506 be leaf - connected with connecting map @xmath509 . then we define inductively @xmath518 for all @xmath519 . from now on , we assume that @xmath506 is leaf - connected with connecting map @xmath509 . for @xmath520 , we define @xmath521 and @xmath522 . moreover , we set @xmath523 and @xmath524 . [ lemmaproperties1 ] let @xmath406 , @xmath506 be leaf - connected and @xmath508 . 1 . the representation @xmath525 is indecomposable . there is a short exact sequence @xmath526 . if @xmath527 are regular , then @xmath525 is regular . for @xmath423 , just note that @xmath528 is balanced for @xmath529 , hence @xmath528 is regular . [ lemmatau ] let @xmath530 be regular indecomposable . then the following statments are equivalent . 1 . there is @xmath253 such that @xmath531 . 2 . there is @xmath532 such that @xmath533 . 3 . there is @xmath534 such that @xmath535 . we only show @xmath326 . let @xmath253 , such that @xmath536 . then @xmath537 hence we find @xmath532 such that @xmath538 . [ corollaryproperties ] let @xmath406 , @xmath506 be leaf - connected and @xmath539 regular , then 1 . @xmath540 . 2 . @xmath541 . note that we have a filtration of @xmath542 by regular subrepresentations @xmath543 with @xmath544 regular for all @xmath520 , where @xmath545 . + @xmath415 let @xmath546 and @xmath547 . consider the short exact sequence @xmath548 now @xmath432 such that @xmath549 . then there exists @xmath239 and @xmath253 such that @xmath550 . hence we find @xmath532 with @xmath551 for some @xmath532 . left - exactness of @xmath552 ensures that @xmath553 and therefore we find @xmath253 with @xmath554 . hence @xmath555 and therefore @xmath556 . + if @xmath557 for some @xmath253 and @xmath239 , then we get find @xmath532 with @xmath558 . by @xcite we find @xmath559 with @xmath560 . hence there is @xmath253 with @xmath561 . hence @xmath562 . + @xmath417 note for @xmath563 that @xmath564 . hence @xmath565 has a filtration @xmath566 now apply @xmath415 and note that @xmath567 for each regular indecomposable representation @xmath20 . a balanced representation @xmath455 is called small if @xmath568 . note that @xmath455 being balanced always implies @xmath569 and @xmath570 . [ definitionmetric ] denote with @xmath571 the underlying graph of @xmath40 . then @xmath572 obtains the structure of a metric space , where @xmath573 denotes the length of the unique path in @xmath571 connecting vertices @xmath88 and @xmath101 . let @xmath309 be a finite subtree . @xmath312 is called small if 1 . @xmath312 has leaves in @xmath319 and @xmath291 , 2 . for all @xmath311 , we have @xmath574 , 3 . if @xmath575 then @xmath576 or @xmath577 . let @xmath146 and @xmath578 with @xmath579 . we denote with @xmath580 a @xmath17small@xmath18 subtree of the following form : [ every node/.style = fill , circle , inner sep = 1pt ] in 1 , ... , at ( , 0 ) @xmath120 ; at ( , -0.3 ) @xmath581 ; in 1 , ... , ( 2*+ 0.2,0 ) to ( 2*+ 1 - 0.2,0 ) ; ( 2 * - 0.2,0 ) to ( 2 * - 1 + 0.1,0 ) ; at ( 4 * 1 - 1,1 ) @xmath120 ; at ( 4 * 2 - 1,1 ) @xmath120 ; at ( 4 * 3 - 0.2,1 ) @xmath120 ; at ( 4 * 1 - 1,1 + 0.3 ) @xmath582 ; at ( 4 * 2 - 1,1 + 0.3 ) @xmath583 ; at ( 4 * 3 - 0.2,1 + 0.3 ) @xmath584 ; at ( + 1.8,-0.3 ) @xmath585 ; at ( + 2.8,-0.3 ) @xmath586 ; at ( 16.6,-0.3 ) @xmath587 ; ( 4 * 1 - 1,1 - 0.2 ) to ( 4 * 1 - 1,0 + 0.2 ) ; ( 4 * 2 - 1,1 - 0.2 ) to ( 4 * 2 - 1,0 + 0.2 ) ; ( 11 + 0.8,1 - 0.2 ) to ( 11 + 0.8,0 + 0.2 ) ; at ( + 0.4,0 ) ... ; at ( + 0.8,0 ) @xmath120 ; at ( + 1.8,0 ) @xmath120 ; at ( + 2.8,0 ) @xmath120 ; at ( + 3.8,0 ) @xmath120 ; ( 2 * 5 + 0.6 , 0 ) to ( 2 * 5,0 ) ; ( 2 * 5 + 1 , 0 ) to ( 2 * 5 + 1.6,0 ) ; ( 2 * 5 + 2.6 , 0 ) to ( 2 * 5 + 2,0 ) ; at ( 13.2,0 ) ... ; at ( 13.6,0 ) @xmath120 ; at ( 14.6,0 ) @xmath120 ; at ( 15.6,0 ) @xmath120 ; at ( 16.6,0 ) @xmath120 ; ( 14.4,0 ) to ( 13.8,0 ) ; ( 14.8,0 ) to ( 15.4,0 ) ; ( 16.4,0 ) to ( 15.8,0 ) ; assume @xmath588 is an indecomposable representation with small tree @xmath589 . then @xmath588 is small . since @xmath589 is small with @xmath590 , @xmath588 is balanced and therefore regular . + let @xmath432 be such that @xmath591 . by @xmath592 and @xmath593 we find @xmath239 and @xmath253 with @xmath594 . fix @xmath595 non - zero . + we assume that @xmath596 . by @xmath597 , @xmath598 is a monomorphism and @xmath599 . let @xmath211 be a sink of @xmath600 . by @xmath597 , we have @xmath601 . since @xmath589 is small we get @xmath602 . hence @xmath603 now let @xmath604 be sinks . then @xmath65 yields @xmath605 since @xmath589 is small we get @xmath606 or @xmath607 . since @xmath40 has bipartite orientation and @xmath608 is connected , it follows @xmath606 . hence @xmath608 contains exactly one sink @xmath211 . write @xmath609 . since @xmath610 , @xmath611 implies @xmath612 . hence @xmath613 can be considered as a representation of the dynkin diagram @xmath614 with unique sink @xmath211 such that all linear maps are surjective . it follows that @xmath615 for all @xmath616 . hence @xmath617 and @xmath618 is indecomposable . but the only indecomposable representation @xmath619 with dimension vector @xmath620 is injective . this is a contradiction since @xmath441 is regular . therefore @xmath621 and @xmath622 . for the other inequality note that @xmath589 is small if and only if @xmath623 is small and @xmath624 . [ lemmabochen ] let @xmath169 be an indecomposable representation with dimension vector @xmath625 , @xmath626 . then @xmath48 is a regular and quasi - simple representation . that @xmath48 is regular follows immediatly from @xcite . by @xcite it suffices to show that @xmath627 is not a common divisor of @xmath628 and @xmath368 for all @xmath629 , where @xmath630 is the dimension vector of the preprojective indecomposable representation @xmath631 @xmath17see figure @xmath275 . but this is trivial since @xmath632 . [ theoremmaintheorem ] let @xmath43 be balanced in the regular component @xmath51 and @xmath633 . there is @xmath634 such that for each @xmath635 there is a regular component @xmath66 with @xmath636 moreover @xmath66 contains a balanced quasi - simple representation @xmath637 with @xmath638 or @xmath639 and @xmath640 for @xmath641 . write @xmath642 . after dualising @xmath48 we can assume @xmath643 . set @xmath644 and @xmath645 . now let @xmath646 with @xmath647 . consider an indecomposable and thin representation @xmath588 @xmath17i.e . @xmath648 for all @xmath207 such that @xmath649 is leaf - connected and @xmath650 is of type @xmath651 . let @xmath253 be such that @xmath652 is leaf - connected . we conclude for @xmath653 , @xmath654 , and @xmath655 that @xmath656 by @xmath657 @xmath637 is a regular indecomposable representation and by @xmath658 @xmath659 is quasi - simple . therefore @xmath637 is quasi - simple in a regular component @xmath66 . we conclude with @xmath660 @xmath661 i.e. @xmath662 . by the same token we have @xmath663 and conclude @xmath664 it follows immediatly from the construction that the regular components are pairwise distinct , since @xmath665 , @xmath666 are non - isomorphic and satisfy @xmath667 and @xmath668 for @xmath641 . [ corollaryspecialmodule ] let @xmath669 be balanced and @xmath633 . then there exists a balanced and quasi - simple representation @xmath670 in a regular component @xmath61 such that @xmath671 . moreover there is a leaf @xmath486 with @xmath672 for the unique element @xmath673 . fix @xmath635 in the proof of the theorem and set @xmath674 . the last claim follows since @xmath588 is a thin representation of type @xmath675 which has @xmath676 leaves in @xmath319 @xmath17see figure @xmath677 . of type @xmath678,scaledwidth=90.0%,height=75 ] the aim of this section is to construct for each @xmath38 a regular component @xmath51 with @xmath679 . although each indecomposable representation has a leaf , it is in general not true that a regular representation has leaves in @xmath319 and @xmath291 . for example if @xmath48 is indecomposable in @xmath41 , then each leaf of @xmath48 is a sink by @xmath373 . the next results shows that self - dual representations have leaves in @xmath319 and @xmath291 . [ lemmadual ] let @xmath43 be indecomposable such that @xmath680 , then @xmath48 is balanced . since @xmath681 is finite there exists a leaf @xmath88 of @xmath48 . without loss of generality we assume that @xmath486 . we get @xmath682 . therefore we find @xmath532 such that @xmath683 since @xmath684 @xmath17see @xmath685 is a leaf of @xmath686 the claim follows . we denote with @xmath687 the quiver obtained by changing the orientation of all arrows in @xmath40 . note that @xmath688 and @xmath689 . we denote by @xmath690 the composition of the bernstein - gelfand - ponomarev reflection functors ( * ? ? ? * vii 5.5 . ) for all the sources of @xmath40 . @xmath690 is a well - defined functor @xmath691 @xmath17see @xcite@xmath18 . by the same token we have a functor @xmath692 given by the composition of the reflection functors for all the sources of @xmath687 . then @xmath693 satisfies @xmath694 for @xmath43 indecomposable and non - injective @xcite , ( * ? ? ? * vii 5.8 . ) . therefore statements @xmath415 and @xmath417 of the next lemma follow immediately from the definition of the reflection functors . [ lemmaleavesexistence ] let @xmath48 be in @xmath252 indecomposable and not injective . 1 . for each @xmath486 we have @xmath695 2 . for each @xmath487 we have @xmath696 3 . let @xmath113 be an almost split sequence with @xmath697 indecomposable . if @xmath698 is a leaf of @xmath103 , then @xmath697 has a leaf in @xmath291 . 4 . let @xmath113 be an almost split sequence with @xmath697 indecomposable . if @xmath699 is a leaf of @xmath103 and @xmath700 satisfies @xmath701 , then @xmath368 is a leaf of @xmath697 . @xmath423 consider a path @xmath702 such that @xmath703 are not in @xmath704 as illustrated in figure @xmath705 . since @xmath706 is in @xmath319 we get with @xmath415 that @xmath707 now let @xmath708 . then @xmath709 . hence we get that @xmath710 hence @xmath711 is a leaf of @xmath712 and since @xmath713 we conclude that @xmath714 is a leaf of @xmath697 . + .,scaledwidth=35.0%,height=55 ] @xmath715 application of @xmath415 yields @xmath17see figure @xmath716 @xmath717 now fix @xmath718 , then @xmath714 . let @xmath719 then @xmath720 for all @xmath721 , since @xmath722 . hence we get @xmath723 we conclude @xmath724 we have shown that @xmath725 . since @xmath726 we get @xmath727 . since @xmath728 the vertex @xmath699 is a leaf of @xmath729 . since @xmath697 is indecomposable we have @xmath730 . .,scaledwidth=32.0%,height=60 ] let @xmath731 be regular with @xmath732 . we define inductively a sequence of indecomposable representations in the regular component @xmath733 of @xmath734 . the representation @xmath734 is quasi - simple . assume that @xmath735 is already defined . if @xmath736 is odd , then @xmath737 is the unique indecomposable representation with irreducible epimorphism @xmath738 ; if @xmath736 is even , then @xmath737 is the unique indecomposable representation with irreducible monomorphism @xmath739 . the component @xmath733 is shown in figure @xmath740 . we have @xmath741 ( * ? ? ? * example 3.3 ) . [ theoremmaintheorem2 ] let @xmath1 . 1 . for each @xmath742 there is @xmath743 and a family @xmath744 of regular components with @xmath745 and @xmath66 contains a quasi - simple representation @xmath746 with @xmath747 . 2 . @xmath748 . 3 . @xmath749 . recall that @xmath750 and @xmath751 for the regular component @xmath752 containing @xmath753 . fix @xmath754 . then we have @xmath755 by ( * ? ? * example 3.3 ) . by @xmath756 @xmath757 is balanced . moreover we have @xmath758 . hence @xmath759 yields @xmath634 and an infinite family of components @xmath760 of width @xmath761 . by corollary @xmath762 we find a balanced and quasi - simple representation @xmath763 that satisfies the assumption of @xmath764 . consider the ar sequence @xmath113 . then @xmath697 is balanced by @xmath765 , @xmath766 and @xmath767 . by @xmath759 we get an infinite family of components of width @xmath768 . this proves @xmath415 and @xmath417 . for @xmath423 observe that there exist regular components @xmath769 with @xmath770 @xcite . this section is motivated by the following result by kerner and lukas . @xcite [ corollarykerneruncountable ] assume that @xmath9 is uncountable and @xmath103 is a wild hereditary algebra with @xmath771 simple modules . then the number of regular component of @xmath103 with quasi - rank @xmath772 is uncountable . moreover the set of components of quasi - rank @xmath773 for the kronecker algebra is uncountable . the proof of the second statement uses the first statement for the path algebra @xmath103 of the wild quiver @xmath774 with @xmath775 simple modules and the existence of a regular tilting module @xmath776 in @xmath106 that induces a bijection @xmath777 with @xmath778 for all @xmath779 . to generalize the arguments from @xmath773 to @xmath780 for @xmath781 one would need the existence of bricks of arbitrary quasi - length . unfortunately for each hereditary algebra there is a finite upper bound for the quasi - length of regular bricks given by the number of simple modules @xmath772 , which is in our case @xmath782 . we show how to circumvent this obstacle by considering an action of the general linear group @xmath68 on @xmath69 . @xcite denote with @xmath783 the group of invertible @xmath784-matrices with coefficients in k which acts on @xmath785 via @xmath786 for @xmath787 , @xmath788 . for @xmath789 , let @xmath790 the algebra homomorphism with @xmath791 , @xmath792 and @xmath793 , @xmath794 . for a @xmath3-module @xmath48 denote the pullback of @xmath48 along @xmath795 by @xmath796 . the module @xmath48 is called @xmath68-stable if @xmath797 for all @xmath798 , in other words if @xmath799 . 1 . the simple representations of @xmath11 are @xmath68-stable and by @xcite every preinjective and every preprojective representation is @xmath68-stable . 2 . there are @xmath68-stable representations that are regular @xcite . in this case all representations in the same component are also @xmath68-stable . recall that the preinjective representation @xmath800 has dimension vector @xmath801 . let @xmath48 be in @xmath252 with @xmath802 . the support of @xmath48 for @xmath803 is shown in figure @xmath804 . let @xmath805 be the vertex @xmath806 . + = [ sibling angle=120 , level distance=1.4 cm ] = [ sibling angle=60,level distance=1.2 cm ] = [ sibling angle=45 ] = [ ] = [ ] + child [ color = black ] foreach in 1,1,1 node edge from parent[- > ] child [ color = black ] foreach in 1,1 node edge from parent[<- ] child [ color = black ] foreach in 0,0 node edge from parent[- > ] ; + ( 0,0 ) circle ( 40pt ) ; ( 0,0 ) circle ( 68pt ) ; ( 0,0 ) circle ( 100pt ) ; + plot [ smooth cycle ] coordinates ( 1.1,0 ) ( 2.5,0.8 ) ( 2.5,-0.8 ) ; at ( 3,0 ) @xmath807 ; + the underlying tree of @xmath681 is symmetric in the following sense . the quiver @xmath808 is not connected and consists of @xmath809 isomorphic trees @xmath810 . moreover for @xmath811 the sum @xmath812 for @xmath813 with distance @xmath814 is independent of @xmath815 . we will prove that this is not a coincidence . we show that every representation @xmath48 such that @xmath49 is @xmath68-stable , has a central point . denote by @xmath816 the symmetric group on @xmath307 . then each for each @xmath817 there is an induced bijection on @xmath818 given by @xmath819 which extends in a natural way to the set of equivalence classes of walks in @xmath11 . by abuse of notation we denote by @xmath820 the induced quiver automorphism . let @xmath821 \to [ \gamma_i w]$ ] be an arrow in @xmath40 , then by definition @xmath822 is the unique arrow @xmath823 ) \to [ \gamma_{\sigma(i ) } ] \sigma([w])$ ] . note that @xmath824 and @xmath825 . + now let @xmath43 be an indecomposable representation . we define @xmath826 to be the indecomposable representation with @xmath827 we say that @xmath48 is @xmath816-stable if for each @xmath817 there is @xmath828 with @xmath829 . this definition is motivated by the following obvious result : let @xmath43 be an indecomposable representation . if @xmath49 is @xmath68-stable then @xmath48 is @xmath816-stable . we let @xmath830 be the permutation matrix given by @xmath831 , i.e. @xmath832 if and only if @xmath833 and @xmath834 otherwise . now we assume that @xmath49 is @xmath68-stable . then we get for each @xmath817 that @xmath835 hence we find @xmath836 such that @xmath837 note that @xmath838 ) = [ \epsilon_1]$ ] and since @xmath243 acts freely @xmath40 , the element @xmath839 is uniquely determined . in the following we study the quiver automorphisms @xmath840 . a group g is said to have property fa ( * ? ? ? * i.6.1 ) if every action of g on a tree @xmath312 by graph automorphisms ( which do not invert an edge ) has a global fixed point @xmath841 , i.e. @xmath842 for all @xmath253 . it is known ( * ? ? ? * i.6.3.1 ) that all finitely generated torsion groups have the property fa . in particular , for every group action of a finite group acting on a quiver which underlying graph is a tree , there is a global fixed point . for @xmath486 and @xmath794 denote by @xmath843 the connected component of @xmath844 containing @xmath845 , where @xmath846 is the unique arrow with @xmath847 . + let @xmath48 in @xmath252 be indecomposable and @xmath848 , then we define @xmath849 . note that @xmath850 . [ propositioncenter ] let @xmath43 be @xmath816-stable . then there is @xmath851 such that 1 . @xmath852 for all @xmath817 . 2 . for each @xmath38 the number @xmath64 divides @xmath853 . since @xmath829 , we have @xmath854 we assume that @xmath855 , otherwise there is nothing to show . let @xmath309 be the finite subtree @xmath856 . then @xmath857 is a quiver automorphism of @xmath312 . since @xmath312 is finite , @xmath858 is finite and there exists a vertex @xmath859 with @xmath860 for all @xmath861 . we assume that @xmath862 . for @xmath794 we let @xmath863 be the unique arrow with @xmath394 and set @xmath864 . since @xmath855 and @xmath48 is indecomposable , we can assume w.l.o.g . that @xmath865 . since every automorphism of @xmath40 respects the metric @xmath17see @xmath866 we get @xmath867 in particular , @xmath868 . now fix @xmath869 and @xmath870 . then we have @xmath871 . since @xmath872 for all @xmath253 , we get @xmath873 and conclude @xmath874 . hence @xmath875 are non - empty isomorphic quivers . for each @xmath38 and @xmath239 we define @xmath876 . let now @xmath877 , then we have @xmath878 since @xmath879 and @xmath880 moreover we have @xmath881 . it follows @xmath882 and we conclude @xmath883 [ corollarynotgradable ] assume that @xmath48 is @xmath816-stable . if @xmath884 then @xmath64 divides @xmath368 or @xmath706 . let @xmath851 be as in @xmath885 . then we have @xmath886 if @xmath887 , then @xmath888 and @xmath889 otherwise . hence @xmath64 divides @xmath706 or @xmath368 . as an application we get the following result for components of the kronecker quiver @xmath11 . [ lemmaexistencenostable ] let @xmath890 , then there exists a regular component @xmath2 with @xmath891 and no representation in @xmath2 is @xmath68-stable . let @xmath742 . by @xmath759 there exists @xmath634 such that for each @xmath635 there is a regular component @xmath66 with @xmath892 and @xmath893 contains a quasi - simple representation @xmath894 with @xmath895 and @xmath896 . since @xmath1 , we find @xmath897 @xmath17even infinitely many@xmath18 such that @xmath64 does not divide @xmath898 and @xmath899 . hence @xmath900 is not @xmath816-stable and @xmath901 not @xmath68-stable . therefore no representation in @xmath902 is @xmath68-stable by @xcite . a locally closed set is an open subset of a closed set . a constructible set is a finite union of locally closed sets . [ lemmainjection ] let @xmath169 with @xmath903 . there is an injection @xmath904 . by @xcite @xmath905 is a closed subgroup of @xmath68 and by @xcite an @xmath906 an algebraic variety . hence we find an affine variety @xmath907 with @xmath908 . since @xmath68 is irreducible we have @xmath909 . let @xmath910 \to k[v]$ ] be a noether - normalization and @xmath911 be the comorphism . then @xmath912 is dominant . hence there is a dominant morphism @xmath913 . by chevalley s theorem @xmath914 is constructible and hence finite or cofinite . since @xmath914 is dense in @xmath915 , @xmath914 is not finite and therefore cofinite . that means @xmath916 is finite . since @xmath9 is infinite we have @xmath917 . it follows @xmath918 [ theoremmaintheorem3 ] let @xmath890 . there is a bijection @xmath919 . it is well known that @xmath920 , see ( * ? ? ? * xviii 1.8 ) . in particular there is an injection @xmath921 by @xmath922 there is a regular component @xmath2 with @xmath923 such that no representation in @xmath2 is @xmath68-stable . for @xmath60 in @xmath2 the map @xmath924 is well defined and injective . since the number of representations in a regular component with given dimension vector @xmath925 is @xmath926 ( * ? ? ? * xiii.1.7 ) and @xmath68 acts via auto equivalances we get with @xmath927 an injection @xmath928 by the schrder - bernstein theorem we get the desired bijection . note that we restrict ourselves to components of width @xmath929 , since we do nt know whether components in @xmath252 of width @xmath930 exist . also the examples @xcite of components of width @xmath930 in @xmath69 are @xmath68-stable . for the case @xmath931 we argue as follows . let @xmath1 , then there exists an bijection @xmath932 . the proof of @xcite yields an injective map @xmath933 , where @xmath934 are the indecomposable objects in a category @xmath935 equivalent to the category of finite dimensional modules over the power - series ring @xmath936 in non - commuting variables @xmath937 and @xmath629 . now let @xmath938 and consider the indecomposable module @xmath939 given by @xmath940 , @xmath941 and @xmath942 for @xmath943 . then @xmath944 for @xmath945 and we have an injection @xmath946 . the claim follows as in @xmath947 . let @xmath1 , then for each @xmath38 there are exactly @xmath71 regular components such that @xmath948 $ ] . assume that @xmath9 is uncountable and @xmath781 . the set of components of quasi - rank @xmath780 in @xmath141 is uncountable . the results of this article are part of my doctoral thesis , which i am currently writing at the university of kiel . i would like to thank my advisor rolf farnsteiner for his continuous support and helpful comments . i also would like to thank the whole research team for the very pleasant working atmosphere and the encouragement throughout my studies . in particular , i thank christian drenkhahn for proofreading . + furthermore , i thank claus michael ringel for fruitful discussions during my visits in bielefeld .

a model of mobile , charged ion channels embedded in a biomembrane is investigated . the ion channels fluctuate between an opened and a closed state according to a simple two state reaction scheme whereas the total number of ion channels is a conserved quantity . local transport mechanisms suggest that the ion channel densities are governed by electrodiffusion - like equations that have to be supplemented by a cable type equation describing the dynamics of the transmembrane voltage . it is shown that the homogeneous distribution of ion channels may become unstable to either a stationary or an oscillatory instability . the nonlinear behavior immediately above threshold of an oscillatory bifurcation occuring at finite wave number is analyzed in terms of amplitude equations . due to the conservation law imposed on ion channels large - scale modes couple to the finite wave number instability and have thus to be included in the asymptotic analysis near onset of pattern formation . a modified ginzburg landau equation extended by long - wavelength stationary excitations is established and it is highlighted how the global conservation law affects the stability of traveling ion channel density waves .

nlin0602033

spatiotemporal pattern formation is omnipresent in physical , chemical as well as biological systems driven out of equilibrium and can be be ascribed to a finite number of universality classes @xcite . cardiac spiral waves as well as the propagation of nerve signals along the axon of a neuron are prominent examples of bioelectric nonstationary patterns @xcite ; other examples might include the emergence of aster - like patterns in filament - motor - systems @xcite or the nematic ordering of biopolymers such as actin and microtubule complexes @xcite . futhermore spontaneous pattern formation of ion channels is well - known to occur in cell membranes @xcite that are a ubiquitous building block in biology . in most biological systems the pattern forming processes are more elaborate than in fluid dynamical ones . hitherto a qualitative , especially a quantitative understanding of universal aspects of pattern formation has mainly been achieved in fluid systems @xcite . biological systems , such as the model investigated in the present work , give however rise to new universal features of pattern formation . the global conservation law imposed upon the number of ion channels embedded into the cell membrane leads within the considered model system to a modified set of amplitude equations describing the generic properties of traveling ion channel density waves and alters moreover their stability behavior . the considerable amount of information on membranes gained throughout the years has been synthesized into the fluid - mosaic model of a biomembrane @xcite which includes the following main features : the phospholipids serve as a solvent for proteins and as a permeability regulator , the latter by adopting an energetically very effective bilayer configuration ; membrane proteins are free to migrate within the lipid - bilayer ; the conductance of the membrane layer is mainly made up of discrete ion channels formed by protein molecules @xcite . freely movable ion channels embedded in a fluid - mosaic membrane have been observed to self - organize if the concentration of salt across the membrane exceeds a certain threshold @xcite . the basic fluid - mosaic model underwent however considerable refinement as it neglects first of all that ion channels may fluctuate between opened and closed states due to interactions with signal molecules @xcite and secondly that channel proteins may become immobilized due to binding to the cytoskeleton @xcite . a compound system of a membrane with embedded ion channels is typically driven out of equilibrium due to ionic concentration gradients and transmembrane fluxes : opened ion channels induce ionic transmembrane currents looped to the transmembrane voltage . therefore thermodynamic models have to be extended by ion transport mechanisms based upon nernst - plank theory and by electrodynamical aspects @xcite . the resulting electrodiffusive models can be realized by considering a membrane seperating a narrow cleft of electrolyte from an electrolytic bath of ions @xcite , a configuration often met in biological situations : closely neighboring cells like post - synaptic membranes of a neuronal synapse or membrane cables forming dendrites and axons of neurons are important examples of such geometries . the transmembrane voltage corresponding to the voltage difference across the thin layer of electrolyte is then given by a one- or two - dimensional cable - type equation @xcite . it has been shown that the coupled dynamics of channel proteins and a binding - release reaction of the latter ones with the cytoskeleton leads to spatial peridoc patterns @xcite described near onset of pattern formation by a ginzburg - landau equation @xcite . a model of two different , noninteracting species of ion channels has been studied in refs . a generic evolution equation for nonstationary patterns generated by two different ion channel species assumed to remain permanently in an opened state was given in ref . @xcite wherein a constant number of opened ion channels was presumed . in the present work we will consider ion channels that may switch between an opened and a closed state @xcite according to a simple two state reaction scheme with constant rates , the diffusion coefficients as well as the effective charges of the channel proteins being state dependant . if compared to the models discussed in refs . @xcite , the system at hand may be interpreted as one comprising two types of ion channels with a reaction stimulating transition between the different channel species while imposing the constraint of a conserved total number of ion channels . after introducing the model in sec . [ sec2 ] the linear instability analysis performed in sec . [ sec3 ] evinces that either a stationary or an oscillatory bifurcation is met and is concluded by a phase diagram in sec . [ neutralcurves ] . immediately above the threshold of a supercritical bifurcation the amplitudes of the unstable pattern forming modes emerging from the homogeneous basic state are small enough to apply the powerful perturbational technique of amplitude equations . oscillatory instabilities have been broadly discussed in literature as long as systems lack global conservation laws and their nonlinear evolution is governed by the most prominent class of amplitude equations , the so - called complex ginzburg - landau equations @xcite . the influence of a conserved quantity on pattern forming instabilities , whether stationary or oscillatory , has been , starting from symmetry arguments , a subject of continuous interest during the recent years @xcite but was never discussed for a specific microscopic model . as stationary bifurcations have already been studied for the present biological system in ref . @xcite , we will focus in the remaining sections on hopf - bifurcations to traveling ion channel density waves . within the context of a weakly nonlinear approach near onset of pattern formation in sec . [ sec4 ] , the complex ginzburg - landau equation known to describe stripe patterns is altered through the coupling to an evolution equation for a large - scale , real mode evoked by the presence of a global conservation law as shown in sec . [ genericamplitueeq ] . in sec . [ nonlinearwaves ] we investigate how the implication of the large - scale mode reflecting the conservation law modifies the stability regions of supercritically bifurcating traveling ion channel density waves obtained in sec . [ bifstructcgl ] . it turns out that the latter ones become always unstable with respect to an instability at finite wave number and that observing stable stripe patterns becomes less probable the more ion channels contribute to the total conductance of the membrane layer . analytical calculations are confirmed by numerical simulations in sec . [ numerics ] . concluding remarks and an outlook on future work are given in sec . [ conclusion ] . we consider a model system constituted of a membrane and embedded ion channels which may be encountered either in an opened or in a closed state . the dynamical equations of opened ( @xmath0 ) and closed ( @xmath1 ) ion channel densities @xmath2 respectively @xmath3 are coupled via a global conservation law @xmath4 ( wherein @xmath5 denotes the area of the considered biomembrane ) implemented by a simple local reaction scheme as well as to the cable equation governing the evolution of the transmembrane voltage . the switch between the opened and closed state of an ion channel , referred to as opening closing reaction , is governed by the following monomolecular reaction scheme @xmath6 with constant rates @xmath7 and @xmath8 . according to the homogeneous , stationary ion channel densities @xmath9 channel dynamics are best captured in terms of the deviations @xmath10 and @xmath11 @xmath12 from their respective homogeneous mean densities @xmath13 , @xmath14 . local transport mechanisms as well as the mentioned conversion kinetics suggest the electrodiffusion - like equations of motion to be [ devdyn ] @xmath15 wherein the current densities @xmath16 depend on the chemical potential @xmath17 and the electrophoretic drift @xcite via the transmembrane voltage @xmath18 as follows @xmath19 the occuring mobilities @xmath20 of opened and closed ion channels @xmath21 are proportional to the respective diffusion constants @xmath22 . the stated charges @xmath23 of opened and closed ion channels are considered to be effective charges taking screening as well as electroosmotic effects into account @xcite . if excluded volume interactions between opened and closed channels together with nonlinear effects are neglected , the chemical potential takes the simple form @xmath24 presuming a one dimensional system substitution of the current @xmath16 according to eq . ( [ current ] ) in eqs . ( [ devdyn ] ) gives rise to the following equations [ fundeq ] @xmath25 -r_c \tilde{n}_o + r_o \tilde{n}_c\,,\\ \partial_t \tilde{n}_c & = d_c \nabla_x^2 \tilde{n}_c + \frac{d_c q_c}{k_b t } \nabla_x \left[(\tilde{n}_c + \bar{n}_c ) \nabla_x v \right ] -r_o \tilde{n}_c + r_c \tilde{n}_o\end{aligned}\ ] ] describing the spatiotemporal evolution of the ion channel densities . those equations have nevertheless to be supplemented by an equation governing the transmembrane voltage @xmath18 . as a biomembrane is electrically equivalent to a cable whose repetitive basic unit is formed of a capacitor modeling the pecularities of lipid molecules , a voltage source reflecting the properties of an opened ion channel and a leak conductivity taking all remaining effects into account , the kelvin cable equation may be considered to determine the transmembrane potential @xmath18 on condition that the caracteristic length scales of lateral patterns are large compared to the width of the electrolyte layer . a derivation of the one - dimensional cable equation @xcite @xmath26 from the general nernst - plank theory was given in @xcite . herein @xmath27 denotes the membrane capacitance , @xmath28 the resistance of the thin membrane layer , @xmath29 the leak conductance and @xmath30 the conductance of the opened ion channels per unit length . finally @xmath31 stands for the nernst potential , i.e. the resting potential of the membrane layer . in most cases we consider the limit of a very quick response of the transmembrane voltage , i.e. formally @xmath32 holds , which is characteristic for physiological conditions @xcite . we rescale eqs . ( [ fundeq ] ) by introducing dimensionless coordinates for space @xmath33 and time @xmath34 with a typical length scale of an electrical perturbation @xmath35^{-1/2}$ ] and the time constant of displacement @xmath36 . as normalized particle densities we opt for @xmath37 the rescaled voltage reading @xmath38 , wherein @xmath39 denotes the resting voltage and @xmath40 $ ] the density parameter measuring to what extent the ion channels contribute to the total conductance of the membrane layer . through defining a normalized relaxation time @xmath41 , we obtain on the basis of the additional abbreviations @xmath42 the following scaled equations [ scaleq ] @xmath43 \nonumber \\ & & \hspace{2 cm } -\beta_c ( n_o - n_c)\ , , \\ \label{scaleq2 } \partial_t n_c & = & d\nabla_x^2 n_c + dr\nabla_x\left[(1+n_c)\nabla v\right ] \nonumber \\ & & \hspace{2 cm } + \beta_o ( n_o - n_c ) \,,\\ \label{scaleq3 } \tau_v \partial_t v & = & ( \nabla_x^2 - 1 ) v-\alpha(1-\alpha)\varepsilon n_o - \alpha n_o v \label{scaleqcabl}\end{aligned}\ ] ] wherein the primes of the new coordinates @xmath44 and @xmath45 have been suppressed for the sake of simplicity . it should be noticed that the spatially homogeneous basic state of these scaled equations is @xmath46 . in a certain parameter range the spatially homogeneous basic state , namely @xmath46 , becomes unstable against infinitesimal perturbations and a spatially inhomogeneous pattern is likely to emerge from this very state . in order to determine this interval of instability we transform the linear part of the partial differential equations eqs . by means of the ansatz @xmath47 into a set of homogeneous linear equations for the amplitudes @xmath48 , @xmath49 and @xmath50 . these equations can be further simplified by assuming an instantaneously adapting transmembrane voltage @xmath50 , i.e. presuming @xmath51 ( @xmath52 , @xmath53 , @xmath54 @xcite ) or equivalently @xmath32 . within this limit the voltage perturbation @xmath50 may be expressed in the linear regime in terms of the opened ion channel density @xmath48 @xmath55 with @xmath56 playing the role of the control parameter @xmath57 and may thus be adiabatically eliminated from the other equations . for the two remaining linear and homogeneous equations governing the pure ion channel dynamics @xmath58 the solubility condition provides a quadradic polynomial for @xmath59 @xmath60 the two coefficients @xmath61 and @xmath62 reading @xmath63\,\ , , \\ \label{defb } b & = & \beta_o + d\left ( \beta_c + k^2 \right ) -e \,\dfrac{\beta_o + d(\beta_c r + k^2)}{1+k^2}\,.\nonumber \\\end{aligned}\ ] ] depending on the model parameters the two solutions @xmath64 of eq . are either real or complex . provided that the real part of only one of those solutions @xmath64 becomes positive in the neighborhood of a wave number @xmath65 , the basic state @xmath46 is said to be unstable : in case of a vanishing imaginary part @xmath66 one has to deal with a _ stationary _ instability @xcite whereas for finite imaginary parts , i.e. @xmath67 , an oscillatory instability , a so called _ hopf - bifurcation _ @xcite , is met . forthcoming discussions are considerably clarified by introducing the conversion rate ratio @xmath68 the real parts @xmath69 of @xmath70 , sketched in fig . [ dispall ] for @xmath71 and @xmath72 , exhibit typical shapes as a function of the wave number @xmath65 . beyond the threshold of pattern formation these different shapes of @xmath73 may lead in the weakly nonlinear regime to different universality classes of amplitude equations as discussed in the following paragraphs . it should be noted that for typical nernst potentials with absolute values reaching from @xmath74 to @xmath75 mv ( equilibrium potentials of the most important voltage gated ion channels for the mammalian skeletal muscle : @xmath76 mv in case of @xmath77 channels and @xmath78 mv in case of @xmath79 channels @xcite ) , the control parameter @xmath56 defined in eq . takes values in the range @xmath80 $ ] under physilogical conditions . thus all instabilities presented in fig . [ dispall ] may play a role in a real biological system as electrophoretic charges are usually of a few elementary units . since the number of ion channels is a conserved quantity for the investigated model , i.e. @xmath81 , the largest real part labelled @xmath82 is diffusive in the long - wavelength limit @xmath83 with @xmath84/\left(\beta_o+\beta_c\right)$ ] as confirmed by all examples shown in fig . [ dispall ] . in part @xmath85 of fig . [ dispall ] the real parts of both eigenvalues , @xmath73 , are drawn as a function of the wave number @xmath65 for a parameter combination that leads to a stable homogeneous state . part @xmath86 shows an example for the same charge and conversion rate ratio @xmath87 respectively @xmath88 , with a positive curvature of the real part in reference to the wave number @xmath65 @xmath89 and a maximum of the dispersion relation at a finite value @xmath90 . if the charge ratio @xmath91 is reduced while the conversion rate ratio @xmath92 remains the same , the homogeneous state may become stable with respect to long - wavelength perturbations as for the parameter set used in part @xmath93 , whereas the curvature @xmath94 is negative . the homogeneous state gets nevertheless unstable against perturbations with a finite wave number in the vicinity of @xmath95 as can be foreseen from the extremum condition @xmath96 . as the charge ratio @xmath91 is further diminished , the two real parts @xmath73 approach each other as depicted in part @xmath97 . decreasing @xmath91 even further leads to a pair of complex eigenvalues @xmath64 instead of real ones [ cf . [ dispall ] @xmath98 , meaning that the homogeneous basic state becomes simultaneously unstable against a pair of traveling waves ( solid branch ) with a critical wave number of @xmath99 and a spatially periodic modulation ( dashed branch ) with a wave number of @xmath100 . as @xmath91 gets even smaller the latter stationary branch , i.e. the dashed one visible in part @xmath101 , becomes strongly negative and only a hopf bifurcation to traveling waves remains as depicted in part @xmath102 . when one of the dispersion curves in fig . [ dispall ] hits the @xmath103axis as a function of the control parameter @xmath56 , the threshold of the respective instability is met . the real parts of the two solutions @xmath70 of eq . are shown as a function of the wave number @xmath65 for different values of the charge ratio @xmath91 while the conversion rate ratio @xmath104 with @xmath105 and the diffusion ratio @xmath71 are kept constant . dashed branches correspond to stationary bifurcations , while solid ones to oscillatory instabilities . in part @xmath85 the charge ratio is @xmath87 and the control parameter has a subcritical value @xmath106 , whereas @xmath56 is always supercritical in the remaining cases . in part @xmath86 the ratio @xmath87 and @xmath107 , in @xmath93 @xmath108 and @xmath109 , in part @xmath97 @xmath110 and @xmath111 , in part @xmath101 @xmath112 and @xmath113 and finally in part @xmath102 @xmath114 with @xmath115 . ] the threshold @xmath116 for a stationary bifurcation follows from the neutral stability condition @xmath117 , which is equivalent to claiming @xmath118 in eq . . from @xmath118 the wave number dependent neutral curve @xcite @xmath119\left(1+k^2\right ) } { d\left(k^2 + r\beta_c\right ) + \beta_o}\ ] ] may be derived . this neutral curve takes its minimum either at a vanishing critical wave number @xmath120 @xmath121 or the minimum becomes even lower than the latter one for a finite value @xmath122 of the wave number @xmath123 } } { \sqrt{d } } \,,\ ] ] where @xmath122 is given by the extremum condition @xmath124 . the threshold value of the control parameter @xmath56 at @xmath122 reads @xmath125 \left [ d\left(r\beta_c -1\right)+\beta_o-\gamma \right ] } { d\gamma}\ ] ] wherein the abbreviation @xmath126}\ ] ] has been introduced . at onset of a hopf bifurcation the real parts @xmath127 of the complex conjugate pair of eigenvalues @xmath64 vanish , while the imaginary parts @xmath128 remain nevertheless finite . assuming complex eigenvalues @xmath64 the polynomial in eq . may be decomposed into its real and imaginary part . since both parts have to vanish separately in order to satisfy the solubility condition eq . , they provide two conditions allowing to determine the neutral curve of the oscillatory instability @xmath129 as well as the frequency @xmath130 with @xmath62 from eq . . both quantities are shown as a function of the wave number @xmath65 for a typical parameter combination in fig . [ hopfstatct ] where the stationary and the hopf - branch have a similar threshold as expected from fig . [ dispall]@xmath101 . if parameters corresponding to fig . [ dispall]@xmath102 are chosen , the shape of the neutral curves depicted in fig . [ hopfstatct ] would not be altered but the hopf - branch would have the lowest threshold . the neutral curve specified in eq . takes its minimum at @xmath131 and the corresponding critical value of the control parameter at @xmath132 is @xmath133 whereas the hopf frequency evaluates to @xmath134 with @xmath135 neutral curves for a stationary ( @xmath136 ) and oscillatory instability ( @xmath137 ) as a function of the wave number @xmath65 ( on the left hand side ) and the frequency along these neutral curves ( on the right hand side ) as given by eq . for the parameter set used in fig . [ dispall]@xmath101 . ] there are obviously three different types of instabilities of the homogeneous state occuring in different areas of the @xmath91-@xmath92-plane , @xmath92 being the conversion rate ratio defined in eq . . one has either a cahn hilliard - like @xcite instability characterized by @xmath138 and @xmath139 ( @xmath140-bifurcation @xcite ) , i.e. the long - wavelength excitations dominate ; a stationary instability at a finite wave number @xmath65 with @xmath141 as well as @xmath142 ( @xmath143-bifurcation @xcite ) ; or a hopf bifurcation distinguished by @xmath144 and @xmath145 ( @xmath146-bifurcation @xcite ) . these three regions of existence are shown in fig . [ phasedia1 ] for @xmath147 and @xmath71 , wherein the dashed line is determined by the condition that the two stationary thresholds agree @xmath148 whereas the solid line is given by the codimension-2 condition @xmath149 the dispersion relation corresponding to the point labelled @xmath93 in fig . [ phasedia1 ] is given by fig . [ dispall]@xmath93 . the parameter set corresponding to @xmath101 in fig . [ phasedia1 ] lies in the immediate vicinity of the codimension-2 line and the neutral curves as well as the change of frequency along those curves are drawn as a function of the wave number @xmath65 in fig . [ hopfstatct ] , the associated dispersion relation being depicted in fig . [ dispall]@xmath101 . the figure shows the respective ranges of a stationary bifurcation either of the cahn hilliard - type with a critical wave number @xmath150 or with a finite critical wave number @xmath151 and of a hopf - bifurcation @xmath152 . the symbols @xmath85 to @xmath102 correspond to the parameters in the @xmath91-@xmath92-plane for whom the dispersion relations are plotted in fig . [ dispall ] , parameters being @xmath71 and @xmath147 . ] if the diffusion ratio @xmath153 is reduced compared to @xmath71 in fig . [ phasedia1 ] , while the conversion rate @xmath154 is nevertheless kept constant at @xmath147 , the area wherein a @xmath143 stationary bifurcation occurs is mostly affected as depicted in fig . [ phasedia2 ] . for large values of the diffusion ratio , such as @xmath155 in fig . [ phasedia2](a ) , a stationary bifurcation at finite wave number is encountered in a broad band of the @xmath91-@xmath92-plane bounded by the dashed line corresponding to the border between the @xmath140- and @xmath143-bifurcation and the solid line visualizing the codimension-2 line . reducing @xmath153 induces the region of existence of the @xmath143-bifurcation to shrink significantly . for very small diffusion ratios , such as @xmath156 in fig . [ phasedia2](b ) , a broadening of the @xmath143-bifurcation - region occurs for large values of @xmath92 or equivalently for strongly negative values of @xmath91 , which is not so pronounced for intermediate diffusion ratios @xmath153 , while its area of existence remains strongly confined for slightly negative charge ratios . for @xmath157 the solid line seperates in case of @xmath156 [ fig . [ phasedia2](b ) ] a hopf - bifurcation from a @xmath140-bifurcation instead of a @xmath143-bifurcation otherwise , such an interval of @xmath92 values can always been determined as long as @xmath158 . furthermore it should be noticed from fig . [ phasedia2 ] that with decreasing values of @xmath153 the slopes of the codimension-2 line and of the phase limit between the @xmath140- and @xmath143-bifurcation become steeper which progressively shifts the domain of emergence of an oscillatory bifurcation , lying beneath the codimension-2 line , to more negative charge ratios @xmath91 . the boundary between a @xmath143-bifurcation and a @xmath140-bifurcation ( dashed line ) as well as the one corresponding to the transition from a stationary to an oscillatory bifurcation ( solid line ) are shown for different diffusion ratios @xmath155 in ( a ) and @xmath156 in ( b ) while the conversion rate @xmath147 . ] if the diffusion ratio @xmath153 is however kept constant while the conversion rate @xmath154 is varied , the relative positions of the phase boundaries remain almost unchanged as it may be observed in fig . [ phasedia3 ] wherein @xmath71 . the latter figure evinces that the area of occurence of a @xmath143 stationary bifurcation located in between the two phase boundaries gets insignificantly smaller for increasing values of @xmath154 . hence the value given to the conversion rate @xmath154 is not determinant while focusing on hopf bifurcations in the forthcoming sections . phase diagram in the @xmath91-@xmath92-plane for different values of the conversion rate @xmath154 given by eq . ( [ convratio ] ) . the codimension-2 lines ( solid lines ) as well as the phase boundaries between a @xmath143-bifurcation and a @xmath140-bifurcation ( dashed lines ) are displayed for @xmath159 , @xmath147 and @xmath160 . the directed arrows visualize the shrinking of the area wherein a @xmath143-bifurcation is encountered with growing values of the conversion rate @xmath154 . the diffusion ratio @xmath153 is assumed to evaluate to 1 . ] we will focus in the present section on the oscillatory bifurcation regime while a partial analysis of the @xmath143-bifurcation may be found in ref . @xcite . immediately beyond the threshold of a supercritical bifurcation , where the amplitude of a pattern emerging from the homogeneous state is still small , the concept of the so - called amplitude equations is a very successful one to characterize the nonlinear behavior of patterns as it is exemplified for several physical , chemical and biological systems in refs . @xcite . due to the presence of the global conservation law , namely @xmath161 , all dispersion relations depicted in fig . [ dispall ] behave diffusively in the long - wavelength limit . therefore large - scale modes are marginal at the considered bifurcation point and have to be included in the reduced dynamics near onset of pattern formation . restricting our investigation to traveling ion channel density waves , the complex ginzburg landau equation @xcite normally used to describe traveling waves has to be extended by an evolution equation for the large - scale mode that is evoked by the aforementioned global conservation law . this set of coupled amplitude equations , is obtained in sec . [ genericamplitueeq ] by a systematic perturbation expansion from the scaled equations eqs . , whereby the coefficients occuring in these amplitude equations reflect the dependance of the pattern on the parameters of the underlying system . the symmetry consitstency of the latter equations is checked in sec . [ genericamplitueeq ] as well . after having studied the bifurcation structure by means of the complex ginzburg landau equation in sec . [ bifstructcgl ] , it will be shown in sec . [ nonlinearwaves ] that the implication of the large - scale mode dramatically changes , in case of an infinitely extended system , the results on stability of traveling waves acquired in sec . [ bifstructcgl ] . finally the validity of the performed amplitude expansion is checked numerically in sec . [ numerics ] . the small parameter used in the perturbative derivation of the coupled amplitude equations is the relative distance @xmath162 to the threshold @xmath163 or likewise @xmath164 @xmath165 given a system presenting a global conservation law , as encountered in this work , the generating field @xmath166 of traveling waves in the vicinity of the bifurcation point comprises a complex , propagating mode @xmath61 and a real , large - scale mode @xmath167 stipulated by the constraint of conserved number of ion channels @xmath168 wherein @xmath169 denotes the complex conjugate of the preceding terms . comparable to the adiabatic elimination of the transmembrane voltage @xmath170 performed via eq . ( [ vlinear ] ) in sec . [ sec3 ] within the linear regime , @xmath170 can be similarly expressed in terms of the channel densities within the weakly nonlinear regime , the field @xmath171 becoming thus implicitly dependant of the membrane voltage . @xmath172 in the latter ansatz designates the critical wave number defined in eq . and the cited eigenvectors are @xmath173 the second component of the eigenvector @xmath174 reading @xmath175 the amplitudes @xmath61 and @xmath167 of the fluctuations @xmath171 depend on the following critically slow time scales @xmath176 and @xmath177 as well as on the scaled spatial variable @xmath178 , thus suggesting a standard multiscale perturbative approach @xcite , that has to be combined with the aforesaid adiabatic elimination of the transmembrane voltage as sketched for convinience in appendix [ appa ] . the expansion of the fundamental equations eqs . and of the pattern forming fields @xmath179 , @xmath180 as well as @xmath170 with respect to powers of the small parameter @xmath181 results on the basis of the ansatz specified in eq . in the subsequent system of coupled amplitude equations [ amplitudeeq ] @xmath182 provided that space and time are once more rescaled to @xmath183 and @xmath184 . the analytical expressions of the coefficients appearing in the latter equations eqs . as functions of the parameters of the underlying biological model are rather lengthy . instead of giving their analytical forms we therefore plot them in fig . [ coeffplots ] as functions of the conversion rate ratio @xmath92 while keeping the remaining parameters constant . likewise amplitude equations , wherein the real , however damped , mode @xmath167 plays the role of a concentration field , were used to describe localized traveling - wave trains observed in experiments on binary mixture convection @xcite and to clarify the influence of a long - wave mode on solitary waves @xcite . the amplitude equations discussed in the latter references were recovered by a reaction - diffusion system @xcite . in ref . @xcite the nonlinear behavior of traveling waves arising in a supercritical hopf - bifurcation , coupled to a real , weakly damped mode @xmath167 that is advected by the waves and thus affects the stability of the latter ones , yielded envelope equations resembling to eqs . ( [ amplitudeeq ] ) . in all the previous examples the real mode @xmath167 was nevertheless damped and not a true zero mode , i.e. a conserved quantity : slightly different amplitude equations , motivated by work on magnetoconvection as well as rotating convection , with a true conserved quantity were derived by symmetry arguments in ref . the amplitude equations of ref . @xcite as well as those found for the present biological model will be further discussed within the scope of a two - component reaction - diffusion model which also describes the coupling of a traveling wave to an oscillatory long - wavelength mode instead of a stationary one and leads thus to a generalized version of eqs . ( [ amplitudeeq ] ) @xcite . the coefficients of the coupled amplitude equations eqs . are plotted as functions of the conversion rate ratio @xmath92 for the parameter combination @xmath185 , @xmath186 , @xmath187 and @xmath147 . ] let us now check if the amplitude equations eqs . ( [ amplitudeeq ] ) obtained by the multiscale perturbative approach are consistent with the symmetries of the underlying model system . the electrodiffusive model equations ( [ fundeq ] ) or equivalently the scaled basic equations ( [ scaleq ] ) present translational as well as reflectional symmetry . accordingly amplitude equations describing standing waves generated by two complex counterpropagating linear modes @xmath188 and @xmath189 coupled to a real , large - scale field @xmath190 evoked by the conservation law inherit the following invariance under spatial and temporal translation : [ transsym ] @xmath191 where @xmath192 , @xmath193 as well as under reflection : @xmath194 due to the scalar constraint of a total conserved number of ion channels the translational symmetry , c.f . ( [ transsym1 ] ) , stipulates that the leading - order coupling term that extends the complex ginzburg landau equations @xcite @xmath195 is proportional to @xmath196 ( i=1,2 ) . possible higher order terms that might be included in extended ginzburg landau equations comprise @xmath197 or @xmath198 with @xmath199 . in order to satisfy the reflectional symmetry given by eq . the complex coefficents of the latter two terms need to be antisymmetric with respect to reflections as they involve first order spatial derivatives . in absence of eventual coupling terms the real mode @xmath167 fulfills a diffusion equation as suggested by the conserved quantity and the reflectional symmetry specified in eq . ( [ refsym ] ) . the translational symmetries cited in eqs . ( [ transsym ] ) then claim that coupling terms from the propagating modes @xmath200 and @xmath201 likely to occur in the evolution equation of @xmath167 involve equal numbers of @xmath202 and their associated complex conjugates @xmath203 . therefore the leading coupling term consistent with reflectional symmetry reads @xmath204 , the generic evolution equation for @xmath167 finally being a diffusion equation with nonlinear forcing from @xmath205 . within the limit @xmath206 and @xmath207 , the standing wave patterns reduce to traveling waves and the symmetry consistent amplitude equations happen to be exactly those , namley eqs . , obtained by a standard multiscale approach from the inital model equations eqs . in the previous section . furthermore the coefficients @xmath208 , @xmath209 and @xmath210 appearing in eqs . are pseudoscalars as required by the reflectional symmetry of the underlying model system . since the large scale mode @xmath167 only affects stability properties of traveling waves , we may neglect @xmath167 while determining their general bifurcation structure . the problem at hand then corresponds to a simple traveling wave of complex amplitude @xmath61 and group velocity @xmath211 described by the amplitude equation @xmath212 by changing to a coordinate frame moving with the group velocity @xmath211 , i.e. by performing a gallilei transform , eq . can be reduced to the prominent complex ginzburg landau equation @xcite @xmath213 wherein the linear imaginary part @xmath214 is easily eliminated by the transformation @xmath215 . for the complex ginzburg landau equation it is well known that traveling waves bifurcate supercritically if the real part of the coefficient of the cubic term @xmath216 in eq . is positive , namely @xmath217 . the tricritical line , along which the coefficient @xmath218 vanishes , marks the border between super- and subcritically bifurcating waves in parameter space @xcite . [ tricrit2 ] shows how the tricritical point , namely the root of the coefficient @xmath218 , moves at a strongly negative charge ratio @xmath187 towards greater values of the conversion rate ratio @xmath92 as @xmath153 becomes larger . for a given value of the diffusion ratio , such as @xmath185 , and a fixed conversion rate @xmath147 the tricritical line is only slightly shifted within the @xmath91-@xmath92-plane while varying the density parameter @xmath219 over the permitted interval ( @xmath220 $ ] ) as may be concluded from fig . [ tricrit1 ] . traveling waves always bifurcate supercritically beneath the tricritical line ( dashed and dotted lines in fig . [ tricrit1 ] ) and a change to a subcritical bifurcation takes place while approaching the codimension-2 line visualized by the solid line in fig . [ tricrit1 ] . quite a similar behavior has been found in other systems , for instance in binary fluid convection @xcite . the coefficient @xmath218 appearing in the amplitude equations eqs . is drawn as a function of the conversion rate ratio @xmath92 for different diffusion ratios @xmath153 , with @xmath221 from left to right . the zero points illustrate how the tricitical point is moving for a given charge ratio @xmath187 to increasing values of @xmath92 . further parameters are @xmath147 , @xmath186 . ] if @xmath153 is in- or decreased compared to @xmath185 in fig . [ tricrit1 ] , the tricritical lines are moved along the corresponding codimension-2 lines towards more negative charge ratios while simultaneously abuting even further to the associated codimension-2 lines . thus a broadening of the domain of supercritically bifurcating waves is observed for slightly negative charge ratios . the codimension-2 line ( solid line ) as well as the tricritical lines are plotted in the @xmath91-@xmath92-plane for the density parameters @xmath222 ( dashed line ) and @xmath186 ( dotted line ) . furthermore @xmath185 and @xmath147 . ] among the classes of solutions admitted by the complex ginzburg landau equation eq . ( [ ginzlandau ] ) is the family of plane wave solutions @xmath223 with the frequency @xmath224 existing for @xmath225 in case of a supercritical bifurcation . one easily sees that solutions of the type ( [ planewavesol ] ) are long - wavelength stable for a range of wave numbers surrounding the homogeneous @xmath226 state as long as the benjamin feir criterion , @xmath227 , is satisfied @xcite . the positions of the codimension-2 line ( solid line ) , the tricritical line ( dashed line ) and the line caracterizing marginally long - wavelength stable plane waves ( dotted line ) are displayed in the @xmath91-@xmath92-plane . traveling waves bifurcate supercritically in the area located underneath the tricritical line and plane wave solutions given by eq . ( [ planewavesol ] ) are long - wavelength unstable in the region lying in - between the tricritical line and the dotted line , wherein @xmath228 holds . parameters used are @xmath229 , as well as @xmath186 . ] provided that the charge ratios @xmath91 are sufficiently negative , traveling waves are , according to fig . [ benjfeir1 ] , linearly stable in a large area of the @xmath91-@xmath92-plane underneath the dotted line , determined by the marginal stability criterion @xmath230 , and become unstable in - between the latter line and the tricritical one visualized by the dashed line . additionally a small unstable stripe , wherein the benjamin feir criterion is violated , emerges for slightly negative charge ratios in the neighborhood of the codimension-2 line depicted by the solid line in fig . [ benjfeir1 ] wherein @xmath229 and @xmath186 . whereas the extension of the instability region at strongly negative charge ratios is barely affected by the choice of the density parameter @xmath219 , the one located at mildly negative values of @xmath91 utterly depends on @xmath219 as shown exemplarily in fig . [ benjfeir2 ] for @xmath231 , the diffusion ratio @xmath153 as well as the conversion rate @xmath154 still evaluating to @xmath232 . the marginal stability condition @xmath230 being plotted in fig . [ benjfeir2 ] as a function of the conversion ratio @xmath92 up to the associated codimension-2 point , it turns out that this unstable stripe vanishes for small density parameters . if @xmath219 is in contrast kept constant at @xmath233 while the diffusion ratio @xmath153 is varied , fig . [ benjfeir3 ] proves that a diffusion ratio of @xmath185 induces a maximal broadening of this very unstable band situated at slightly negative charge ratios . for strongly negative charge ratios the region of benjamin feir - stable traveling waves is widened in a similiar manner to the one of supercritically bifurcating waves presented in fig . [ tricrit2 ] . the benjamin feir criterion @xmath234 for marginally long - wavelength stable plane wave solutions is plotted as a function of the conversion rate ratio @xmath92 while assuming a slightly negative charge ratio @xmath231 . along the directed arrow the density parameter @xmath219 is progressively increased from 0.2 to 0.8 by steps of 0.1 and evinces that for @xmath235 the domain of unstable plane waves lying in the immediate neighborhood of the codimension-2 line is incessantly aggrandized . remaining parameters read @xmath229 . ] the effect that traveling waves become unstable via benjamin feir resonance , which may eventually lead to spatiotemporal chaos , while approaching the tricritical point lying in the vicinity of a codimension-2 point has been predicted in a general context @xcite and discussed within the scope of binary fluid convection @xcite . the dependance of the marginal benjamin feir criterion @xmath234 on the diffusion ratio @xmath153 as a function of @xmath92 for a density parameter @xmath186 , the charge ratio @xmath91 evaluating to @xmath236 . increasing the diffusion ratio from @xmath237 via @xmath238 to @xmath185 along the arrow leads to a maximal enlargement of the unstable regime of wave solutions , the latter effect being once again reversed if @xmath239 . furthermore @xmath147 . ] in the present section we consider once more traveling wave solutions of the form eq . ( [ planewavesol ] ) and study , in contrast to sec . [ benjfeir ] , their stability by implicating the long - wavelength stationary excitation @xmath167 , that is by taking into account the global conservation law imposed upon the ion channel densities . to analyse the stability of the aforesaid traveling waves we set [ nonlinwavesol ] @xmath240\,e^{\dot{\imath}(qx-\omega t)}\ , , \\ c&=v_2(x , t)\ , , \end{aligned}\ ] ] substitute these expressions into the amplitude equations eqs . ( [ amplitudeeq ] ) and linearizing with respect to the small perturbation terms @xmath241 and @xmath242 yields in case of the homogeneous @xmath226 state the following coupled partial differential equations [ nonlineqs ] @xmath243v_1 & = \left[(\eta+\dot{\imath } a)+\xi_0 ^ 2(1+\dot{\imath}c_0)\partial_x^2\right ] v_1 \nonumber\\ & + f\left[g_5(1+\dot{\imath}c_5)\partial_x- g_2(1+\dot{\imath}c_2)\right]\,v_2\nonumber\\ & -g_1(1+\dot{\imath}c_1)(v_1^{\star}+2 v_1)f^2\ , , \\ \partial_t\,v_2 & = d_1\partial_x^2\,v_2 -b f\partial_x ( v_1 + v_1^{\star})\end{aligned}\ ] ] with @xmath244 and @xmath28 denoting amplitude respectively frequency of the plane wave solution eq . ( [ planewavesol ] ) . it is beneficial to consider sideband instability by setting [ nonlinansatz ] @xmath245 where @xmath246 designate complex amplitudes . from the coupled equations eqs . ( [ nonlineqs ] ) the aforementioned ansatz then generates a set of homogeneous , linear equations whose solubility condition leads to the growth rate @xmath247 as a function of the perturbation wave number @xmath248 . the expansion of the latter growth rate with respect to @xmath248 @xmath249 shows that in the limit of small perturbation wave numbers @xmath248 the benjamin feir criterion of sec . [ benjfeir ] remains always unchanged when taking into account the large - scale mode @xmath167 and switching to a frame moving with group velocity @xmath211 . since the large - scale mode @xmath167 leaves the stability of traveling waves untouched with respect to long - wavelength perturbations , it might alter their stability by triggering an instability at finite perturbation wave number which is assured by fig . [ nichtlindispersio]@xmath86 . stability regions of nonlinear traveling waves given by eqs . ( [ nonlinwavesol ] ) within the @xmath91-@xmath92-plane for @xmath229 and different values of the density parameter @xmath219 . the solid line visualizes the codimension-2 line , traveling waves being stable in the area lying underneath the lines of marginal stability , namely the dashed - dotted one corresponding to @xmath250 , the dashed one to @xmath251 and the dotted one to a density parameter of @xmath186 . the distance to the bifurcation point was chosen to be @xmath252 . ] compared to the case @xmath253 studied in sec . [ benjfeir ] exemplarily for @xmath229 and @xmath186 through fig . [ benjfeir1 ] , fig . [ nonlinstab1 ] clearly evinces a strong influence of the large - scale mode @xmath167 due to the conservation law on the stability of traveling waves : regions of the @xmath91-@xmath92-plane , wherein stable plane wave solutions occur , are significantly restricted and shifted towards more negative charge ratios . [ nonlinstab1 ] shows those stability domains limited , depending on the choice of the density parameter @xmath219 , to the area located below the dashed - dotted , the dashed or the dotted line of marginally stable nonlinear traveling waves satisfying @xmath254 . the real part of the latter growth rate @xmath247 is plotted in fig . [ nichtlindispersio]@xmath86 as a function of the perturbation wave number @xmath248 presuming the density parameter @xmath219 evaluates to 0.8 . in account with the stability diagram , i.e. fig . [ nonlinstab1 ] , the tuple ( @xmath187 , @xmath255 ) corresponds to wave solutions becoming unstable with respect to finite perturbation wave numbers @xmath248 as expected from the power series of @xmath247 performed in eq . ( [ sigmaent ] ) . to hightlight the effect of the large - scale mode @xmath167 evoked by the conservation law , the real part @xmath256 of the same growth rate is depicted in fig . [ nichtlindispersio]@xmath85 in case of a vanishing real mode @xmath167 . the stability analysis of @xmath253 in sec . [ benjfeir ] has shown by means of the stability diagram fig . [ benjfeir1 ] that for the parameter set ( @xmath257 , @xmath258 , @xmath186 ) traveling waves are long - wavelength unstable and thus the observed instability in fig . [ nichtlindispersio]@xmath85 is generically different from the one encountered in the presence of a conserved quantity whose typical shape is given by fig.[nichtlindispersio]@xmath86 . the typical shape of the real part @xmath256 corresponding to the growth rate @xmath247 specified by the ansatz eq . ( [ nonlinansatz ] ) is plotted as a function of the perturbation wave number @xmath248 for @xmath229 , @xmath186 as well as @xmath252 in @xmath86 . @xmath187 and @xmath255 leads , according to the stability diagram in fig . [ nonlinstab1 ] , to wave solutions unstable with respect to finite perturbation wave numbers @xmath248 . on the left hand side however the tuple @xmath259 induces , in the absence of the large - scale field @xmath167 characterizing the conservation law , generic long - wavelength unstable traveling waves in accordance with fig . [ benjfeir1 ] . ] moreover the sensitivity of the former unstable stripe at slightly negative values of @xmath91 ( cf . enlargement in fig . [ benjfeir1 ] ) on the density parameter , depicted in fig . [ benjfeir2 ] , is amplified as it may be concluded from fig . [ nonlinstab1 ] as well as from fig . [ nonlinstab2 ] , wherein the points of marginally stable long - wavelength nonlinear plane wave solutions are plotted as a function of @xmath219 for a constant conversion ratio @xmath92 . the latter figure also stipulates that , for a given value of @xmath92 , there exists a certain density parameter @xmath219 generating a maximal outspread of the area of stable traveling waves . for the given parameter set @xmath260 and @xmath252 a density parameter of @xmath261 , meaning the ion channels contribute up to @xmath262 to the total conductance of the membrane layer , realizes the largest extension of the stable traveling wave regime . points of marginal stability for a fixed conversion ratio @xmath263 as a function of the density parameter @xmath219 , the diffusion ratio @xmath153 and the conversion rate @xmath154 evaluating to @xmath232 as in fig . [ nonlinstab1 ] . furthermore @xmath252 . ] a major problem always encountered when dealing with amplitude equations is that their validity range around the threshold @xmath264 remains a priori unpredictable . therefore we have determined the amplitude of a stripe solution by numerical simulations of the basic eqs . ( [ scaleq ] ) as a function of the relative distance @xmath162 to the bifurcation point and compared these numerical results with the analytical solution @xmath265 in fig . [ simulstripe ] . up to @xmath266 we find for the particle density @xmath180 of the closed ion channels a fairly good agreement between analytical and numerical results whereas the same holds up to @xmath267 for the opened ion channel density @xmath179 , which gives a reasonable estimate of the quantitative validity range of the preceding results obtained in terms of amplitude equations . the amplitudes @xmath179 and @xmath180 of traveling ( opened and closed respectively ) ion channel densities obtained by numerical simulations ( crosses ) of the microscopic model given by eqs . ( [ scaleq ] ) are compared to the analytical solution @xmath268 . the parameters used are @xmath269 with the individual conversion rates being @xmath147 and @xmath270 . ] besides its biological motivation , the model presented in this work exhibits interesting generic properties concerning pattern formation that are for the first time discussed within the scope of a specific model system . the homogeneous ion channel density distribution on the biomembrane may become unstable with respect to either a stationary or an oscillatory bifurcation , whereas we focused exclusively on analyzing the latter bifurcation type . as the total ion channel density is a conserved field , the supercritical hopf - bifurcation occuring at finite wave number shows nonlinear properties largely differing from those emerging in systems lacking conserved order parameters . in presence of a global conservation law , such as the one encountered in our model , it can be expected that the asymptotic description of pattern formation by means of amplitude equations is altered through the coupling of known forms of envelope equations to long - wavelength modes . this allegation has been confirmed by the perturbation expansion of the initial model equations that yielded a set of coupled amplitude equations in case of traveling ion channel density waves rather than a single complex ginzburg - landau equation . the validity range of the latter reduced dynamics has been checked by simulations of the inital model equations . this coupling to stationary , long - wavelength excitations has however been completely neglected during the discussion of the stationary bifurcation scenario in ref . @xcite which gives consequently a distorted picture of the nonlinear bifurcation behavior . detailed phase diagrams highlight in which parameter space hopf - bifurcations at finite wavelengths take place . neither the threshold value of the control parameter nor the supercriticality of the oscillatory bifurcation are modified by implicating the long - wavelength mode evoked by the conserved ion channel density field into the asymptotic description near onset of instability . the stability of traveling ion channel density waves is however strongly affected by the latter real mode for infinitely extended systems . it turns out that the long - wavelength mode , reflecting the conserved density field , triggers an instability at finite perturbation wave number whereas plane wave solutions would be long - wavelength unstable in the absence of the large - scale mode . if compared to a case without a real mode , stable traveling wave solutions are generally shifted towards strongly negative ratios of opened and closed ion channel charges respectively and thus the domain , wherein spatiotemporal chaos is likely to be seen , significantly broadens . if finally all model parameters are fixed the observation of stable traveling ion channel density waves becomes less probable the more the ion channels contribute to the total conductance of the considered membrane layer . the dynamics of traveling waves in finite geometries and consequently their stability properties naturally differ from those in infinitely extended systems @xcite . how the finite size of the system further influences the studied traveling ion channel density waves at hand , already affected by a large - scale mode , is postponed to future work . the formation of dissipative patterns of ion channels on a membrane that were discussed in the present work is most likely to be observed in in vitro systems as the one proposed in ref . @xcite where a linear array of 48 field - effect transistors in silicon was coated with a bimolecular layer of lipid and gramicidin . nevertheless the investigated electrodiffusive mechanism seems to be also relevant for in vivo systems as shown by experiments on the effect of electric fields on clustering of acetylcholine receptors as cited in ref . @xcite . couplings between a long - wavelength mode and a traveling wave are expected in other models from cell biology as well , e.g. in mixtures of cytoskeletal filaments interacting with molecular motors @xcite . the investigation of an oscillatory instability occuring at finite wave number coupling to a stationary , long - wavelength mode , that was motivated by the biological model at hand , will be extended within the scope of a reaction - diffusion system in a forthcoming work to oscillatory , long - wavelength excitations influencing on oscillatory instabilities at finite wave number @xcite . * acknowledgments * fruitful discussions with m. hilt and f. ziebert were very much appreciated . according to the slow time and spatial scales @xmath271 , @xmath272 and @xmath273 introduced in sec . [ genericamplitueeq ] all time and space derivatives occuring in the scaled equations eqs . may be substituted by in order to accomplish a perturbation expansion of eqs . . within the adiabatic limit , i.e. if instantaneously adapting membrane voltage @xmath170 is assumed , meaning formally @xmath32 or @xmath276 , eq . becomes time independant and may be solved successively by means of the resulting equation hierarchy in @xmath277 [ viter ] @xmath279\ , , \\ v^2 & = \mathcal{f}^{-1}\left[-e_c^h n_o^2 + ( 2\partial_x\partial_x -\alpha n_o^1 ) v^1\right]\ , , \\ v^3 & = \mathcal{f}^{-1}\left[-e_c^h ( n_o^1+n_o^3 ) + ( \partial_x^2 -\alpha n_o^2)v^1\right.\nonumber\\ & \hspace{1.2cm}\left.+(2\partial_x\partial_x -\alpha n_o^1)v^2\right]\end{aligned}\ ] ] wherein the operator @xmath280 has been introduced . + since the desired solutions @xmath281 and @xmath282 with @xmath283 are harmonic functions of @xmath284 , the occuring operator @xmath285 in the antecedent expressions for @xmath286 may be replaced in the forthcoming calculations as @xmath287 with @xmath288 by performing the perturbation expansion of the scaled equations eqs . , governing the spatiotemporal evolution of the ion channel densities and taking into account eqs . as well as eq . one ends up with a hierarchy in @xmath181 @xmath291\mathcal{f}^{-3 } & 0\\ dqe_c^h\left[\left(\partial_x^2 + \partial_x^2\right)\,\mathcal{f}^2 + 5\partial_x^2\partial_x^2\,\mathcal{f } + 4\partial_x^4\partial_x^2\right]\mathcal{f}^{-3 } & -d\partial_x^2 \end{array}\right)\ , , \\ { { \mathcal n}}_1 = & -\left(\begin{array}{c } 1\\dq \end{array}\right ) \left[-\mathcal{f}^{-1}\partial_x^2\left(n_o^1\,\mathcal{f}^{-1 } n_o^1\right)\right]\alpha e_c^h \renewcommand{\arraystretch}{2}-\left(\begin{array}{c } \partial_x n_o^1\,\mathcal{f}^{-1 } n_o^1 + n_o^1\,\mathcal{f}^{-1}\partial_x^2 n_o^1 \\ dq\left[\partial_x n_c^1\,\mathcal{f}^{-1 } n_o^1 + n_c^1\,\mathcal{f}^{-1}\partial_x^2 n_o^1\right ] \end{array}\right)e_c^h\ , , \\ { { \mathcal n}}_2 = & -\left(\begin{array}{c } 1\\dq \end{array}\right ) \left[-\mathcal{f}^{-1}\partial_x^2\left(n_o^2\,\mathcal{f}^{-1 } n_o^1\right ) -2\mathcal{f}^{-1}\partial_x\partial_x(1+\partial_x^2)\left(n_o^1\,\mathcal{f}^{-1 } n_o^1\right ) -m_1\right]\nonumber\\ & \qquad - e_c^h\renewcommand{\arraystretch}{2}\left(\begin{array}{c } m_2 \partial_x n_o^1 + \left(\partial_x n_o^1+\partial_x n_o^2 + n_o^2\partial_x\right)\mathcal{f}^{-1}\partial_x n_o^1 + m_3 n_o^1\\ dq\left[m_2\partial_x n_c^1 + \left(\partial_x n_c^1+\partial_x n_c^2 + n_c^2\partial_x\right)\mathcal{f}^{-1}\partial_x n_o^1+m_3 n_c^1\right ] \end{array}\right)\end{aligned}\ ] ] with the abbreviations @xmath292\ , , \\ m_2 = & \left(2\partial_x\partial_x\mathcal{f}^{-2}+\partial_x\mathcal{f}^{-1}\right)n_o^1 + \mathcal{f}^{-1}n_o^2\ , , \\ m_3 = & \left(2\partial_x^3\partial_x\mathcal{f}^{-2}+2\partial_x\partial_x\mathcal{f}^{-1 } \right)n_o^1 + \partial_x^2\mathcal{f}^{-1}n_o^2\ , .\end{aligned}\ ] ] the aforementioned hierarchy , eqs . , has to be solved successively for @xmath293 with @xmath283 . the first of these three equations is fulfilled if the ansatz given in eq . is chosen for @xmath171 . the determined solution may be inserted into equation eq . which leads to a linear , inhomogeneous equation of the type @xmath294 . since @xmath295 is not invertible , as it has a zero eigenvalue , the only possibility for the latter equation to have nontrivial solutions is that @xmath296 is orthogonal to the kernel of @xmath297 which yields the solvability condition @xmath298 with @xmath299 denoting the left eigenvector of @xmath295 . terms proportional to @xmath300 , linked to the large - scale mode @xmath167 , stipulate a second fredholm s alternative @xmath301 where @xmath302 designates the left eigenvector spanning the adjungated kernel of @xmath303 . inserting the solution @xmath304 of eq . as well as @xmath171 into eq . gives rise to another linear , inhomogeneous equation of the form @xmath305 . to the latter equation are associated two solvability equations @xmath306 from fredholm s alternatives at order @xmath162 , namley eqs . and , as well as at order @xmath307 , that is eqs . and , the final amplitude equations can be obtained after rescaling space and time according to @xmath308 as suggested by .

we have analyzed _ asca _ and _ rosat _ pspc spectra and images of the galaxy cluster abell 2029 . the _ asca _ spectra of the cluster indicate that the gas temperature declines with radius . the pspc image shows that the cluster is very regular and smooth . also , there is no significant evidence for any irregularities in the temperature distribution in the cluster , as would be produced by a subcluster merger . these results suggest that a2029 is a relaxed cluster , and that the gas is in hydrostatic equilibrium . we use the assumption of equilibrium to determine the gravitational mass of the cluster as a function of radius . at a radius of ( mpc ; km s mpc ) , the gravitational mass is , while the mass of gas is . the gas fraction is found to increase with radius ; within a spherical radius of , the fraction is . the iron abundance in the gas is found to be of solar . there is no significant evidence for any variation in the abundance with position in the cluster . the global x - ray spectra , central x - ray spectra , and _ rosat _ surface brightness all require a cooling flow at the cluster center . the global x - ray spectrum implies that the total cooling rate is yr . the global x - ray spectra are consistent with the galactic value for the soft x - ray absorption toward the cluster . the _ rosat _ pspc spectra of the central regions of the cluster are inconsistent with the large value of foreground excess absorption found by white et al . ( 1991 ) based on the _ einstein _ sss spectrum . the upper limit on excess foreground absorption is . however , the spectra do not rule a significant amount of intrinsic absorbing gas located within the cooling flow region .

astro-ph9712046

abell 2029 ( a2029 ) is a richness class ii , bautz - morgan class i cluster of galaxies at a redshift of @xmath11 ( abell , corwin , & olowin 1989 ) . with an x - ray flux of @xmath12 ( 210 kev ) @xmath13 ergs @xmath10 s@xmath3 ( david et al . 1993 ) , it is one of the brighter x - ray clusters . this flux corresponds to an x - ray luminosity of @xmath14 ( 210 kev ) @xmath15 ergs s@xmath3 , and a bolometric x - ray luminosity of @xmath16 ergs s@xmath3 ( @xmath2 km s@xmath3 mpc@xmath3 ) . a2029 is one of the most regular rich clusters known , particularly in its x - ray properties ( slezak , durret , & gerbal 1994 ; david , jones , & forman 1995 ; buote & tsai 1996 ) . as a result , it seems likely that the intracluster gas in a2029 is in hydrostatic equilibrium out to a large radius . thus , a2029 is an ideal candidate for the determination of the total gravitational mass of the cluster at large radii using the assumption of hydrostatic equilibrium . this requires that the gas temperature and density be determined as a function of radius the cluster . previous observations using the @xmath17 monitor proportional counter ( mpc ) found an average gas temperature of @xmath18 kev ( david et al . however , this observation does not provide any information on the distribution of temperatures . david et al . ( 1995 ) derived the mass of a2029 from _ rosat _ position sensitive proportional counter ( pspc ) observations assuming that the gas was isothermal . this paper presents new _ asca _ x - ray spectra of a2029 . we use these spectra to constrain the temperature variation in the cluster . we also analyze archival _ rosat _ pspc observations of the x - ray image and spectrum of a2029 . we use the _ asca _ derived temperature gradients and _ rosat _ derived x - ray surface brightness to determine the variation of the total gravitational mass and gas mass with radius in the cluster . as a test of the hydrostatic assumption , we search for asymmetries in the the temperature distribution which would be indicative of recent dynamical activity , such as a subcluster merger . we also constrain any abundance gradients in the intracluster gas in the cluster . the use of _ asca _ spectra to determine the spatial variation of the spectral properties of the x - ray emitting gas in a2029 requires that the effects of the energy - dependent point spread function ( psf ) be corrected . this is particularly difficult when the x - ray surface brightness is strongly peaked to the cluster center , as is true in a2029 . a2029 contains a very strong cooling flow , which is centered on the large central cd galaxy ( ugc 9752 = ic 1101 ) . the presence of a cooling flow has been established based both on analyses of the x - ray surface brightness profile ( arnaud 1989 ; sarazin , oconnell , & mcnamara 1992 ) , and on analyses of the x - ray spectrum of the central cluster region ( white et al . the _ einstein _ imaging proportional counter ( ipc ) observations of the surface brightness gave a cooling rate of @xmath19 yr@xmath3 within a radius of @xmath20 kpc . ( arnaud 1989 ) , while the _ rosat _ high resolution imager ( hri ) gave @xmath21 yr@xmath3 within @xmath22 kpc ( sarazin et al . the _ einstein _ solid state spectrometer ( sss ) spectrum gave a cooling rate of @xmath23 yr@xmath3 within its aperture of 3@xmath24 ( white et al . however , a2029 is nearly unique among strong , well - studied cooling flows in that it shows no evidence for optical emission lines or blue stellar continuum color within its central regions ( e.g. , mcnamara & oconnell 1989 ) . white et al . ( 1991 ) reanalyzed the @xmath17 sss spectra of the central regions ( 3 arcmin radius ) of a number of cooling flow clusters , and found evidence for large amounts of soft x - ray absorption , in excess of that expected from the interstellar gas in our galaxy . in a2029 , white et al . found an excess absorbing column of @xmath25 @xmath10 , if the excess absorber was modeled as foreground galactic material . because the galactic hydrogen column towards a2029 is relatively small , @xmath26 @xmath10 ( stark et al . 1992 ) , this excess absorption should have a profound effect on the soft x - ray spectrum of the central regions of a2029 . we extract the central _ asca _ and _ rosat _ x - ray spectra of a2029 to study the spectral properties of the cooling flow gas . in particularly , we use the soft x - ray spectra to constrain the amount of absorbing material in excess of that expected from our galaxy . in [ sec : obs ] , we describe the _ asca _ and _ rosat _ observations and the treatment of the data . the _ rosat _ pspc image of the cluster field is extracted in [ sec : image ] ; it is needed to analyze the spectral data and to determine the x - ray surface brightness profile . the global x - ray spectrum is analyzed using _ asca _ and _ rosat _ data in [ sec : global ] . in [ sec : spatial ] , we discuss the spatial variation of the x - ray spectrum of a2029 . here , we determine the radial temperature and abundance gradients , and search for asymmetries in the temperature distribution . the central cooling flow spectrum is extracted in [ sec : cflow ] . using the x - ray temperature and surface brightness profiles , we determine the distribution of the total gravitational mass and gas mass in [ sec : masses ] . finally , our conclusions are summarized in [ sec : conclude ] . unless otherwise noted , all uncertainties quoted in this paper are 90% confidence regions . the abundances given are all relative to the `` cosmic '' values in anders & grevesse ( 1989 ) . all distance - dependent values in this paper assume @xmath27 km s@xmath3 mpc@xmath3 and @xmath28 ; we have tried to indicate the distance dependence by parameterizing the hubble constant as @xmath2 km s@xmath3 mpc@xmath3 . the cluster was observed with _ asca _ on 1994 february 1920 . the solid - state imaging spectrometer ( sis ) detectors operated with two chips per detector being active ( s0c0,1 and s1c2,3 ) , and the observation included both bright and faint mode sis data . there were a number of problems with the attitude control during a portion of this observation , which resulted in the original standard attitude file being incorrect . for the first 1180 seconds of the integration , the attitude was about 20 arcmin off . we discarded this portion of data . the attitude settled into an accurate and steady value for the remainder of the observation . the gas imaging spectrometer detector 3 ( gis3 ) data was taken during the period when the events were recorded with a bit error . this error was corrected , although this data has fewer spectral channels as a result of this . the data were screened using relatively strict versions of the standard cleaning criteria . when these screening operations were complete , the exposures were 29,612 , 29,710 , 34,365 and 34,162 seconds for the sis0 , sis1 , gis2 , and gis3 , respectively . the sis and gis spectra were corrected for background using blank sky observations which were screened in the same manner as the data . for the gis , the background was constructed by weighting blank sky observations with differing values of the cut - off rigidity ( @xmath29 ) by the actual distribution of @xmath29 values during the screened exposure time . for the sis , a long observation with the same minimum cut - off rigidity ( @xmath30 ) was used . for the analysis of the spatial variation of the x - ray spectrum ( [ sec : spatial ] ) , we added a systematic uncertainty of 20% in the background to the other uncertainties in deriving confidence intervals for parameters . a2029 was observed with the _ rosat _ position sensitive proportional counter ( pspc ) for 3,236 seconds on 1992 january 24 ( observation wp800161 ; u. briel , p.i . ) and for 12,550 seconds on 1992 august 10 - 11 ( observation rp800249 ; c. jones , p.i . ) . the first observation was discussed previously in slezak et al . ( 1994 ) , while the second was presented in david et al . the two observations were combined , and were aligned using the positions of two bright point sources . the pspc data was screened for periods of high background based on a master veto rate @xmath31 170 ( plucinsky et al . 1993 ) , for other times of high background based on the light curve of counts in the background region used for the extraction of the spectrum ( see below ) , for periods of 15 seconds after switching on the high voltage , and for periods with an uncertain aspect solution . the analysis of the pspc was done with a combination of the iraf / pros software , the xspec / xanadu software , the xselect program , and snowden s sxrb package for the analysis of diffuse x - ray emission ( snowden 1995 ) . the screened data contained a total exposure of 12,905 seconds and the average master veto rate was 87.05 . the data were corrected for particle background , scattered solar x - ray background , the long term enhancements ( lte ) in this background , and after - pulses . with these backgrounds removed , the image was corrected for exposure and vignetting . the _ rosat _ pspc spectra and x - ray surface brightness profiles were corrected for background using four regions at radii from 24 to 44 arcmin which were free of strong sources or obstruction by the instrument ribs . because of calibration uncertainties at the softest and hardest energies , we restricted the pspc spectra to energies of 0.22.2 kev ( channels 6 - 32 ) . we fit the observed spectra with thermal emission models for a low density plasma in collisional ionization equilibrium . where possible , we used models based on both the raymond - smith code ( raymond & smith 1977 ) and the mekal code ( e.g. , kaastra et al . the mekal code includes a better treatment of the iron l lines near 1 kev , based on measurement and calculations of the atomic parameters by liedahl and his colleagues ( e.g. , liedahl et al . we did not fit mekal models during the spatial analysis of the spectra in [ sec : spatial ] , as the mekal spectra were not available in a form which could be used with the programs employed to correct for the energy - dependent point - spread - function ( psf ) . however , at higher temperatures , we found that the raymond - smith and mekal spectra gave almost exactly the same results . larger differences are found between the raymond - smith and mekal spectra at lower temperatures , and these become more important when the spectrum includes the cooling flow regions at the center of the spectra . however , we still found that the spectral parameters from fitting raymond - smith and mekal models were in good agreement . because the mekal model generally gave slightly better fits , we report the results using this model in [ sec : global ] and [ sec : cflow ] . we considered two simple models for the thermal distribution of the gas in each region where the spectrum was extracted . the first model assumed thermal emission at a single temperature ; we refer to this as a `` single temperature '' model . in table [ tab : spectra ] , these models labeled as `` 1 t . '' this model is characterized by two spectral parameters , the temperature of the gas and the ratio of the abundances in the gas to the solar values , and by the overall emission integral of the gas which normalizes the spectra . as discussed above ( [ sec : intro ] ) , there is considerable previous evidence from x - ray spectra and images for a cooling flow at the center of a2029 . thus , we also considered a simple spectral model for ambient gas and a cooling flow . the gas in the cooling flow was assumed to cool from the temperature of the ambient gas to very low temperatures due to its own radiation . the abundances in the cooling flow gas were taken to be those in the ambient gas , as expected if the cooling gas originated as intracluster gas . the cooling flow gas was assumed to cool at constant pressure ; more detailed spectral models have indicated that this is a reasonably accurate approximation ( wise & sarazin 1993 ) . the x - ray spectrum of this cooling flow component is given by @xmath32 where @xmath33 is the photon energy , @xmath34 is the luminosity emitted in the photon energy range @xmath35 , @xmath36 is the mean mass per particle in terms of the proton mass @xmath37 , @xmath38 is the ambient gas temperature in the cluster from which the gas is assumed to cool , @xmath39 is its current temperature , the emissivity of the gas per unit volume and photon energy is @xmath40 , @xmath41 is the mass density of the gas , and the total emissivity ( cooling rate ) of the gas is @xmath42 ( e.g. , wise & sarazin 1993 ) . we will refer to this model as the `` single temperature plus cooling flow model . '' in table [ tab : spectra ] , these models labeled as `` cf . '' this model is characterized by a single additional parameter , the cooling rate @xmath43 . the global cluster x - ray spectrum ( [ sec : global ] ) and the spectrum of the inner cooling flow region ( [ sec : cflow ] ) were fit to these models using xspec version 9 . because this version of xspec did not contain a model for the cooling flow spectrum using the mekal emission code , we constructed such a model . the analysis of the spatial variation of the spectrum , correcting for the effects of the energy - dependent psf of _ asca _ , was done with the program of markevitch et al . the _ rosat _ pspc data in the hard band of 0.5 to 2.0 kev ( pi channels 52 to 201 ) were used to construct an image of the region of the _ asca _ observation . this hard band is useful because it reduces the background in the image without much loss of counts from the hot intracluster gas . this image was corrected for particle and solar x - ray background , exposure , and vignetting as discussed above . the resulting image was adaptively smoothed with a variable smoothing kernel so that each smoothing beam had a signal to noise ratio of at least five ( huang & sarazin 1996 ) . contours of the x - ray surface brightness for the central portion of the resulting image are shown in figure [ fig : pspc_image ] . 2.7truein 0.2truein the _ rosat _ pspc image of a2029 contains a large number of presumably unrelated point sources . there are two moderately strong point sources located approximately 4 arcmin ( src . a ) and 9 arcmin ( src . b ) from the center of the cluster . src . a is known to be a background active galaxy nucleus ( agn ; bregman , private communication ) . b has not been identified to our knowledge . these sources are labeled on figure [ fig : pspc_image ] , and are discussed further in [ sec : spatial ] , where they are included in the determination of the variation in the x - ray spectrum of a2029 . when these sources are removed , the contours of the x - ray surface brightness of a2029 in figure [ fig : pspc_image ] appear elliptical and quite regular . this agrees with the previous conclusions reached by slezak et al . ( 1994 ) using wavelet techniques applied to the shorter _ rosat _ pspc exposure , and by buote & tsai ( 1996 ) using the power ratio technique applied to the longer _ rosat _ pspc exposure . there is general agreement that a2029 is regular and that it lacks any strong substructure . this is also consistent with the tendency for clusters with large cooling flows to be regular ( e.g. , buote & tsai 1996 ) . the regularity of a2029 makes it a particularly attractive candidate for applying the hydrostatic method to determine the mass distribution . at the faintest levels , the _ rosat _ pspc observation shows x - ray emission from a2029 extending out to a radius of approximately 20 arcmin from the cluster center . in [ sec : spatial ] , we will analyze the x - ray spectrum of a2029 out to a radius of 16 arcmin . all of this region is included in the fields of view of both of the gis detectors . however , the sis detectors only cover a fraction of the cluster emission . in figure [ fig : pspc_image ] , the dashed squares give the approximate fields of view of the two chips used on the sis0 detector ( s0c0 and s0c1 ) . the corresponding regions for the sis1 detector chips ( s1c2 and s1c3 ) are similar and have about the same orientation , but are displaced about 1 arcmin to the west . the _ asca _ and _ rosat _ observations were used to determine the integrated spectrum of the cluster . for this purpose , we determined the spectrum within 16@xmath24 ( @xmath1 mpc ) of the center of the cluster . this region is shown as the outermost dashed circle in figure [ fig : pspc_slices ] . this region extends nearly to the bright ring of enhanced background near the edge of each of the _ asca _ gis detectors , and lies well within the ribs on _ rosat _ pspc detector . since the _ asca _ sis spectra cover only a portion of this region ( fig . [ fig : pspc_image ] ) , we do not include the sis data in this determination of the global spectrum . later , we will find that the cluster has significant spatial variations in its x - ray spectrum , due to a cooling flow at the cluster center ( [ sec : cflow ] ) and due to a radial temperature gradient ( [ sec : spatial_t ] ) . thus , it would be difficult to directly compare the sis spectra with those from the gis and pspc , which cover the same larger region of the cluster . the spectra were binned to insure that there were at least 20 counts per spectral channel , so that the @xmath44 test gave reliable results . we excluded appropriately sized regions around each of the two point sources labeled a and b in figure [ fig : pspc_image ] . the results of fitting `` single temperature '' and `` single temperature plus cooling flow '' models to the observed spectral are summarized in table [ tab : spectra ] . the first column enumerates the rows for reference in the text . the second column indicates the region of a2029 being considered , and the third column lists the instrument(s ) used . the fourth column gives the model being fit to the data . columns 58 give the galactic hydrogen absorbing column @xmath45 , the ambient gas temperature @xmath46 , the heavy element abundance , and the cooling rate ( the latter is only given for the cf models ) . column 9 gives the resulting @xmath44 and number of degrees of freedom ( d.o.f . ) for the best - fit model . the final column gives the probability that the value of @xmath44 would be this large if the data were drawn from the best - fit model with only statistical uncertainties . because it is likely that there are systematic errors in the responses of the instruments , these numbers may not be accurate . the uncertainties in the parameters give the 90% confidence region . because of their nearly disjoint spectral bands , we find that the _ rosat _ pspc ( 0.22 kev ) and _ asca _ gis s ( 0.511 kev ) provide nearly orthogonal but complementary information of the spectrum of a2029 . we first discuss the results from _ rosat _ and _ asca _ separately , and then jointly fit all of the spectra . llcccccccc row&region&instr.&model & @xmath45&@xmath46&abund.&@xmath47&@xmath44/d.o.f.&prob.&&&&(@xmath48 @xmath10)&(kev)&(@xmath49)&(@xmath50 yr@xmath3 ) & & ( % ) 1&whole cluster&pspc&1 t & @xmath51&@xmath52&@xmath53 & &@xmath54&6.442&@xmath55&pspc&cf&@xmath56&@xmath316.23&@xmath310.34 & @xmath57&@xmath58&32.313&&gis&1t&@xmath594.43&@xmath60&@xmath61 & &@xmath62&5.774&&gis&cf&@xmath596.04&@xmath63&@xmath7 & @xmath59609&@xmath64&6.055&&pspc , gis&1t&@xmath65&@xmath66&@xmath67 & &@xmath68&0.646&&pspc , gis&cf&@xmath69&@xmath70&@xmath7 & @xmath71&@xmath723.87&&&&&&&&7&@xmath73&asca&cf&3.07&@xmath74&0.40 & @xmath75&@xmath76&8&@xmath77&asca&1t&3.07&@xmath78&0.40&&&9&@xmath79&asca&1t&3.07&@xmath80&0.40&&&10&@xmath81&asca&1t&3.07&@xmath82&0.40&&&&&&&&&&11&@xmath83&asca&1t&3.07&@xmath84&0.40 & & @xmath85&12&@xmath86&asca&1t&3.07&@xmath87&0.40 & & &13&@xmath88&asca&1t&3.07&@xmath89&0.40 & & &14&@xmath90&asca&1t&3.07&@xmath91&0.40 & & &&&&&&&&&15&center&gis&1t&@xmath92&@xmath93 & @xmath94&&@xmath95&2.5316&@xmath96&sis&1t&@xmath97&@xmath98&@xmath99& & @xmath100&@xmath590.0117&&pspc&1t&@xmath101&@xmath102 & @xmath103&&@xmath104&25.1718&&gis&cf&@xmath105&@xmath106&@xmath94 & @xmath107&@xmath108&3.4419&&sis&cf&@xmath109&@xmath110&@xmath99 & @xmath111&@xmath112&@xmath590.0120&&pspc&cf&@xmath113&@xmath316.36&@xmath310.30&@xmath114 & @xmath115&57.6821&&all&cf&@xmath116&@xmath117&@xmath118&@xmath5932 & @xmath119&@xmath590.01 & & & & & & & & the _ rosat _ pspc spectrum of the inner 16@xmath24 of a2029 was extracted and corrected for background as discussed above and in [ sec : pspc_data ] . first , a single temperature model was fit to this spectrum ; the results are shown in the first row of table [ tab : spectra ] . the fit was not acceptable , with a @xmath120 for 23 d.o.f . this unacceptable fit gave a well - defined value for the absorbing column , but fairly uncertain values for the temperature and abundance of the gas . for the high ambient gas temperature in a2029 , the low energy band of _ rosat _ makes it relatively insensitive to changes in the temperature . moreover , at these high temperatures , the strongest spectral feature due to heavy elements is the 7 kev iron k line complex , which is beyond the band of _ rosat_. on the other hand , the low energy band of the pspc does make it very sensitive to the galactic absorbing column . the absorbing column required by the pspc spectrum is consistent with the galactic value of @xmath26 @xmath10 ( stark et al . 1992 ) . however , the temperature is lower than that given previously by the @xmath17 mpc ( @xmath18 kev ; david et al . 1993 ) or that determined below from _ asca _ ( @xmath121 kev ; [ sec : global_gis ] ) . an examination of the residuals for this single temperature fit suggested that the poor fit was due to additional soft x - ray emission . the poor fit of a single temperature spectrum to the pspc data and the previous evidence for a cooling flow at the center of the cluster suggest that a cooling flow component be added to the model spectrum . the best - fit single temperature plus cooling flow model is shown in the second row of table [ tab : spectra ] . this provided a considerably improved fit ( a reduction of @xmath122 for one additional d.o.f . ) . in this model , 53% of the counts in the pspc spectrum come from the cooling flow component . because of the soft band of the pspc , the temperature of the gas is very poorly defined in this model . at the 90% confidence limit , only a lower limit on the temperature is determined . as noted before , varying the temperature has little effect on the soft x - ray spectrum in the single temperature model , and even less in the cooling flow model ( eq . [ eq : cflow ] ) . the reason for the latter is that most of the softer emission in the cooling flow model comes from gas at which has cooled to lower temperature , so that the integral in equation ( [ eq : cflow ] ) is nearly independent of the upper limit temperature @xmath38 for @xmath123 . the abundance of heavy elements is also rather poorly determined by the single temperature plus cooling flow fit to spectrum , and only gives a lower limit . as noted above , the fe k lines can not be observed with _ rosat _ to constrain the abundances . the cooling flow component does produce strong line features within the pspc band due to heavy elements , including the l lines of iron and the k lines of c , n , o , ne , and mg . however , the strengths of these lines are not strongly dependent on the abundance of the heavy elements , at least if they are all assumed to vary together and the abundance is not too low . the reason is that , at temperatures @xmath124 kev where these lines are strongest , heavy element line emission dominates the cooling of the gas . thus , if the abundances are increased , the line emission rate is increased but the cooling time is decreased by nearly the same factor . in equation ( [ eq : cflow ] ) , increasing the abundance increases both @xmath125 in the numerator and @xmath126 in the denominator by approximately the same amount . on the other hand , the _ rosat _ pspc spectrum does give an excellent determination of the absorbing column , and a good determination of the cooling rate @xmath43 . as noted above , the absorbing column is reasonably close to the galactic value from stark et al . the cooling rate also agrees well with previous determinations ; we discuss this in more detail below ( [ sec : global_both ] ) . the _ asca _ gis2 and gis3 spectra in the photon energy range of 0.511 kev were determined for the inner @xmath0 of the cluster . we did nt use the _ asca _ sis0 or sis1 spectra , because these instruments did not cover the entire region of the cluster ( figure [ fig : pspc_image ] ) . later , we will find that the cluster has significant spatial variations in its x - ray spectrum , due to a cooling flow at the cluster center ( [ sec : cflow ] ) and due to a radial temperature gradient ( [ sec : spatial_t ] ) . thus , it would be difficult to directly compare the sis spectra with those from the gis and pspc , which cover the same larger region of the cluster . the results of fits to these spectra are summarized in rows 3 and 4 of table [ tab : spectra ] . because of the hard energy band of the gis spectra , these observations have strengths and weaknesses which are the opposites of those of the _ rosat _ pspc . the gis provides accurate measures of the ambient gas temperature and the iron abundance , but do not usefully constrain the soft x - ray absorption or the cooling rate . these two spectra were fit first with the single temperature model . this gave an accurate determination of the temperature and of the iron abundance , but only an upper limit on the absorbing column . the fit was adequate , with a @xmath44 per d.o.f . because of the evidence for a cooling flow in a2029 , we also tried a fit which included a cooling flow . this provided only a marginal improvement in the fit ( @xmath127 for one extra fitting parameter ) , and accordingly , only an upper limit on the cooling rate . because of the complementary strengths and weaknesses of the the _ asca _ and _ rosat _ spectra , and the fact that the best - fit individual spectra gave consistent results ( table [ tab : spectra ] ) , we determined the best - fit spectrum for the combined data sets of _ rosat _ pspc and _ asca _ gis2 and gis3 spectra . we allowed for a variation in the overall calibration between _ asca _ and _ rosat _ , but required that the shape of the spectrum agree . the results of fitting single temperature and single temperature plus cooling flow models to the spectra are shown in rows 5 and 6 of table [ tab : spectra ] . the abundances in these fits are essentially the same as those given by the gis spectra alone , while the temperatures are slightly lower . the absorbing columns are similar to those from the pspc alone , but with smaller uncertainties . the cooling rate in the cf model is basically that same as that given by the pspc alone . the addition of a cooling flow improves the fit significantly ( @xmath128 for one extra fitting parameter ) . this is due almost entirely to the effect of this component on the fit to the pspc spectrum . because the single temperature plus cooling flow model provides fairly good fit to both the _ rosat _ and _ asca _ data , we will adopt the spectrum in this fit as our fiducial global cluster spectrum for a2029 . the upper panel in figure [ fig : whole_asca_spect ] shows the observed global _ gis2 and gis3 spectra and this best - fit model ; the lower panel shows the residuals to the fit . figure [ fig : whole_pspc_spect ] shows the _ rosat _ pspc spectrum , model , and residuals . 4truein the hydrogen absorbing column required by this fit is in reasonable agreement with the galactic value of @xmath26 @xmath10 from stark et al . the temperature in the single temperature plus cooling flow model is higher than the value of @xmath18 kev found by david et al . ( 1993 ) with the _ einstein _ mpc , although the error bars overlap . however , david et al . fit a single temperature model to their spectrum . if one compares to the single temperature joint fit to the _ asca _ and _ rosat _ data ( row 5 in table [ tab : spectra ] ) , the agreement is quite good . the total cooling rate is in remarkably good ( and probably fortuitous ) agreement with the values of @xmath19 yr@xmath3 derived with the _ einstein _ ipc ( arnaud 1989 ) , and @xmath21 yr@xmath3 derived from the _ rosat _ hri ( sarazin et al . both of these determinations were based on the surface brightness profile of the cluster , rather than the x - ray spectrum . the cooling rate was also derived using the x - ray spectrum from the _ einstein _ sss by white et al . their value of @xmath23 yr@xmath3 is consistent with ours within their very large uncertainties . 4truein the x - ray flux in these models is @xmath12 ( 210 kev ) @xmath129 ergs @xmath10 s@xmath3 . the uncertainty shown is statistical , and is dominated by the good statistics in the _ asca _ gis spectra . however , the systematic uncertainty in the absolute calibration of these _ asca _ spectra is much larger . this flux is considerably larger than that found by david et al . ( 1993 ) , @xmath12 ( 210 kev ) @xmath130 ergs @xmath10 s@xmath3 , using the _ einstein _ mpc , which had a larger field of view than _ asca_. the _ asca _ observations were used to study the spatial variation in the x - ray spectrum , and derive the radial variation in the gas temperature and in the abundances . both sis and gis data were used . the determination of the spatial variation in the spectrum is complicated by the broad , energy - dependent point - spread - function ( psf ) of the _ asca _ mirrors . we used the technique of markevitch et al . ( 1996 ) to correct for the redistribution of x - ray photons produced by the _ asca _ psf . a set of `` model '' regions on the sky were selected , and we assumed that the shape of the x - ray spectrum was constant within each of these regions . a corresponding set of `` image '' regions were defined in the focal plane of _ asca_. generally , the image regions and model regions were identical , except that the outermost model region was larger than the outermost image region . this was done to avoid the unphysical effects of assuming that the cluster emission ended abruptly at the outer edge of the last image region . also , the image regions were restricted to the area within @xmath131 of the center of each of the gis detectors . the observed _ asca _ x - ray spectra were accumulated within each image region . for a given set of spectral parameters for the model regions , we determined the predicted spectra in the image regions , including the detector response and the redistribution of photons among image regions by the energy - dependent _ asca _ psf . the parameter of the spectra in the model regions were varied until the best - fit model ( minimum @xmath44 ) for the observed spectra in the image regions was found . an annealing technique ( press et al . 1992 ) was used to avoid false minima . we were careful to use spectral bins which were large enough that the gaussian approximation to poisson uncertainties was valid , so that the @xmath44 test could be applied . the energy - dependent _ asca _ psf was approximated by interpolation between gis images of cyg x-1 taken at various locations in the focal plane ( takahashi 1995 ) . the psf is poorly known at photon energies below 2 kev . for photon energies between 2.0 and 2.5 kev , the effective area of _ asca _ is affected by a poorly calibrated absorption edge from gold in the mirrors . thus , we have chosen to fit only photon energies above 2.5 kev in the spatial analysis . the lower energy cutoff should not affect the determinations of the high temperatures found for the gas . excluding lower energies does reduce the sensitivity of _ asca _ to the central cooling flow ( [ sec : cflow ] ) , but also reduces the calibration problems encountered with _ asca _ at low energies ( mainly below 1 kev ; [ sec : cflow_general ] ) . a relative systematic psf uncertainty ( 5% for circularly symmetric image regions , 15% for other region shapes ) was added to the statistical uncertainties . the instrumental contributions to the psf are different for the sis and gis . for modeling of the sis data , the gis cyg x-1 calibration images were corrected for the energy dependence of the intrinsic gis detector blurring by additional smoothing resulting in a final constant resolution ( gaussian @xmath132 ) . the cluster sis data also were smoothed to the same resolution . because of its superior spatial resolution , the _ rosat _ pspc image ( fig . [ fig : pspc_image ] ) was used to determine the distribution of projected emission measure within the cluster . errors due to photon statistics and the background correction in the _ rosat _ pspc image where included in the spectral analysis . at each step in the iteration to find the best - fit _ asca _ spectra , the spectrum in each model region was used to determine the emission within the band of the _ rosat _ pspc image including the _ rosat _ pspc response , and the surface brightness in the pspc image was converted into a distribution of emission measure on the sky . because there is some uncertainty in the overall _ rosat _ to _ asca _ calibration , we allow this to vary , and only use the _ rosat _ pspc image to fix the relative emission measure variation across the sky . this emission measure distribution and the spectral parameters were used to determine the predicted _ asca _ spectra within the image regions , including the redistribution by the psf . because a small miss - alignment of the _ rosat _ and _ asca _ images can affect the spectral fit , it is important to align the _ rosat _ and _ asca _ data accurately . we did this , but we also vary the offset between the two satellite data sets , and include an estimate of the resulting uncertainties in the alignment on our spectral fits . the use of the _ rosat _ pspc image to constrain the emission measure distribution in the _ asca _ only works well if the same spectral component provides the bulk of the emission in these two disjoint spectral bands ( 0.52.0 kev for the _ rosat _ image , and 2.511 kev for the _ asca _ data ) . this is probably a reasonable assumption in the outer parts of a2029 , but not in the inner cooling flow region . there , most of the _ rosat _ photons come from cooling gas , while much of the _ asca _ spectrum is contributed by hotter , ambient gas . to avoid errors due to this , we free the normalization of the central _ asca _ spectrum relative to the outer regions during the spectral fitting . two bright x - ray point sources are seen on the _ rosat _ pspc image ( fig . [ fig : pspc_image ] ) at about 4 and 9 arcmin from the cluster center . a region around each of these sources was excluded from the _ asca _ cluster spectra . however , because of the broad _ asca _ psf , some emission from each of these point sources may be included in the adjoining cluster image regions . thus , we accumulate the _ asca _ x - ray spectrum from the image regions around each of these sources , and simultaneously fit a power - law spectrum to each of these sources . ( thermal spectra were found to provide poorer fits than power - law spectra . ) we found that contamination from these sources only has a small effect on the derived spectral properties of the cluster regions except for the outermost cluster regions . because these sources may have varied between the time of the _ rosat _ and _ asca _ observations , we allow the overall normalization of the flux of these sources to vary . in fact , the _ asca _ and _ rosat _ fluxes of source a ( fig . [ fig : pspc_image ] ) agree within the uncertainties when corrected for the different spectral bands , but source b is about 3 times fainter in the _ asca _ observation than in the _ rosat _ data . the photon power - law indices of sources a and b were @xmath133 and @xmath134 , respectively , in the best - fit models . the uncertainties in derived spectral quantities are 90% confidence regions from 200 monte carlo simulations of fitting the data with random statistical and systematic uncertainties added . the radial temperature variation in a2029 was determined by accumulating the _ asca _ sis and gis spectra in four adjoining annuli with boundary radii of 0 , 1.5 , 5 , 10 , and 16 arcmin . the outer radius of the outermost model annulus was taken to be 20 arcmin . the spectra were grouped into eight spectral channels with boundary photon energies of 2.5 , 3.0 , 3.5 , 4.0 , 5.0 , 6.0 , 7.0 , 8.5 , 11.0 kev in the gis detectors . in the sis detectors , the last two of these channels were combined into a single channel from 7 to 11 kev . for the purposes of determining the temperature variation , the abundances of the heavy elements were fixed at 0.40 of the solar value , which was the best - fit abundance for the entire cluster spectrum ( [ sec : global ] ) . we initially tried to fit the spectra assuming a single temperature model for the spectrum in all of the cluster annuli . however , this did not provide a very good fit to the central circular region . even when we freed the normalization of this region ( i.e. , no longer set the normalization of this region relative to the others from the _ rosat _ pspc image ) , the fit was only improved very slightly . examination of the residuals indicated that the model was unable to provide enough soft flux in the central region . previous x - ray spectral ( e.g. , white et al . 1991 ) and spatial observations ( e.g. , sarazin et al . 1992 ) , and the central _ rosat _ pspc and _ asca _ spectra presented in this paper ( [ sec : cflow ] ) , all indicate that a2029 contains a cooling flow in its central regions . therefore , we added a cooling flow component to the central circle region in all of our spatially - resolved spectral fits . the gas was assumed to cool from the ambient temperature in this region ( the temperature of the raymond - smith component ) , and have the same abundances as the intracluster gas . the addition of this cooling flow component lowered @xmath44 by 32 for only one additional parameter . this model did provide an acceptable fit . the best - fit spectral parameters for this model for the cluster annuli are shown in rows 710 of table [ tab : spectra ] . note that the best - fit value of @xmath44/d.o.f . is significantly less than unity . this is due to the inclusion of systematic uncertainties in the evaluation of @xmath44 , as discussed above . as a result of this , we do not give the probabilities associated with these fits . the best - fit temperatures in annuli are also plotted in figure [ fig : asca_temp ] . there is some evidence for a decrease in the temperatures with radius . this trend was found in all of the fits to the _ asca _ spectra , including the fits which did not correct for the two unrelated sources or include the effects of the central cooling flow . to assess the significance of the temperature gradient , the spectra in annuli were fit assuming that the temperature was the same at all radii . the horizontal dashed line in figure [ fig : asca_temp ] gives this best - fit isothermal model for all of the spatially - resolved spectra ( @xmath135 kev ) . note that this temperature is slightly different than the best - fit temperature for the whole cluster ( also including the effects of a cooling flow ) , @xmath136 kev , for several reasons . first , this was a much worse fit to the spectra . based on the f - test and the simulations of the uncertainties in the fits , this isothermal fit can be rejected at the @xmath31 96% confidence level . to parameterize the temperature gradient , we also fit the spectra in annuli assuming that the temperature was a linear function of the radius . the slanting dashed line in figure [ fig : asca_temp ] is the best - fit linear function of radius , @xmath137 it is an improvement over the isothermal temperature distribution , with a reduction in @xmath44 of 9 for only one additional parameter . 2.5truein 0.2truein fits to the gis data alone showed the same declining trend of temperatures with radius and consistent values for the temperatures . in fits with the sis data alone , the temperatures in the inner two annuli were similar to those in the combined fit . the temperatures in the outer two annuli were very poorly determined , due to the small portion of these regions included on the sis chips ( figure [ fig : pspc_image ] ) . however , the temperatures were consistent within the uncertainties , which were very large in the outer two annuli . we also derived the abundances in the same radial rings which were used in the previous section . the values of the derived abundances are given in figure [ fig : asca_abund ] . unfortunately , the uncertainties in the individual abundances are quite large , and only upper limits are found for the outer two annuli . the is not really due to the statistical uncertainties on individual spectra . instead , it is largely a consequence of the energy dependent psf of _ asca_. the psf is particularly wide at high energies , so the fe k line complex , which is the primary spectral feature determining the abundance , can be strongly affected . when combined with the very strong central surface brightness of the fe k line complex in this cooling flow cluster , this implies that there is a very large correction for scattering of the central emission in the fe k line spectra of the outer regions . this results in very large uncertainties for the outer abundances , and causes the uncertainties in the outer abundances to be anticorrelated with the uncertainties in the central abundance . 2.5truein 0.2truein figure [ fig : asca_abund ] shows some slight evidence for either a decline in the iron abundance with radius or a higher abundance in the center cooling flow region than in the rest of the cluster . however , the difference between the spectral fit with individual abundances and a fixed overall abundance is not statistically significant ( a decrease in @xmath44 of 3.3 for three additional parameters ) . one finds that the hypothesis that the abundance is constant can only be rejected at the 51% confidence level ( i.e. , not at all ) . for this reason , we have assumed that the abundance is constant at the value set by the global cluster spectrum in the determinations of the cluster temperature variation . we also determined the azimuthal variation in the x - ray temperature in a2029 . as shown in figure [ fig : pspc_image ] , the x - ray emission in a2029 is elongated at a position angle ( @xmath138 ) of approximately @xmath139 ( measured from the north to the east ) . we divided the region of the cluster between radii of 3@xmath24 and 16@xmath24 into four annular sections , each @xmath140 wide , with the first centered on @xmath141 . the temperature was determined independently for the central 3 arcmin radius circle and for each of these annular sections . these regions are shown in figure [ fig : pspc_slices ] . as before , regions were used to isolate the two point sources a and b. the abundance was fixed at 0.40 of solar , the best - fit value for the cluster spectrum derived above . the relative systematic psf uncertainty was increased to 15% because of the shape of the regions used . 2.7truein 0.2truein the resulting temperatures for the four regions are listed in rows 11 - 14 of table [ tab : spectra ] , and are shown as a function of the position angle in figure [ fig : tslices ] . there is some evidence that the gas is hotter to the northwest , and cooler to the southeast . we also fit a single temperature to all four sectors ; this is shown as a dashed horizontal line in figure [ fig : tslices ] . formally , the measured temperatures disagree with this single temperature fit at almost the 90% confidence level . however , the presence of the two point sources within the cooler sector and increased systematic uncertainties in fitting the spectra of regions which are not azimuthally symmetric argues for caution in interpreting these results . we conclude that the uniformity of the gas temperature with azimuthal angle can not be ruled out at the greater than 90% confidence level . we also extracted the spectrum of the inner region of the cluster to study the properties of the cooling flow in more detail . the sis and gis spectra were extracted for the inner 3 arcmin ( 360 kpc ) in radius of the cluster . this region just fits on the s0c1 and s1c3 chips , which are the `` best '' chips on the two instruments and the ones normally used to determine the spectra of point sources . we note that our treatment of the central cooling flow spectra differed from that of the previous analysis of the global cluster spectrum ( [ sec : global ] ) or of the spatially resolved spectra ( [ sec : spatial ] ) in several ways . first , we included the spectra from the two sis instruments ; these were excluded in the analysis of the global spectrum because the cluster extended beyond the field of view of the sis . second , we include the lower energy channels in the spectra , which were not included in the spatial analysis in [ sec : spatial ] . because our interest here is in the soft x - ray emitting cooling flow , we extend the spectral analysis down to energies of 0.35 kev in the sis and 0.55 kev in the gis . 2.5truein 0.2truein the energy - dependent psf of _ asca _ can cause problems in determining the spectrum of any component of an extended x - ray source like a cluster . in [ sec : spatial ] , we corrected for these effects while simultaneously fitting simple models to the x - ray spectrum in several regions of the the cluster . unfortunately , we will fit more complicated models to the spectrum of the cooling flow , and such a simultaneous fit is not practical with these models . also , to study the cooling flow , it is important to extend the spectral analysis down to lower energies where the psf is not well - understood . we expect that the problems associated with the energy - dependent psf will be much smaller for the cooling flow region than for the outer cluster regions , because of the large surface brightness peak associated with the cooling flow . to test this , we have determined the contribution of outer x - ray emission scattered into the 3 arcmin circle from outer regions , assuming the _ rosat _ pspc image for the cluster and the best - fit spectral parameters as a function of position from [ sec : spatial ] . in the gis , we find that scattered light from outer regions contributes roughly 8% to the flux in the central region . the values are somewhat smaller for the sis . there is also an energy - dependent loss of flux from the central region due to the psf . during the spectral analysis , the program ascaarf corrects for this effect . we also extracted the _ rosat _ pspc spectrum from the same region . all of the spectra were binned to insure that there were at least 20 counts per spectral channel , so that the @xmath44 test should give reliable results . single temperature model fits to the gis , sis , and pspc spectra are shown in rows 1517 of table [ tab : spectra ] . individually , these are probably acceptable fits . however , we could not find an acceptable common fit to all of the spectra . there are two inconsistencies between the fits to the different instruments . first , the absorbing column required by the sis spectrum was about twice that found with the pspc or the gis . second , the temperatures found by each of the instruments disagreed at a level beyond the uncertainties . it is interesting that the temperature increased with the hardness of the spectral response of the instrument . this suggests that multiple temperature components are present , and that each instrument shows the gas which emits more strongly in its spectral band . the three instruments did give consistent values for the heavy element abundance , and the values for the central spectra are in good agreement with those determined for the entire cluster . this agrees with our previous conclusion in [ sec : spatial_ab ] that the abundances are nearly constant within a2029 . because of the inconsistent temperatures found with the different instruments and the other evidence for a cooling flow at the center of a2029 ( [ sec : intro ] , [ sec : global ] , & [ sec : spatial_t ] ) , we also fit the single temperature plus cooling flow model to the spectra from each of the instruments . the best - fit spectral parameters are listed in rows 1820 of table [ tab : spectra ] . the data , best - fit spectral models , and residuals for the three instruments are shown in figures [ fig : center_gis_spect ] , [ fig : center_sis_spect ] , and [ fig : center_pspc_spect ] . in all three cases , the addition of a cooling flow significantly improved the fit . the abundances are all consistent with one another and with the single temperature models . the temperatures from the three instruments are now all marginally consistent within the uncertainties . the cooling rates are also all consistent , but the uncertainties are rather large . a cooling rate of about 250 @xmath50 yr@xmath3 within the central 3 arcmin would be most consistent with the three individual spectral fits . 4truein 4.1truein 4.1truein unfortunately , the required absorbing columns disagree even more strongly for the cooling flow model than they did in the single temperature model . when we tried to simultaneously fit the spectra from all three instruments with a single model , we were unable to find any acceptable fit . we allowed the normalization of the model to vary from instrument to instrument to account for any error in the overall calibration of the instruments , and only required that the model have the same spectral shape . still , no acceptable common fit was found . the last row in table [ tab : spectra ] shows the best - fit common single temperature plus cooling flow model fit to all of the spectra . it is a rather poor fit . examination of the residuals to such fits showed that the problem was primarily due to a difference in the shape of the spectrum and the fluxes at low photon energies ( @xmath1421 kev ) between the _ asca _ sis and the _ rosat _ pspc . we tried a number of different models for the origin of the soft x - ray absorption ( e.g. , partial covering models , intrinsic absorption models , and models with intrinsic absorption applied only to the cooling flow ) . we also considered various alternative models for the cooling flow spectra ( e.g. , non - isobaric cooling or a finite limiting temperature for the cooling ) . none of these provided an acceptable common spectral fit or a significant improvement on the unacceptable common cooling flow fit . these problems suggest that there are significant calibration errors in the sis data at low energies at present . similar calibration problems and overestimated absorbing columns with the sis have been found by other observers ( e.g. , mukai 1995 ) . these low energy sis calibration uncertainties should not affect the previous results ( [ sec : global ] & [ sec : spatial ] ) , because low energy sis data was not used there . white et al . ( 1991 ) reanalyzed the @xmath17 observatory solid state spectrometer ( sss ) spectrum of the central 3 arcmin radius of a2029 ( essentially the same region as used to extract the _ asca _ and _ rosat _ pspc spectra ) . white et al . found an excess absorption column of @xmath25 @xmath10 ( if the excess absorber is modeled as galactic material ) . in their fits to the spectrum , the absorber was treated as lying in the foreground of all of the x - ray emission . such a large foreground excess column is completely inconsistent with the _ rosat _ pspc spectrum . the 90% upper limit on the foreground excess absorbing column in the inner 3 arcmin of a2029 from the pspc spectrum is @xmath143 @xmath10 . figure [ fig : center_pspc_white_spect ] shows that best - fit single temperature plus cooling flow model for the pspc spectrum of the inner 3 arcmin radius of a2029 , with the absorption fixed at the value from white et al . this is a terrible fit to the spectrum , with a @xmath44 of 2417 for 23 d.o.f . the strong soft x - ray emission in the spectrum from the center of the a2029 cluster is inconsistent with any large amount of excess foreground absorption . the white et al . value for the excess foreground absorption is also inconsistent with the _ asca _ sis and gis spectra of the central 3 arcmin of a2029 ( rows 18 and 19 in table [ tab : spectra ] ) , albeit at a lower level of significance . if the excess absorption is strongly concentrated to the center of the cooling flow , then its effects would be diluted in the 3 arcmin aperture used to collect the cooling flow spectrum . in order to provide a stronger constraint on excess absorption , we also extracted the _ rosat _ pspc spectrum from the inner 1 arcmin radius of the cooling flow . to provide a conservative upper limit on any foreground excess absorption , we compared the absorption derived from this inner region with both the galactic column of @xmath26 @xmath10 from stark et al . ( 1992 ) , and the galactic absorption derived from the pspc spectrum of the outer parts of the cluster ( the annulus from 316 arcmin in radius ) . we adopted the larger value of the upper limit on the absorption for these two determinations of the galactic column . the excess absorber was treated as a foreground screen , but at the redshift of the cluster ( this increases slightly the limits on the excess column ) . this gave a 90% upper limit on the column of foreground excess absorption of @xmath144 4.3truein the detection of excess absorption toward the central regions of cooling flow clusters of galaxies would suggest that the absorber is physically located with the cooling flow region ( white et al . 1991 ; allen et al . 1993 ; irwin & sarazin 1995 ) . in this case , it is much more difficult to detect or constrain the excess absorption . to illustrate this , we consider a model in which the excess absorption is applied only to the cooling flow component of the spectrum , but not to the ambient cluster emission . if this model is fit to the pspc spectra of the inner 3 arcmin of a2029 , the best - fit value of the excess absorption is still zero . however , the 90% upper limit on the excess absorption is very large , @xmath145 the reason for this very weak limit is that the value of the excess absorption and the cooling rate are very strongly correlated in the fits . if one increases the amount of soft x - ray emission by increasing the cooling rate @xmath43 , one can increase the excess soft x - ray absorption @xmath146 if it is applied only to this cooling flow emission . our conclusion is that the pspc spectra ( and , to a lesser extent , the _ asca _ spectra ) rule out the large foreground absorption claimed by white et al . however , an absorber confined to the cooling flow region is much more difficult to detect or constrain with pspc spectra . in principle , _ asca _ sis spectra could give a stronger constraint , but the inconsistency between the _ asca _ sis and _ rosat _ pspc spectra in a2029 suggests that there are still significant calibration uncertainties with the _ asca _ sis spectra at low energies . for the purposes of determining the x - ray gas pressure in the inner region where the radio source is located ( huang & sarazin 1998 ) , we also determined the best - fit pspc temperature for the inner arcmin . this gave @xmath147 kev . we used the information on the temperature and temperature distribution from the _ asca _ spectra to determine the gas and total mass distribution in a2029 . the first step in this process was to determine the x - ray surface brightness distribution , using the _ rosat _ pspc image ( fig . [ fig : pspc_image ] ) . the x - ray count rate was determined in concentric circular annuli , excluding regions around each of the point sources in the field . the surface brightness was corrected for background using the same regions used for the background correction of the spectra ( [ sec : pspc_data ] ) . the resulting x - ray surface brightness profile for the spectral band 0.52.0 kev is shown in figure [ fig : pspc_profile ] . a sufficient number of counts were collected in each annulus to insure that gaussian statistics applied . possible faint emission is seen extending to a radius of at least 22 arcmin ( @xmath148 mpc ) , but the uncertainties in the surface brightness are very large , due to the low surface brightness and the effect of the instrumental support structure in this region . in the subsequent analysis ( and in fig . [ fig : pspc_profile ] ) , we consider only the surface brightness at radii @xmath149 arcmin ( @xmath150 mpc ) where the uncertainties are more reasonable . 3.75truein we fitted the x - ray surface brightness in figure [ fig : pspc_profile ] with the isothermal `` beta '' model @xmath151^{-3 \beta + 1/2 } \ , . \label{eq : beta_model}\ ] ] the model was convolved with the pspc psf and compared to the data . when all of the data points in figure [ fig : pspc_profile ] were included , no acceptable fit could be found . the best - fit model , shown as a dashed curve in figure [ fig : pspc_profile ] , had @xmath152 for 27 d.o.f . as has generally been found with cooling flows clusters ( e.g. , jones & forman 1984 ) , the beta model does not fit the sharply peaked central surface brightness within the cooling radius . thus , we fit the surface brightness with the beta model , progressively removing the interior points until an acceptable fit ( @xmath153d.o.f @xmath154 ) was found . this required that the data interior to 164@xmath155 ( @xmath156 mpc ) be eliminated . when these points were removed , an acceptable fit with @xmath157 arcsec and @xmath158 was found . our value of @xmath159 is in very good agreement with that given by david et al . ( 1995 ) of @xmath160 . our fit , which gave @xmath161 for 16 d.o.f . ( @xmath153d.o.f @xmath162 ) , is shown as the solid curve in figure [ fig : pspc_profile ] . because the derived core radius is within the region of data excluded from the fit and where the beta - model does nt fit , the value of the core radius and central surface brightness @xmath163 are probably not given reliably by this fit . however , this model does appear to provide a reasonable fit for the surface brightness beyond about 160 arcsec . because we are most interested in the mass at large radii , we will use this fit to describe the gas distribution . we adopt the linear fit to the radial temperature in the gas derived from the _ asca _ spectra ( eq . [ eq : t_linear ] ) . assuming the abundances of the heavy elements are 0.40 of the solar value , which was the best - fit abundance for the global cluster spectrum , we inverted the best - fit beta - model to determine the density of the gas @xmath164 and its pressure @xmath165 . the density of the gas was integrated over the interior volume to give the interior gas mass as a function of the radius , @xmath166 . the total gravitational mass was determined from the assumption that the gas is in hydrostatic equilibrium , which implies that @xmath167 the uncertainties in @xmath166 and @xmath168 were determined from the uncertainties and covariances in the coefficients of the fit to the temperature as a function of radius ( eq . [ eq : t_linear ] and fig . [ fig : asca_temp ] ) and in the pspc surface brightness fit ( eq . [ eq : beta_model ] and fig . [ fig : pspc_profile ] ) . 5.0truein the resulting mass profiles are shown in figure [ fig : masses ] . at a radius of 16@xmath24 ( @xmath1 mpc , the largest region used for the temperature determination ) , the masses are @xmath5 and @xmath4 , and the ratio of masses is @xmath169 . the fraction of the total cluster mass which is due to the intracluster gas is rising steeply with radius at the largest radii . a very similar result was found for a2029 by david et al . ( 1995 ) assuming an isothermal gas profile , and for abell 2256 by markevitch & vikhlinin ( 1997 ) using an _ asca _ temperature profile . we have analyzed the _ asca _ and _ rosat _ pspc x - ray observations of the rich cluster abell 2029 . the _ asca _ gis and _ rosat _ pspc are in good agreement on the global x - ray spectrum , and give an average ambient gas temperature of @xmath70 kev ( including the effects of a cooling flow ) . if the gas temperature is assumed to be constant in the analysis of the spatially resolved _ asca _ sis and gis , the average temperature is found to be 8.6 kev . the iron abundance in the gas is @xmath7 of the solar value . there is no significant evidence for any variation in the abundance with position in the cluster . the global x - ray spectra , central x - ray spectra , and _ rosat _ surface brightness all require a cooling flow at the cluster center . the global x - ray spectrum implies that the total cooling rate is @xmath8 yr@xmath3 . the global x - ray spectra are consistent with the galactic value for the soft x - ray absorption toward the cluster . the _ rosat _ pspc spectra of the central regions of the cluster of completely inconsistent with the large value of foreground excess absorption found by white et al . ( 1991 ) based on the _ einstein _ sss spectrum . the upper limit on excess foreground absorption is @xmath9 @xmath10 . however , the spectra do not rule out a significant amount of intrinsic absorbing gas located within the cooling flow region . one concern with modeling the cooling flow spectrum and absorption in a2029 is that the _ asca _ sis and _ rosat _ pspc spectra do not appear to be consistent . this may indicate that there are calibration problems with the _ asca _ sis spectra at low energies . the pspc image shows that the cluster is elliptical , but is very regular and smooth . this agrees with previous analyses of the optical and x - ray distribution of the cluster ( slezak et al . 1994 ; david et al . 1995 ; buote & tsai 1996 ) . we also find that there is no significant evidence for any irregularities in the temperature distribution in the cluster , as would be produced by a subcluster merger . structure in the x - ray surface brightness and particularly the gas temperature distribution has been interpreted as evidence for mergers and other hydrodynamic activity in many other clusters ( e.g. , henry & briel 1995 ; markevitch et al . 1996,1997 ) . the lack of structure in the x - ray properties of a2029 suggests that the cluster is relaxed and that the gas is in hydrostatic equilibrium . we use the assumption of hydrostatic equilibrium to determine the gravitational mass of the cluster as a function of radius . within a spherical radius of @xmath0 ( @xmath1 mpc ) , the total gravitational mass is @xmath4 , while the mass of gas is @xmath5 . the gas fraction of the cluster at this radius is @xmath169 , and the gas fraction is is increasing with radius at the largest radii . thus , a2029 is a particularly strong example of the so - called `` baryon catastrophe '' in clusters ( e.g. , david et al . 1995 ) . we thank the referee , joel bregman , for a number of helpful suggestions . c. l. s. was supported in part by nasa rosat grants nag 51891 , nag 53308 , nasa asca grant nag 5 - 2526 , and nasa astrophysical theory program grant 5 - 3057 . m. l. m. was supported by nasa grant nag5 - 2611 .

we prove a version of the negative norm theorem in orlicz - sobolev spaces . a study of continuity properties of the -operator between orlicz spaces is a crucial step , of independent interest , in our approach . applications to the problem of pressure reconstruction for non - newtonian fluids governed by constitutive laws , which are not necessarily of power type , are presented . a key inequality for a numerical analysis of the underlying elliptic system is also derived .

1401.4885

assume that @xmath0 is a domain , namely a connected open set , in @xmath1 , with @xmath2 , and let @xmath3 . the negative sobolev norm of the distributional gradient of a function @xmath4 can be defined as @xmath5 here , @xmath6 stands for the divergence operator , and @xmath7 , the hlder conjugate of @xmath8 . moreover , in , and in similar occurrences throughout the paper , we tacitly assume that the sumpremum is extended over all functions @xmath9 which do not vanish identically . observe that the notation @xmath10 is consistent with the fact that the quantity on the right - hand side of agrees with the norm of @xmath11 , when regarded as an element of the dual of @xmath12 , where @xmath13 denotes the sobolev space of @xmath1-valued functions in @xmath0 with zero traces . definition goes back to neas @xcite , who showed that , if @xmath0 is regular enough a bounded lipschitz domain , say and @xmath14 , then the @xmath15 norm of a function is equivalent to the @xmath16 norm of its gradient . namely , there exist positive constants @xmath17 and @xmath18 , such that @xmath19 for every @xmath20 , where @xmath21 the mean value of @xmath22 over @xmath0 , and @xmath23 denotes the lebesgue measure of @xmath0 . this result is known as neas negative norm theorem . in the present paper we are concerned with a version of this theorem when a negative norm is introduced with the lebesgue space @xmath15 replaced with a general orlicz space . loosely speaking , orlicz spaces extend lebesgue spaces in that the role of the power @xmath24 in their definition is played by a more general convex function ( a precise definition is recalled in the next section ) . clearly , a full analogue of can not hold for arbitrary orlicz spaces , since fails , for instance , in the borderline cases when either @xmath25 , or @xmath26 . our main result in this connection asserts that , however , an inequality in the spirit of still holds if , on the leftmost side , an orlicz norm appears which , in general , has to be slightly weaker than that on the rightmost side . a precise balance between the relevant norms for a conclusion of this kind to hold is the content of theorem [ thm:3.2 ] , section [ main ] . a key step in our approach is an analysis , of possible independent interest , of the divergence equation in orlicz spaces , via boundedness properties of the ( gradient of the ) operator in these spaces . @xmath27 is introduced with the lebesgue space @xmath15 is replaced with a more general orlicz space @xmath28 associated with a young function @xmath29 , and , accordingly , the sobolev space @xmath13 replaced with the orlicz - sobolev space @xmath30 , associated with the conjugate function @xmath31 to @xmath29 . a full analogue of can not hold , since fails , for instance , in the borderline cases when @xmath25 or @xmath26 . our main motivation for a discussion of negative orlicz norms are applications to a mathematical model for non - newtonian fluids . in the stationary case , the relevant model tells us that the velocity field @xmath32 and the pressure @xmath33 of a fluid solve the following system of partial differential equations : @xmath34 see , for instance , @xcite . here , @xmath35 is a positive constant , whose physical meaning is the density of the fluid , the operation @xmath36 denotes tensor product , the function @xmath37 describes the given volume forces , and the stress deviator @xmath38 is related to @xmath9 via a constitutive law . of course , the physically relevant dimensions are @xmath39 and @xmath40 . in a common model for fluids with non - newtonian behavior , the dependence of @xmath41 on @xmath9 is through a nonlinear function of the symmetric part @xmath42 of its @xmath43-valued gradient , defined as @xmath44 , where @xmath45 stands for transpose . lebesgue and usual sobolev spaces provide an appropriate functional framework for the study of existence , uniqueness and regularity of solutions to when this nonlinear function is of power type . on the other hand , if nonlinearities of non - polynomial type are allowed , such as in the eyring - prandtl model @xcite , the more flexible orlicz and orlicz - sobolev spaces have to be called into play . in particular , in section [ fluids ] we show how our negative orlicz norm theorem applies in the description of a suitable space for the pressure @xmath46 in . a related result of use in a finite elements method for a numerical analysis of is also established . many problems in pdes are related to distributions defined on sobolev - spaces . for instance on @xmath47 , with @xmath48 bounded and @xmath49 , one considers the dual space @xmath50 . it is isometrically isomorphic to the closure of @xmath15 w.r.t . the norm @xmath51 the norm introduced above is called negative norm of a lebesgue function . of course it is naturally defined also for its distributional gradient by @xmath52 neas @xcite showed the equivalence of the @xmath53-norm of a function and the negative norm of its gradient , i.e. @xmath54 for every @xmath55 with @xmath56 . inequality ( [ eq:1.1 ] ) is known as neas negative norm theorem . the main application he had in mind was korn s inequality : @xmath57 for every @xmath58 . korn s inequality , which involves the symmetric gradient @xmath59 of a function @xmath32 , has important consequences in fluid mechanics and in the mechanics of solids ( see @xcite and @xcite ) . due to the formula @xmath60 it follows easily by ( [ eq:1.1 ] ) . + a further application of the negative norm theorem is the pressure reconstruction in fluid mechanics . in the stationary case one has to find the velocity field @xmath32 and the pressure @xmath33 of a fluid motion as solution of the following set of pdes : @xmath61 here @xmath37 is a given system of volume forces and ( in case of a generalized newtonian fluids ) one has a relation between the stress deviator @xmath62 and @xmath42 ( see section 4 for more details ) . it is common to split the problem in two parts . first one has to find a sobolev function @xmath9 such that @xmath63 holds in the sense of distributions on @xmath64 . ] by de rahms theorem one can find a distribution @xmath46 such that @xmath65 if @xmath66 for some @xmath67 we obtain from ( [ eq:1.1 ] ) that @xmath46 can be represented by a @xmath68-function and @xmath69 a generalization of ( [ eq:1.1 ] ) is shown in @xcite . let @xmath70 be a young function with @xmath71 and @xmath72-condition ( see section 2 for definitions ) then a version of ( [ eq:1.1 ] ) holds on the orlicz space @xmath73 . it is straightforward to extend the concepts of negative norms and boundary values to the fashion of orlicz - spaces in this situation since it is reflexive and smooth functions are dense . the results follow from an extrapolation argument and ( [ eq:1.1 ] ) . however this is only possible for `` nice '' young functions and the limit cases are not included . those are not only interesting in order to complete the mathematical theory but also for the applications in fluid mechanics . we will explain in detail in section 4 why the cases @xmath74 and @xmath75 are important for the study of the pressure in the prandtl - eyring fluid model invented in @xcite . + the aim of this paper is the extension of the negative norm concept to general orlicz spaces and relation between a young function and a its distributional gradient . without the @xmath71 and @xmath72-condition we can not expect that it can be described by meanings of the same orlicz space this is also the case for singular integral operators and korn s inequality ( see @xcite and @xcite ) . in fact we will give conditions for young functions @xmath29 and @xmath76 such that @xmath77 holds ( see theorem [ thm:3.2 ] ) . in our proof we use the connection between the solution to the divergence - equation given by the -operator @xmath78 and negative norms . hence in theorem [ thm:3.1 ] we first give conditions for young functions such that @xmath79 this result is also interesting by itself since the @xmath80 does not satisfy the conditions supposed on singular integral operators in @xcite . + our main application is connected to the problem of finding the pressure for stationary navier - stokes equations . given a function @xmath81 such that @xmath82 we give conditions for a young function @xmath76 such that there is @xmath83 with @xmath84 we also investigate the same problem on a discrete level using finite elements . precisely we give conditions for an inf - sup condition in orlicz spaces which allows the reconstruction of the pressure . in the present section we collect some definitions and fundamental results from the theory of orlicz and orlicz - sobolev spaces . we refer to @xcite for a comprehensive treatment of this topic . a function @xmath85 $ ] is called a young function if it is convex , left - continuous , and neither identically equal to @xmath86 , nor to @xmath87 . thus , with any such function , it is uniquely associated a ( nontrivial ) non - decreasing left - continuous function @xmath88 $ ] such that @xmath89 the young conjugate @xmath90 of @xmath29 is the young function defined by @xmath91 note the representation formula @xmath92 where @xmath93 denotes the ( generalized ) left - continuous inverse of @xmath94 . one has that @xmath95 moreover , @xmath96 for any young function @xmath29 . then @xmath97 to see this , recall that @xmath98 thus , @xmath99 if @xmath29 is any young function and @xmath100 , then @xmath101 as a consequence , if @xmath100 , then @xmath102 where @xmath103 denotes the ( generalized ) right - continuous inverse of @xmath29 . a young function @xmath29 is said to satisfy the @xmath71-condition if there exists a positive constant @xmath104 such that @xmath105 we say that @xmath29 satisfies the @xmath72-condition if there exists a constant @xmath106 such that @xmath107 for @xmath108 . if [ resp . ] just holds for @xmath109 for some @xmath110 , then @xmath29 is said to satisfy the @xmath71-condition [ @xmath72-condition ] near infinity . we shall also write @xmath111 [ @xmath112 to denote that @xmath29 satisfies the @xmath71-condition [ @xmath72-condition ] . one has that @xmath113 [ near infinity ] if and only if @xmath114-condition [ near infinity ] . a young function @xmath29 is said to dominate another young function @xmath76 [ near infinity ] if there exists a positive constant @xmath104 @xmath115\,.}\ ] ] the functions @xmath29 and @xmath76 are called equivalent [ near infinity ] if they dominate each other [ near infinity ] . let @xmath0 be a measurable subset of @xmath1 , and let @xmath29 be a young function . the luxemburg norm , associated with @xmath29 , is defined as @xmath116 for any measurable function @xmath117 . the collection of all functions @xmath22 for which such norm is finite is called the orlicz space @xmath28 , and is a banach function space . the subspace of @xmath28 of those functions @xmath22 such that @xmath118 will be denoted by @xmath119 . a hlder type inequality in orlicz spaces takes the form @xmath120 for every @xmath121 . if @xmath29 dominates @xmath76 , then @xmath122 with embedding norm depending on the constant @xmath104 appearing in . when @xmath123 , embedding also holds if @xmath29 dominates @xmath76 just near infinity , but , in this case , the embedding constant also depends on @xmath76 , @xmath124 and @xmath23 . the decreasing rearrangement @xmath125 $ ] of measurable function @xmath117 is the ( unique ) non - increasing , right - continuous function which is equimeasurable with @xmath22 . thus , @xmath126 the equimeasurability of @xmath22 and @xmath127 implies that @xmath128 for every @xmath129 . the lebesgue spaces @xmath15 , corresponding to the choice @xmath130 , if @xmath131 , and @xmath132 for @xmath133 $ ] and @xmath134 for @xmath135 , if @xmath26 , are a basic example of orlicz spaces . other customary instances of orlicz spaces are provided by the zygmund spaces @xmath136 , and by the exponential spaces @xmath137 . if either @xmath138 and @xmath139 , or @xmath25 and @xmath140 , then @xmath141 is the orlicz space associated with a young function equivalent to @xmath142 near infinity . given @xmath143 , @xmath144 denotes the orlicz space built upon a young function equivalent to @xmath145 near infinity . the orlicz space @xmath146 of @xmath1-valued measurable functions on @xmath0 is defined as @xmath147 , and is equipped with the norm given by @xmath148 for @xmath149 . the orlicz space @xmath150 of @xmath151 matrix - valued measurable functions on @xmath0 is defined analogously . assume now that @xmath0 is an open set . the orlicz - sobolev space @xmath152 is the set of all weakly differentiable functions in @xmath28 whose gradient also belongs to @xmath28 . it is a banach space endowed with the norm @xmath153 we also define the subspace of @xmath152 of those functions which vanish on @xmath154 as @xmath155 in the case when @xmath130 for some @xmath156 , and @xmath157 is regular enough , such definition of @xmath73 can be shown to reproduce the usual space @xmath47 defined as the closure in @xmath158 of the space @xmath159 of smooth compactly supported functions in @xmath0 . in general , the set of smooth bounded functions is dense in @xmath160 only if @xmath29 satisfies the @xmath161-condition ( just near infinity when @xmath162 ) , and hence , for arbitrary @xmath29 , our definition of @xmath73 yields a space which can be larger than the closure of @xmath163 in @xmath73 even for smooth domains . on the other hand , if @xmath0 is a lipschitz domain , then @xmath164 where @xmath165 is defined as usual . recall that an open set @xmath0 is called a lipschitz domain if it is bounded and there exists a neighborhood @xmath166 of each point of @xmath157 such that @xmath167 is the subgraph of a lipschitz continuous function of @xmath168 variables . an open set @xmath0 is said to have the cone property if there exists a finite cone @xmath169 such that each point of @xmath0 is the vertex of a finite cone contained in @xmath0 and congruent to @xmath169 . clearly , any lipschitz domain has the cone property , but the converse is not true in general . a poincar type inequality in orlicz - sobolev spaces tells us that , if @xmath0 is a lipschitz domain , then there exists a constant @xmath104 , depending on @xmath170 and on the lipschitz constant of @xmath0 , such that @xmath171 for every @xmath172 . inequality is established in ( * ? ? ? * lemma 4.1 ) in the special case when @xmath0 is a ball . its proof makes use of a rearrangement type inequality for the norm @xmath173 which holds , in fact , for sobolev functions @xmath22 on any lipschitz domain @xmath0 ( * ? ? ? * lemma 4.1 and inequality ( 3.5 ) ) . the same proof then applies to any lipschitz domain , and one can verify that the constant in the resulting poincar inequality has the form claimed in . the orlicz - sobolev space @xmath174 of @xmath1-valued functions is defined as @xmath175 , and equipped with the norm @xmath176 . the space @xmath177 is defined accordingly . let @xmath29 be a young function , and let @xmath0 be a bounded domain in @xmath1 . we define the negative orlicz - sobolev norm associated with @xmath29 of the distributional gradient of a function @xmath178 as @xmath179 the alternative notation @xmath180 will also occasionally be employed to denote the negative orlicz - sobolev norm @xmath181 associated with the orlicz space @xmath28 . our orlicz - sobolev space version of the negative norm theorem involves pairs of young functions @xmath29 and @xmath76 which obey the following balance conditions : @xmath182 and @xmath183 for some positive constant @xmath184 . let us mention that assumptions and also come into play in a version of the korn inequality for the symmetric gradient in orlicz spaces @xcite . [ thm:3.2 ] let @xmath29 and @xmath76 be young functions fulfilling and . assume that @xmath0 is a bounded domain with the cone property in @xmath1 , @xmath2 . then there exist constants @xmath185 and @xmath18 such that @xmath186 for every @xmath178 . here , @xmath184 denotes the constant appearing in and . [ 3.2bis ] inequality continues to hold even if conditions and are just fulfilled for @xmath187 for some @xmath188 , but with constants @xmath189 and @xmath190 depending also on @xmath29 , @xmath76 , @xmath191 and @xmath23 . indeed , the young functions @xmath29 and @xmath76 can be replaced , if necessary , with young functions equivalent near infinity , and fulfilling and for every @xmath192 . owing , such replacement leaves the quantities @xmath193 , @xmath194 and @xmath195 unchanged , up to multiplicative constants depending on @xmath29 , @xmath76 , @xmath191 and @xmath23 . as recalled in section [ intro ] , the standard negative - norm theorem expressed by breaks down in the borderline cases when @xmath25 or @xmath26 . this shows that , in general , equation can not hold with @xmath196 on the left - hand side . in fact , condition , or fails , with @xmath196 , if , loosely speaking , the norm @xmath197 is close " to @xmath198 , or to @xmath199 , respectively . note that , if either or holds , then @xmath29 dominates @xmath76 globally ( * ? ? ? * proposition 3.5 ) . in a sense , assumptions and provide us with a quantitative information on if , and how much , the norm @xmath200 has to be weaker than @xmath197 for a version of the negative - norm theorem to be restored in orlicz - sobokev spaces . however , if @xmath111 , then certainly holds with @xmath196 ( * ? ? ? * theorem 1.2.1 ) . hence , in this case , assumption can be dropped in theorem [ thm:3.2 ] . on the other hand , if @xmath201 , then @xmath202 and hence holds with @xmath196 , and assumption can be dropped in theorem [ thm:3.2 ] . in particular , if @xmath203 , then both conditions and are fulfilled with @xmath196 . hence , we have the following corollary which also follows from the results of @xcite . [ cor:3.2 ] assume that @xmath0 is a bounded domain with the cone property in @xmath1 , @xmath2 . let @xmath29 be a young function in @xmath204 . then there exist a constants @xmath205 and @xmath18 such that @xmath206 for every @xmath178 . [ exlog ] assume that @xmath207 is a young function equivalent to @xmath208 near infinity , where either @xmath138 and @xmath209 , or @xmath25 and @xmath210 . hence , if @xmath162 , then @xmath211 assume that @xmath0 is a bounded domain with the cone property in @xmath1 . if @xmath138 , then @xmath212 , and hence corollary [ cor:3.2 ] tells us that @xmath213 for every @xmath178 . however , if @xmath25 , then @xmath214 , but @xmath215 . an application of theorem [ thm:3.2 ] now yields @xmath216 for every @xmath178 . in particular , @xmath217 for every @xmath178 . [ exexp ] let @xmath143 , and let @xmath207 be a young function equivalent to @xmath218 near infinity . then @xmath219 if @xmath162 . one has that @xmath220 , but @xmath221 . theorem [ thm:3.2 ] ensures that , if @xmath0 is a bounded domain with the cone property in @xmath1 , then @xmath222 for every @xmath178 . moreover , @xmath223 for every @xmath178 . [ ex1 ] if @xmath138 and @xmath224 , then @xmath225 for every @xmath226 . if @xmath227 , then @xmath228 for every @xmath229 . [ exrn ] let @xmath29 and @xmath76 be young functions such that @xmath230 where either @xmath231 and @xmath232 , or @xmath233 and @xmath234 , and either @xmath235 and @xmath224 , or @xmath25 and @xmath236 , and @xmath237 near @xmath86 , and @xmath238 near infinity . then @xmath239 for every compactly supported function @xmath240 . [ ex2 ] assume that @xmath138 and @xmath224 . then @xmath241 for every @xmath242 . if @xmath227 , then @xmath243 for every @xmath244 . [ ex3 ] assume that @xmath143 . then @xmath245 for every @xmath246 . our proof of theorem [ thm:3.2 ] relies upon an analysis of the divergence equation @xmath247 in orlicz spaces . this is the objective of the next result . in what follows , we set @xmath248 and @xmath249 [ thm:3.1 ] assume that @xmath0 is a bounded domain with the cone property in @xmath1 , @xmath2 . let @xmath29 and @xmath76 be young functions fulfilling and . then there exists a bounded linear operator @xmath250 such that @xmath251 and @xmath252 for every @xmath253 . in particular , there exists a constant @xmath254 such that @xmath255 and @xmath256 for every @xmath253 . here , @xmath184 denotes the constant appearing in and . the proof of theorem [ thm:3.1 ] in turn makes use of a rearrangement estimate , which extends those of ( * ? ? ? * theorem 16.12 ) and @xcite , for a class of singular integral operators of the form @xmath257 for an integrable function @xmath258 . here , the kernel @xmath259 fulfils the following properties : ( i ) @xmath260 ( ii ) @xmath261 ( iii ) for every @xmath262 , there exists a constant @xmath189 such that @xmath263 where @xmath264 denotes the unit sphere , centered at @xmath86 , in @xmath265 , and @xmath266 stands for the @xmath267-dimensional hausdorff measure ; \(iv ) there exists a constant @xmath190 such that @xmath268 and , if @xmath269 , then @xmath270 let @xmath0 be a bounded open set in @xmath1 , and let @xmath271 be a kernel satisfying . if @xmath272 and @xmath273 in @xmath274 , then the singular integral operator @xmath275 given by is well defined for a.e . @xmath276 , and there exists a constant @xmath277 such that @xmath278 , the boundedness of singular integral operators given by between orlicz spaces associated with young functions @xmath29 and @xmath76 fulfilling and can be established . [ singularcor ] let @xmath0 , @xmath279 and @xmath275 be as in theorem [ singular ] . assume that @xmath29 and @xmath76 are young functions satisfying and . then there exists a constant @xmath280 such that @xmath281 and @xmath282 for every @xmath283 . here , @xmath184 denotes the constant appearing in and . * proof*. by ( * ? ? ? * lemma 1 ) , we have that , if @xmath29 and @xmath76 are young functions satisfying , then there exists a constant @xmath284 such that @xmath285 for every @xmath286 . moreover , if @xmath29 and @xmath76 fulfill , then there exists a constant @xmath284 such that @xmath287 for every @xmath286 . combining , and , and making use of property yield inequality . as far as is concerned , observe that , inequalities and continue to hold , with the same constant @xmath184 , if @xmath29 and @xmath76 are replaced with @xmath288 and @xmath289 , where @xmath290 is any positive constant . thus , inequality continues to hold , with the same constant @xmath104 , after this replacement , whatever @xmath279 is , namely @xmath291 for every @xmath292 . now , given any such @xmath293 , choose @xmath294 . the very definition of luxemburg norm tells us that @xmath295 . hence , by , @xmath296 . the definition of luxemburg norm again implies that @xmath297 , namely . * proof of theorem [ singular]*. let @xmath298 be such that @xmath299 , the ball centered at @xmath86 , with radius @xmath300 . fix a smooth function @xmath301 such that @xmath302 in @xmath303 $ ] and @xmath304 in @xmath305 . define @xmath306 by properties of @xmath271 , one has that : @xmath307 @xmath308 for every @xmath262 , there exists a constant @xmath309 such that @xmath310 where @xmath189 is the constant appearing in ; there exists a constant @xmath311 such that @xmath312 and , if @xmath276 , @xmath313 and @xmath269 , then @xmath314 @xmath315 where @xmath190 is the constant appearing in . define @xmath316 inequality will follow if we prove that @xmath317 for some constant @xmath318 , and for every @xmath272 such that @xmath319 in @xmath320 . a proof of inequality can be accomplished along the same lines as that of theorem 1 of @xcite , which in turn relies upon similar techniques as in @xcite . for completeness , we sketch such proof hereafter . the key step in the derivation of consists in showing that , for every @xmath321 , there exists a constant @xmath322 such that @xmath323 for every @xmath272 such that @xmath319 in @xmath324 . fix @xmath325 , and define @xmath326 then , there exists an open set @xmath327 such that @xmath328 . by whitney s covering theorem , there exist a family of disjoint cubes @xmath329 such that @xmath330 , @xmath331 , and @xmath332 the operator @xmath333 is of weak type ( 1,1 ) , namely , there exists a constant @xmath334 such that @xmath335 for @xmath272 , as proved in ( * ? ? * proof of lemma 6.3 ) . we shall now show that there exists a constant @xmath336 such that @xmath337 fix any @xmath338 , choose @xmath339 such that @xmath340 and denote by @xmath341 the cube , centered at @xmath342 , with @xmath343 . define @xmath344 so that @xmath345 . if we prove that there exist constants @xmath346 and @xmath347 such that @xmath348 and @xmath349 then follows with @xmath350 . consider first . let @xmath347 be such that @xmath351 . since @xmath352 for @xmath353 , an application of with this choice of @xmath354 tells us that @xmath355 , @xmath356 indeed , since @xmath357 , we have that @xmath358 , and hence implies . we may thus focus on . fix @xmath359 , and set @xmath360 . observe that @xmath361 . given any @xmath353 , define @xmath362 . one has that @xmath363 observe that , if @xmath364 , then @xmath365 and hence @xmath366 . thus , owing to , @xmath367 moreover , @xmath368 . therefore , there exists a constant @xmath369 such that @xmath370 on the other hand , @xmath371 where the first inequality holds since @xmath372 in @xmath373 if @xmath374 , and trivially holds ( with equality ) if @xmath375 . since @xmath376 in the last integral in , and @xmath293 vanishes in @xmath324 , by @xmath377 hence , @xmath378 for some constant @xmath379 . note that , in the first inequality , we have made use of the inclusion @xmath380 , which holds since @xmath381 , and @xmath382 . combining inequalities yields . inequality is fully established . via summation in @xmath383 , we obtain from that @xmath384 coupling with the inequality @xmath385 whence follows , by the very definition of decreasing rearrangement . starting from inequality leads to , via the same iteration argument as in the proof of ( * ? ? ? * theorem 1 ) . [ galdi ] let @xmath0 be a bounded domain with the cone property in @xmath1 , with @xmath2 . then there exist @xmath386 and a finite family @xmath387 of domains which are starshaped with respect to balls , such that @xmath388 . moreover , given @xmath389 , there exist @xmath390 , @xmath391 , such that @xmath392 in @xmath393 , @xmath394 and @xmath395 for some constant @xmath396 . * proof , sketched*. any bounded open set with the cone property can be decomposed into a finite union of lipschitz domains ( * ? ? ? * lemma 4.22 ) . on the other hand , any lipschitz domain can be decomposed into a finite union of open sets which are starshaped with respect to balls ( * ? ? ? * lemma 3.4 , chapter 3 ) . this proves the existence of the domains @xmath387 as in the statement . the same argument as in the proof of ( * ? ? ? * lemma 3.2 , chapter 3 ) then enables one to construct the desired family of functions @xmath397 on @xmath0 , @xmath398 , according to the following iteration scheme . we set @xmath399 , @xmath400 , and , for @xmath401 , @xmath402 and @xmath403 observe that , since @xmath0 is connected , we can always relabel the sets @xmath404 in such a way that @xmath405 for @xmath406 . finally , we define @xmath407 the family @xmath408 satisfies the required properties . the only nontrivial one is . to verify the latter , fix @xmath409 , and observe that , by , the second inequality in , inequality , and inequality @xmath410 on the other hand , by and a chain similar to , one has that @xmath411 from , and an iteration of , one infers that @xmath412 and follows . * proof of theorem [ thm:3.1 ] * by lemma [ galdi ] , it suffices to prove the statement in the case when @xmath0 is a domain starshaped with respect to a ball @xmath76 , which , without loss of generality , can be assumed to be centered at the origin and with radius @xmath413 . in this case , we shall show that the ( gradient of the ) bogovskii operator @xmath78 , defined at a function @xmath414 as @xmath415 where @xmath416 is any ( nonnegative ) function in @xmath417 with @xmath418 , agrees with a singular integral operator , whose kernel fulfills , plus two operators enjoying stronger boundedness properties . to be more precise , we set @xmath419 and claim that @xmath420 , and @xmath421 where @xmath422 is the linear operator defined at @xmath293 as @xmath423 for @xmath424 . here , @xmath425 is the kernel of a singular integral operator satisfying the same assumptions as the kernel @xmath279 in theorem [ singular ] , and the kernels @xmath426 satisfy @xmath427 to verify this assertion , recall that , if @xmath428 , then @xmath429 , and moreover equations and hold for every @xmath430 ( * ? ? ? * proof of lemma iii.3.1 ) . consider next the general case when @xmath431 . owing to , @xmath432 , since @xmath433 grows at least linearly near infinity , and hence @xmath207 dominates the function @xmath434 near infinity . since the space @xmath435 is dense in @xmath436 , there exists a sequence of functions @xmath437 such that @xmath438 in @xmath439 . one has that @xmath440 ( in fact , @xmath78 is also bounded into @xmath441 ) . furthermore , @xmath442 as a consequence of and of a special case of theorem [ singularcor ] , with @xmath443 and @xmath444 . thus , @xmath445 in @xmath446 and @xmath447 in @xmath448 . this implies that @xmath449 , and and hold . by theorem [ singularcor ] , the singular integral operator defined by the first addend on the right - hand side of is bounded from @xmath450 into @xmath451 . by inequality , the operator defined by the second addend on the right - hand side of has ( at least ) the same boundedness properties as a riesz potential operator with kernel @xmath452 . such an operator is bounded in @xmath448 and in @xmath453 , with norms depending only on @xmath23 and on @xmath170 . an interpolation theorem by calderon ( * ? ? ? * theorem 2.12 , chap . 3 ) then ensures that it is also bounded from @xmath28 into @xmath450 , and hence , a fortiori , from @xmath28 into @xmath451 , with norm depending on @xmath170 and @xmath23 . finally , the operator given by the last addend on the right - hand side of is pointwise bounded ( in absolute value ) by @xmath454 . thus , it is bounded from @xmath28 into @xmath450 , and hence from @xmath28 into @xmath451 . equations and are thus established . inequality can be derived from via a scaling argument analogous to that which leads to from see the proof of theorem [ singularcor ] . we need a last preliminary result in preparation for the proof of theorem [ thm:3.2 ] . [ supapprox ] let @xmath0 be an open subset in @xmath265 such that @xmath455 , and let @xmath29 be a young function . assume that @xmath456 . then : @xmath457 and @xmath458 note that equation is well known under the assumption that @xmath459 near infinity , namely @xmath460 near infinity , since @xmath461 is dense in @xmath462 in this case . equation also easily follows from this property when @xmath463 near infinity . the novelty of proposition [ supapprox ] is in the arbitrariness of @xmath29 . * proof of proposition [ supapprox]*. consider first . it clearly suffices to show that @xmath464 and @xmath465 given any @xmath466 , define , for @xmath338 , the function @xmath467 as @xmath468 clearly , @xmath469 , and @xmath470 a.e . in @xmath0 as @xmath471 . hence , @xmath472 by the monotone convergence theorem for integrals , and , by the fatou property of the luxemburg norm , @xmath473 thus , since @xmath474 equation follows . as far as is concerned , consider an increasing sequence of compact sets @xmath475 such that @xmath476 , @xmath477 for @xmath338 , and @xmath478 . moreover , let @xmath479 be a family of ( nonnegative ) smooth mollifiers in @xmath265 , such that @xmath480 and @xmath481 for @xmath338 . given @xmath482 , define @xmath483 as @xmath484 and @xmath485 as @xmath486 classical properties of mollifiers ensure that @xmath487 thus , if @xmath488 , then @xmath489 by the dominated convergence theorem for integrals . moreover , @xmath490 indeed , by dominated convergence and the definition of luxemburg norm , @xmath491 in particular , for every @xmath359 , there exists @xmath492 such that @xmath493 hence , by the arbitrariness of @xmath494 and the definition of luxemburg norm , @xmath495 we also have that @xmath496 indeed , assume that fails . then , there exists @xmath497 and a subsequence of @xmath498 , still denoted by @xmath498 , such that @xmath499 a contradiction . equation follows from and . coupling with yields . the proof of is complete . the proof of follows along the same lines , and , in particular , via the equations @xmath500 and @xmath501 on defining , for any @xmath502 , the sequence of functions @xmath503 as @xmath504 for @xmath338 , where @xmath505 is given by , one can prove equation via a slight variant of the argument employed for . here , one has to use the fact that @xmath506 as @xmath507 . similarly , equation can be established similarly to on replacing , for any given @xmath508 , the sequence @xmath509 defined by with the sequence @xmath510 defined as @xmath511 where @xmath512 is any function in @xmath513 such that @xmath514 . note that now , for every @xmath359 , there exists @xmath515 such that @xmath516 , provided that @xmath517 . * proof of theorem [ thm:3.2]*. let @xmath518.then @xmath519 note that the inequality in holds since , by the first inequality in , @xmath520 and the last equality in relies upon . by theorem [ thm:3.1 ] , applied with @xmath29 and @xmath76 replaced with @xmath521 and @xmath31 , respectively , there exists a constant @xmath522 such that @xmath523 the first inequality in follows from and . the second inequality is trivial , since @xmath524 for some constant @xmath525 . in many customary mathematical models , the stationary flow of a homogeneous incompressible fluid in a bounded domain @xmath526 is described by a system with the structure . with a slight abuse of notation with respect to , we also denote by @xmath527 the function , acting on the symmetric gradient @xmath42 of the velocity field @xmath9 , which yields the stress deviator of the fluid . thus , we shall consider systems of the form @xmath528 in the simplest case of a newtonian fluid , the function @xmath41 is linear , and @xmath529 , the laplacian of @xmath9 . however only fluids with an easy molecular structure , such as water , oil , and several gases are governed by this low . more complex liquids are not , and are called non - newtonian fluids see e.g. @xcite . the most common nonlinear model among rheologists is the power law model , corresponding to the choice @xmath530 here , @xmath531 and @xmath532 are constants , and @xmath533 is an exponent which need to be specified via physical experiments . an extensive list of specific @xmath8-values for different fluids can be found in @xcite . in various instances of interest in applications , the term @xmath536 is negligible in , compared with the other terms appearing in the first equation . this is the case , for example , if the modulus of the velocity @xmath9 is small . another situation where the role of the term is immaterial is that of plastic or pseudo - plastic fluids . indeed , accounts for the inner rotation in the fluid flow , and for such fluids the impact of this term is very limited . dropping the term reduces to the simplified system @xmath537 a standard approach to ( [ eq:4.1 ] ) or ( [ eq:4.1 ] ) consists in two steps . firstly , a velocity field @xmath9 is exhibited such that @xmath538 for every @xmath539 , where either @xmath540 or @xmath541 according to weather the convective term @xmath542 is included in the model or not . here , : " stands for scalar product between matrices , and @xmath543 denotes the space of compactly supported , infinitely differentiable @xmath265-valued functions whose divergence vanishes in @xmath0 . the function @xmath9 belongs to a proper sobolev type space depending on the constitutive law underlying the definition of the function @xmath41 . secondly , the pressure @xmath46 is reconstructed . a discussion of the first issue falls beyond the scopes of the present paper , and will not be addressed here . let us just mention that the standard power type model has been investigated in the classical contributions @xcite , and in the recent papers @xcite . stationary flows of fluids whose constitutive law satisfies ( [ eq:4.4 ] ) with a young - function @xmath544 are studied in @xcite . an unconventional constitutive law , where @xmath41 has the form with @xmath545 near infinity , and hence @xmath546 near infinity , was introduced by eyring in @xcite , where it is assumed that @xmath547 . similar results are due to prandtl ( see e.g. @xcite for an overview on this kind of models ) . an analysis of the simplified system for the eyring - prandtl model is the object of @xcite , whereas the complete system ( [ eq:4.1 ] ) , in the case @xmath39 , is considered in @xcite . in the remaining part of this paper , we focus , instead , on the second question , namely the reconstruction of the pressure @xmath46 in a correct orlicz space . in case of fluids governed by a general constitutive low of the form , the function @xmath548 belongs to some orlicz space @xmath549 . if @xmath550 , then @xmath551 as well . however , in general , one can only expect that @xmath46 belongs to some larger orlicz space @xmath552 . the balance between the young functions @xmath29 and @xmath76 is determined by conditions and , as stated in the following result . a very general constitutive law for non - newtonian fluids is @xmath553 with a young function @xmath29 . the natural space for the symmetric gradient of the velocity is @xmath28 so that ( [ ineqa1 ] ) implies @xmath554 . using theorem [ thm:4.1 ] we can introduce the pressure in the correct orlicz space @xmath552 . + let us have a look at the most interesting cases : * if @xmath29 satisfies a @xmath71- and the @xmath72-condition we can choose @xmath555 . + in @xcite , @xcite and @xcite the model ( [ eq:4.4 ] ) is considered where @xmath556 satisfies the @xmath71- and the @xmath72-condition . note that this includes power law fluids where @xmath557 or @xmath558 with @xmath533 . the existence of a weak solution @xmath559 ( together with regularity properties ) to ( [ eq:3.1weak ] ) with @xmath41 given by ( [ eq:4.4 ] ) is shown . now we have @xmath560 and can reconstruct the pressure in the same space by theorem [ thm:4.1 ] . * if @xmath561 we can choose @xmath562 for all @xmath563 . + in the case of slow flows of prandtl - eyring fluids we have @xmath564 and hence @xmath565 . * if @xmath566 we can choose @xmath567 for all @xmath568 . + considering the complete equation of motion for prandtl - eyring fluids in 2d we have @xmath569 and hence @xmath570 . this was already shown in @xcite ( end of section 3 ) using continuity properties of the -operator in @xmath571-spaces . [ thm:4.1 ] let @xmath29 and @xmath76 be young functions fulfilling and . let @xmath0 be a bounded domain with the cone property in @xmath1 , @xmath2 . assume that @xmath572 and satisfies @xmath573 for every @xmath539 . then there exists a unique function @xmath574 such that @xmath575 for every @xmath576 . moreover , there exists a constant @xmath254 such that @xmath577 and @xmath578 here , @xmath184 denotes the constant appearing in and . in particular , theorem [ thm:4.1 ] reproduces , within a unified framework , various results appearing in the literature . for instance , when the constitutive relation is in force , the function @xmath207 is just a power @xmath579 , where the exponent @xmath580 , and depends on @xmath8 , on @xmath581 , and on whether the system or is taken into account . in any case , @xmath582 agrees with the lebesgue space @xmath583 , and theorem [ thm:4.1 ] recovers the fact that @xmath46 belongs to the same lebesgue space @xmath584 . as far as the simplified system for the eyring - prandtl model is concerned , under appropriate assumptions on @xmath581 one has that @xmath585 . hence , via theorem [ thm:4.1 ] , we infer the existence of a pressure @xmath586 . more generally , if @xmath587 for some @xmath143 , one has that @xmath588 . the complete system for the eyring - prandtl model , in the @xmath589-dimensional case , admits a weak solution @xmath9 such that @xmath590 and hence @xmath591 @xcite . again , one can not expect that the pressure @xmath46 belongs to the same space . in fact , theorem [ thm:4.1 ] yields the existence of a pressure @xmath592 , thus reproducing a result from @xcite . in general , if @xmath593 for some @xmath210 , then we obtain that @xmath594 . * proof of theorem [ thm:4.1]*. by de rahms theorem , in the version of @xcite , there exists a distribution @xmath595 such that @xmath596 for every @xmath597 . replacing @xmath598 with @xmath599 in , where @xmath600 , yields @xmath601 for every @xmath602 . we claim that the linear functional @xmath603 is bounded on @xmath604 equipped with the @xmath453 norm . indeed , by ( [ 1.1 ] ) , one has that @xmath605 . moreover , by a special case of theorem [ thm:3.1 ] , @xmath606 . thus , since @xmath607 and @xmath608 are orlicz spaces built upon young functions which are conjugate of each other , @xmath609 for every @xmath602 , where @xmath610 and @xmath611 . hence , the relevant functional can be continued to a bounded linear functional on @xmath612 , with the same norm . now , as a consequence of riesz s representation theorem , there exists a radon measure @xmath595 such that @xmath613 for every @xmath614 . fix any open set @xmath615 . by theorem [ thm:3.1 ] again , there exists a constant @xmath104 such that @xmath616 one can verify that the norm @xmath617 is absolutely continuous , in the sense that for every @xmath359 there exists @xmath618 such that @xmath619 if @xmath620 , and since any lebesgue measurable set can be approximated from outside by open sets , inequality implies that the measure @xmath621 is absolutely continuous with respect to the lebesgue measure . hence , @xmath621 has a density with respect to the lebesgue measure . so @xmath595 can be represented by a function @xmath622 fulfilling ( [ eq : thm4.1 ] ) holds . the function @xmath46 is uniquely determined if we assume that @xmath623 . by this assumption , theorem [ thm:3.2 ] , and equation ( [ eq : thm4.1 ] ) we have that @xmath624 where @xmath522 . this proves inequality . inequality follows from , on replacing @xmath29 and @xmath76 with @xmath288 and @xmath289 , respectively , with @xmath625 , via an argument analogous to that of the proof of . let us turn to another constitutive relation where the situation is not that clear anymore : the prandtl - eyring fluid model . it was invented by eyring @xcite who supposes @xmath626 with physical constants @xmath627 . similar results are due to prandtl ( compare @xcite for an overview ) . fuchs and seregin @xcite study the slow flow of prandtl - eyring fluids ( cf . system ( [ eq:4.1 ] ) ) . the natural function space for weak solutions is @xmath628 note that the function @xmath629 is equivalent to @xmath207 . the existence of a function @xmath630 satisfying @xmath631 with @xmath41 given by ( [ eq:4.3 ] ) can be shown by the direct method in the calculus of variations . under appropriate assumptions on @xmath581 we have that @xmath632 belongs to the space @xmath633 . now it is not obvious at all which is the correct space for the pressure . since we are in a limiting situation we do not expect that it belongs to the same space than @xmath548 . if @xmath634 near infinity for some @xmath563 , we can choose @xmath635 near infinity in theorem [ thm:4.1 ] . hence , we obtain by theorem [ thm:4.1 ] the existence of a pressure with @xmath586 . + in @xcite the full equation is considered in case @xmath39 . the existence of a weak solution @xmath636 to ( [ eq:4.1 ] ) is shown and also that @xmath637 . this means we are now in the opposite limit case and have @xmath638 . again we do not expect that it is possible to obtain the pressure in the same space . if @xmath566 we can choose @xmath567 for every @xmath568 . considering the complete equation of motion for prandtl - eyring fluids we get by theorem [ thm:4.1 ] the existence of a pressure with @xmath592 . this was already shown in @xcite ( end of section 3 ) using continuity properties of the -operator in @xmath571-spaces . + a very general constitutive law for non - newtonian fluids is @xmath553 with a young function @xmath29 . the natural space for the symmetric gradient of the velocity is @xmath28 so that ( [ ineqa1 ] ) implies @xmath554 . using theorem [ thm:4.1 ] we can introduce the pressure in the correct orlicz space @xmath552 . + let us have a look at the most interesting cases : * if @xmath29 satisfies a @xmath71- and the @xmath72-condition we can choose @xmath555 . + in @xcite , @xcite and @xcite the model ( [ eq:4.4 ] ) is considered where @xmath556 satisfies the @xmath71- and the @xmath72-condition . note that this includes power law fluids where @xmath557 or @xmath558 with @xmath533 . the existence of a weak solution @xmath559 ( together with regularity properties ) to ( [ eq:3.1weak ] ) with @xmath41 given by ( [ eq:4.4 ] ) is shown . now we have @xmath560 and can reconstruct the pressure in the same space by theorem [ thm:4.1 ] . * if @xmath561 we can choose @xmath562 for every @xmath563 . + in the case of slow flows of prandtl - eyring fluids we have @xmath564 and hence @xmath565 . * if @xmath566 we can choose @xmath567 for every @xmath568 . + considering the complete equation of motion for prandtl - eyring fluids in 2d we have @xmath569 and hence @xmath570 . this was already shown in @xcite ( end of section 3 ) using continuity properties of the -operator in @xmath571-spaces . let us turn to a further consequence of theorem [ thm:4.1 ] , which is related to a numerical analysis of problem , or of its simplified version . we shall adopt the scheme of the finite element method for the @xmath8-stokes system exploited in @xcite in the special case when @xmath41 is given by . in what follows , we assume that @xmath0 is a polyhedron . the goal is to compute an approximate solution to problem via discretization . to this purpose , one needs a triangulation @xmath639 of @xmath0 into simplices of diameter bounded by @xmath640 . recall that a simplex in @xmath265 is the convex hull of @xmath641 points which do not lie on the same hyperplane . we also need that such a triangulation is regular enough for the lipschitz constant of the functions , which locally represent the boundaries of the relevant simplices , to be uniformly bounded in @xmath642 . we denote by @xmath643 the space of those functions in @xmath0 whose restriction to each simplex of @xmath639 is constant , and by @xmath644 its subspace of those functions from @xmath645 whose mean value over @xmath0 is zero . also , for @xmath646 , we denote by @xmath647 the space of first - order weakly differentiable functions in @xmath0 whose restriction to each simplex of @xmath639 is a polynomial of degree @xmath648 , and by @xmath649 its subspace of those functions from @xmath647 which vanish on @xmath154 . finally @xmath650 stands for the space of all functions from @xmath647 whose mean value over @xmath0 is zero . clearly , given any young function @xmath29 , one has that @xmath651 and @xmath652 the spaces @xmath653 , @xmath654 and @xmath655 of @xmath265-valued functions are defined accordingly . an important tool for the numerical analysis of problem is a projection operator @xmath656 which , for given @xmath657 and @xmath658 , is such that @xmath659 , and enjoys the following properties : + ( i ) @xmath656 preserves zero boundary values , i.e. @xmath660 ( ii ) @xmath656 is divergence preserving , namely , @xmath661 ( iii ) @xmath656 is continuous in the @xmath662-sense , i.e. there exists a constant @xmath396 such that @xmath663 for every @xmath664 , and every @xmath665 . here @xmath666 denotes the diameter of @xmath667 , and @xmath668 the union of @xmath667 and all its direct neighbors in the triangulation @xmath639 . the existence of such an operator for suitable couples of @xmath669 and @xmath290 is standard see e.g. ( * ? ? ? * appendix ) . for instance , the choice @xmath670 and @xmath671 is admissible , whereas @xmath672 and @xmath671 is not . this can again be accomplished in two steps . first , on setting @xmath677 and @xmath678 one has to find @xmath679 such that @xmath680 the existence of a unique function @xmath681 satisfying follows easily from the theory of monotone operators . if @xmath41 has variational structure one may equivalently solve the corresponding strictly convex minimizing problem on the finite dimensional function space @xmath682 . next , the pressure has be to reconstructed . precisely , one has to find @xmath683 such that @xmath684 this is the discrete analogone of the problem from theorem [ thm:4.1 ] . as far as existence is concerned , we only need to solve an algebraic linear system . the following questions then arise : the answer to all this questions follows from the so - called inf - sup condition . such a condition , in the standard case when @xmath41 has a power type growth as in , reads as @xmath688 for some positive constant @xmath104 independent of @xmath642 . in fact , the uniqueness of @xmath685 does not even require @xmath104 to be independent of @xmath642 . [ thm:4.2 ] let @xmath29 and @xmath76 be young functions fulfilling and . assume that @xmath0 is a polyhedron in @xmath1 , @xmath2 , and that @xmath657 and @xmath658 are such that there exists an operator @xmath689 satisfying ( [ itm : a1 - 2])-([itm : a2 ] ) . then , @xmath690 for some positive constant @xmath254 , where @xmath184 is the constant appearing in and * proof*. as a first step , we show that the operator @xmath656 is continuous in every orlicz space , in the sense that there exists a constant @xmath396 such that , for every young function @xmath29 , @xmath691 inequality has been established in ( * ? ? ? 4.5 ) and in ( * ? ? ? 3.2 ) under the additional assumption that @xmath29 fulfils a global @xmath71-condition . a variant of those proofs shows that , in fact , this assumption can be dropped . we outline the argument hereafter . since @xmath692 belongs to a finite dimensional function space , it follows from ( [ itm : a2 ] ) that @xmath693 for some constants @xmath396 and @xmath694 , and for every @xmath695 . hence , owing to jensen s inequality and the convexity of @xmath29 , @xmath696 for every @xmath697 . now , given any @xmath698 , we deduce from the convexity of @xmath29 , and latexmath:[\[\begin{aligned } \dashint_{\mathcal s}a(h_\mathcal s|\nabla\bfu-\nabla \pi^h\bfu|)\,dx & \leq \frac12\dashint_{\mathcal s}a(2h_\mathcal s|\nabla(\bfu-\bfq ) |)\,dx+ \frac 12\dashint_{\mathcal s}a(2h_\mathcal s|\nabla \pi^h(\bfu-\bfq)|)\,dx\nonumber\\ & \leq \dashint_{m_{\mathcal s}}a(ch_\mathcal s|\nabla(\bfu-\bfq ) for some constant @xmath396 and for every @xmath700 . finally we choose @xmath701 . by inequality and a scaling argument as in the proof of theorem [ thm:3.1 ] one can show that there exists a constant @xmath702 such that @xmath703 for every @xmath697 . inequalities and imply that @xmath704 for every @xmath697 . the convexity of @xmath29 and inequality ( [ eq : appr ] ) yield @xmath705 for some constant @xmath396 , and for every @xmath700 . this implies , on replacing @xmath706 with @xmath707 and summing up over all simplexes @xmath667 . note , in this connection , that the number of neighbours of each @xmath667 only depends on @xmath170 . we are now in a position to prove inequality . by theorem [ thm:3.2 ] , there exists a constant @xmath708 such that @xmath709 for every @xmath710 . owing to and to properties and of the operator @xmath656 , @xmath711 for some constants @xmath522 and @xmath712 , and for every @xmath713 . hence , follows . since the functions @xmath714 belong to finite dimensional spaces it would be possible to replace @xmath715 in theorem [ thm:4.2 ] by @xmath716 . however , the constant then depends on the dimension of @xmath717 and hence on @xmath642 . in that form , the result would be of no use in the study of the convergence of the finite element method approximation . * there exists a unique function @xmath683 such that @xmath721 for every @xmath722 . * there exists a constant @xmath104 , independent of @xmath642 , such that @xmath723 * there exists a constant @xmath104 , independent of @xmath642 , such that , if @xmath724 and @xmath725 satisfy @xmath726 for every @xmath727,then @xmath728 * proof*. ( i ) consider a basis @xmath729 , @xmath730 of @xmath731 and a basis @xmath732 , @xmath733 of @xmath734 . note that @xmath735 . then the problem @xmath736 for every @xmath722 is equivalent to the algebraic linear system @xmath737 , where @xmath738 such system has a solution , provided that @xmath739 belongs to the image of the matrix @xmath740 , or , equivalently , if it is orthogonal to @xmath741 . now , observe that a vector @xmath742 belongs to @xmath741 if and only if @xmath743 whence , @xmath744 . thus , owing to ( [ cor:4.1.0 ] ) , @xmath745 for every @xmath746 . hence , @xmath747 . next , we show that @xmath740 is injective , whence the uniqueness of @xmath685 follows . to verify the injectivity of @xmath740 , assume that @xmath748 is such that @xmath749 . hence , on setting @xmath750 , we have that @xmath751 for every @xmath409 . let us apply theorem [ thm:4.2 ] with any pair of young functions @xmath29 and @xmath76 fulfilling and , for instance @xmath752 for @xmath753 ( this means that we are in fact applying with @xmath754 ) . we then obtain that @xmath755 whence @xmath756 . + ( ii ) by theorem [ thm:4.2 ] and ( [ b.5 ] ) @xmath757 for some constants @xmath254 and @xmath712 . this proves inequality . + ( iii ) the triangle inequality and the embedding @xmath758 ensure that , for every @xmath759 , @xmath760 for some constant @xmath284 . by theorem [ thm:4.2 ] , @xmath761 for some constant @xmath762 , and for every @xmath763 . moreover , by , , and an approximation argument for functions in @xmath764 via functions in @xmath765 , we have that @xmath766 for every @xmath759 and @xmath767 . hence , via hlder s inequality in orlicz spaces , @xmath768 for some constant @xmath525 . altogether , we conclude that @xmath769 for some constant @xmath104 independent of @xmath642 , namely . this research was partly supported by leopoldina ( german national academy of science ) , by the research project geometric properties of partial differential equations and related topics " 2008 of the italian miur ( ministry of education , university and research ) , and by gnampa of the italian indam ( istituto nazionale di alta matematica ) . m. e. bogovski , solutions of some problems of vector analysis , associated with the operators @xmath771 and @xmath772 . in _ theory of cubature formulas and the application of functional analysis to problems of mathematical physics ( russian ) _ , pages 540 , * 149 * , akad . nauk sssr sibirsk . inst . mat . , novosibirsk , 1980 . j. frehse , j. malek & m. steinhauer , an existence result for fluids with shear dependent viscosit steady flows , _ nonlinear anal . _ * 30 * ( 1997 ) 30413049 . j. frehse , j. malk & m. steinhauer , on analysis of steady flows of fluids with shear - dependent viscosity based on the lipschitz truncation method , _ siam j. math . * 34 * ( 2003 ) 10641083 . o. a. ladyzhenskaya , _ the mathematical theory of viscous incompressible flow _ , gorden and breach , 1969 . o. a. ladyzhenskaya , on some new equations describing dynamics of incompressible fluids and on global solvability of boundary value problems to these equations , _ trudy steklov s math . institute _ * 102 * ( 1967 ) 85104 . o. a. ladyzhenskaya , on some modifications of the navier - stokes equations for large gradients of velocity , _ zap . nauchn . sem . otdel . mat . inst . steklov ( lomi ) _ * 7 * ( 1968 ) 126154 . j. neas , sur les normes quivalentes dans @xmath778 et sur la coecivit des formes formellement positives . sminaire des mathmatiques suprieures , in : _ equations aus drives partielles _ * 19 * , 102128 , les presses de luniversit de montral , montral ( 1965 )

using derivative expansion applied to the wigner transform of the two - point green function we analyse the anomalous quantum hall effect ( aqhe ) , and the chiral magnetic effect ( cme ) . the corresponding currents are proportional to the momentum space topological invariants . we reproduce the conventional expression for the hall conductivity in d. in d our analysis allows to explain systematically the aqhe in topological insulators and weyl semimetals . at the same time using this method it may be proved , that the equilibrium cme is absent in the wide class of solids , as well as in the properly regularized relativistic quantum field theory .

1603.03665

momentum space topology is becoming the important tool for the study of the ground states of condensed matter systems ( for the review see @xcite ) . in particular , the momentum space topological invariants protect gapless fermions on the boundaries of topological insulators @xcite . topological invariants in momentum space protect also the bulk gapless fermions in dirac and weyl semi - metals @xcite . the large variety of topological defects and textures exist in the fermionic superfluids , and the gapless fermions associated with these objects are described by momentum space topology @xcite . momentum space topology was also discussed in the context of relativistic quantum field theory ( qft ) @xcite . in @xcite the topological invariants in momentum space have been considered for the lattice regularization of qft with wilson fermions . appearance of the massless fermions at the intermediate values of bare mass parameter was related to the jump of the introduced momentum space topological invariant . this invariant may actually be used for the description of a certain class of topological insulators . in @xcite the model with overlap fermions has been considered on the same grounds . in particular , the possible physical meaning of the zeros of the green function has been discussed . the appearance of zeros of the green function , in turn , has been discussed in the context of condensed matter physics ( see , for example , @xcite ) . the momentum space topological invariants are expressed in terms of the green functions . therefore , they are applicable both to the non - interacting and to the interacting systems @xcite . suppose , that we start from the model without interactions . when the interactions are turned on , the value of the topological invariant is not changed until the phase transition is encountered . this means , that the properties of the system described by the given topological invariant are robust to the introduction of interactions . the more simple non - interacting model may be investigated in order to describe such properties of the complicated interacting system . in the present paper we apply momentum space topology to the description of the anomalous quantum hall effect ( aqhe ) in topological insulators and weyl semimetals . in @xcite we considered the chiral magnetic effect ( cme ) ( mainly , in the framework of relativistic quantum field theory ) using the approach based on momentum space topology . here we briefly repeat our proof of the absence of the equilibrium bulk cme with the emphasis in the application to the solid state systems . actually , momentum space topology with topological invariants expressed through the green functions represents the alternative to the less powerful but more popular technique of berry curvature proposed to describe the qhe in @xcite and developed later in a number of publications ( see , for example , @xcite and references therein ) . the technique of berry curvature may be applied to the noninteracting condensed matter systems with green function of the form @xmath2 , where @xmath3 is the imaginary frequency while @xmath4 is the hamiltonian . unfortunately , this formalism does not allow to deal in a similar way with the interacting systems with more complicated dependence of the green function on the imaginary frequency . besides , in spite of all its advantages , the berry curvature formalism does not allow to describe the response of the system to the external magnetic ( rather than electric ) field . finally , the bulk - boundary correspondence remains out of this formalism . those three points are improved in the technique that utilizes the topological invariants composed directly of the green functions . it was proposed first by g.e.volovik ( see @xcite and references therein ) . following @xcite in the present paper we develop this technique and give the relation between the electromagnetic response of electric current and the topological invariants in momentum space of both @xmath0d and @xmath1d systems . in the present paper we emphasise , that those topological invariants are constructed of the wigner transform of the two - point green functions , which allows to apply the proposed methodology to the direct description of bulk - boundary correspondence . as for the berry curvature formalism , we demonstrate that it follows from our approach as a particular case . the family of the non - dissipative transport effects related to chiral anomaly has been widely discussed recently both in the context of the high energy physics and in the context of condensed matter theory @xcite . in particular , the possible appearance of such effects in the recently discovered dirac and weyl semimetals has been considered @xcite . besides , the possibility to observe those effects in relativistic heavy - ion collisions was proposed @xcite . the chiral magnetic effect ( cme ) is the generation of electric current in the presence of external magnetic field and chiral chemical potential @xcite . the quantum hall effect in the @xmath0 d and the @xmath1 d systems is the appearance of electric current in the direction orthogonal to the direction of the applied electric field . the anomalous quantum hall effect ( aqhe ) is such an effect that takes place due to the internal properties of the system rather than due to the external magnetic field . recently its possible appearance in weyl semimetals has been widely discussed @xcite . the aqhe was also discussed in the @xmath1d topological insulators @xcite . in the present paper we consider aqhe within the lattice regularized quantum field theory and within the tight - binding models of the solid state physics . the expression of the hall current ( through the topological invariant in momentum space @xmath5 ) in the quasi two dimensional condensed matter systems is well - known @xcite . we reproduce this result basing on the technique of the derivative expansion applied to the wigner transform of the two - point green function . next , we apply the same technique to the aqhe in the @xmath1 d lattice models . the obtained expression relates the aqhe current to the new topological invariant . the technique of its calculation is developed . our methodology allows to predict the appearance of the aqhe both in the weyl semi - metals and in certain topological insulators . we represent here the consideration of the equilibrium cme , which was based on the same technique . it appears , that the resulting current is also proportional to the topological invariant in momentum space . unlike the case of the naive continuum fermions for the considered lattice models with regular fermionic green function the value of the mentioned topological invariant does not depend on the chiral chemical potential . we consider the case , when the lattice dirac fermions are massive . however , the limit , when the physical dirac mass tends to zero , does not change our conclusion . this indicates , that the equilibrium bulk cme current is absent for the lattice regularized quantum field theory and in the certain class of solids . notice , that our conclusions on aqhe and cme are in accordance with the recent consideration of the particular lattice model of weyl semimetals @xcite . the conclusion on the absence of the equilibrium bulk cme is in line with the recent numerical calculations made using the particular lattice models @xcite . it also does not contradict to the consideration of the effective continuum field theoretic description of weyl / dirac semimetals @xcite and the consideration of dirac semimetals in the framework of the semi - classical approach @xcite . besides , this conclusion is in accordance with the bloch theorem @xcite . we incorporate the slowly varying external gauge field directly to the momentum space representation of the lattice model@xcite . in this representation it appears as a pseudo - differential operator @xmath6 , where @xmath7 determines the functional dependence of the gauge field on coordinates , while @xmath8 is the derivative over momentum . such an introduction of the gauge field to the lattice model deforms it at small distances ( of the order of the lattice spacing ) . however , the gauge invariance remains intact , and at large distances the theory is indistinguishable from the lattice model with the external gauge field introduced in a conventional way . therefore , the given way to introduce the electromagnetic field satisfies all requirements to be imposed on the lattice regularization of the qft . we also consider this appropriate to apply this method to the lattice models of solid state physics if the long range phenomena are to be investigated . we demonstrate explicitly , that in the particular case of the noninteracting condensed matter system our formalism allows to reproduce the expression for the hall conductivity as an integral over the berry curvature ( the green function of such system has the form @xmath2 , where @xmath3 is the imaginary frequency while @xmath4 is the hamiltonian ) . the paper is organized as follows . in sections [ sectgaugelat ] , [ sectwignlat ] , [ sectlinlat ] we describe the formalism proposed to be used @xcite . in sect . [ sectgaugelat ] we describe the way to introduce the slowly varying external gauge field to the momentum space formulation of lattice model . in sect . [ sectwignlat ] we consider the wigner transformation of the two - point green function in momentum space . the linear response of electric current to external field strength is considered in sect . [ sectlinlat ] . in sect . [ secthall ] we reproduce the expression for the qhe current in the @xmath0 d systems using the developed methodology . in sect . [ sectberry ] we demonstrate , how the formalism of berry curvature follows from our expressions . in sect . [ secthall3d ] we consider the aqhe in the @xmath1d systems . in sect . [ bulkboundary ] we discuss bulk - boundary correspondence for the topological insulators with aqhe . the considerations of sect . [ secthall3d ] are illustrated by two particular models of insulators in sect . [ sectins ] , and by two particular models of weyl semimetal in sect . [ sectweylsem ] . in sect . [ sectcme ] we analyse using this methodology the cme , and demonstrate , that it is absent in the considered systems . in sect . [ sectconcl ] we end with the conclusions . we consider the @xmath9 dimensional lattice model of non - interacting fermions . in this way the tight - binding models of solid state physics as well as the lattice regularization of the continuum quantum field theory ( qft ) may be described . in momentum space the fermionic green function @xmath10 depends on the @xmath11 vector @xmath12 of euclidean momentum . momentum space is supposed to have the form of the product @xmath13 or @xmath14 , where @xmath15 is the compact @xmath16 - dimensional brillouin zone . here @xmath17 is the line of the imaginary frequencies @xmath18 . in the lattice regularization of qft the values of @xmath18 correspond to @xmath19 . in condensed matter physics the representation of the theory with momentum space @xmath13 appears if the evolution in time is discretized . in particular , momentum space of such form appears in the application of the numerical lattice methods to the calculation of the functional integrals in the solid state physics . in the conventional lattice models defined on the hypercubic lattice @xmath15 has the form of the @xmath16 - dimensional torus does not appear as the eigenvalue of the operator @xmath20 . ] . in the solid state physics the form of the brillouin zone is typically more complicated . in the absence of the external electromagnetic field the partition function of the theory defined on the infinite lattice may be written as @xmath21 where @xmath22 is the volume of momentum space @xmath23 , while @xmath24 and @xmath25 are the grassmann - valued fields defined in momentum space @xmath23 . for example , the model with @xmath1 d wilson fermions that describes qualitatively a certain class of topological insulators corresponds to @xmath26 that has the form @xmath27 where @xmath28 are euclidean dirac matrices while @xmath29 and @xmath30 are the real - valued functions ( @xmath31 ) given by @xmath32 the fields in coordinate space are related to the fields in momentum space as follows @xmath33 at the discrete values of @xmath34 corresponding to the points of the lattice this expression gives the values of the fermionic field at these points , i.e. the dynamical variables of the original lattice model . however , eq . ( [ psip ] ) allows to define formally the values of fields at any other values of @xmath34 . the partition function may be rewritten in the form @xmath35_{{\bf r}={\bf r}_n } \big)\label{z2}\ ] ] here the sum in the exponent is over the discrete coordinates @xmath36 . however , the operator @xmath20 acts on the function @xmath37 defined using eq . ( [ psip ] ) . in order to derive eq . ( [ z2 ] ) we use identity @xmath38 in the particular case of wilson fermions we may rewrite the partition function in the conventional way as @xmath39 with @xmath40 + ( m^{(0 ) } + 4 ) \delta_{{\bf x}{\bf y}}\ ] ] here @xmath41 is the unity vector in the @xmath42 - th direction . gauge transformation of the lattice field takes the form @xmath43 in case of wilson fermions the @xmath44 gauge field is typically introduced as the following modification of operator @xmath11 : @xmath45 + ( m^{(0 ) } + 4 ) \delta_{{\bf x}{\bf y}}\ ] ] here @xmath46 is the gauge field attached to the links of the lattice . in the same way the gauge field is typically incorporated into the models of solid state physics . below we propose the alternative way of the introduction of gauge field into the lattice model . first of all , eq . ( [ gt ] ) may be understood as the gauge transformation of the field @xmath25 defined for any values of @xmath34 : we simply extend the definition of the function @xmath47 to the function , which is defined at any values of @xmath34 and take the original values at the discrete lattice points . this prompts the following way to introduce the external gauge field to our lattice model . suppose , that we need to put on the lattice the gauge field , which in continuum theory has the form of the function @xmath7 of the continuum coordinate @xmath34 . we consider the partition function of the form @xmath48_{{\bf r}={\bf r}_n } + ( h.c.)\big]\big)\label{z3}\end{aligned}\ ] ] here by @xmath49 we denote the hermitian conjugation , which is defined as follows . first of all , it relates the components of grassmann variable @xmath25 with the corresponding components of @xmath24 . besides , it inverses the ordering of operators and the variables @xmath50 , and substitutes each operator by its hermitian conjugated . for example , a conjugation of @xmath51 for a certain operator ( in internal space ) @xmath52 is given by @xmath53\hat{b}^+ \psi$ ] . as well as in continuum theory operators @xmath54 and @xmath55 do not commute for @xmath56 . therefore , we should point out the way of their ordering inside @xmath57 . we choose the following way for definiteness : each product @xmath58 in the expansion of @xmath59 is substituted by the symmetrized product @xmath60 . this method of introducing the gauge field to the lattice model differs from the more conventional ways , but it is manifestly gauge invariant , i.e. the exponent in eq . ( [ z3 ] ) is invariant under the transformation of eq . ( [ gt ] ) if we transform the gauge field as @xmath61 where @xmath62 is the extension to the continuous values of @xmath34 of the function of eq . ( [ gt ] ) defined on the discrete lattice points @xmath36 . besides , the partition function of eq . ( [ z3 ] ) is obviously reduced to the conventional continuum partition function with the minimal connection of fermion field with the gauge field in the naive continuum limit . thus , the proposed way of the incorporation of the external gauge field into the lattice theory satisfies all requirements to be fulfilled by the introduction of the gauge field in lattice regularization of quantum field theory . in case of the models of the solid state physics the proposed introduction of the external gauge field deforms the theory at the interatomic distances . however , the gauge invariance remains intact , and the long distance behavior of the theory should remain the same . therefore , we consider this appropriate to incorporate the slow varying gauge field to the solid state lattice models in this way . now let us come back to momentum space . one can easily check , that eq . ( [ z3 ] ) may be rewritten as @xmath63 here @xmath64 while the pseudo - differential operator @xmath6 is defined as follows . first , we represent the original gauge field @xmath7 as a series in powers of coordinates @xmath65 . next , variable @xmath65 is substituted in this expansion by the operator @xmath66 . in order to prove eq . ( [ z4 ] ) we introduce the function @xmath67 that is constructed of @xmath59 as follows . we represent @xmath57 as a series in powers of @xmath20 and @xmath68 such that in each term @xmath68 stand right to @xmath20 . for example , we represent @xmath69 as @xmath70 . next , we substitute the argument of @xmath71 by @xmath72 and @xmath20 by @xmath73 . correspondingly , @xmath74 is defined with the inverse ordering . with these definitions one may easily prove , that eq . ( [ z3 ] ) is equivalent to @xmath75\big)\label{z40}\end{aligned}\ ] ] after the substitution of eq . ( [ psip ] ) . since the commutators @xmath76 = i \delta_i^j$ ] and @xmath77 = i \delta_i^j$ ] are equal to each other , the actual expression for @xmath78 $ ] is given by eq . ( [ calqm ] ) . finally , the green function of our system in momentum space satisfies equation @xmath79 according to the proposed above way to incorporate gauge field into the lattice model the green function in momentum space appears as a correlator @xmath80 it obeys equation eq . ( [ qgl ] ) . wigner transform @xcite of the green function may be defined as @xmath81 in terms of the green function in coordinate space this green function is expressed as : @xmath82 where @xmath83_{{\bf r}={\bf r}_n } + ( h.c.)\big]\big)\nonumber\end{aligned}\ ] ] in appendix a we prove , that this green function obeys the groenewold equation ( see also appendix b of @xcite ) : @xmath84 as well as in continuum case the weyl symbol @xcite of operator @xmath85 is given by function @xmath86 that depends on the real numbers rather than on the operators . as it is explained in appendix a , if @xmath85 has the form of a function @xmath59 of the combination @xmath87 with a gauge potential @xmath88 , i.e. @xmath89 then we have @xmath90 ^ 2)\label{qf}\ ] ] here @xmath91 ^ 2)$ ] does not contain terms independent of the derivatives of @xmath92 and the terms linear in those derivatives , i.e. it is higher order in derivatives . in certain particular cases the restrictions on the term @xmath93 ^ 2)$ ] may be more strong , or it may even vanish at all @xcite . in particular , from sect . 4 ( eqs . ( 1.23)-(1.27 ) ) of @xcite it follows , that @xmath91 ^ 2)$ ] vanishes for the case of wilson fermions with the green function @xmath26 given by eqs . ( [ g10 ] ) and ( [ gwilson ] ) . notice , that the star product entering eq . ( [ i d ] ) is widely used in deformation quantisation @xcite and also in some other applications ( see , for example , @xcite and references therein ) . let us apply the gradient expansion to the wigner transform of the green function . for this we expand exponent in powers of its arguments in eq . ( [ i d ] ) . this gives @xmath94^{-1}}{\partial p_i } \tilde g^{(0 ) } \frac{\partial \big[\tilde g^{(0)}\big]^{-1}}{\partial p_j } \tilde g^{(0 ) } a_{ij } ( { \bf r})\nonumber\end{aligned}\ ] ] here @xmath95 is defined as the green function with the field strength @xmath96 neglected . it is given by @xmath97 next , suppose , that we modified the external gauge field as @xmath98 . the response to this extra contribution to gauge potential gives electric current . let us calculate this response basing on the description of the system given by eq . ( [ z4 ] ) : @xmath99\psi({\bf p } ) \nonumber\\ & = & -\int_{\cal m } \frac{d^d { \bf p}}{|{\cal m}| } \ , { \rm tr } \ , \big [ \delta \hat{\cal q}(i{\partial}_{{\bf p}_1},{\bf p}_1)\big]g({\bf p}_1,{\bf p}_2)\big|_{{\bf p}_1 = { \bf p}_2 = { \bf p}}\nonumber\\ & = & -\sum_{{\bf r}={\bf r}_n}\int_{\cal m } \frac{d^d { \bf p}}{|{\cal m}| } \ , { \rm tr } \ , \big [ \delta \hat{\cal q}(i{\partial}_{{\bf p}}+i{\partial}_{{\bf p}}/2 , { \bf p}+{\bf p}/2)\big]\ , e^{-i { \bf p}{\bf r } } \tilde{g}({\bf r},{\bf p})\big|_{{\bf p } = 0 } \label{j4}\end{aligned}\ ] ] in appendix b we discuss the weyl symbol @xmath100 of the operator @xmath85 entering eq . ( [ i d ] ) . notice , that @xmath101 and @xmath102 enter the expression inside @xmath85 in a symmetric way . this allows to use eq . ( [ corrl ] ) . the form of eq . ( [ j4 ] ) demonstrates , that the above expression for the electric current may also be written through the function @xmath86 : @xmath103\ , \nonumber\\&&e^{-i { \bf p}{\bf r } } \tilde{g}({\bf r},{\bf p})\big|_{{\bf p } = 0}\nonumber\\&= & -\sum_{{\bf r}={\bf r}_n}\int_{\cal m } \frac{d^d { \bf p}}{|{\cal m}| } \ , { \rm tr } \ , \big [ \delta { \cal q}({\bf r } , { \bf p}+{\bf p}/2)\big]\ , \nonumber\\&&e^{-i { \bf p}{\bf r } } \tilde{g}({\bf r},{\bf p})\big|_{{\bf p } = 0 } \label{j428}\end{aligned}\ ] ] according to the notations of appendix b the arrows above the derivatives mean , that those derivatives act only outside of @xmath104 , and do not act on the arguments of @xmath86 , i.e. @xmath105 acts on the function equal to @xmath106 standing left to the function @xmath86 while @xmath107 acts on the exponent @xmath108 . as a result of the above mentioned manipulations we come to the following simple expression for the electric current per unit volume of coordinate space , which follows from the relation @xmath109 : @xmath110^{-1}\label{j423}\end{aligned}\ ] ] here by @xmath111 we denote the volume of the unit cell understood as the ratio of the total volume of the system to the number of lattice points at which the field @xmath25 is defined . for the ordinary hypercubic lattice the product of the two volumes is obviously equal to @xmath112 . one might think , that for the lattices of more complicated symmetry the product of the momentum space volume and the defined above volume of the lattice cell may differ from this expression . nevertheless , this is not so , and in general case the given product is always equal to @xmath113 exactly . in general case of an arbitrary crystal the direct proof is rather complicated . however , the result for the product of the two volumes may be found from the simple field theoretical correspondence : the limit of the microscopic model described by the effective low energy theory should correspond to the product of the two volumes equal to @xmath112 . notice , that the construction of the unit cell in the original lattice should be performed with care . one has to count only those sites of the original crystal lattice , at which the dynamical variables of the model described by eq . ( [ z3 ] ) are incident . let us apply the gradient expansion to eq . ( [ j423 ] ) . it results in the following expression for the electric current : @xmath114^{-1}}{\partial p_k}\label{j142}\end{aligned}\ ] ] notice , that the second row of this expression represents the topological invariant as long as we deal with the system with regular green functions , which do not have poles or zeros , i.e. this expression is unchanged while we are continuously deforming the green function . we will not need this expression below since it does not contain the linear response to the external field strength . in the @xmath1 d systems the contribution to this current originated from @xmath115 is given by @xmath116 \label{nug}\end{aligned}\ ] ] in the linear response theory we should substitute @xmath117 into the expression for @xmath118 . therefore , in eq . ( [ nug ] ) we substitute @xmath26 instead of @xmath119 . @xmath118 is the topological invariant , i.e. it is robust to any variations of the green function @xmath120 as long as the singularities are not encountered ( for the proof see appendix b ) . although eq . ( [ calm ] ) was derived for the case of compact momentum space , the non - compact case may always be obtained as a limit of the compact case , when a certain parameter of the theory tends to infinity . therefore , in eq . ( [ calm ] ) the integration may be over compact or over non - compact space depending on the type of the given system . in particular , we will encounter in the upcoming sections the non - interacting condensed matter systems with the green functions of the form @xmath2 , where @xmath121 is the imaginary frequency while @xmath4 is the hamiltonian . momentum space of such a system is non - compact and has the form of @xmath122 , where @xmath15 is the compact brillouin zone . the linear response of electric current to external electromagnetic field is still given in such systems by eq . ( [ calm ] ) . the derivation is analogous to the one we gave above for the case of compact momentum space . the considerations of the previous section may easily be applied to the @xmath0 d systems . then instead of eq . ( [ calm ] ) we arrive at @xmath123^{-1}}{\partial p_i } \frac{\partial \big[\tilde g^{(0)}({\bf r},{\bf p})\big]}{\partial p_j } \frac{\partial \big[\tilde g^{(0)}({\bf r},{\bf p})\big]^{-1}}{\partial p_k } \big ] \label{calm2d}\end{aligned}\ ] ] in this section we give the derivation of the well - known expression for the hall current in the gapped system through the topological invariant in momentum space . for the references to the other derivations see , for example , @xcite . let us consider the @xmath124 d model with the gapped fermions and the green function @xmath10 that depends on the three - vector @xmath125 of euclidean momentum . in order to obtain expression for the hall current let us introduce into eq . ( [ calm2d ] ) the external electric field @xmath126 as @xmath127 ( the third component of vector corresponds to imaginary time ) . this results in the following expression for the hall current @xmath128 where the topological invariant ( denoted by @xmath5 according to the classification of @xcite ) is to be calculated for the original system with vanishing background gauge field : @xmath129 eq . ( [ n3a ] ) defines the topological invariant ( this is proved in appendix b ) . its appearance explains the quantization of hall conductance . in appendix c we demonstrate how to calculate the invariant @xmath5 entering eq . ( [ hallj ] ) for the models with @xmath130 green functions , and give the example of the model , where the corresponding value is nonzero . ( for the other examples of such models see @xcite . ) in the considered example the green function has the form @xmath131 , where the hamiltonian is @xmath132 for @xmath133 we have @xmath134 while @xmath135 for @xmath136 and @xmath137 for @xmath138 . let us consider briefly the above mentioned system with the hamiltonian eq . ( [ ham0 ] ) in the presence of boundary . it gives rise to the potential step of electric potential and the step in @xmath139 , which becomes the function of @xmath140 . say , if the boundary is directed along the @xmath141 axis , the potential step results in the extra electric field directed towards the interior of the sample @xmath142 where @xmath143 is the jump of electric potential . we may suppose , that @xmath144 then at @xmath145 we deal with the aqhe system with @xmath146 while at @xmath147 there is the ordinary insulator with @xmath148 . the step in @xmath149 is important for the demonstration of the appearance of massless edge states . namely , at @xmath150 the green function @xmath119 has the pole , which corresponds to the massless excitation . such excitations have definite chirality , and in general case their number is proportional to the jump of the topological invariant @xmath5 across the boundary ( if we substitute into the expression for @xmath5 the @xmath151 - dependent function @xmath139 ) . for the derivation of this index theorem the reader is advised to consult chapter 22 of @xcite . the similar derivation for the @xmath1 d case will be given in sect . [ bulkboundary ] . in this section we demonstrate how the topological invariant @xmath5 is expressed through berry curvature in the @xmath0 d models . this correspondence works for the green function of the non - interacting model that has the form @xmath152 with the hermitian hamiltonian @xmath4 . in the presence of interactions , when the green function receives a more complicated form this correspondence looses its sense . ( this is the advantage of the formalism , that utilizes the green functions . ) nevertheless , at the present moment the approach to the description of qhe based on berry curvature is more popular . therefore , we feel this instructive to derive it from our expressions . thus , we start from @xmath153 and substitute into it @xmath154 here @xmath155 is the @xmath156 - th eigenvalue of the hamiltonian depending on to the lattice momentum @xmath157 . the corresponding eigenvector of the hamiltonian is denoted by @xmath158 . we have : @xmath159 integral over the imaginary frequency @xmath3 may be taken as a residue at the negative value of @xmath160 , and gives @xmath161 this expression is the starting point of @xcite , where the expression for the hall conductivity through the berry curvature has been derived for the first time . after some algebra we arrive at @xmath162 with the following definition of the momentum space gauge field @xmath163 we arrive at @xmath164 where the sum is over the occupied states while berry curvature is given by @xmath165 now let us turn to the @xmath1d systems . similar to the two - dimensional models the hall current is given by @xmath166 where @xmath167 is given by @xmath168 \label{nughall}\end{aligned}\ ] ] notice , that in the case of the non - interacting condensed matter system with the green function @xmath169 expressed through the hamiltonian @xmath4 we may derive ( similar to the above considered case of the @xmath0 d system ) the representation of the components of topological invariant @xmath170 with @xmath171 through the berry curvature given by eqs . ( [ ac ] ) , ( [ bc ] ) : @xmath172 here the sum is over the occupied branches of spectrum . let us demonstrate how @xmath174 may be calculated . let us consider first ( as in appendix c ) the case , when the green function has the form @xmath175 now we have @xmath176 , and @xmath177 is to be identified with the imaginary frequency . we denote @xmath178 , @xmath179 , and introduce the parametrization @xmath180 where @xmath181 while @xmath182 , and @xmath183 $ ] . we suppose , that @xmath184 on the boundary of momentum space @xmath185 . this gives @xmath186 now @xmath187 is the small vicinity of line @xmath188 in momentum space , where vector @xmath189 is undefined . along these lines @xmath190 . we have @xmath191 here we use the notations of @xcite , and @xmath192 is the integer number , the corresponding integral is along the infinitely small surface @xmath193 , which is wrapped around the line @xmath194 near to the given point of this line . let us consider the more complicated systems with the green function of the form @xmath196 where @xmath28 are euclidean dirac matrices while @xmath29 , @xmath197 are the real - valued functions , @xmath31 . we define @xmath198 in chiral representation as @xmath199 . let us represent @xmath59 as follows : @xmath200 where @xmath201 let us assume , that @xmath202 never equals to zero , and that the green function does not have poles . then since @xmath174 is the topological invariant , we may deform continuously the green function to the form @xmath203 under this deformation @xmath174 remains invariant if on the boundary of momentum space @xmath204 { \cal g } ] d { \cal g}^{-1}\wedge d { \cal g}\right ) = 0\label{condtop0}\ ] ] this follows from appendix b. this condition is fulfilled if on @xmath205 @xmath206 now @xmath174 is equal to @xmath207 where @xmath208 is to be calculated using the @xmath173 green functions @xmath209 thus the problem is reduced to the one considered above in sect . [ sect22 ] . it is worth mentioning , that the given way of calculation may easily be extended to the green functions of some other forms . in particular , the same expression for the topological invariant is valid for @xmath210 . we are able to use the wigner transform of the green function in order to describe the bulk - boundary correspondence ( for the description of the method see chapter 22 in @xcite and @xcite ) . let us suppose , that in the given solid ( @xmath140 is inside the bulk ) there are no poles of @xmath211 as functions of @xmath212 . let us also suppose , that a surface divides space into the two parts with different values of @xmath174 . for simplicity let us consider the situation , when momentum space has the form of torus , @xmath170 is nonzero for @xmath213 only , while this surface is situated in the @xmath214 plane , and @xmath211 depends on @xmath215 only . the set @xmath216 with @xmath217 constitues four - dimensional space . the following integrals over the two hyper - surfaces in this space with @xmath218 ( with small @xmath219 ) may be deformed safely to a closed hyper - surface : @xmath220 \label{nughall21}\end{aligned}\ ] ] if the resulting integral is nonzero , then inside it there are @xmath221 topologically protected poles of @xmath119 , which correspond to massless states concentrated in a small vicinity of boundary . there is also the question about the relation of these poles of @xmath119 with the gapless excitations in the effective theory that describes fermions on the boundary . such a relation was discussed in chapter 22 of @xcite and in @xcite ) . the output of this study is that the chirality of those @xmath222 dimensional fermions is given by the sign of @xmath221 . this chirality is encoded in the quantity @xmath223 here the contour is in the plane @xmath224 , while the sum is over all possible ( discrete ) values of @xmath225 . again , we use eq . ( [ i d ] ) and expand exponent in powers of its arguments , which gives @xmath226^{-1}}{\partial p_3 } \tilde g^{(0 ) } \frac{\partial \big[\tilde g^{(0)}\big]^{-1}}{\partial z } \tilde g^{(0 ) } \nonumber\\ & & -\frac{i}{2 } \tilde g^{(0 ) } \frac{\partial \big[\tilde g^{(0)}\big]^{-1}}{\partial z } \tilde g^{(0 ) } \frac{\partial \big[\tilde g^{(0)}\big]^{-1}}{\partial p_3 } \tilde g^{(0 ) } \nonumber\end{aligned}\ ] ] here @xmath95 is given by eq . ( [ q0 ] ) . let us substitute eq . ( [ gexp2 ] ) into eq . ( [ n1 ] ) . assuming , that @xmath120 is slowly varying we substitute @xmath227 by @xmath228 . this way we come to the integral of the form @xmath229 \label{nughall22}\end{aligned}\ ] ] here the integral is over the surface @xmath230 , where @xmath19 corresponds to the values of @xmath231 while @xmath17 corresponds to the values of @xmath225 . since there are no poles of @xmath120 inside the bulk we are able to deform this hyper - surface into the form of two hyper - planes @xmath232 placed at @xmath218 , which gives the value of @xmath233 . in turn , for the case of noninteracting system with @xmath234 we have @xmath235 where the sum is over the eigenstates of the hamiltonian enumerated by index @xmath156 and the values of conserved momenta @xmath236 . let us denote @xmath237 . then eq . ( [ n1 ] ) may be represented as @xmath238 this expression gives the sum of the chiralities of the states enumerated by @xmath156 . only those states which are localized at the boundary may have gapless excitations because according to our supposition in the bulk of the system there are no poles of the green function . thus we come to conclusion , that @xmath233 enumerates the gapless boundary modes , and its sign corresponds to their chirality . notice , that the similar conclusion is given in the lattice theory of domain - wall fermions ( see , for example , @xcite ) . now let us describe the dependence of this pattern on @xmath239 . @xmath233 can not depend on @xmath239 because we assumed , that there are no poles in the bulk . therefore , we come to the conclusion , that the topologically protected gapless surface states exist , which form lines in momentum space ( the dispersion does not depend on @xmath239 ) . the number of such fermi lines is equal to the jump of @xmath221 accross the boundary and is related to the jump of @xmath240 . the generalization of this consideration to the values of @xmath174 for arbitrary @xmath241 is straightforward . the generalization to the case of the brillouin zones of arbitrary form is more involved and will be omitted here . it is worth mentioning , that in section [ sectweylsem ] we will see , that for the weyl semimetals , when there are fermi points in the bulk , the described above boundary fermi lines become the fermi arcs connecting the bulk fermi points . in this section we will consider the two particular examples of the @xmath1d insulators , for which the topological invariant @xmath174 may be calculated using the technique developed above . let us start from the model with the hamiltonian @xmath242 for @xmath243 , and @xmath244 we deal with the insulator . here we use the system of units , in which the lattice spacing @xmath245 is equal to unity . therefore , lattice momentum @xmath246 is dimensionless . the @xmath173 green function has the form @xmath131 while the brillouin zone has the form of torus . we will be using the following expression for the green function : @xmath247 we have @xmath248 and @xmath249 therefore @xmath250 while @xmath251 one can see , that we deal with the aqhe current @xmath252 we restored the value of the lattice spacing @xmath245 in this expression in order to keep the track of dimensionality . notice , that the same expression may be read off from eq . ( a.5 ) of @xcite . ( [ nughall ] ) represents the topological invariant if on the boundary of momentum space @xmath253 { \cal g } ] \ } d { \cal g}^{-1}\wedge d { \cal g}\right ) = 0\label{condtop}\ ] ] this follows from appendix b. the interactions , that keep eq . ( [ condtop ] ) can not affect the value of the aqhe current until the phase transition is encountered . our second example corresponds to the system with the @xmath195 green function of the form of eq . ( [ g202 ] ) with @xmath254 this system is able to describe qualitatively a certain class of topological insulators @xcite . depending on the identification of matrices @xmath255 entering matrices @xmath256 of eq . ( [ g202 ] ) with the operators of spin , isospin , or bogolyubov spin , eq . ( [ ins2 ] ) may describe different classes of topological insulators , including those with broken time reversal symmetry and/or broken particle - hole symmetry . we assume , that the parameters entering the expression for the green function satisfy @xmath257 , @xmath258 while @xmath259 and @xmath260 we will use eqs . ( [ mpm ] ) and ( [ g12d23 ] ) . let us denote @xmath261 and @xmath262 , where @xmath263 ^ 2}$ ] . then we have @xmath264 the above mentioned ranges of parameters guarantee , that @xmath265 we are able to use eq . ( [ mpm ] ) if eq . ( [ cond00 ] ) is fulfilled at @xmath266 . this can be checked directly . we find , that again @xmath250 while @xmath251 as well as for the considered above system with the @xmath173 green function we deal with the aqhe current @xmath252 ( here @xmath245 is the lattice spacing . ) as it was mentioned above , the nontrivial topology in the bulk of the topological insulators , which possess the aqhe , manifests itself in the appearance of the surface fermi lines . let us suppose , that the given insulator sample has the cubic form . then the fermi lines appear on the boundaries that belong to planes @xmath267 and @xmath268 . those fermi lines are placed at @xmath269 . on the boundaries that belong to @xmath270 plane there is the fermi point at @xmath271 . this pattern may easily be derived using the wigner transform of the green function @xmath211 , which depends on coordinate denending parameter @xmath272 . it is assumed , that this parameter vanishes when we approach boundary . let us extend the above consideration to the models of weyl semimetals . first , let us start from the toy model of weyl semimetal with the green function @xmath131 and the hamiltonian of the form @xmath273 this is the modification of the model i1 considered above . for @xmath274 the system contains the two fermi points @xmath275 although the green function contains singularities , the integral in eq . ( [ nughall ] ) is convergent as we will see below . to be explicit , we first integrate over the momentum space with the small vicinities of the poles subtracted , and then consider the limit , when those vicinities are infinitely small . we will be using the following expression for the green function : @xmath276 we have @xmath277 and @xmath278 therefore @xmath250 while @xmath279 thus we come to the expression for the aqhe current @xmath280 from the above expression it is obvious , that the contributions of all zeros of @xmath281 ( @xmath282 ) listed in eq . ( [ listzerosw1 ] ) are important . nevertheless , our result coincides with the one of the naive low energy effective field theory @xcite . the coefficient in eq . ( [ hallj3dp ] ) has also been calculated using the technique of berry curvature in @xcite . it is also worth mentioning , that in the particular case of the weyl semimetal , when the green function contains poles eq . ( [ nughall ] ) is not the topological invariant . it still gives the aqhe current , though . but we should remember , that turning on interactions we change the value of @xmath283 , thus leading to the continuous renormalization of the value @xmath284 that marks the position of the fermi point . notice , that even in the presence of a pole eq . ( [ nughall ] ) remains invariant under those continuous changes of the green function , which do not alter the positions of the fermi points and for which @xmath285 { \cal g } ] \ } d { \cal g}^{-1}\wedge d { \cal g}\right ) = 0\ ] ] where the integral is taken along the boundary of the small vicinity of the pole . this follows immediately from the derivation of appendix b. it is worth mentioning , that the first order in the derivative expansion does not allow to calculate the derivative of electric current . that s why we can not simply differentiate eq . ( [ hallj3dp ] ) in order to calculate such a derivative . both dependence of @xmath284 on @xmath151 and the second order in the derivative expansion should be applied to calculate @xmath286 ( that should vanish in equilibrium due to the gauge invariance ) . let us consider the more complicated model of weyl semimetal , which represents the deformation of the insulator model i2 with the green function of the form of eq . ( [ g202 ] ) . we imply , that the functions @xmath287 and @xmath288 entering eq . ( [ g202 ] ) have the form of eq . ( [ ins2 ] ) . but we assume different relations between the parameters of the model . now we require , that the parameters entering the expression for the green function satisfy @xmath289 while @xmath290 @xmath291 @xmath292 the given model contains the two fermi points @xmath293 instead of eq . ( [ listzerosi2 ] ) we have @xmath294 in eq . ( [ nughall ] ) we represent @xmath118 as an integral over the component @xmath295 of the topological invariant of the @xmath0 d system , which depends on @xmath295 as on parameter . in the presence of the pole the integral is ill - defined at the positions of poles and requires regularization . we consider the integration over @xmath295 with the small vicinities of the @xmath241 - th coordinates of the poles subtracted . for each value of @xmath295 , which does not belong to this interval the integral over the other components of @xmath73 is well - defined . therefore , we are able to use eq . ( [ mpm ] ) if eq . ( [ cond00 ] ) is fulfilled at @xmath266 ( which may be checked easily ) . the values of @xmath296 are given by the integrals in eq . ( [ calmproj ] ) over @xmath297 , where as it was explained above , the small vicinities @xmath298 of the @xmath299 - th coordinates of the poles @xmath300 , @xmath301 are subtracted . at the end of the calculations we take the limit @xmath302 . this way we find , that @xmath250 while @xmath303 the aqhe current reads @xmath304 in order to illustrate how the bulk - boundary correspondence works here , let us consider what happens to the fermi points when we aproach the boundary of the sample . this approaching is described by the wigner transform of the green function @xmath211 , which depends on the coordinates @xmath140 through the dependence of parameter @xmath288 on @xmath140 . we assume , that out of the bulk of the semimetal @xmath305 , and this parameter is increased inside the semimetal . in this situation the derivative expansion in the presence of external electric field becomes more complicated near boundary than inside of the bulk . however , we do not discuss here the corresponding expression for the electric current . as for the fermi points , one can easily see , that the two fermi points @xmath306 approach each other , and finally annihilate while we are aproaching boundary . since this process occurs in the small vicinity of the boundary , it results in the appearance of surface states of zero energy that connect the two bulk fermi points . these surface states are the fermi arcs . the conventional expression for the cme current reads @xmath307 , where @xmath308 is the chiral chemical potential . one may expect , that such an expression will follow from eq . ( [ calm ] ) : we need to substitute @xmath117 into eq . ( [ nug ] ) in the linear response approximation . then one might expect that @xmath309 for the system of a single massless dirac fermion and a similar expression may take place for the other systems with gapless fermions ( or , nearly gapless fermions ) . let us consider the situation , when vector gauge field @xmath310 has the nonzero components with @xmath282 that do not depend on imaginary time . there may exist many different definitions of @xmath308 . the most straightforward way is to consider the following expression for the fermion green function : @xmath311 in the limit of vanishing chiral chemical potential it is reduced to the form of eq . ( [ g10 ] ) , where @xmath28 are euclidean dirac matrices while @xmath29 and @xmath30 are the real - valued functions , @xmath31 . we define @xmath198 in chiral representation as @xmath199 . we may substitute @xmath312 of eq . ( [ g2 ] ) into eq . ( [ calm ] ) instead of @xmath313 while dealing with the linear response to the external magnetic field . we will apply our methodology to the investigation of the theory with compact brillouin zone . we imply , that momentum space can be represented as @xmath314 or @xmath315 , where @xmath15 is the compact @xmath316 brillouin zone . first , we assume , that the green functions do not have zeros or poles , which means , that the fermions are gapped ( massive ) . however , at the end of the calculations the limit of vanishing mass may always be considered . we need , though , that the inclusion of the chiral chemical potential does not produce poles in the green function . for the green function of the form of eq . ( [ g2 ] ) this may be proved as follows . the poles of the green function appear as the solutions of the following equation @xmath317 one can easily see , that if functions @xmath30 and @xmath318 never vanish simultaneously , then the chiral chemical potential of eq . ( [ g2 ] ) can not lead to the appearance of the poles of @xmath26 . to illustrate this general pattern let us consider the green function for the toy model of topological insulator inspired by the lattice regularization with free wilson fermions : @xmath319 one can see , that for the positive values of bare mass parameter @xmath320 the values of @xmath30 do not vanish . @xmath321 is topological invariant , i.e. it is robust to any variations of the green function as long as the singularities are not encountered ( for the proof see appendix b ) . the introduction of chiral chemical potential is the particular case of such a variation . actually , for the green function of the form of eq . ( [ g10 ] ) @xmath322 , which may be proved easily using the properties of gamma matrices . therefore , if the nonzero value of @xmath308 does not cause the appearance of the poles of the green function , then @xmath322 for @xmath323 . the mentioned above consideration is sufficient to prove , that there is no cme in the lattice regularized relativistic quantum field theory because the wilson fermions with @xmath324 represent the typical lattice regularization . however , in the solid state systems the situation may be different from the one considered above . this may be illustrated by the model of eq . ( [ wilsonlike ] ) for @xmath325 . the corresponding value of the topological invariant @xmath326 introduced in @xcite is nonzero . in this case the nonzero value of @xmath308 causes the appearance of the fermi lines . those fermi lines appear as the solution of the system of equations latexmath:[\ ] ] that s why we proved that @xmath5 and @xmath321 are the topological invariants . in the similar way it may be proved , that @xmath118 is the topological invariant for @xmath171 if @xmath394 , and @xmath26 tends to zero sufficiently fast at @xmath395 . also it may be easily found that for any @xmath241 the value of @xmath118 is topological invariant if we deal with any compact momentum space . in this section we demonstrate how the value of the topological invariant @xmath5 may be calculated . below we calculate @xmath5 for the case , when the green function has the form @xmath396 where @xmath397 are pauli matrices while @xmath29 and @xmath318 are the real - valued functions , @xmath398 . let us define @xmath399 where @xmath178 , and @xmath179 . then @xmath400 where @xmath401 and @xmath28 are euclidean dirac matrices in chiral representation , @xmath198 in chiral representation is given by @xmath199 . this gives @xmath402 let us introduce the parametrization @xmath403 where @xmath404 while @xmath405 , and @xmath183 $ ] . let us suppose , that @xmath184 on the boundary of momentum space @xmath185 . this gives @xmath406 in the last row @xmath187 is the small vicinity of point @xmath407 of momentum space , where vector @xmath189 is undefined . the absence of the singularities of @xmath408 implies , that @xmath190 at such points . let us illustrate the above calculation by the consideration of the particular example of the system with the green function @xmath131 , where the hamiltonian has the form @xmath412 this gives @xmath413 the boundary of momentum space corresponds to @xmath414 . we have @xmath415 for example , for @xmath416 we have @xmath417 therefore , we get immediately @xmath418 in the similar way @xmath135 for @xmath419 and @xmath137 for @xmath420 . hasan and c.l . kane , topological insulators , rev . 82 * , 30453067 ( 2010 ) . xiao - liang qi and shou - cheng zhang , topological insulators and superconductors , rev . mod . phys . * 83 * , 10571110 ( 2011 ) . volovik , quantum phase transitions from topology in momentum space , in : quantum analogues : from phase transitions to black holes and cosmology , eds . w.g . unruh and r. schutzhold , springer lecture notes in physics 718/2007 , pp . 31 - 73 ; cond - mat/0601372 . nielsen , m. ninomiya : absence of neutrinos on a lattice . i - proof by homotopy theory , nucl . b * 185 * , 20 ( 1981 ) ; absence of neutrinos on a lattice . ii - intuitive homotopy proof , nucl . b * 193 * , 173 ( 1981 ) . m.a.zubkov , g.e.volovik , momentum space topological invariants for the @xmath422 relativistic vacua with mass gap , nucl . b , 860 ( 2012 ) 295 - 309 doi:10.1016/j.nuclphysb.2012.03.002 , arxiv:1201.4185 k. b. dave , p. w. phillips and c. l. kane , `` absence of luttinger s theorem due to zeros in the single - particle green function , '' phys . * 110 * ( 2013 ) 9 , 090403 doi:10.1103/physrevlett.110.090403 [ arxiv:1207.4201 [ cond-mat.str-el ] ] . mikhail zubkov , `` on the absence of equilibrium chiral magnetic effect . '' @xmath423hal-01275180v2@xmath424 https://hal.archives-ouvertes.fr/hal-01275180v2/document k. landsteiner , e. megias and f. pena - benitez , `` anomalous transport from kubo formulae , '' lect . notes phys . * 871 * ( 2013 ) 433 [ arxiv:1207.5808 [ hep - th ] ] . m. n. chernodub , a. cortijo , a. g. grushin , k. landsteiner , and m. a. vozmediano , _ `` a condensed matter realization of the axial magnetic effect '' _ , phys . b * 89 * , 081407(r ) ( 2014 ) [ arxiv:1311.0878 ] . e. v. gorbar , v. a. miransky , i. a. shovkovy and p. o. sukhachov , `` chiral separation and chiral magnetic effects in a slab : the role of boundaries , '' phys . b * 92 * ( 2015 ) 24 , 245440 doi:10.1103/physrevb.92.245440 [ arxiv:1509.06769 [ cond-mat.mes-hall ] ] . zhou , jian - hui ; jiang , hua ; niu , qian ; shi , jun - ren , `` topological invariants of metals and the related physical effects '' , chinese physics letters , volume 30 , issue 2 , article i d . 027101 ( 2013 ) , arxiv:1211.0772 p. v. buividovich , m. puhr and s. n. valgushev , `` chiral magnetic conductivity in an interacting lattice model of parity - breaking weyl semimetal , '' phys . b * 92 * ( 2015 ) 20 , 205122 doi:10.1103/physrevb.92.205122 [ arxiv:1505.04582 [ cond-mat.str-el ] ] . p. v. buividovich , `` spontaneous chiral symmetry breaking and the chiral magnetic effect for interacting dirac fermions with chiral imbalance , '' phys . d * 90 * ( 2014 ) 125025 doi:10.1103/physrevd.90.125025 [ arxiv:1408.4573 [ hep - th ] ] . y. chen , d. bergman , and a. burkov , _ `` weyl fermions and the anomalous hall effect in metallic ferromagnets '' _ , phys . rev . b * 88 * , 125110 ( 2013 ) [ arxiv:1305.0183 ] ; david vanderbilt , ivo souza , and f. d. m. haldane phys . rev . b * 89 * , 117101 ( 2014 ) [ arxiv:1312.4200 ] . a. a. zyuzin and a. a. burkov , _ `` topological response in weyl semimetals and the chiral anomaly , '' _ phys . b * 86 * ( 2012 ) 115133 [ arxiv:1206.1868 [ cond-mat.mes-hall ] ] . pallab goswami , sumanta tewari , _ axionic field theory of ( 3 + 1)-dimensional weyl semi - metals , _ phys . rev . b 88 , 245107 ( 2013 ) , arxiv:1210.6352 d. e. kharzeev , j. liao , s. a. voloshin and g. wang , `` chiral magnetic effect in high - energy nuclear collisions a status report , '' arxiv:1511.04050 [ hep - ph ] . d. e. kharzeev , `` chern - simons current and local parity violation in hot qcd matter , '' nucl . phys . a * 830 * ( 2009 ) 543c doi:10.1016/j.nuclphysa.2009.10.049 [ arxiv:0908.0314 [ hep - ph ] ] . h. b. nielsen and m. ninomiya , `` adler - bell - jackiw anomaly and weyl fermions in crystal , '' phys . b * 130 * ( 1983 ) 389 . doi:10.1016/0370 - 2693(83)91529 - 0 d. e. kharzeev and h. j. warringa , `` chiral magnetic conductivity , '' phys . d * 80 * ( 2009 ) 034028 doi:10.1103/physrevd.80.034028 [ arxiv:0907.5007 [ hep - ph ] ] . hongming weng , rui yu , xiao hu , xi dai , zhong fang , `` quantum anomalous hall effect and related topological electronic states '' , advances in physics , volume 64 , issue 3 , 2015 , pages 227 - 282 , doi:10.1080/00018732.2015.1068524 f. r. klinkhamer and g. e. volovik , `` emergent cpt violation from the splitting of fermi points , '' int . j. mod . phys . a * 20 * ( 2005 ) 2795 doi:10.1142/s0217751x05020902 [ hep - th/0403037 ] . a. p. schnyder , s. ryu , a. furusaki , and a. w. w. ludwig , phys . rev . b 78 , 195125 ( 2008 ) .

high energy physics experiments are currently entering a new era which requires the operation of gaseous particle detectors at unprecedented high rates and integrated particle fluxes . full functionality of such detectors over the lifetime of an experiment in a harsh radiation environment is of prime concern to the involved experimenters . new classes of gaseous detectors such as large - scale straw - type detectors , micro - pattern gas detectors and related detector types with their own specific aging effects have evolved since the first workshop on wire chamber aging was held at lbl , berkeley in 1986 . in light of these developments and as detector aging is a notoriously complex field , the goal of the workshop was to provide a forum for interested experimentalists to review the progress in understanding of aging effects and to exchange recent experiences . a brief summary of the main results and experiences reported at the 2001 workshop is presented , with the goal of providing a systematic review of aging effects in state - of - the - art and future gaseous detectors .

hep-ex0204005

aging effects in proportional wire chambers , a permanent degradation of operating characteristics under sustained irradiation , has been and still remains the main limitation to their use in high - rate experiments @xcite . although the basic phenomenology of the aging process has been described in an impressive variety of experimental data , it is nevertheless difficult to understand any present aging measurement at a microscopic level and/or to extrapolate it to other operating conditions . many chemical processes are expected to occur simultaneously in the gaseous discharges surrounding the wire , and consequently a quantitative description of aging effects , which would require as a minimum a detailed analysis of all gas - phase and gas - surface reaction products , is currently not available . there is much experimental information , well summarized in @xcite-@xcite , which suggests that wire chamber lifetime may be extremely sensitive to the nature and purity of the gas mixture , different additives and trace contaminants , materials used in contact with the gas , geometry of electrodes and configuration of electric field . the classical aging effects , well known since the advent of wire chambers , lead to the formation of deposits , conductive or insulating , on the electrode surfaces and manifest themselves as a decrease of the gas gain due to the modification of electric field , excessive currents , self - sustained discharges , or sparking . traditionally , the aging rate has been parameterized as a normalized gas gain loss : @xmath0 , where @xmath1 is the initial gas gain , @xmath2 is the loss of gas gain after collected charge @xmath3 per unit length @xcite . however , the assumption that the aging rate is only a function of the total accumulated charge has not been confirmed for gaseous detectors operated in high - rate environments . in reality , the rate of polymer formation depends upon many microscopic variables such as cross - sections of electron and photon processes and their energy distributions in gas avalanches , molecular dissociation energies , as well as densities of electrons , ions and free radicals . consequently , one may expect that the aging rate could also be affected by macroscopic parameters , such as gas gain , ionization density and radiation intensity , which are directly related to the basic microscopic variables . several results presented at the 2001 workshop clearly indicate that such dependencies do exist . some of the conclusions from the 1986 workshop are still valid in 2001 . however , the dramatic increase in charge ( up to 1.0 @xmath4 per year ) , which is expected to be accumulated on sensing electrodes in the new high rate experiments , poses much more stringent constraints on the radiation hardness of materials and gas mixtures , assembly procedures , and basic rules for construction and operation of gaseous detectors , than previously encountered . only a limited choice of gases have been demonstrated to tolerate such doses . moreover , recent experience with straws and honeycomb drift tubes revealed that chemical etching processes leading to a dramatic damage of gold - plating on wires could occur in non - polymerizing @xmath5 mixtures at exceedingly high current densities . these new developments since the 1986 workshop raise a question about the adequacy of using @xmath5-based mixtures for long - term , high - rate applications . the scientific program of the 2001 workshop addressed specific questions which , as reported by many authors , are of a primary interest : classical aging effects , models and insights from plasma chemistry , materials for detectors and gas systems , lessons learned from detector operation at high radiation intensities , new aging effects , experiences with large systems and recommendations for future detectors . about 100 detector experts attended the 4 day workshop , and 10 invited talks , 31 contributed talks and 9 posters were presented in 7 sessions @xcite . the classical aging effects are the result of chemical reactions occurring in avalanche plasmas near anodes in wire chambers leading to formation of deposits on electrode surfaces . during gas avalanches many molecules break up in collisions with electrons , de - excitation of atoms , and uv - photon absorption processes . whereas most ionization processes require electron energies greater than 10 ev , the breaking of covalent molecular bonds and formation of free radicals requires only 3 - 4 ev , and can lead to a higher concentration of free radicals than that of ions in the gaseous discharges . consequently , free - radical polymerization is regarded as the dominating mechanism of wire chamber aging . since free radicals are chemically very active they will either recombine to form the original molecules or other volatile species , or may start to form new cross - linked molecular structures of increasing molecular weight . when the polymerized chain becomes large enough for condensation to occur , it will diffuse to an electrode surface . it is worthwhile to mention that one has to distinguish between formation of polymers in the gas avalanche near the anode wire and their deposition on electrode ( anode or cathode ) surfaces . the polymer deposition mechanism can be viewed as a phenomenon that occurs whenever the gaseous species fails to bounce back after a collision with an electrode surface , including a surface layer of molecules previously formed in the gas discharges . initially , the polymer could be attached to the surface very weakly , unless some additional chemical reactions take place between the polymer atoms and atoms of the wire material . moreover , many free radicals are expected to have permanent or induced dipole moments so that electrostatic attraction to a wire can also play a significant role in the polymer deposition process . for the inert gold - plated anode wires the probability for polymers to stick to the surface is rather small until the creation of the first monolayer of deposits , which may significantly increase further deposition . the influence of surface wire quality on anode aging and a model of polymer film growth is proposed in @xcite . the importance of reactions between the electrode material and polymers produced in avalanches for the deposition mechanism can be illustrated by the following examples : * non - gold anode wires react with fluorine radicals produced in an avalanche to form resistive metal fluorides . many studies have demonstrated excellent aging properties , up to 10 @xmath6 , of @xmath7 ( 80:20 ) gas avalanches @xcite-@xcite , which also has the ability to etch silicon - based and hydrocarbon deposits on previously aged gold - plated wires @xcite . however , extensive deposition was observed on unplated wires irradiated in @xmath8 ( 80:20 ) @xcite . * exceedingly large aging rates were observed in pure @xmath5 and in @xmath9 ( 50:40:10 ) @xcite-@xcite , which are typical etching gases and reluctant to polymerize . this effect was related to the chemical processes at the cathode , where trace fluorocarbon deposits were found resulting in a loss of gas gain and not in a self - sustained malter discharge . the self - sustained discharge ( malter effect ) @xcite , which is due to a thin insulating layer deposited on a conducting cathode by a polymerization mechanism , is one of the most devastating phenomena of all aging effects . the resistivity of the insulating layer defines the maximum rate capability of the detector before the onset of field - emission of electrons from the cathode , which starts if the rate of ionic charge neutralization across the dielectric film is smaller than the rate of ion charge build - up @xcite . there exists evidence that certain metal oxide coatings on the cathode and/or simply the cathode material itself ( eg . carbon - loaded polycarbonate foil ) may not be conducting enough and could cause malter - like breakdowns in the presence of large localized ionization densities @xcite . several other factors may facilitate its ignition , such as highly ionizing particles , sparks , sharp points on electrodes causing corona discharges , or thin anode wires @xcite . it is easy to ignite malter currents in a detector operating with hydrocarbon gases at elevated high voltages @xcite or forcing chambers to breakdown @xcite-@xcite , and in a detector , which has been previously exposed to tmae gas @xcite . the crid rich detector @xcite with an excellent 3-dimensional single electron reconstruction capability allowed the first imaging of the onset of the malter effect , which starts from sporadic bursts of single electrons from a localized cathode spot @xcite . such a positive feedback between electron emission at the cathode and anode amplification will lead to high ionization densities at distinct chamber locations . this , in turn , can initiate the production of new reactive species at much larger rates , thus promoting more deposits to form at the same cathode spot to an extent sufficient to establish a classical self - sustained malter discharge . the most dangerous consequence of this phenomena is that the malter effect could easily spread over a large area , if it goes undetected for a long period of operation , thus causing irreparable damages to the chamber . many experiments have demonstrated that the addition of @xmath10 or alcohols after the insulating layer at the cathode is already formed tends to stabilize the detector operation , but not to cure the malter effect @xcite . when these additives are removed , usually the chamber suffers from malter effect again . recently , it was discovered that the addition of oxygen ( 0.02 - 0.05 @xmath11 ) or @xmath12 ( 5 @xmath11 ) to the damaged chamber , which showed a self - sustained dark current with @xmath13 ( 80:20 ) , could revert or cure a malter breakdown in the presence of high current density @xcite . when the oxygen is removed , the chamber can still operate at a high ionization level ( although it will start to age again without additive ) . it is also worthwhile to mention that the possibility of reanimation of anode wires aged in hydrocarbon gases by means of sputtering was demonstrated in @xmath14 ( 99:1 ) and @xmath15 ( 93:7 ) @xcite . these effects support results from plasma chemistry , where it is known that oxygen reacts with hydrocarbon molecules and the end products are volatile @xmath16 , @xmath12 , @xmath10 and @xmath17 . while the specific reactions responsible for wire chamber aging are extremely complex , some qualitative approach to the aging phenomena in different gases could be obtained from similarities between chemical processes in plasmas of gas avalanches @xcite and those that occur in the better - understood low - pressure ( @xmath18 1 torr ) rf ( 13.6 mhz ) plasmas @xcite . although many parameters ( electric field , gas pressure , electron density , power density ) are vastly different between the two regimes , the electron energies are not so different . also , in both cases the free radicals are most likely the active species involved in polymer formation . in plasma polymerization , the overall mechanism of ` competitive ablation and polymerization ' proves to be a basic principle that describes reactions occurring in a plasma polymerization system . considerable fragmentation of the gas molecules or rearrangement of atoms occurs in the plasma and the extent of the process and the dominating mechanism vary with the types of gases and the discharge conditions . the most important concept here is that both polymer - forming species and species that cause ablation ( physical or chemical etching ) of materials are created in the plasma of the original gas . the significance of this concept is fully established in perfluorocarbon plasmas , which represent the most extreme case of ablation competing with polymer formation . actually , @xmath5-based gases are used for both etching and deposition processes , the distinction being made by the gas and its concentration with which @xmath5 is mixed . in general , the addition of oxygenated species shifts the chemistry of @xmath5 plasmas towards etching , while the addition of hydrogenated species shifts the chemistry towards polymerization @xcite . in the former case , the dissociative products of @xmath5 and @xmath19 are the most desirable active species for the etching processes in plasmas @xcite-@xcite . for the latter case , in the absence of hydrogen , products of the @xmath5 discharge could act as an effective etching gas especially for @xmath20-based deposits , which react with fluorine to form volatile @xmath21 . the addition of hydrogen atoms or molecules to @xmath5 scavenge @xmath22 atoms by the formation of more stable @xmath23 and produces a mixture with carbon - enriched ( @xmath24 ) residues . as the ratio of @xmath25 decreases , perfluorocarbons polymerizes readily , i.e. the balance shifts from ablation to polymerization @xcite-@xcite . for instance , very fast polymer formation was observed in @xmath26 and @xmath27 plasmas @xcite . on the other hand , hydrofluoric acid can chemically attack @xmath23-soluble materials existing in the system . under certain conditions , @xmath20-etching can be accompanied by the polymerization of the etching gas @xmath5 on the @xmath20-substrate @xcite . correspondingly , recent results from wire chamber operation also show that both polymerization and etching phenomena can occur in @xmath5-based gases ( see section v ) . particularly , using the same experimental setup , a lack of apparant aging have been observed in @xmath8 ( 80:20 ) and @xmath8 ( 50:50 ) mixtures , whereas heavy carbonaceous deposits were observed on the gold - plated wires in @xmath8 ( 95:5 ) , @xmath8 ( 20:80 ) and @xmath28 ( 95:5 ) gases @xcite . two other examples , where conclusions from plasma chemistry are qualitatively applicable to wire chambers are : * in plasma chemistry , most organic compounds with oxygen - containing groups are generally reluctant to form polymers . for example , water in plasmas could act as an efficient modifier of the polymer chain - growth mechanism by reacting with polymer precursors and forming volatile species ( up to 50 @xmath11 @xmath10 was added to the plasma feed gas ) @xcite . in wire chambers , the addition of water ( a few hundred to a few thousands @xmath29 of @xmath10 ) has been found to effectively suppress polymerization effects @xcite , to prevent malter breakdown @xcite , or even to restore the original operation in aged counters @xcite . there is more than one mechanism by which @xmath10/alcohols can help in wire chambers @xcite . because of the large dipole moment , these molecules will tend to concentrate near the electrode surfaces , where polymerization takes place . water has an additional advantage in wire chambers since it increases the conductivity of the partially damaged electrodes - a property that can have adverse effects in a msgc @xcite . * in plasmas , the characteristic polymerization rate of @xmath20 is higher than for @xmath30 @xcite . from the viewpoint of wire aging , even minor traces of @xmath20-pollutants in the gas have a much higher tendency to create deposits , than similar amounts of hydrocarbon molecules . it has to be stated , though , that the absence of corresponding systematic studies in plasma chemistry with parameters similar to wire chambers ( atmospheric pressure , power densities , gas mixtures ) does not allow any quantitative comparisons between the plasma chemistry and wire chamber processes . over the last few decades an impressive variety of experimental data has been accumulated from laboratory tests and detectors installed at high energy physics facilities . however , there are many contradictory experiences obtained in seemingly identical conditions , which means that we do not always control all parameters that influence aging effects . it is now well established that even if a low aging rate can be obtained in the laboratory with very pure gas and otherwise clean conditions large - area detectors using the same mixture can fail due to severe aging after a relatively small beam exposure . however , experience from the laboratory , where operating conditions are much better controlled , can be used to understand some general principles and might help to implement these results successfully in large chambers . there are a lot of experiments that clearly indicate premature aging in @xmath32 mixtures exposed to intense radiation @xcite-@xcite . moreover , the aging rate in @xmath32 ( 90:10 ) was found to be mainly a function of current density , i.e. the product of irradiation rate and gas gain , independently from electrode material and purity of methane @xcite . this observation indicates that @xmath33 itself polymerizes in the avalanche plasma due to the hydrogen deficiency of radicals and their ability to make bonds with hydrocarbon molecules @xcite , and similarly for all hydrocarbon gases . under certain conditions the aging rate in @xmath34 with alcohol can be strongly reduced @xcite . however , noble gas / hydrocarbon mixtures are not trustworthy for long - term , high - rate experiments . in order to suppress polymerization of hydrocarbons , oxygen - containing molecules can be added to the mixture . for @xmath35 , measurements have shown that sensitivity to aging decreases with decreasing @xmath33 and increasing @xmath12 content @xcite . dimethylether ( dme ) appeared in the 1986 workshop as a good quencher and a reasonably good radiation - hard gas for wire chambers operated at high intensities . the aging rate in dme tends to be lower than the polymerization rate of ordinary hydrocarbons @xcite and several groups reported the absence of aging effects in wire chambers up to large values of accumulated charge @xcite . however , the aging effects in dme appear to be highly sensitive to traces of pollutants at the ppb level , which are difficult to keep under control in large detectors . there is also evidence of high chemical reactivity of dme , which requires a careful material selection for detector assembly and gas system components @xcite . attempts were made to replace organic quenchers with aging resistant ones , like @xmath12 . the @xmath36/@xmath12 gases could be in principle absolutely radiation resistant under clean conditions ; up to now , there is no well - established mechanism , which could lead to the formation of anode deposits in these mixtures . stable operation up to @xmath37 was reported for @xmath15 @xcite and up to @xmath38 for @xmath39 mixtures @xcite . however , a gradual decomposition of @xmath12 can also occur and the resulting pure carbon can be deposited specifically at the cathode @xcite . sometimes this carbon layer does not affect the performance of drift tubes @xcite . recent systematical aging tests were performed for the atlas muon aluminum drift tubes . in order to guarantee reproducibility of the results and to study aging behavior under different operating conditions , 26 tubes with @xmath15 ( 93:7)+600 @xmath29 @xmath10 mixture have been irradiated with an @xmath40 source up to an accumulated charge of @xmath41 1.3 @xmath4 and 47 tubes with @xmath15 ( 90:10 ) gas were exposed to a @xmath42 source up to @xmath41 0.6 @xmath4 @xcite . all tubes were 100 @xmath11 efficient at the end of these aging runs , however , these measurements represented an average performance of the wire over a length of 3 m and were not sensitive to local inefficiencies . it should be mentioned , that the aging performance of @xmath15 is sensitive to traces of impurities . @xmath20-based pollutants are one of the sources of aging in @xmath15 , probably due to the production of non - volatile @xmath43 @xcite . several other experiments also observed aging effects in @xmath15 , however , the reasons of the gain reduction were not identified @xcite . the identification of radicals and fragments formed in the electron avalanches is a means to understand and eventually overcome the problems related to the aging of gaseous detectors @xcite . a recent investigation of avalanche products has shown that seventeen new compounds were identified in the effluent gas stream from an irradiated proportional counter with @xmath44 ( 50:50 ) mixture @xcite . some of the observed species ( aliphatic hydrocarbons ) contained double or triple bonds which , similar to plasma polymerization , can be easily opened in the discharges and polymerize very aggressively . the systematic analysis of light emission spectra in proportional counters may also provide useful information about basic physics processes in electron avalanches @xcite . for a long time , classical wire chambers of various designs with many thousands of wires have been used for large - area tracking detectors in the ` standard radiation levels ' experiments , i.e.with total collected charges @xmath18 50 mc / cm for the whole period or running . basic rules for the construction and operation of these detectors are known . moreover , many of the large wire chambers were built and demonstrated to work @xcite . nevertheless , recent experience with large systems still shows the appearance of aging effects associated mainly with hydrocarbon polymerization and with presence of pollutants in the gas system . a time expansion chamber filled with @xmath45 ( 80:20 ) mixture successfully operated as a vertex detector of the l3 experiment at lep . after an accumulated charge of @xmath46 collected during 11 years of running , there was no sign of aging @xcite . classical aging effects ( malter effect or / and sense wire deposits ) have been observed in the h1 central jet chamber operated with @xmath34 ( 50:50 ) + 0.1 @xmath11 @xmath10 and in the zeus central tracking detector filled with @xmath47 ( 83:12:5 ) + 0.5 @xmath11 @xmath48 at the hera @xmath49-collider . in the former case , the replacement of 0.1 @xmath50 with 0.8 @xmath11 of ethanol cured the malter effect and stabilized the detector operation @xcite , while for the latter case the aging problem was alleviated by the addition of @xmath10 @xcite . in both systems there was no clear indication that polymerization is ` fed ' by the presence of impurities in the gas system , indicating that hydrocarbons are the likely source of chamber aging . an abundance of literature exists describing the dramatic effect of certain gaseous constituents , which may be either due to the contaminants initially present in the gas system or that result from outgassing of construction materials upon the aging rate of wire chambers . several examples of large systems , where the presence of pollutants increased the aging rate many times , have been reported in @xcite . while laboratory tests with prototype chamber indicated negligible aging rates ( @xmath51 ) , a much larger aging rate ( 1000 @xmath52 ) was observed in the large central tracking chamber of the cdf experiment operated with @xmath34 ( 50:50 ) + 0.1 @xmath11 alcohol . the analysis of the aged sense wires showed the presence of @xmath30 , @xmath53 and @xmath20 elements . after cleaning the gas system components and making changes to reduce aerosols emanating from an alcohol bubbler , the aging rate was greatly reduced allowing the detector to operate without dramatic loss in performance . the presence of @xmath20 is to be pointed out here since silicon has been systematically detected in analysis of many wire deposits , although in many cases the source of @xmath20-pollutant has not been clearly identified . the @xmath20-compounds are found in many lubricants , adhesives and rubber , encapsulation compounds , silicon - based grease , various oils , g-10 , rtv , o - rings , fine dust , gas impurities , polluted gas cylinders , diffusion pumps , standard flow regulators , molecular sieves , and their presence may not necessarily be noted in the manufacturer s documentation @xcite . because of a high specific polymerization rate @xmath20 molecules should be avoided in the detector system at all cost . consequently , if there is a question whether or not some device may incorporate silicon compounds it should be subjected to additional aging test . another example is the d0 wamus muon drift chambers , which suffered fast anode aging when operated with an @xmath54 ( 90:6:4 ) mixture . here , the source of the contaminant was outgassing of glass - steel polyester epoxy resin used in the construction . a cold - trap added to the gas recirculating system reduced the aging rate by a factor of ten , while the extreme method of quickly heating the wires just below their melting point ( ` zapping ' ) succeeded in blowing hardened sheaths of outgassing products off of the wires ; thus completely cleaning the aged gold wires ` in - situ ' @xcite . the chosen examples underline the importance of having control over all detector parameters , but often it is quite difficult to draw final conclusions since nominally identical detectors connected to the same gas circuit may perform very differently @xcite . in some cases it might be possible to eliminate harmful impurities by installing appropriate filters or cold traps in the gas system @xcite . many helpful guidelines for construction and operation of classical wire chambers at ` low ' rates , which have been compiled over the past 40 years , are summarized in @xcite as follows : * create a moderately clean environment during detector construction and clean the gas system components prior to start of operation ; * avoid the presence of ` bad ' molecules in contact with active gas ( @xmath20 , halogens , sulphur , plasticizers , outgassing ) @xcite ; * a huge variety of gases can be successfully used ( noble gases , hydrocarbons , freons , @xmath12 , @xmath55 , @xmath10 , alcohols,`magic gas ' , ... ) ; * hydrocarbons are the most likely source of aging ( effect is more pronounced in presence of contamination or under discharges , sparks , malter effect ) . even with addition of water / alcohol the improvements are still limited and it is problematic to consider them for high rate applications ; * if aging effects are observed despite taking all of the above precautions , add suitable additives and/or identify the source of pollution and clean the gas system . in the 1990 s , resistive plate chambers ( rpc ) were proposed as an economical and proven technology ideally suited for large - area detection systems . for example , both the belle and babar experiments have instrumented their flux returns with rpc s operated in streamer mode . however , high chamber currents started to show up in belle s rpc s almost immediately upon installation . this problem was related to the presence of high levels of water ( @xmath56 ) in the gas , which permeated through the walls of the polyethylene tubing . operating the rpc s with @xmath57 ( 30:8:62 ) plus water in streamer mode led to the formation of hydrofluoric acid that etched the electrodes made from ordinary glass . this resulted in the creation of emission points triggering chamber currents . the problem was finally solved by replacing the plastic tubes with copper ones ( @xmath58 ) @xcite . in the babar rpc s the electrodes are made of bakelite coated with linseed oil . after an initial period of successful operation with an @xmath57 mixture , the rpc s started to show a permanent reduction in efficiency and an increase in dark current . the main conclusion of the subsequent extensive @xmath59 studies related the babar rpc problem to the lack of polymerization of the linseed oil and formation of oil droplets under the influence of high temperature and high currents @xcite . further efficiency deterioration mechanisms , that may play an important role in babar s rpc s have been proposed by j. vavra @xcite . he suggested that this problem could be due to an electrochemical change of resistivity of fresh linseed oil , modulated by the presence of water in the rpc s . a positive example is the l3 rpc at lep , which operated at a very low particle flux over 8 years without significant loss of efficiency @xcite . in contrary to rpc s used in streamer mode at the belle , babar and l3 experiments , the future lhc experiments will operate rpc s in proportional mode , which is desirable in terms of total accumulated charge per particle . however , much higher particle fluxes at the lhc require more systematic @xmath59 studies of the rpc technology , since many processes could degrade their performance under high - rate conditions . recent results of aging tests for atlas , cms and lhc - b rpc s indicate that under the right circumstances rpc can withstand large integrated doses @xcite-@xcite . it should be stressed though that the problems with belle and babar rpc s are not ` classical aging effects ' , but rather unpredictable surface effects , related to the specific choices of materials and operating conditions . gaseous photon detectors used in high energy physics experiments must ensure an efficient way of converting @xmath60 photons to electrons with a subsequent detection of single photoelectrons @xcite . for gaseous converters , systematical aging studies have been carried out with @xmath61 and @xmath62 vapors @xcite-@xcite , which are added to the carrier gas to provide photoionization capability @xcite . tmae is the best material in terms of quantum efficiency , however , the main obstacle of using @xmath61 at high rates is an exceedingly rapid gas gain loss due to deposits on the anode wires @xcite-@xcite . several studies have indicated that anode wire deposits can be removed by heating the wires with elevated currents @xcite ; unfortunately this recovery is followed by a quick drop in gain @xcite . in addition , all photosensitive materials , and , most probably , their various aging products are good insulators and may excite self - sustained currents when deposited on the cathode @xcite . systematic studies with @xmath61 and @xmath62 also allowed to establish basic dependencies between the aging behavior and dissociation energy or wire diameter @xcite : * aging rate for @xmath61 is larger than for @xmath62 ( @xmath61 molecule is more fragile than @xmath62 ) ; * aging rate for @xmath61 and @xmath62 is inversely proportional to the anode wire diameter ; at low rates , the possibility to use gaseous photon detectors on a very large scale at long - term with hydrocarbon / tmae gases has been demonstrated for large 4@xmath63 devices ( e.g. sld crid and delphi rich ) @xcite . it is worthwhile to mention , that the high reactivity of tmae with oxygen and other substances necessitates a very high degree of cleanliness and leak - tightness for gas systems in these detectors . in recent years , there has been considerable work in the field of photon imaging detectors by combining solid photocathodes ( @xmath64 ) and wire chambers or gaseous electron multipliers . however , several aging tests also revealed degradation of @xmath64 photocathode quantum efficiency for very high rate environments ; a collection of existing aging data for these can be found in @xcite-@xcite . the most recent developments in high - energy physics require a dramatic increase of the radiation intensity encountered by gaseous detectors : from @xmath65 for the ` standard radiation level ' detectors up to @xmath66 for the new high - rate experiments of the lhc era ( hera - b , lhc ) . among the most critical items that affect the lifetime of gaseous detectors ( apart from the gas mixtures ) are the materials in contact with gas , assembly procedures , gas mixing and distribution systems , and tubing . in section iii.a we discuss the outgassing properties of several materials and general rules for assembly of high - rate gaseous detectors and gas systems , while sections iii.b - d contain the summary of the recent experience with aging problems in gaseous detectors operated at extremely large particle fluxes . the increasingly challenging requirements for building and testing the next generation of large - area gaseous detectors has demanded a concerted effort towards finding adequate materials for detectors and associated gas systems . many non - metallic ` good materials ' successfully used in the ` standard radiation level ' detectors might nevertheless outgas at a small level , thus causing fast aging under high rate conditions . the lifetime studies of microstrip gas chambers ( msgc ) in high intensity environments , which also had the greatest impact on the understanding of aging phenomena in all types of gaseous detectors , demonstrate that the amount of pollutants in the gas system play a major role in determining the aging properties of the detector @xcite . consequently , the outgassing from materials , epoxies , joints , tubing , etc . has to be carefully controlled . for obvious reasons , the use of glues , plastics and many organic materials is unavoidable in particle detectors . it is therefore very important to choose materials , which are suitable for the practical mechanical and electrical assembly of a gaseous detectors , in terms of their possible outgassing effects and radiation robustness . it is suggested to start by searching for low - outgassing materials in a nasa database , which was originally developed for selecting spacecraft materials @xcite . it contains more than 1600 entries for adhesives , 500 entries for rubbers and elastomers , 800 entries for potting materials . this list can help to pre - select assembly materials before doing tests matching the specific requirements of each detector . a large amount of outgassing data for epoxy compounds , adhesive tapes , leak sealers , rigid materials , o - rings , and plastic pipes have been accumulated in the framework of the rd-10 project at cern , which afterwards was merged with the more specific research on msgc within the rd-28 project . in the rd-10 tests , an outgassing box , placed upstream of the strongly irradiated wire counter , was used to introduce samples of materials into the gas stream , thus allowing a systematic study of outgassing effects on the chamber lifetime . furthermore , the gas flowing from the chamber was analyzed with a gas chromatograph ( gc ) and mass spectrometer ( ms ) or electron capture device ( ecd ) . while for some materials only outgassing properties were verified and materials releasing detectable pollutants were rejected , for other ` radiation hard materials ' full evidence of suitability was obtained in long - term aging tests of a validated clean detector . the long list of candidates recommended to be used in the construction of gaseous detectors can be found in @xcite-@xcite,@xcite-@xcite and was summarized at the 2001 workshop @xcite . the effect of outgassing from materials on the lifetime of gaseous detector can be illustrated by several examples : araldit aw103 epoxy mixed with hy991 hardener did not induce any detectable gas pollutants in the gc / ms and was also validated by irradiating the wire chamber . it is presently used as a glue for the construction of atlas straws @xcite and triple - gem stations of the compass experiment @xcite . the gc / ms analysis of another popular epoxy , araldit106 and hv953 u , extensively used in the assembly of older mwpc s , revealed traces of heavy hydrocarbon molecules in the effluent gas stream , which could be partially responsible for the observation of aging reported in @xcite . in fact , this glue has shown the largest outgassing rate among all tested glues in @xcite . interestingly , outgassing can also be due to an incorrect ratio of hardener to resin or even insufficient curing time ; both factors may largely increase the gas contamination @xcite . gas tubes used for the supply of the active gas have always been the object of primary attention when analyzing aging effects in wire chambers . electro - polished stainless steel and hydrogen - fired copper gas pipes are the best choices for gaseous detectors operated in high - rate environments , since they are free of outgassing and ensure zero gas permeability . however , due to their high price and concerns for the material budget in the active area of the detector , many experiments often use cheaper plastic tubes , although these are susceptible to outgassing , have high gas permeability and can consequently cause severe aging . particularly , pvc , teflon and neoprene rubber tubes contain halogen atoms in molecular chains , which are known to increase drastically the aging rate @xcite . polyethylene tube outgasses water , large alcanes and substituted aromatics @xcite . one of the classical examples , cited by many authors , shows that the introduction of pvc pipe can initiate a gain reduction , which continues with the same rate even after the pvc tube is replaced with the original stainless steel tubes @xcite . this indicates a potentially very serious problem : one can cause permanent damage to a detector by a wrong choice of material even for a limited period of time . therefore , one should use as much steel as possible for gas supply lines , especially in parts exposed to high radiation doses . up to now , there is no strong objection to the use of nylon ( polyamide , rilsan ) tubes if they are not too long . however , plastic pipes usually introduce water to the gas due to the natural outgassing and/or due to the diffusion of air humidity through the walls : as much as @xmath41 1700 @xmath29 of water can be added to the gas by placing 20 @xmath67 of nylon pipe at the chamber inlet @xcite . particularly , nylon tubes were used to introduce water indirectly to the chamber for curing malter breakdown @xcite . however , the presence of water can cause ` bad ' surface chemistry as described above and is therefore extremely dangerous for certain rpc and msgc detectors . the general recommendations concerning the choice of assembly materials and rules for the mechanical construction of high - rate detectors , which includes adequate assembly procedures , personnel training , quality checks , final testing as part of fighting against the aging , have been reported at the workshop in @xcite . there are clearly many ` bad ' and a lot of usable materials . however , a specific material is either adequate or not for a particular detector type and operating conditions - one has to do tests matching the specific requirements of the experiment . finally , no spontaneously - chosen materials should be installed in the detector or gas system in the last minute , before the start of real operation . future high - luminosity experiments have prompted a series of inventions of new high - rate gaseous detectors : msgc , micromegas , gas electron multipliers ( gem ) and many others @xcite . the systematic research of the physical parameters used to manufacture and operate msgc s , such as substrate and assembly materials , metal for the strips , and type and purity of the gas mixtures play a dominant role in determining their long - term stability @xcite . despite the promising performance of msgc s ( high - rate capability , good space accuracy , and excellent multi - track resolution ) , there are several major processes , particularly at high rates , leading to operating instabilities : charge - up of substrates , destructive microdischarges , and surface deposition of polymers @xcite . the influence of glass conductivity has been verified for msgc s : the use of borosilicate glass as a substrate results in rate - dependent modification of gain due to the radiation induced variation of surface resistivity . use of electron - conducting or diamond - coated glass solves the problems of short and long - term instabilities for detectors made on insulating support @xcite . the problem of microdischarges , induced by heavily ionizing particles and destroying the electrode structure , turned out to be a major limitation to all single - stage micro - pattern detectors in hadronic beams @xcite . the nature and resistivity of the strip material affects the developments of sparks @xcite . aluminum electrodes are more robust against gas discharges than gold . however , the use of @xmath68 electrodes led to the appearance of ` bubbles ' or ` craters ' on the cathode strips even at modest collected charges , while no aging effects were observed with strips made from gold @xcite . microstrip chambers have been operated with a large variety of gases ; to prevent fast aging at high rates , convincing evidence suggests again to avoid hydrocarbons in the gas @xcite . under optimal laboratory conditions , absence of any degradation of msgc performance with @xmath69 has been demonstrated by many groups up to large accumulated charges @xcite-@xcite . however , msgc - gem detectors operated with @xmath69 shows fast aging under @xmath70-rays , if the size of the irradiated area is large enough , while identical chambers with @xmath15 showed no aging @xcite . long - term survival without degradation has been also observed for triple - gem and micromegas - gem detectors , operated with @xmath15 @xcite . unfortunately , @xmath15 mixtures have worse quenching properties and are more prone to discharges than @xmath69 . protection against sparking can be significantly improved by adding a small amount ( @xmath71 ) of @xmath10 to the msgc . however , in contrary to wire chambers , such an addition of water led in one case to massive coating on the anode strips both in @xmath69 and @xmath15 mixtures @xcite . these observations underline the importance of careful selection of materials and gas mixtures for high - rate applications and of treating micro - pattern detectors as delicate devices during production and running phases . future high energy and luminosity experiments pose a new challenge for gas mixtures , raising the requirement for their radiation hardness up to @xmath37 per year . under these constraints only a limited choice of gases is available , and from the ` conventional mixtures ' only @xmath72 is demonstrated to tolerate such doses . unfortunately , these mixtures are quite transparent for photons and have a low electron drift velocity , which limits their possible application for high - rate detectors and large drift distances . about twenty years ago , @xmath5 was proposed as the most attractive candidate for high - rate environments @xcite-@xcite . this is primarily due to the high - drift velocity , high primary ionization , low electron diffusion and resistance to aging @xcite . within the broad spectrum of gas mixtures , there is no gas mixture without @xmath5 that is able to tolerate doses @xmath73 . it is believed that , when @xmath5 dissociates in the gaseous discharges into highly reactive @xmath74 and @xmath22 radicals , the atomic fluorine is very effective in suppressing polymerization effects . however , @xmath36/@xmath5 mixtures have rather poor energy and spatial resolution due to the dissociative electron attachment processes in @xmath5 @xcite- @xcite . moreover , the @xmath5 molecule has a small quenching cross - sections of metastable @xmath75-states @xcite and excited @xmath5 molecules emit photons from the far uv to the visible @xcite . this results in an intolerable level of afterpulsing in @xmath76 gases even at moderate gas gains . the advantage of the enhanced drift velocity of @xmath5 for high - rate applications have been realized by the addition of one of the common quenchers ( e.g. @xmath12 , @xmath33 ) to @xmath5 or to @xmath76 . this can also ` cool ' electrons to the extent that attachment does not occur . accelerated aging tests , often carried out with radioactive sources or @xmath70-rays to emulate the long - term lifetime properties of the detector , can be extrapolated to the real experimental conditions directly only if the aging rate depends on the total collected charge and not upon the current density , particle rate , or gas gain at which the dose was accumulated . in reality , many laboratory studies have demonstrated severe gain losses at lower charge accumulation rates , other conditions being held constant @xcite . the lower polymerization rate for higher current densities is attributed to the onset of space charge effects , which reduce the electron energies in the avalanche , thus decreasing the density of ions and radicals in the avalanche plasma . the new generation of high - rate detectors of the lhc era has not only to cope with high dose rates , but also has to survive in a hostile presence of heavily ionizing particles with an average energy deposition 10 - 1000 times larger than for mip s . an exposure of ` large - scale ' gaseous detectors over the full area in such a harsh radiation environment at high ionization densities ( @xmath77 ) can result in greatly enhanced polymer formation : an abundance of aggresive radicals will diffuse for rather long times ( @xmath41 hours ) within the irradiated chamber and will react with other avalanche - produced polymer fragments . according to this naive picture , this mechanism could significantly accelerate polymerization in large systems , whereas in small - scale laboratory tests the aging rate typically decreases with increasing gas flow @xcite . furthermore , polymer deposition in wire chambers likely starts from localized spots and then can spread over the entire irradiated region . since in large , mass - produced systems an extremely high quality for electrode surfaces and cleanliness of the gas systems are hard to reach , any imperfections in the presence of high currents could easily trigger sparks , discharges or malter currents , which will in turn dramatically increase the polymerization rate . recent systematic research clearly demonstrates that the initial stage of radiation tests usually performed in the laboratory may not offer the full information needed to estimate the lifetime of the real detector . strong dependence of aging performance upon * size of the irradiated area , current density , high voltage ( gas gain ) , irradiation rate , particle type and energy * have been observed in high - rate environments . severe anode and cathode aging effects were found in prototype honeycomb drift tubes operated with @xmath78 ( 80:20 ) and @xmath79 ( 74:20:6 ) mixtures in the high - rate hera - b environment ( secondaries from interactions of 920 gev proton with target nucleus ) after a few @xmath80 of accumulated charge @xcite . this effect was surprising since chambers had previously been proven to be immune to very large @xmath70-ray doses up to 5 @xmath81 . an extensive @xmath59 program , carried out at 10 different radiation facilities to resemble hera - b conditions , revealed that @xmath70-rays or electrons were not able to trigger malter currents , while in the large - area modules , irradiated with hadrons above a certain energy , malter effect appeared very rapidly . the aging effects in these honeycomb drift tubes were traced to a combination of several problems ; a solution which uses gold - coated cathode foils and a @xmath54 ( 65:30:5 ) mixture was found to survive in a high - rate hadronic environment up to @xmath82 @xcite . the aging performance of the hera - b muon proportional chambers has been studied with @xmath79 , @xmath78 , and @xmath54 mixtures in a variety of conditions @xcite-@xcite . the aging rate in @xmath79 was found to be more than two orders of magnitude higher in hadronic beams than in the laboratory studies with radioactive sources . in addition , strong dependences of the aging properties on high voltage and progressive deterioration of the gas gain in the direction of the serial gas flow have been observed for large - scale prototypes irradiated with @xmath79 ( 67:30:3 ) in the harsh hera - b environment . aging effects increasing in the direction of the serial gas flow ( even outside the irradiation zone ) have been also reported for a @xmath83 mixture @xcite . a strong dependence of the aging properties on high voltage , irradiation rate , length of irradiation and gas flow rate has been also observed in the atlas muon drift tubes operated with @xmath84 ( 94:3:2:1 ) @xcite . here , the systematic @xmath59 studies have shown a nearly exponential dependence of chamber lifetime on high voltage and on the counting rate within the experimentally investigated parameter range . unfortunately , the high voltage is not the physical quantity directly responsible for aging in wire chambers , therefore , these aging effects could be classified as depending on the gas gain and/or current density , which are related to the density of ions and radicals in the avalanche plasma . dependence of polymer formation on the energy input is well established in plasma polymerization . nearly all organic compounds regardless of their chemical nature can be polymerized under certain conditions . the structure of plasma polymers formed from the same monomer is highly dependent on the actual conditions of polymerization : the energy input level , the size ( cross - sectional area ) of a tube and even on the position within the reactor @xcite . the experimental dependence of chamber lifetime on ionization density , gas gain and irradiation rate , which are also related to the total dissipated energy in the detector from ionizing particles , indicates that the aging behavior can not be solely explained on the basis of the molecule ratios in the mixture ( gas composition ) , without taking into account the actual operating conditions . these results illustrate the need for studying the aging performance of a detector under conditions as close as possible to the real environment . the dependence of the detector lifetime on the size of the irradiated area , in particular , and the increase of the aging rate in the direction of the serial gas flow means that aging should be viewed as a non - local and intensity - dependent phenomenon . these observations seem to be the most critical when trying to extrapolate the aging behavior from laboratory tests to large - scale detectors . some of the long - lived aggressive radicals may diffuse in the direction of the gas flow and react with other avalanche generated contaminants , thus enhancing aging effects with increasing usage of the gas . here it is important to note that due to the increased aging in the direction of the gas flow it is worthwhile to avoid gas distribution systems that supply many chambers by a serial flow . ( 100,40 ) ( -5.0,-42.0 ) ( 85.0,-42.0 ) as reported above , polymer deposition can occur in @xmath78 mixtures , as suggested by plasma chemistry . similarly , the aging properties of @xmath85 gases , which by analogy with plasma experiments should have excellent etching properties , have been widely investigated over the last years . using @xmath54 mixtures under optimal operating conditions , no observable drop in gain due to polymerization has been found for the hera - b honeycomb drift tubes up to 1.5 @xmath81 @xcite , hera - b aluminum proportional chambers up to 0.7 @xmath81 @xcite , cms cathode strip chambers up to 0.4 @xmath81 @xcite , 13 @xmath81 @xcite , lhcb multi - wire proportional chambers up to 0.25 @xmath81 @xcite , compass straw tubes up to 1.1 @xmath81 @xcite and hermes drift tubes up to 9 @xmath81 @xcite . moreover , honeycomb drift tubes which were initially aged with @xmath79 were afterwards successfully recovered in @xmath54 @xcite . however , an analysis of the cathode surfaces at the end of operation with @xmath54 revealed in some cases the presence of fluorine - based deposits on the cathodes , which fortunately did not result in self - sustained currents @xcite . since dissociative products of @xmath5 react violently with many materials and the resultant polymer films at cathodes could provoke aging effects one should seriously consider using materials in high - rate detectors which are very robust to @xmath5 ; gold - plated electrodes @xcite or straw cathode materials @xcite fulfill this requirement . a further advantage of @xmath5-based mixtures is additional resistance against @xmath20-polymerization . this suggestion is based on experiences from plasma polymerization , where discharges of @xmath86 are successfully used for @xmath87 etching , while @xmath88 plasmas selectively etches @xmath20 @xcite . extensive studies performed for the atlas straws partially confirm this hypothesis : @xmath20-deposits have not been observed in irradiation area for large current densities ( @xmath89 ) . on the other hand , @xmath90-deposits were found at the edges and even outside of the irradiated area . the resulting balance between @xmath20 polymerization and @xmath5 etching processes was found to be very sensitive to the @xmath20-source intensity and ionization density @xcite . in the most recent investigations at extremely high current densities ( @xmath91/@xmath92 ) in @xmath85 mixtures a new ` aging phenomenon ' has appeared - the damage of the gold - plating of wires in straws and honeycomb drift tubes . several years of intensive research at cern with straw tubes under different conditions and for different types of wires indicate that the main components responsible for gold wire damage in @xmath93 ( 70:20:10 ) are harmful radicals , products of @xmath5 disintegration , in connection with @xmath10 . ( it was suggested that @xmath23 acid could be responsible for destruction of gold - plating and formation of @xmath94 deposits ) . no wire damage effects have been observed for water concentrations below 0.1 @xmath11 up to 20 @xmath81 ! @xcite . dedicated studies with straw tubes performed at pnpi with @xmath93 ( 70:20:10 ) and @xmath54 ( 60:30:10 ) mixtures demonstrated that under high dose rates the gold - plating of the wire was cracked , the wire diameter increased and a large amount of oxygen was observed on the tungsten in gold cracks @xcite-@xcite . similar effects have been observed for wires irradiated in a @xmath83 mixture @xcite . the authors propose a model to explain the results - an anode wire ` swelling ' mechanism , where the forces causing the damage to the wire surface develop under the gold layer of the wire @xcite . [ damage ] shows examples of a variety of wires with damaged gold - plating from @xcite . in contrary to the experience with straws , in one test with honeycomb drift tubes irradiated with @xmath54 ( 65:30:5 ) the destruction of gold coating and even rupture of anode wires have been observed only for water concentration below 50 @xmath29 , while for @xmath95 , gold wire damage effects were avoided @xcite . further studies still remain to be done to fully understand the exact mechanism of gold wire damage during operation at high ionization densities . although no @xmath22-based deposits were observed on the anode wires in any of the tests , the chemically reactive dissociative products of @xmath5 most probably initiate the destruction processes of gold - plating . studies of straw proportional tubes with @xmath93 mixture revealed another phenomenon , which might degrade detector performance in high rate experiments . the gas composition was found to be modified in the avalanche plasma of a strongly irradiated straw , presumably due to the production of some long - lived and highly electro - negative species @xcite . these electro - negative radicals could be also responsible for the so - called ` transient aging effect ' observed at high - rates in @xmath93 @xcite , @xmath54 @xcite and @xmath79 @xcite . a ` transient aging effect ' is a temporary gain reduction , which can be restored by an appropriate increase of the gas flow . the very high aggressiveness of dissociative products of @xmath5 and the dynamic modification of the gas composition requires more detailed studies to evaluate the possible consequence of these effects on the long - term performance and stability of large - area gaseous detectors . in view of the aging results described here , one can see that the presence of large amounts of @xmath5 in the mixture does not necessarily ensure good aging properties automatically . the challenge to avoid aging in the new generation of high - rate experiments requires not only a very careful choice of all detector materials , but also forces the gas systems to be of previously unnecessary quality and cleanliness . however , a real system will always contain some degree of imperfection and pollution despite all precautions . it has to be stressed here that in closed - loop recirculation systems , which are required for detectors operated with expensive gases ( @xmath96 ) all impurities and reactive radical fragments will remain in the gas until they are removed by a purification system or deposited elsewhere . therefore , for the construction of large - area gaseous detectors the maximal cleanliness for all processes and quality checks for all system parts are of primary importance . examples of ` clean ' gas systems currently used for high - rate detectors are presented in @xcite . certainly , the definition of the word ` clean ' has changed considerably since the 1986 workshop . aging phenomena obviously constitute one of the most complex and serious problems which could limit the use of gaseous detectors in unprecedently severe radiation environments . the operation in high - intensity experiments of the lhc era demand not only an extraordinary radiation hardness of construction materials and gas mixtures , but also appropriate assembly procedures and quality checks during detector construction and testing . since the 1986 workshop , considerable progress has been made on the understanding of general principles which might help to prevent or at least to suppress the aging rate to an acceptable level . however , a quantitative description of the aging processes , which would require a detailed knowledge of the reaction cross sections of all chemical species in the avalanche plasma , is still not available . after the many years of intensive research and development of radiation - hard gaseous detectors , an impressive variety of experimental data has been accumulated . the radiation hardness and outgassing properties of the various materials used for the construction of detectors and gas systems are among the most crucial items affecting the lifetime of gaseous detector . however , the observed dependences of aging performance on the nature and purity of the gas mixture , different additives and trace contaminants , construction materials , gas flow , size of the irradiated area , irradiation intensity , ionization density , high voltage , particle type and energy , make quantitative comparisons of aging properties under very different conditions very difficult . consequently , this data can serve only as a guideline before the start of long - term studies under conditions as close as possible to the real environment of the experiment . such radiation tests should include an extended study of ` large - scale ' final prototype chambers , exposed over the full area to a realistic radiation profile ( particle type and energy , ionization density , irradiation rate ) . it is of primary importance to vary the operating parameters systematically in order to investigate their possible influence on the aging performance . in order to exclude statistical fluctuations of unknown nature and to provide a reliable estimate for the detector lifetime , the radiation tests should be carried out with several detectors irradiated under identical conditions . this paper is based on the results reported at the desy workshop , and also briefly discusses other experience with gaseous detectors relevant to the present aging problems . transparencies and videos of presentations from the ` international workshop on aging phenomena in gaseous detectors ' ( desy , hamburg ) are available at the workshop s web - page ( http://www.desy.de/agingworkshop ) . the proceedings of the workshop will be published in a special volume of nuclear instruments and methods : section a. we are especially grateful to andreas schwarz ( desy ) , jaroslav vavra ( slac ) and mar capeans ( cern ) for many helpful discussions and careful reading of this manuscript . r. openshaw , r. henderson , w. faszer , d. murphy , m. salomon , g.sheffer , `` tests of wire chamber ageing with @xmath5/isobutane ( 80:20 ) , @xmath34 ( 50:50 ) and @xmath98 ( 48:48:4 ) '' _ ieee trans . 36(1 ) , pp . 567 - 571 , 1989 . _ m.j . kushner , `` a kinetic study of the plasma - etching processes . a model for the etching of @xmath20 and @xmath43 in @xmath99 and @xmath100 plasmas , '' _ 53(4 ) , pp . 2923 - 2938 , 1982 . _ e.a . truesdale , g. smolinsky , t.m . mayer , `` the effect of added acetylene on the rf discharge chemistry of @xmath101 . a mechanistic model for fluorocarbon plasmas , '' _ j. appl . 51(5 ) , pp . 2909 - 2913 , 1980 . _ j.p . dewulf , d. geiregat , p. vilain , f. bergsma , c. busi , a. capone , @xmath97 , `` test results of the streamer - tube system of the charm ii neutrino detector , '' _ nucl . instr . meth . a252 , pp . 443 - 449 , 1986 . _ k. silander , e.p . de lima , m.m . fraga , r.f . marcus , f. fraga , m. salete leite , a.j.p.l . policarpo , `` aging studies with argon / methane based gas mixtures , '' _ nucl . instr . meth . a367 , pp . 298 - 301 , 1995 . _ r. bouclier , m.capeans , c. garabatos , r.d . heuer , m. jeanrenaud , t.c . meyer , f. sauli , k. silander , `` results of wire chamber aging tests with @xmath33- and dme - based gas mixtures , '' _ nucl . instr . meth . a346 , pp . 114 - 119 , 1994 . _ m. jibaly , s. majewski , p. chrusch , r. wojcik , f. sauli , j. gaudaen , `` the aging of wire chambers filled with dimethyl ether : wire and construction materials and freon impurities , '' _ nucl . instr . meth . a283 , pp . 692 - 701 , 1989 . _ g. gavrilov , `` aging studies of cms muon chambers prototypes under high doses '' _ presented at the international workshop on aging phenomena , desy , hamburg , oct . 2 - 5 , 2001 ; and preprint pnpi-2442 ( 2001 ) . _ t. kowalski , `` a study of aging effect in gas monitoring proportional counters of the bac calorimeter of the zeus experiment , '' _ presented at the international workshop on aging phenomena , desy , hamburg , oct . 2 - 5 , 2001 . _ k. kurvinen , `` analysis of organic compounds formed in electron avalanches in a proportional counter filled with @xmath103 , '' _ presented at the international workshop on aging phenomena , desy , hamburg , oct . 2 - 5 , 2001 . sumner , g.k . rochester , p.d . smith , j.p . cooch , r.k . sood , `` scintillating drift chambers - the nature of the emission process in @xmath32 , '' _ ieee trans . 29(5 ) , pp . 1410 - 1414 , 1982 . _ t. marshall , `` restoring contaminated wires , removing gas contaminants and aging studies of drift tube chambers , '' _ presented at the international workshop on aging phenomena , desy , hamburg , oct . 2 - 5 , 2001 . _ j. pyrlik , m. atiya , d. broemmelsiek , th . hamacher , m. ispiryan , s. korpar , @xmath97 , `` aging measurements of tmae based photon detector for the hera - b rich , '' _ nucl . instr . meth . a414 , pp . 170 - 181 , 1998 . _ j. vavra , t. bienz , f. bird , m. gaillard , g. hallewell , y. kwon , @xmath97 , `` construction and initial operation of a proportional wire detector for use in a cherenkov ring imaging system , '' _ ieee trans . 35(1 ) , pp . 487 - 492 , 1988 . _ a. barr , s. bachmann , b. boimska , r. bouclier , a. braem , c. camps , @xmath97 , `` construction , test and operation in a high intensity beam of a small system of micro - strip gas chambers , '' _ nucl . instr . meth . a403 , pp . 31 - 56 , 1998 . _ y. bagaturia , o. baruth , h. dreis , f. eisele , i . gorbunov , w. gradl , @xmath97 , `` studies of aging and hv breakdown problems during development and operation of msgc and gem detectors for the inner tracking system of hera - b , '' _ submitted for publication to nucl . instr . meth . , _ r. bouclier , j. florent , j. gaudaen , g. million , a. pasta , f. sauli , @xmath97 , `` high flux operation of microstrip gas chambers on glass and plastic supports , '' _ nucl . instr . meth . a323 , pp . 240 - 246 , 1992 . _ l. alunni , r. bouclier , g. fara , c. garabatos , g. manzin , g. million , @xmath97 , `` performance of msgc on electronically and ionically conductive substrata in various operational conditions , '' _ nucl . instr . meth . a348 , pp . 344 - 350 , 1994 . _ r. bouclier , m. capeans , c. garabatos , g. manzin , g. million , l. ropelewski , @xmath97 , `` development of micro - strip gas chambers for high - rate operation , '' _ nucl . a367 , pp . 168 - 172 , 1995 . _ j. fischer , a. hrisoho , v. radeka , p. rehak , `` proportional chambers for very high counting rates based on gas mixtures of @xmath5 with hydrocarbons , '' _ nucl . a238 , pp . 249 - 264 , 1985 . _ anderson , j.c . armitage , e. dunn , j.g . heinrich , c. lu , k.t . mcdonald , @xmath97 , `` electron attachment , effective ionization coefficient , and electron drift velocity for @xmath5 gas mixtures , '' _ nucl . instr . meth . , vol . a323 , pp . 273 - 279 , 1992 . _ m. danilov , l. laptin , i. tichomirov , m. titov , yu . zaitsev , `` aging tests of the proportional wire chambers using @xmath79 ( 74:20:6 ) , @xmath79 ( 67:30:3 ) and @xmath54 ( 65:30:5 ) mixtures for the hera - b muon detector , '' _ hep - ex/0107080 , preprint itep-43 - 00 , 2000 . _ m. danilov , yu . gilitsky , t. kvaratschellia , l. laptin , i. tichomirov , m. titov , yu . zaitsev , `` aging studies of large area proportional chambers under high rate irradiation with @xmath5-based mixtures , '' _ hep - ex/0111077 and hep - ex/0111078 , preprint itep-15 - 01 , 2001 . _ g. gavrilov , a. krivchitch , e. kuznetsova , v. lebedev , l. schipunov , e. lobachev , `` aging properties of the straw drift - tubes operating with gas mixture containing @xmath27 , '' _ preprint pnpi-2407 , 2001 . _ t. ferguson , g. gavrilov , a. egorov , a. krivchitch , e. kuznetsova , v. lebedev , @xmath97 , `` anode wire swelling - a possible phenomenon in anode wire aging under high - accumulated doses , '' _ preprint pnpi-2406 , 2001 . _

the general low - energy theory of electrons interacting via repulsive short - range interactions on graphene s honeycomb lattice at half filling is presented . the exact symmetry of the lagrangian with local quartic terms for the dirac four - component field dictated by the lattice is identified astime reversal , where is the dihedral group , and is a subgroup of the `` chiral '' group of the non - interacting lagrangian , that represents translations in dirac language . the lagrangian describing spinless particles respecting this symmetry is parameterized by six independent coupling constants . we show how first imposing the rotational , then lorentz , and finally chiral symmetry to the quartic terms - in conjunction with the fierz transformations - eventually reduces the set of couplings to just two , in the `` maximally symmetric '' local interacting theory . we identify the two critical points in such a lorentz and chirally symmetric theory as describing metal - insulator transitions into the states with either time - reversal or chiral symmetry being broken . the latter is proposed to govern the continuous transition in both the thirring and nambu - jona - lasinio models in 2 + 1 dimensions and with a single dirac field . in the site - localized , `` atomic '' , limit of the interacting hamiltonian , under the assumption of emergent lorentz invariance , the low - energy theory describes the continuous transitions into the insulator with either a finite haldane s ( circulating currents ) or semenoff s ( staggered density ) masses , both in the universality class of the gross - neveu model . the simple picture of the metal - insulator transition on a honeycomb lattice emerges at which the residue of the quasiparticle pole at the metallic , and the mass - gap in the insulating phase both vanish continuously as the critical point is approached . in contrast to these two critical quantities , we argue that the fermi velocity is non - critical as a consequence of the dynamical exponent being fixed to unity by the emergent lorentz invariance near criticality . possible effects of the long - range coulomb interaction , and the critical behavior of the specific heat and conductivity are discussed .

0811.0610

two - dimensional honeycomb lattice of carbon atoms may be viewed as the mother of all other forms of carbon . its crucial electronic property , which arises as a consequence of the absence of the inversion symmetry around the lattice site , is that the usual fermi surface is reduced to just two points . the electronic dispersion may be linearized around these two points , after which it becomes isotropic and dependent on the single dimensionful parameter , fermi velocity @xmath4 . the pseudo - relativistic nature of the electronic motion in graphene has since its synthesis placed this material at the center stage of condensed matter physics . many qualitatively novel phenomena that take , or may take place in such a system of `` dirac '' electrons are actively discussed in the rapidly growing literature on the subject . @xcite in this paper we discuss the low - energy theory and the metal - insulator quantum phase transitions of the _ interacting _ dirac electrons on the honeycomb lattice , building upon and expanding significantly the earlier work by one of us . @xcite in the first approximation , _ all _ weak interactions of dirac electrons in graphene may be neglected at half filling , when the fermi surface consists of the dirac points . this is because short - range interactions are represented by local terms which are quartic in the electron fields , which makes them irrelevant near the non - interacting fixed point by power counting . the same conclusion turns out to apply to the long - range tail of the coulomb interaction , which remains unscreened in graphene , although only marginally so . @xcite nevertheless , if strong enough , the same interactions would turn graphene into a gapped mott insulator . as an example , at a strong on - site repulsion the system is likely to be the usual nel antiferromagnet . @xcite it is not _ a priori _ clear on which side of this metal - insulator transition graphene should be . with the standard estimate for the nearest - neighbor hopping in graphene of @xmath5 and the hubbard interaction of @xmath6 , it seems that the system is below yet not too far from the critical point estimated to be at @xmath7 . @xcite if sufficiently weak , the electron - electron interactions only provide corrections to scaling of various quantities , which ultimately vanish at low temperatures or frequencies . at , what is probably a more realistic , an intermediate strength , the flow of interactions and the concomitant low - energy behavior may be influenced by the existence of metal - insulator critical points . it is possible that some of the consequences of such interaction - dominated physics have already been observed in the quantization of the hall conductance at filling factors zero and one . @xcite as we argued elsewhere , the anomalously large value of the minimal conductivity in graphene @xcite may be yet another consequence of the coulomb repulsion between electrons . @xcite the above discussion raises some basic questions . what is the minimal description of interacting electrons in graphene at `` low '' energies ? what is the symmetry of the continuum interacting theory , and how does it constrain the number of coupling constants ? what kinds of order may be expected at strong coupling , and what is the nature of the metal - insulator quantum phase transition ? in this paper we address these and related issues . in the rest of the introduction we give a preview of our main results . the simplest prototypical system that exhibits the physics of interacting dirac fermions which we seek to understand is the collection of spinless electrons interacting via short - range interactions , at half - filling . for present purposes an interaction may be considered as `` short - ranged '' if its fourier transform at the vanishing wavevector is finite . @xcite the least irrelevant quartic terms one can add to the non - interacting dirac lagrangian will then be local in space - time , and of course quartic in terms of the four - component dirac fields that describe the electronic modes near the two inequivalent dirac points at wavevectors @xmath8 at the edges of the brillouin zone . the most general local quartic term in the lagrangian would be of the form @xmath9 where @xmath10 and @xmath11 are four - dimensional hermitian matrices . the symmetry alone , however , immediately drastically reduces the number of independent couplings from the apparent @xmath12 to just fifteen . although the point group of the honeycomb lattice is @xmath13 , the exact spatial discrete symmetry of the lagrangian is only the _ dihedral group _ @xmath1 , or the _ vierergruppe _ , which consists of the reflections through the two coordinate axis shown in fig . 1 , and the inversion through the origin . such a small symmetry results from the very choice of two inequivalent dirac points out of six corners of the brillouin zone , which reduces the symmetry to the simple exchange of the two sublattices ( reflection around a axis ) , the exchange of dirac points ( reflection around b axis ) , and their product ( the inversion through the origin ) . @xmath1 , the time - reversal , and the translational invariance are shown to leave fifteen possible different local quartic terms in the lagrangian . fortunately , not all of these still numerous quartic terms are independent , and there are linear constraints between them implied by the algebraic fierz identities . @xcite the fierz transformations are rewritings of a given quartic term in terms of others , and we provide the general formalism for determining the number and the type of independent quartic couplings of a given symmetry . for the case at hand we find that spinless electrons interacting with short - range interactions on honeycomb lattice are in fact described by only _ six _ independent local quartic terms . the inclusion of electron spin would double this number to twelve . the linearized noninteracting lagrangian for dirac electrons , @xmath14 as well - known , exhibits the lorentz and the global @xmath3 ( `` chiral '' ) symmetry . the latter , generated by @xmath15 , with @xmath16 , is nothing but the `` rotation '' of the `` pseudospin '' , or `` valley '' , corresponding to two inequivalent dirac points . @xcite a general quartic term allowed by the lattice symmetry , on the other hand , has a much smaller symmetry , as already mentioned . nevertheless , we will argue that near the metal - insulator quantum critical points , all , or nearly all of the larger symmetry possessed by the non - interacting part of the lagrangian gets restored . this conclusion is supported by the , admittedly uncontrolled , but nevertheless quite informative one - loop calculation . first , we find three distinct critical points in the theory , all of which have not only the rotational , but the full lorentz - symmetric form . this immediately implies that the dynamical critical exponent is always @xmath17 . this is quite remarkable in light of the fact that the microscopic theory is not even rotationally invariant , and that the critical points in question are purely short - ranged . @xcite the fact that @xmath17 has important implications for several key physical observables near the critical point , as we discuss shortly . furthermore , we find that two out of three critical points in the theory exhibit a full chiral symmetry as well . we identify the three fixed points in the theory as corresponding to three possible order parameters , or `` masses '' that develop in the insulating phase at strong coupling . \1 ) @xmath18 , which preserves chiral , but breaks time - reversal symmetry . microscopically , this order parameter may be understood as a specific pattern of circulating currents , as discussed in the past . @xcite \2 ) @xmath19 , which preserves the time - reversal symmetry , and the single chiral generator @xmath20 , which will be shown to correspond to translational invariance . this order parameter describes a finite staggered density , i. e. the difference between the average densities on the two sublattices of the honeycomb lattice . @xcite \3 ) @xmath21 , which preserves the time - reversal , but breaks translational invariance ( @xmath20 ) . this order parameter can be understood as the specific `` kekule '' modulation of the nearest - neighbor hopping integrals . @xcite in one - loop calculation all three critical points have the same correlation length exponent @xmath22 , which we believe is an artifact of the quadratic approximation . the result that the dynamical critical exponent @xmath17 is , on the other hand , possibly exact . if we denote the relevant interaction parameter with @xmath23 , the fermi velocity near the transition scales as @xmath24 so the above value of @xmath25 would simply imply that it stays regular at the transition . this appears to be in agreement with the picture of the transition as the opening of the relativistic `` mass '' in the spectrum . the mass - gap in the insulating phase scales as usual @xcite as @xmath26 the transition on the metallic side is manifested as vanishing of the residue of the quasiparticle pole @xcite @xmath27 where we assumed @xmath17 . ( a more general power - law is discussed in the text . ) at one - loop the fermion anomalous dimension @xmath28 vanishes , but in general it is a positive , small , and critical - point - dependent number . the overall picture of the metal - insulator transition that emerges is presented in fig . 2 . for graphene s @xmath29-orbitals well localized on carbon sites , a further significant simplification takes place . all the terms without the equal number of creation and annihilation operators for each of the two sublattices must vanish . assuming again the emergent lorentz symmetry at low energies this allows one to finally write the simplest internally consistent interacting theory as @xmath30 this lagrangian provides the minimal low - energy description of interacting spinless electrons on honeycomb lattice . it has two critical points , corresponding to transitions into insulators 1 ) and 2 ) in the above , both corresponding to the gross - neveu criticality in 2 + 1 dimensions . we discuss the internal consistency and the sufficiency of this lagrangian and some of the peculiarities of the ensuing phase diagram . the rest of the paper is organized as follows . we discuss the point symmetry , translational symmetry , and the time - reversal symmetry of the interacting lagrangian as dictated by the microscopic hamiltonian for the system in the next section . in section iii it is shown how further enlargements of the symmetry would reduce the number of coupling constants . we introduce the notion of `` maximally symmetric '' theory , which shares the full lorentz and chiral symmetry with the quadratic term in the lagrangian . the general formalism of fierz transformations is developed and applied to the cases of interest in section iv . the change of the coupling constants with the ultraviolet cutoff in the theory is studied in section v. the atomic limit of the general interacting theory is described in section vi , and the critical exponents are discussed in section vii . in section viii we discuss the scaling of the electron propagator and the power - laws for various quantities of interest . the discussion of the long - range coulomb interaction and the critical behavior of the specific heat and the optical conductivity are given in section ix , and the summary in section x. finally , in appendixes we present some of the requisite technical details : the fierz transformation , the spectral form of the asymmetric matrix needed in section iv , and an alternative implementation of the renormalization group in presence of linear constraints . as the simplest microscopic model that contains the relevant physics we may consider the tight - binding hamiltonian on the graphene s honeycomb lattice , defined as @xmath31 where @xmath32 and @xmath33 are the electron annihilation operators at two triangular sublattices of the honeycomb lattice . here , @xmath34 denotes sites of the sublattice generated by linear combinations of basis vectors @xmath35 , @xmath36 , whereas @xmath37 are the sites on the second sublattice , with @xmath38 being @xmath39 , or @xmath40 , and @xmath41 is the lattice spacing . within the framework of the tight - binding model the energy spectrum is doubly degenerate @xmath42|$ ] , and becomes linear and isotropic in the vicinity of six dirac points , at the edge of the brillouin zone , among which only two , hereafter chosen to be at @xmath43 with @xmath44 , are inequivalent . retaining only the fourier components in the vicinity of these two inequivalent points , the quantum mechanical action corresponding to @xmath45 at low energies can be written in the form @xmath46 , with the free lagrangian density @xmath47 defined as in eq . ( 2 ) , with @xmath48 as the imaginary time and @xmath49 is the temperature . matrices @xmath50 satisfy the clifford algebra @xmath51 , @xmath52 , @xmath53 . the summation over repeated space - time indices is assumed hereafter . the fermionic field @xmath54 is defined as @xmath55 here , the reference frame is conveniently rotated so that @xmath56 , @xmath57 , @xmath58 are the fermionic matsubara frequencies , @xmath59 is a high - energy cutoff , and we set @xmath60 , where @xmath61 is the fermi velocity . choosing @xmath62 implies @xmath63 the two remaining anticommuting matrices may then be taken as @xmath64 this defines the `` graphene representation '' of @xmath65-matrices . @xcite @xmath66 are the standard pauli matrices . considering a more general model with further hoppings or weak anisotropies @xcite can be seen not to destroy the dirac points , but only to shift them in energy . as this can always be compensated by a shift of the chemical potential , the lagrangian ( 2 ) provides the low - energy description of the general free electronic hamiltonian on a honeycomb lattice , with the chemical potential tuned to the dirac point . note that the free lagrangian besides the lorentz symmetrt also possesses another chiral , pseudospin or `` valley '' , global @xmath3 symmetry , generated by @xmath15 . both the lorentz and the chiral symmetry of the free lagrangian are emerging only at low energies , however , and the term quadratic in derivatives in @xmath47 , for example , would spoil it . as will be shown shortly , both symmetries are also violated by the leading irrelevant quartic couplings introduced by the interactions . let us now consider the electron - electron interactions . the hamiltonian of a general four - fermion interaction has the form @xmath67 where @xmath68 or @xmath33 are fermionic annihilation operators , and the matrix element corresponding to the interaction potential @xmath69 is given by @xmath70 here we can take @xmath71 to be a localized @xmath29-orbital on the site @xmath72 , so that it belongs to either one of the two sublattices of the honeycomb lattice . in general , there is no restriction on the overlap of the wave - functions , and all the matrix elements @xmath73 are in principle finite . their relative sizes , however , may be rather different , and we discuss important simplifications that follow in the limit of well localized orbitals in sec . vi . in the following we will consider general `` short - ranged '' interactions , which are defined by the interaction @xmath74 with a regular fourier component @xmath75 . without a loss in generality one may then take the interacting lagrangian for spinless fermions corresponding to @xmath76 as in eq . ( 1 ) where @xmath10 and @xmath11 are some constant @xmath77 hermitian matrices . the interacting lagrangian contains therefore at most @xmath78 independent real coupling constants . however , the number of couplings in @xmath79 is severely reduced by the lattice symmetries , as we discuss next . two obvious discreet symmetries of the honeycomb lattice that have not been broken by our choice of the dirac points are the reflection symmetries through the lines a and b in fig . 1 . let us consider the former symmetry first . it exchanges the two sublattices , but not the two dirac points . the low - energy lagrangian thus has to be invariant under the exchange of the spinor components belonging to different sublattices , @xmath80 . consequently , the symmetry operator acting on the four - component dirac spinor defined in eq . ( [ dirac - spinor ] ) has the form @xmath81 since under this reflection @xmath82 and @xmath83 , @xmath47 is evidently invariant under @xmath84 . the invariance of @xmath79 under this reflection symmetry requires both matrices @xmath10 and @xmath11 to either commute or anticommute with the operator @xmath84 : @xmath85=[s , m_2]=0,\ ] ] or @xmath86 similarly , the reflection symmetry through the line @xmath87 exchanges the two dirac points , while not exchanging the sublattice labels . it corresponds therefore to @xmath88 recalling that under this transformation @xmath89 and @xmath90 , it is evident that the free lagrangian in graphene representation is indeed invariant under @xmath49 as well . demanding the interacting lagrangian @xmath79 to be invariant under the action of the operator @xmath49 on the dirac spinor , implies that both matrices @xmath10 and @xmath11 either commute or anticommute with @xmath49 as well . in other words , both matrices @xmath10 and @xmath11 have to be either even or odd with respect to @xmath49 : @xmath91=[t , m_2]=0,\ ] ] or @xmath92 together with the combination of the two reflections s and t , and the identity operation , the two symmetry operations form the _ dihedral group _ ( or klein s _ vierergrouppe _ , in older literature ) @xmath1 : @xmath93 , the symmetry group of a rectangle . note that the transformation @xmath94 is just the space inversion , and that the rotation by @xmath95 _ does not _ belong to @xmath1 . one may now classify all the four - dimensional matrices into four categories , according to their transformation under the two reflection symmetries @xmath84 and @xmath49 , respectively : even - even , @xmath96 , even - odd , @xmath97 , odd - even , @xmath98 , and odd - odd , @xmath99 . the interacting lagrangian symmetric under the @xmath1 is thus restricted to be of the following form @xmath100 where @xmath101 , @xmath102 , @xmath103 , are real and symmetric . the maximal number of independent real parameters specifying the allowed couplings is thus already reduced to forty , since each @xmath101 has ten independent components . the generator @xmath104 of the chiral symmetry plays a special role . it is in fact the generator of translations . to see this recall that under a translation by @xmath105 the electron fields transform as @xmath106 where @xmath107 . the dirac field under the same transformation thus changes as @xmath108 or , in real space , @xmath109 translational invariance requires therefore that @xmath79 is a scalar under the transformations generated by @xmath20 , which we will denote as @xmath110 . it is easy to see that this is precisely the same as demanding the conservation of momentum in the interaction terms . the reader is also invited to convince herself that the terms with the higher - order derivatives in @xmath47 would also be invariant under the @xmath110 . first , we observe that there are eight linearly independent bilinears that are _ scalars _ under the @xmath110 : @xmath111 where @xmath112 and @xmath113 . the remaining eight bilinears can be grouped into four _ vectors _ under the same @xmath110 : @xmath114 @xmath115 @xmath116 @xmath117 the invariance under @xmath110 implies therefore that the interacting lagrangian has the following form @xmath118 the number of possible independent couplings is down to eighteen . the set of the allowed couplings is further reduced by the time - reversal symmetry . the microscopic interacting hamiltonian ( [ hint ] ) is invariant under the time - reversal , and therefore the corresponding low - energy lagrangian has to possess the same invariance . the time - reversal symmetry requires that @xmath119 , where @xmath120 is the antiunitary operator representing the time - reversal symmetry , and thus has the form @xmath121 , with @xmath122 representing the unitary part of @xmath120 and @xmath123 is the complex conjugation . to find the form of @xmath120 let us consider first the _ massive _ dirac hamiltonian @xmath124 with the mass @xmath125 describing the imbalance in the chemical potential on the two sublattices . @xcite recalling that momentum changes sign under the time - reversal , @xmath126 , in the graphene representation the invariance of the above hamiltonian under the same transformation implies @xmath127=[u,\gamma_0]=0,\end{aligned}\ ] ] and hence @xmath128 . within the simplest framework of the tight - binding model with uniform hopping the time - reversal operator is not uniquely determined . we thus consider a generalized tight - binding model with anisotropic hopping defined as @xmath129 where @xmath130 represents a non - uniform hopping , and @xmath131 . @xcite on a lattice , this particular set of hoppings generates the so - called kekule texture . near the two dirac points the hamiltonian @xmath132 reads @xmath133 where @xmath134 and @xmath135 . the two masses @xmath136 and @xmath137 therefore provide the low - energy representation of a completely real microscopic hamiltonian , so that we postulate that @xmath132 is also time - reversal symmetric . in graphene representation this requires the unitary part of the time - reversal operator to obey the following algebra : @xmath138=\{u , i{\gamma_0}{\gamma_5}\}=0.\ ] ] the matrix @xmath139 satisfies conditions ( [ condt ] ) and thus the unitary part of the operator @xmath120 acting on the spinless dirac field ( [ dirac - spinor ] ) is @xcite @xmath140 with @xmath141 as the @xmath142 unity matrix . the unitary part of the time - reversal operator thus simply exchanges the components of the dirac spinor @xmath54 with different valley indices , as expected . it also happens to be the same as one of the two matrices representing the reflection operators from @xmath1 . another way of arriving at the same form for the time - reversal operator is to postulate that an arbitrary chiral transformation of the dirac hamiltonian in eq . ( 29 ) yields a time - reversal invariant hamiltonian . alternatively , our derivation may be understood as a demonstration of commutativity of the chiral and time - reversal transformations . since we have already used the invariance under @xmath49 to restrict the interacting lagrangian , time - reversal invariance will be observed if the remaining terms are even under complex conjugation . all the terms @xmath143 and @xmath144 are thus automatically invariant under time - reversal , but among the remaining six mixed terms , the terms @xmath145 , @xmath146 , and @xmath147 are odd . time - reversal invariance implies therefore that @xmath148 which leaves then at most fifteen independent couplings . we found that the exact symmetries of the microscopic hamiltonian , @xmath1 , translational , and the time - reversal , leave at most fifteen independent short - range couplings . anticipating some of the results , it is interesting to deduce the further reductions of the number of couplings if one by hand imposes larger symmetries onto the interaction lagrangian @xmath79 . since the rotation by @xmath95 is not a member of the @xmath1 , the matrices @xmath149 and @xmath150 appear asymmetrically in @xmath79 . if we demand that they appear symmetrically , @xmath79 becomes fully rotationally invariant . this is achieved if @xmath151 let us call the interacting lagrangian with the rotational invariance imposed this way @xmath152 . it would be described by at most nine couplings . imposing further the lorentz invariance would require that on top of the above restrictions one also has that @xmath153 with the restrictions in the previous two equations the lagrangian has only six couplings constants , and may be cast in a manifestly lorentz invariant form , worth displaying : @xmath154 + g_\gamma [ ( \bar{\psi } \gamma_\mu \gamma_3 \psi)^2 + ( \bar{\psi } \gamma_\mu \gamma_5 \psi)^2 ] . \end{aligned}\ ] ] finally , the maximally invariant interacting lagrangian would be with the full , i. e. both the lorentz and the chiral , symmetry of the non - interacting part . this is achieved by setting in the last equation @xmath155 the interacting lagrangian can in this case be written as @xmath156 where the participating bilinears in dirac fields , @xmath157 @xmath158 @xmath159 @xmath160 are the scalar ( vector ) , scalar ( scalar ) , vector ( scalar ) , and vector ( vector ) under the chiral ( lorentz ) transformation . the last form makes the lorentz and the chiral symmetry of the lagrangian @xmath161 manifest . such a _ maximally symmetric _ lagrangian contains therefore at most only four coupling constants . the number of independent couplings is further reduced by the existence of algebraic identities between seemingly different quartic terms . the derivation of the so - called fierz transformation , which allows one to write a given local quartic term in terms of other quartic terms is provided in appendix a. a systematic application of this transformation allows one to reduce the number of independent couplings for a given symmetry of the interacting lagrangian . the application of fierz identity to the set of quartic terms allowed by the assumed symmetry in principle leads to the set of linear constraints of the form @xmath162 where @xmath163 is a real typically asymmetric matrix , and @xmath164 is a column ; the elements of which are the quartic terms allowed by the symmetry . of course , only the quartic terms which share the same symmetry may be related by fierz transformations . for example , in the maximally symmetric case @xmath165 . when the number of couplings is small it is easy to discern the linearly independent combinations of the original terms , but when it is not , as the case is for @xmath166 microscopic symmetry of @xmath79 , one needs a more general method of doing so . in appendix b we show that an asymmetric matrix such as the fierz matrix @xmath163 can be written in the diadic form @xcite as @xmath167 where @xmath168 is the real spectrum of the related symmetric matrix @xmath169 . in the eigenbasis of @xmath170 we can write , in dirac notation , @xmath171 so that the above linear equations can be written as @xmath172 since the vectors @xmath173 also form a basis , it must be that either : \a ) for @xmath174 , @xmath175 , or \b ) @xmath176 , so that @xmath177 . the first set provides us then with the linearly independent constraints , and the second with the set of remaining linearly independent quartic terms . since the matrices @xmath163 and @xmath170 obviously have the same kernels , the number of independent coupling constants allowed by the symmetry is simply the dimension of the kernel of the appropriate fierz matrix let us consider the simplest example of the quartic term with the full lorentz and chiral symmetry , @xmath178 first . defining the vector @xmath164 as in the above leads to the fierz matrix @xmath179 with the two - dimensional kernel with the zero - eigenstates @xmath180 @xmath181 and @xmath182 . the remaining two eigenvalues are @xmath183 , and @xmath184 . the general method explained above implies that the general maximally symmetric interacting lagrangian can be written as @xmath185 with the `` physical '' couplings @xmath186 @xmath187 the two remaining linearly independent combinations vanish due to fierz identity . so the maximally symmetric interacting theory is specified by only two quartic coupling constants , which may be chosen to be any linearly independent combinations of the above @xmath188 and @xmath189 . we may then proceed to find the independent couplings for the next case in order in complexity , @xmath190 , with the chiral symmetry broken down to @xmath110 . if we define the six - dimensional vector @xmath191 the fierz matrix is found to be @xmath192 with the three - dimensional kernel spanned by @xmath193 @xmath194 @xmath195 @xmath196 can now be written in terms of only three linearly independent quartic terms , in complete analogy with the maximally symmetric case . the symmetry ladder may now be climbed back to the @xmath166 minimally symmetric lagrangian . first , reducing lorentz to rotational symmetry increases the number of independent couplings to four . removing the rotational symmetry finally increases the number of couplings to six . we may note in passing that there are two independent fierz identities between the mixed terms that obey the time - reversal symmetry in eq . ( 28 ) : @xmath197 @xmath198 so that the three mixed terms in fact contribute a single independent coupling . having determined the independent coupling constants for each symmetry , we now proceed to study their changes with the decrease of the upper cutoff @xmath199 . we will be particularly interested in fixed points of such renormalization group transformations , as they will provide the information on the quantum metal - insulator transitions that can be induced by increase in interactions . let us again begin with the maximally symmetric lagrangian , @xmath200 . there are only two coupling constants to consider in this case , and we choose them to be @xmath201 and @xmath202 , which correspond to @xmath203 and @xmath204 quartic terms , respectively . if any of the other two terms would become generated by the renormalization transformation we would use the fierz identity to rewrite it in terms of @xmath203 and @xmath204 . alternatively , one may wish to renormalize the theory as written in terms of physical couplings in eq . this procedure , completely equivalent to what is pursued here , is described in appendix c. as we integrate the fermionic modes lying in the 2 + 1 dimensional momentum shell @xcite from @xmath205 to @xmath199 , with @xmath206 , to quadratic order in coupling constants we find @xmath207 @xmath208 we rescaled the couplings here as @xmath209 . to this order no other types of quartic terms actually get generated , and the fierz transformation turns out not to be necessary . the limit of the above equations that survives the extension to a large number of dirac fields also agrees with the previous calculation . @xcite the above flows , besides the gaussian , exhibit three fixed points at finite couplings ( fig . the first critical point ( a ) is at @xmath210 , @xmath211 , and the second critical point ( c ) is at @xmath212 , @xmath213 . there is also a bicritical fixed point ( b ) that separates the domains of attraction of the two critical points , at @xmath214 , @xmath215 . the physical interpretation of the critical point a is obvious . since we can tune through it by keeping @xmath211 and increasing @xmath201 over a certain negative value , it should describe the transition into the insulator with the gap that breaks the time - reversal symmetry , described by @xmath216 this state is obviously favored at a large and negative @xmath201 . note that since @xmath20 commutes with @xmath50 , the line @xmath211 is invariant under rg . in fact , the perturbative @xmath217-function along this line has to be identical as the one in the gross - neveu model . we can therefore simply use the already existing higher - order estimates @xcite and the numerical results @xcite to find the critical exponents describing this particular metal - insulator transition . we return to this fixed point shortly . the physical interpretation of the critical point c is less obvious , but we can think of it as follows . first , note that the line @xmath218 , @xmath219 , which describes the thirring model , @xcite belongs to the domain of attraction of c. also , the fierz transformations in eq . ( 50 ) imply @xmath220 the line @xmath221 for @xmath219 , which we name the nambu - jona - lasinio ( njl ) line , @xcite also falls into the domain of attraction of the critical point c. along this line , however , there should be a transition into an insulating state with @xmath222 where @xmath223 is a unit vector . such a state is clearly favored at a large and positive @xmath202 along the njl line , and breaks the chiral @xmath3 symmetry down to @xmath224 . we therefore identify c as the metal - insulator critical point governing the chiral - symmetry breaking transition in _ both _ thirring and njl models with a single dirac field . the picture suggested by the above one - loop calculation in the maximally symmetric theory appears quite natural . there are two possible insulating phases , each breaking either chiral or time - reversal symmetry , which correspond to possible `` masses '' for the dirac fermions . both metal - insulator transitions are continuous , and are described by different critical points . of course , the true low - energy theory on the honeycomb lattice is much less symmetric than the one studied in this section . nevertheless , we will argue that the two identified critical points may in fact be stable at least with respect to weak manifest breaking of the lorentz and chiral symmetries . the explicit breaking of lorentz symmetry down to the rotational symmetry can be easily implemented by adding to @xmath225 a small symmetry breaking term @xmath226 which by virtue of being only rotational symmetric is guaranteed to be linearly independent of @xmath203 and @xmath204 . a weak perturbation @xmath227 to the lowest order in couplings @xmath201 and @xmath202 then flows according to @xmath228 we thus find the critical point c to be stable with respect to weak lorentz symmetry breaking to one loop , the bicritical point b to be unstable , and a marginal . we suspect that this result , although clearly an outcome here of an uncontrolled approximation , may be indicative of the true state of affairs . hereafter we will assume that the critical points a and c are stable with respect to weak breaking of the lorentz symmetry in the lagrangian . it may also be worth mentioning that the complete one - loop @xmath217-functions for @xmath227 , @xmath202 and @xmath201 , which we have computed but have not shown , do not lead to any new critical points at @xmath229 . the simplest quartic term with the full lorentz symmetry , and only @xmath110 subgroup of the full chiral symmetry may be written as @xmath230 the fierz transformation matrix given above implies that these three quartic terms are indeed linearly independent . when @xmath231 , the lagrangian @xmath196 acquires the full chiral @xmath232 symmetry , and may be rewritten as @xmath178 . using the fierz transformation , and after a convenient rescaling of the couplings as @xmath233 , to the quadratic order one finds @xmath234 @xmath235 @xmath236 the two chirally symmetric critical points from the previous section now appear at @xmath237 , @xmath238 ( a ) , and @xmath239 , @xmath240 ( c ) , and both remain critical , even in absence of chiral symmetry in the lagrangian . there is , however , an additional critical point ( e ) at @xmath241 , @xmath242 . note also that the plane @xmath243 is invariant under the renormalization group , but whereas the fixed point a in that plane is critical , the fixed point ( d ) at @xmath218 , @xmath244 is bicritical . one also finds that the line @xmath218 , @xmath245 , @xmath246-infinitesimal and positive intersects the critical surface which contains the point e , whereas for @xmath246-infinitesimal and negative the critical behavior is governed by c. for weak @xmath246 therefore there is a crossover from the fixed point at d toward either c or e , depending on the sign . interestingly , for negative @xmath246 chiral symmetry becomes fully restored at the transition , at least within our one - loop calculation . motivated by the one - loop results , we will assume hereafter that the lorentz symmetry becomes restored at long distances in the domain of interest , and that we need only consider @xmath190 with the three couplings from the last section . the situation however , can then be simplified even further , as we discuss in this section . consider the interaction hamiltonian in eq . if the @xmath29-orbitals are well localized on their corresponding lattice sites , we may neglect the matrix elements with @xmath247 or @xmath248 . keeping only the remaining , dominant matrix elements then one obtains the `` atomic limit '' of the general interaction hamiltonian @xmath249 where @xmath250 is the electron number operator at site @xmath72 . the class of hamiltonians @xmath251 is evidently still rather broad , and would for example include all lattice interacting hamiltonians . writing the lattice hamiltonian @xmath251 in terms of the dirac fields , however , imposes yet another restriction on the coupling constants . since @xmath251 is written in terms of lattice - site particle number operators , any dirac quartic term evidently must contain an equal number of @xmath252 ( @xmath253 ) and @xmath32 ( @xmath33 ) fields . on the other hand , the @xmath254-term from above in momentum space can be written schematically as @xmath255 and thus contains the terms forbidden in the atomic limit,@xcite such as @xmath256 . the index 1 and 2 refers here to the two dirac points . this implies that for any lattice hamiltonian @xmath251 we must have @xmath257 note that the plane @xmath258 is invariant under the change of cutoff in the above one - loop calculation . it is easy to see that this feature of the @xmath217-functions for @xmath201 , @xmath259 and @xmath246 is in fact true to all orders in perturbation theory . the matrices @xmath20 and @xmath260 in the remaining two terms in @xmath196 commute with the dirac propagator , and therefore an arbitrary diagram containing @xmath201 and @xmath259 terms can contribute only to the renormalized @xmath201 and @xmath259 couplings . so imposing @xmath258 at an arbitrary cutoff guarantees its vanishing at all others . it is therefore not only physically justified but also internally consistent to consider only the two couplings @xmath201 and @xmath259 in the lorentz symmetric , but chirally asymmetric low - energy theory . the one - loop result in this plane is depicted in fig . the transition is either into the time - reversal - symmetry - broken , or into chiral - symmetry - broken insulator . several features of this flow diagram that should be generally valid are worth mentioning . \1 ) there should be two critical points , both unstable in a single direction . bicriticality of a , for example , would imply that the transition for negative @xmath201 at a weak positive @xmath259 is first order . this , however , seems unlikely on physical grounds , and , also , it is not found in the explicit large - n generalization of the theory , when the two @xmath217-functions are known to decouple . \2 ) the two critical points have identical critical behavior . this is because the term @xmath261 , in graphene representation , under the transformation @xmath262 \psi,\ ] ] @xmath263,\ ] ] goes into @xmath264 , and vice versa , while @xmath47 remains invariant . this also means that the two @xmath217-functions are symmetric under the exchange @xmath265 . both critical points are thus in the universality class of the gross - neveu model . \3 ) at the line @xmath266 the single @xmath217-function becomes @xmath267 i. e. @xmath201 flows according solely to its canonical dimension . this is because @xmath268 where @xmath269 . since all @xmath50 are block - diagonal , @xmath270 and @xmath271 components at this line decouple . the partition function factorizes into a product of two gross - neveu partition functions , each containing a single _ two - component _ dirac fermion . along this line the system is believed to have the metal - insulator transition , possibly continuous , @xcite but the @xmath217-function vanishes at least to the order @xmath272 . each identified metal - insulator transition is characterized by a set of critical exponents . we will here focus on the three already mentioned in the introduction : the correlation length exponent @xmath273 , the dynamical exponent @xmath25 , and the dirac fermion anomalous dimension @xmath28 . the other exponents can then be obtained from the usual scaling relations . @xcite first , since all the identified critical points exhibit lorentz symmetry , @xmath274 we also find that the exponent @xmath273 is unity at all critical points as well , but this is clearly an artifact of the one - loop calculation . in general , @xmath273 is expected to be different at different critical points . the same goes for @xmath28 , which vanishes in one - loop calculation , but will be finite in general . in the atomic limit , when under the assumed lorentz invariance we need only two coupling constants , the values of the critical exponents are better known . first , in a perturbative calculation in powers of coupling constants , the exponents at critical points @xmath275 and @xmath276 will be identical . we may thus expect that in a lattice theory with short - range repulsion the transition is either into the time - reversal symmetry or chiral symmetry broken insulator , in either case with @xcite @xmath277 @xmath278 note that since there is only a single dirac field involved the numerical values of the exponents @xmath273 and @xmath28 differ significantly from the large - n values of unity and zero , respectively . this raises hope that this non - trivial critical behavior may be observable in numerical simulations of lattice models . although not of immediate relevance to graphene , it would still be of interest to determine the critical exponents at the chirally symmetric critical point c , which we proposed to control the critical behavior of the thirring and the njl model . we , however , are not aware of any analytical nor numerical study of the njl model that goes beyond the leading order in large - n calculation . the critical exponents , as usual , govern for example the critical behavior of the gap on the insulating side of the transition , as mentioned in the introduction . in the present case , however , there are massless fermionic excitations on the metallic side , and one may wonder if and how the approach to the critical point is reflected onto these . let us therefore generalize slightly and provide the support for the results already announced in the introduction . first , the usual scaling @xcite implies that at the cutoff @xmath279 electron s two - point correlation function near the critical point and at zero temperature satisfies @xmath280 where @xmath281 is the matsubara frequency and @xmath282 is the transition s tuning parameter . setting @xmath283 we thus find the usual scaling law @xmath284 where @xmath285 is a universal scaling function . from here we can extract the scaling of the fermi velocity and quasiparticle residue as follows . first , if upon the analytical continuation to real frequencies @xmath286 has a pole at @xmath287 the scaling relation immediately dictates that @xmath288 let us next set @xmath289 , take @xmath290 , and let @xmath291 . in this limit @xmath292 and therefore @xmath293 . in order to cancel the @xmath294-dependence of the prefactor in eq . ( 85 ) in this limit it must be that @xmath295 where @xmath296 is the ( momentum ) anomalous dimension . analogously , assuming that for @xmath297 , @xmath298 , one finds that also @xmath299 on the other hand , in the opposite limit @xmath300 and @xmath301 , in the metallic phase at low energies we have fermionic quasiparticles . this implies that , for example , @xmath302 , i. e. @xmath303 with @xmath304 . combining with the previous relation , the quasiparticle pole s residue behaves as @xmath305 for @xmath17 , the two anomalous dimensions are the same , @xmath306 , where @xmath307 is the dirac fermion s anomalous dimension , and the scaling announced in the introduction follows . the special form of this relation for large number of dirac components when besides @xmath17 , it is also @xmath22 , was previously proposed . @xcite the quasiparticle residue vanishes upon the approach to the metal - insulator transition , as proposed long - ago by brinkman and rice , but here in a decidedly non - mean - field fashion . there are at least two obvious generalizations important for real graphene : the addition of spin , and the inclusion of the long - range tail of coulomb repulsion . adding spin would simply double the number of couplings for each symmetry , since each independent quartic term would then require a separate coupling in the singlet and in the triplet channels . the minimal internally consistent low - energy theory would then be the generalization of the lorentz invariant lagrangian in eq . ( 70 ) , with @xmath258 : @xmath308 where the sum goes over @xmath309 , and @xmath310 , with @xmath311 , and @xmath312 . the lagrangian with @xmath313 would represent the extended hubbard model with on - site and nearest neighbor repulsion , considered before in the limit of large number of dirac fermions in ref . the interplay between the various instabilities in the theory equivalent to the above lagrangian was recently studied in @xcite , where it was pointed out that the second - nearest - neighbor repulsion implies a negative coupling @xmath314 for example . the form of the above minimal spinful lagrangian facilitates a systematic study of the metal - insulator transition in the hubbard model , which will be a subject of a separate publication . few comments on the importance of long - range tail of coulomb interaction are also in order . weak coulomb ( @xmath315 ) interaction is an ( marginally ) irrelevant perturbation at the gaussian fixed point , and this remains true at the metal - insulator critical point at large - n as well @xcite . furthermore , the entire @xmath217-function for the charge coupling @xmath316 can be computed at large - n , and it does not exhibit any non - trivial zeroes . @xcite on the other hand , several calculations show that by increasing the coupling @xmath316 beyond certain point and for small enough @xmath317 the system can be tuned through a metal - insulator transition at which the chiral symmetry becomes spontaneously broken.@xcite the nature of such a putative metal - insulator transition is in our mind an open question at the moment . whereas it is possible that it is described by new `` charged '' critical point @xcite corresponding to the non - trivial zero of the @xmath318 , it also seems conceivable that the charge is always irrelevant and that the transition is still in the universality class of the critical point c , in our nomenclature . yet another possibility is a discontinuous transition . more work is obviously needed in order to be able to address this issue more conclusively . it may also be interesting to note that in the related bosonic problem , when a systematic expansion near four dimension is readily available , there are no charged critical points in the theory to the leading order . @xcite this may also be contrasted with the well - known example of ( albeit lorentz invariant ) scalar higgs electrodynamics , for which the critical points , when they exist , are always charged . @xcite probably the central message of this work is that provided lorentz invariance becomes emergent near criticality , for the @xmath29-orbitals well localized on carbon atoms the lagrangian may be taken to contain only two ( or with the physical spin , four ) coupling constants . if there are no intervening first - order transitions , one can infer then that there are two possible continuous metal - insulator transitions , both governed by the same gross - neveu model in 2 + 1 dimensions , into the states that break either time - reversal or chiral symmetry . the residue of the quasiparticle pole on the metallic side plays the role of the metal s order parameter , and it vanishes continuously with a small critical exponent proportional to the fermion s anomalous dimension . in contrast , the specific heat coefficient @xmath319 being dependent on the fermi velocity only , at the transition vanishes discontinuously from a finite value on the metallic side . near the critical point and for the temperatures much below the bandwidth we may assume that the specific heat obeys the scaling relation @xmath320 for small arguments the universal scaling function @xmath321 behaves as @xmath322 , so that in the metallic phase one finds the usual quadratic temperature dependence @xmath323 recognizing the proportionality of the specific heat coefficient as @xmath324 gives us yet another way to deduce eq . ( 87 ) . at the criticality , on the other hand , @xmath325 with @xmath326 expected to be finite . when @xmath17 , the specific heat coefficient near criticality jumps therefore from @xmath327 in the metallic phase to @xmath326 at the critical point , and finally to zero in the insulating phase.@xcite similarly , the optical conductivity near the metal - insulator transition will obey the scaling relation for @xmath300 @xmath328 with @xmath329 as a universal function , and with @xmath330 as the familiar universal dc conductivity per dirac field in the metallic phase . @xcite in contrast to the specific heat , there is no ( non - universal ) dimensionful quantity such as @xmath331 in the scaling expression for conductivity , and consequently @xmath332 is constant and universal in the entire metallic phase . right at the transition then @xmath333 so the dc conductivity , while still universal , at the criticality should be different than in the metallic phase . finally , in the insulator the dc conductivity vanishes , so that the dc conductivity , similar to the specific heat coefficient , in principle should show two universal discontinuities at the metal - insulator transition . one obstacle to experimental observation of these predictions is that it has not been possible yet to tune the parameter @xmath334(interaction / bandwidth ) in graphene , and sample different phases of the system . the application of the magnetic field , however , changes this , since the kinetic energy becomes completely quenched , and infinitesimal interaction immediately induces a finite gap . if the parameters of the system place it not too far from the metal - insulator transition , the gap @xmath335 would obey @xcite @xmath336 where @xmath337 is the magnetic length , @xmath338 is the lattice constant , and @xmath294 is the tuning parameter . @xmath339 is a ( universal ) scaling function . the computation of the scaling function in the large - n limit , and the consequences of this scaling relation for experiment were discussed at length before . @xcite here we only wish to underline that the emergent lorentz invariance of the metal - insulator critical point , via its consequence that @xmath17 , implies precise proportionality between the interaction gap and the landau level separation at the criticality , @xmath340 where @xmath341 is a universal number . such a square - root magnetic field dependence of the gap is well known to arise from the long - range tail of the coulomb interactions , but the above derivation serves to show that its origin may in principle lie in purely short - range interactions as well . we have presented the theory of electrons interacting via short - range interactions on honeycomb lattice , and in particular , determined the number and types of independent quartic terms in the low - energy lagrangian . metal - insulator quantum critical points and the concomitant quantum critical behavior were discussed , with the particular attention paid to the consequences of the emergent lorentz invariance . the minimal internally consistent local lagrangian for spinless fermions is shown to contain only two gross - neveu - like quartic terms . generalizations that would include long - range coulomb interaction or spin of electrons were briefly considered . we also discussed the critical behavior of several key physical quantities on the metallic side of the transition , such as the fermi velocity , the residue of the quasiparticle pole , specific heat , and the frequency dependent conductivity . this work was supported by the nserc of canada . for completeness , we provide the derivation of the fierz identity . assume a basis @xmath342 in the space of four - dimensional matrices , and choose @xmath343 . then any hermitean matrix @xmath344 can be written as @xmath345 with the summation over repeated indices assumed . this can be rewritten as @xmath346 and therefore it follows that @xmath347 applying this identity to the product of two matrix elements then yields @xmath348 finally , this leads to the expansion of a quartic term as @xmath349 which is used in the text for @xmath350 . the minus sign in the last line derives from the grassmann nature of the fermionic fields . any real @xmath317-dimensional matrix @xmath344 can obviously be written as @xmath351 where @xmath352 , and @xmath353 . in dirac notation , @xmath354 and @xmath355 is the transposed matrix . there exists such a representation of the matrix @xmath344 in any basis of vectors @xmath356 , as can be seen by multiplying @xmath344 from the right with @xmath357 . let us now form a related symmetric matrix @xmath358 . being symmetric , it can be written in the usual spectral form @xmath359 where @xmath360 . we can now write , however , the matrix @xmath344 in the particular eigenbasis of the associated symmetric matrix @xcite @xmath361 @xmath362 from the definition of @xmath84 and its spectral form we see that @xmath363 , and therefore @xmath364 where @xmath365 , and @xmath366 . for a general asymmetric matrix , the basis @xmath367 and @xmath273 are different , and the last equation generalizes the more familiar form for a symmetric matrix , where they are the same . here we provide an alternative formulation of the renormalization group transformation in presence of constraints imposed by the fierz identity . let us demonstrate this method on the simplest example of the maximally symmetric theory . instead of choosing two independent couplings and using fierz transformation at intermediate stages of the calculation to transform any other generated quartic terms back into the chosen ones , one may use the kernel of the fierz matrix to write the lagrangian in terms only of the physical couplings from the outset , as in eq . the advantage of doing this is that no other quartic term besides the ones corresponding to the physical couplings can ever get generated then by the renormalization transformation . the set of couplings @xmath188 and @xmath189 is therefore closed under renormalization . the computation to the quadratic order then yields @xmath368 @xmath369 the connection to eqs . ( 63)-(64 ) in the text can be established as follows . since the fierz transformations in this case imply @xmath370 @xmath371 the lagrangian @xmath178 in eq . ( 41 ) can obviously also be written as @xmath372 in the text we therefore have simply named the entire first bracket @xmath201 , and the second @xmath202 . but these can be recognized as particular linear combinations of the physical couplings @xmath188 and @xmath189 : @xmath373 @xmath374 such a connection is of course completely general , and in particular may be established between the three chosen couplings in the eq . ( 70 ) and the `` physical couplings '' determined by the vectors in eqs . ( 58)-(60 ) . 99 for reviews , see v.p . gusynin , s.g . sharapov , j.p . carbotte , int . j. of mod . b , * 21 * , 4611 ( 2007 ) ; a. h. castro neto , f. guinea , n. m. r. peres , k. s. novoselov , a. k. geim , rev . phys . * 81 * , 109 ( 2009 ) . i. f. herbut , phys . . lett . * 97 * , 146401 ( 2006 ) . j. gonzalez , f. guinea , m. a. h. vozmediano , nucl . b424 * , 595 ( 1994 ) ; phys . b * 59 * , 2474 ( 1999 ) . o. vafek , phys . 98 * , 216401 ( 2007 ) . i. f. herbut , phys . lett . * 99 * 206404 , ( 2007 ) . s. sorella and e. tosatti , europhys . lett . * 19 * , 699 ( 1992 ) . l. m. martelo , m. dzierzawa , l. siffert , and d. baeriswyl , z. phys . b * 103 * , 335 ( 1997 ) . t. paiva , r. t. scalettar , w. zheng , r. r. singh , and j. oitmaa , phys . rev . b * 72 * , 085123 ( 2005 ) . y. zhang , z. jiang , j. p. small , m. s. purewal , y .- w . tan , m. fazlollahi , j. d. chudow , j. a. jaszczak , h. l. stormer , and p. kim , phys . lett . * 96 * , 136806 ( 2006 ) ; z. jiang , y. zhang , h. l. stormer , and p. kim , phys . lett . * 99 * , 106802 ( 2007 ) . joseph g. checkelsky , lu li , n. p. ong , phys . * 100 * , 206801 ( 2008 ) ; preprint arxiv:0808.0906 . d. v. khveshchenko , phys . rev . lett . * 87 * , 206401 ( 2001 ) ; ibid . * 87 * , 246802 ( 2001 ) . v. p. gusynin , v. a. miransky , s. g. sharapov , and i. a. shovkovy , phys . b * 74 * , 195429 ( 2006 ) ; e. v. gorbar , v. p. gusynin , v. a. miransky , preprint arxiv:0710.3527 . i. f. herbut , phys . rev . b * 75 * , 165411 ( 2007 ) ; ibid . * 76 * , 085432 ( 2007 ) ; i. f. herbut and b. roy , * 77 * , 245438 ( 2008 ) . j. alicea and m. p. a. fisher , phys . b * 74 * , 075422 ( 2006 ) ; r. l. doretto and c. morais - smith , phys . b * 76 * , 195431 ( 2007 ) . for a review , see a. k. geim and k. s. novoselov , nature materials * 6 * , 183 ( 2007 ) . i. f. herbut , v. jurii ' c , and o. vafek , phys . lett . * 100 * 046403 ( 2008 ) ; i. f. herbut , v. jurii ' c , o. vafek , m. j. case , preprint , arxive:0809.0725 . l. fritz , j. schmalian , m. mueller , and s. sachdev , phys . b * 78 * , 085416 ( 2008 ) . by this definition the interaction that at large particle separation decays as a power law @xmath375 is short ranged provided that @xmath376 . see for example , y. takahashi , in _ progress in quantum field theory _ , ed . by h. ezawa and s. kamefuchi ( north holland , 1986 ) . for the closely related chiral symmetry of d - wave superconductors , see i. f. herbut , phys . b * 66 * , 094504 ( 2002 ) ; phys . lett . * 88 * , 047006 ( 2002 ) ; phys . * 94 * , 237001 ( 2005 ) ; d. j. lee and i. f. herbut , phys . b * 66 * , 094512 ( 2002 ) ; b. seradjeh and i. f. herbut , phys . rev . b * 66 * , 184507 ( 2002 ) ; z. teanovi ' c , o. vafek , and m. franz , phys . rev . b * 65 * , 180511 ( 2002 ) ; m. franz , t. pereg - barnea , d.e . sheehy , and z. teanovi ' c , phys . b * 68 * , 024508 ( 2003 ) ; i. o. thomas and s. hands , phys . rev . b * 75 * , 134516 ( 2007 ) . at the critical point at a finite coulomb interaction in two - dimensions @xmath25 is always unity ; see i. f. herbut , phys . lett . * 87 * , 137004 ( 2001 ) . f. d. m. haldane , phys . lett . * 61 * , 2015 ( 1988 ) . g. w. semenoff , phys . lett . * 53 * , 2449 ( 1984 ) . hou , c. chamon , and c. mudry , phys . * 98 * , 186809 ( 2007 ) . i. herbut , _ a modern approach to critical phenomena _ , ( cambridge university press , cambridge , 2007 ) . w. f. brinkman and t. m. rice , phys . b * 2 * , 4302 ( 1970 ) . y. hasegawa , r. konno , h. nakano , m. kohmoto , phys . b * 74 * , 033413 ( 2006 ) . i. f. herbut , phys . b. * 78 * , 205433 ( 2008 ) . muicki and b. mili ' c , _ matematicke osnove teorijske fizike _ , ( univerzitet u beogradu , beograd , 1984 ) , sec . 5.3 ( in serbian ) . here we treat the frequency integrals in the loops analogously to momentum integrals . for an alternative , see ref . 2 , for example . k. kaveh and i. f. herbut , phys . b * 71 * , 184519 ( 2005 ) . a. m. vasilev , s. e. derkachov , n. a. kilev , and a. s. stepanenko , teor . mat . fiz . * 92 * , 486 ( 1992 ) ; _ ibid . _ * 97 * , 364 ( 1993 ) ; j. a. gracey , int . j. mod . phys . a * 9 * , 727 ( 1994 ) . l. krkkinen , l. lacaze , p. lacock , and b. petersson , nucl . b * 415 * , 781 ( 1994 ) ; * 438 * 650(e ) ( 1995 ) . l. rosa , p. vitale , and c. wetterich , phys . * 86 * , 958 ( 2001 ) ; f. hfling , c. novak , and c. wetterich , phys . b * 66 * , 205111 ( 2002 ) . s. christofi , s. hands , c. strouthos , phys.rev . d * 75 * , 101701 ( 2007 ) , and references therein . y. nambu and g. jona - lasinio , phys . rev . * 122 * , 345 ( 1961 ) . in the purely bosonic @xmath377-theories , the @xmath378 symmetry is currently believed not to emerge at the critical point out of @xmath379 , in three dimensions . the critical behavior is governed by the `` biconal '' fixed point , which however , appears to be extremely close the @xmath378-symmetric point in the coupling space . see , p. calabrese , a. pelissetto , e. vicari , phys . b * 67 * , 054505 ( 2003 ) . the terms allowed in the atomic limit are already contained in the terms @xmath203 and @xmath380 . see , nevertheless , @xcite for numerical results on thirring models with more than one dirac fermion . s. raghu , xiao - liang qi , c. honerkamp , s .- c . zhang , phys . lett . * 100 * , 156401 ( 2008 ) . i. l. aleiner , d. e. kharzaev , and a. m. tsvelik , phys . b * 76 * , 195415 ( 2007 ) d. t. son , phys . rev . b * 75 * , 235423 ( 2007 ) ; j. e. drut and d. t. son , phys . rev . b * 77 * , 075115 ( 2008 ) . e. v. gorbar , v. p. gusynin , v. a. miransky , and i. a. shovkovy , phys . rev . b * 66 * , 045108 , ( 2002 ) . d. v. khveshchenko and h. leal , nucl . phys . b * 687 * , 323 ( 2004 ) . j. e. drut and t. a. lahde , phys . lett , * 102 * , 026802 ( 2009 ) . s. hands and c. strouthos , phys b * 78 * 165423 , 2008 . o. vafek and m. j. case , phys . b * 77 * , 033410 ( 2008 ) . see the ref . 21 , and the problem 8.10 in ref . i. f. herbut and z. teanovi ' c , phys . * 76 * , 4588 ( 1996 ) ; ibid . * 78 * , 980 ( 1997 ) ; i. f. herbut , j. phys . a : math . gen . * 30 * , 423 ( 1997 ) . for spinless fermions considered here , both insulators break the discrete ising symmetry , so there are no nambu - goldstone bosons .

we report an extensive set of two - dimensional mhd simulations exploring the role and evolution of magnetic fields in the dynamics of supersonic plasma clumps . we examine the influence of both ambient field strength and orientation on the problem . of those two characteristics , field orientation is far more important in the cases we have considered with . that is due to the geometry - sensitivity of field stretching / amplification from large - scale shearing motions around the bullet . when the ambient magnetic field is transverse to the bullet motion , even a very modest field , well below equipartition strength , can be amplified by field line stretching around the bullet within a couple of bullet crushing times so that maxwell stresses become comparable to the ram pressure associated with the bullet motion . the possibility is discussed that those situations might lead to large , induced electric potentials capable of accelerating charged particles . when the ambient field is aligned to the bullet motion , on the other hand , reconnection - prone topologies develop that shorten the stretched field and release much of the excess energy it contains . in this geometry , the maxwell stresses on the bullet never approach the ram pressure level . in both cases , however , the presence of a field with even moderate initial strength acts to help the flow realign itself around the bullet into a smoother , more laminar form . that reduces bullet fragmentation tendencies caused by destructive instabilities . eddies seem less effective at field amplification than flows around the bullet , because fields within eddies tend to be expelled to the eddy perimeters . similar effects cause the magnetic field within the bullet itself to be reduced below its initial value over time . for oblique fields , we expect that the transverse field cases modeled here are more generally relevant . what counts is whether field lines threading the face of the bullet are swept around it in a fashion that folds them ( leading to reconnection ) or that keeps them unidirectional one each side of the bullet . in the second instance , behaviors should resemble those of the transverse field cases . we estimate that this , quasi - transverse , behavior is appropriate whenever the angle , , between the motion and the field satisfies , where is the bullet mach number . from these simulations , we find support in either field geometry for the conclusions reached in previous studies that nonthermal radio emission associated with supersonic clumps is likely to be controlled largely by the generation of strong magnetic fields around the perimeters of the clumps , rather than local particle acceleration and field compression within the bow shock . in addition , since the magnetic pressure on the nose of the bullet likely becomes comparable to the ram pressure and hence the total pressure behind the bow shock , the gas pressure there could be substantially lower than that in a gasdynamical bullet . that means , as well , that the temperature in the region on the nose of the bullet would be lower than that predicted in the gasdynamical case . that detail could alter expectations of the thermal emission , including x - rays and uv - ir lines . = cmmi10 scaled1 = cmmi10 = cmmi7 = cmmib10 scaled1 = cmmib10 = cmmib10 scaled 833 = cmbsy10 scaled1 = cmbsy10 = cmbsy10 scaled 833 1 = 1 = 1 = 9 = 9 = 9 = 10 = 10 = 10 =

astro-ph9606106

in recent years observations have revealed the presence of dense , high - velocity gas clumps within and sometimes outside young supernova remnants and also associated with the outflows from young stars ( , @xcite ; @xcite ; @xcite ; @xcite ) . these `` cosmic bullets '' or , in some scenarios , `` wind - swept clumps '' can shine prominently in either thermal emission ( , @xcite ; @xcite ; @xcite ; @xcite ) or nonthermal emission ( , @xcite ; @xcite ) . sometimes in supernova remnants they show elemental compositions indicating formation exclusively from material ejected out of the depths of the progenitor supernova . the exact origins of the bullets are still unclear , but suggestions range from explosive ejection ( @xcite ; @xcite ) to wind - swept , pre - existing clumps ( @xcite ) , and dynamical instabilities in winds ( @xcite ) . whatever the manner of creation , cosmic bullets seem capable of surviving for remarkable lengths of time , thus transporting energy and mass over large distances , and have been suggested as sites for acceleration of high energy particles ( , @xcite ; @xcite ; @xcite ) . yet , simple arguments and computer simulations suggest that gasdynamical instabilities may be highly disruptive of these bullets ( , @xcite ; @xcite ) , limiting their existence to only a few times the interval required for the applied ram pressure to crush them . that would limit their propagation distance to only a few times the length @xmath4 , where @xmath5 is the density contrast with the environment and @xmath6 is the bullet radius . so , only the densest clumps could survive to propagate more than a few times their own size scale . however , in many cases one expects that both the clouds and the surrounding material are mostly ionized plasmas , so that embedded magnetic fields ought to be largely frozen in and should play a role . in a recent paper we emphasized the possibility , based on numerical simulations involving a passive magnetic field , that bullets may be effective in amplifying weak ambient magnetic fields as a side effect of the destructive interactions with their environments ( @xcite ) . it was emphasized there that field amplification could be more important in generation of associated nonthermal synchrotron emission , for example , than local particle acceleration in the shocks formed by the bullets ( @xcite ; @xcite ; @xcite ) . at the same time amplified magnetic fields could act to stabilize the kelvin - helmholtz and rayleigh - taylor instabilities , thereby limiting bullet fragmentation . that role might help account for their ability to survive , or at least for significant pieces to penetrate much farther than gasdynamical calculations suggest they might . @xcite carried out two - dimensional mhd simulations of a dense plasma cloud overrun by a plane shock , demonstrating in that related case how a magnetic field might be amplified and also how it may prevent the complete destruction of the cloud in the postshock flow . although shocked clouds and supersonic bullets are qualitatively similar objects in some ways , there are also significant differences between them , as well . in particular , as long as the impinging shock is not radiative , the material surrounding the shocked cloud is mostly moving sub - sonically with respect to the cloud , because the cloud is overrun by a hot , postshock flow . the bullet , of course , is ( by definition ) interacting with a supersonic flow , at least in the beginning . this difference influences especially the flow to the rear and on the sides of the cloud and could influence instabilities and magnetic field behavior ( see @xcite ) . furthermore , @xcite considered primarily shocked clouds with a magnetic field aligned along the symmetry axis of the flow , except for one simulation involving a transverse field that was not reported in much detail . that aligned field orientation was mandated by their use of cylindrical symmetry for all but the one simulation . the other simple field geometry possible with cylindrical symmetry ; namely , a toroidal one , is not a very natural one for this problem . on the other hand , the geometry of the magnetic field ought to be very important to its evolution in and around the dense cloud and also to its dynamical influence on the cloud . we may also expect the influence of the magnetic field to depend on its strength . these issues in the bullet context have not been addressed before in the literature . in this study we present results from two - dimensional , full mhd simulations of the evolution of dense supersonic plasma bullets with magnetic field geometries both aligned with the bullet motion and also transverse to it . in order to have freedom in the field geometry , the simulations have been done in cartesian coordinates rather than in cylindrical coordinates . also we have simulated flows using a range of magnetic field strengths and using a range of numerical resolutions . one can envision the bullets either as ejecta or as condensations being swept by a supersonic wind . our objectives are to understand better how the field in and around the bullet is influenced by the bullet and also how the magnetic field acts on the bullet dynamically . 2 outlines the important dynamical issues , while 3 describes the numerical methods we employ . in 4 we describe our results , while a brief summary of the principal conclusions is presented in 5 . the basic issues associated with the interaction between a gas bullet and its surroundings are well - known , so readers are referred to the extant literature for detailed discussions ( , @xcite ; @xcite and references therein ) or for comparisons with shocked clouds ( , @xcite ; @xcite ; @xcite ; @xcite ) . in this section we present only a brief outline of the issues to facilitate later evaluation of the particular new issues raised in this study . as in previous studies we consider the bullet to be initially in pressure equilibrium with the ambient medium and the initial magnetic field to be uniform through both the bullet and the ambient medium . the plasma is assumed throughout to be highly conducting , so that magnetic field is frozen in . at @xmath7 the bullet is set into motion at speed @xmath8 with respect to the ambient medium . if such complications as radiative cooling , thermal conduction , and gravity can be neglected , the bullet dynamics is determined by a minimum of the following six parameters and the bullet shape : the sonic mach number of the bullet through the ambient medium , @xmath9 ; the gas adiabatic index , @xmath10 ; the density contrast between the bullet and the ambient background , @xmath11 ; as well as the magnetic field strength , @xmath12 , and orientation ( generally , two angles ) . the initial strength of the magnetic field can be expressed either in terms of the alfvnic mach number @xmath13 , where @xmath14 is the alfvn speed , or in terms of @xmath15 . the field orientation can be specified by the angle , @xmath1 , between the field , @xmath16 , and the bullet velocity , @xmath17 , as well as between @xmath16 and any distinct symmetry axis or plane for the bullet . fortunately , as previous shocked - cloud studies emphasized ( @xcite ; @xcite ) , in the limit @xmath18 , the sonic mach number can be scaled out of the problem . @xcite found in three - dimensional gasdynamical simulations of shocked clouds that the initial morphology of the cloud does make a quantitative difference in its evolution , but that qualitative features are not very geometry - sensitive . thus , we will consider only a single initial mach number and a single bullet morphology . because three - dimensional mhd simulations of sufficient numerical resolution are still too expensive to follow magnetic fields adequately in this problem , we have used two - dimensional symmetry . we are concerned , on the other hand , with influences of different magnetic field geometries with respect to the bullet and its motion , so we have chosen a cartesian coordinate system , which enables us to set up a uniform magnetic field with arbitrary orientation . in this geometry the simplest initial cloud morphology is a circular cylinder , with its axis out of the computational plane and thus orthogonal to the motion of the bullet . although we have actually carried out test simulations in which the magnetic field has a component out of the plane ( and , thus , aligned with the bullet axis ) , we present here only cases for which the magnetic field is entirely in the computational plane and either aligned to the direction of the bullet motion or transverse to it . we have found no qualitatively different behaviors for the fields with a component out of the plane . the characteristic evolutionary timescale of a bullet is the so - called `` bullet crushing time '' , @xmath19 where @xmath6 is the initial bullet radius . it allows one to subsume the values of the initial density ratio , @xmath5 , and speed , @xmath8 , into one parameter . it is traditional in studies such as this to consider a single , convenient value for each of these , and to discuss the evolution as though it depended only on @xmath20 and not on @xmath5 , for example . we will follow this procedure , but also comment on its limits . in addition we set @xmath21 . we are left , then , with a two - dimensional parameter space to explore ; namely , we need to define the strength and the orientation of the magnetic field relative to the bullet motion in the plane . once set in motion , the bullet history begins through formation of a shock pair ; namely , a strong bow shock that surrounds the bullet and an equally strong , internal or `` bullet shock '' , that propagates through the bullet from nose to tail compressing it along the direction of motion . the time , @xmath22 , defined in eq . [ tcrush ] measures the interval required for the ram - pressure - driven bullet shock to cross the bullet diameter . note that our definition for @xmath22 contains an additional factor of two compared to that normally used in discussions of shocked clouds . before @xmath22 , the bow shock is established and flows along the bullet boundary may lead to some stripping due to small - scale kelvin - helmholtz ( k - h ) instabilities . on a timescale comparable to or exceeding @xmath22 , both k - h and rayleigh - taylor ( r - t ) instabilities with scales comparable to the size of the bullet can begin to develop . gasdynamical calculations mentioned earlier show r - t instability - induced bubbles with low density , ambient plasma penetrating through the bullet body , beginning a fragmentation process . k - h instabilities along boundaries allow small pieces of the bullet to be stripped , so eventually one expects the gas bullet to be broken into a `` mist '' and then dispersed . estimates for this `` dispersal '' time are consistent for both bullet and shocked cloud simulations . although the dispersal timescale is difficult to define uniquely , all gasdynamical simulations show significant signs of bullet disruption by about @xmath23 ( using our definition for the time unit ) . before the bullet is fully disrupted , ram pressure decelerates it substantially ( or if it is seen as a wind - swept clump , accelerates it in the direction of the wind ) . that deceleration , of course , is the origin of the r - t instability . the deceleration is considerably enhanced by lateral expansion of the bullet coming from the penetration of the bubbles induced by r - t instabilities . the simulations presented by @xcite and @xcite showed that weak magnetic fields in the ambient medium can be stretched and thus strengthened through sheared gas motions around clouds . compression can also increase field strength , but that process is generally much less significant in complex flows than stretching . @xcite concluded from their shocked cloud simulations that the fields can sometimes be amplified until @xmath24 . field enhancement through stretching takes place in three situations . @xcite emphasized the importance in their axial geometry of field lines aligned along the symmetry axis being stretched along the symmetry - axis , behind their shocked cloud , while a mach disk carries away ( conducting ) material from the cloud as it catches up to the originally incident shock in recession . @xcite emphasized in their transverse field geometry the importance of field lines being stretched around the cloud perimeter as they envelope the cloud . in addition , for either field geometry , vortices that form , especially in the immediate wake of the cloud , could stretch entrained field lines and , thus , amplify them . field amplification within vortices is complex , however , because it involves field lines that are wrapped around until they are adjacent to fields of opposite direction . that is unstable to reconnection , so that within the vortices field line reconnection will limit the field growth and effectively expel magnetic flux ( , @xcite ; @xcite ) to the vortex perimeter . for numerical simulations of ideal mhd , the resistivity that allows reconnection to take place comes from numerical truncation and the associated dissipation , so it is locally resolution dependent . that means one needs to be cautious in understanding the convergence properties of fields . we emphasize , however , that with the fully conservative methods we employ ( including mass , momentum , total energy and magnetic flux ) , the reconnective process is a necessary consequence , not an added feature . @xcite found from their shocked cloud simulations that inclusion of a dynamical magnetic field did significantly enhance the durability of their clouds , and concluded that this was primarily due to the influence of the magnetic field in inhibiting the vortices that form along the slip surfaces on the sides and rear of the cloud . although @xcite defined their initial field strength so that @xmath25 for most of the runs they did , this was based on the _ preshock _ ambient plasma . behind the incident ( mach 10 ) shock , where the interactions are taking place , @xmath26 . thus , the field interacting with the clouds is initially relatively weak . a similar consideration applies to the mhd bullet problem , especially in the cases with the field aligned to the bullet motion , since the value of @xmath27 behind the bow shock of the bullet is also increased by a factor @xmath28 . our simulations have been carried out using a code for compressible , ideal mhd based on a conservative , explicit tvd method as described in @xcite and @xcite . the scheme , utilizing an approximate mhd riemann solver , is second order accurate in both space and time and cleanly captures all the various families of mhd discontinuities . we have used the multidimensional , cartesian version of the code that was previously applied successfully to the study of the nonlinear mhd k - h instability in modestly compressible conditions ( @xcite ) , a problem that demonstrates the code s ability to follow accurately smooth but complex mhd flows . the code maintains @xmath29 to machine accuracy by employing a simple , corrective transformation at each time step ( @xcite ) utilizing an exact fourier transform relationship as a key feature to maintain both accuracy and speed in this step ( @xcite ) . the simulations are two and a half @xmath30 dimensional ; , vector fields include three components , but all fields are independent of the @xmath31-coordinate . the results presented all involve flows with @xmath32 , however . in each presented computation the bullet is placed with its diameter along the x - axis , and reflection symmetry assumed across that axis , so that only the upper half of the flow structure has been computed . we have carried out several comparison test simulations , however , using the full flow to demonstrate an exact correspondence with the more economical computations presented . the top and right boundaries of the grid are open , while the left boundary is `` flow - in '' to maintain the upstream wind conditions exactly . the bullet initially has a circular cross section with a radius @xmath33 . this , slightly noninteger value resulted from our definitions of the bullet radius in zones ( 50 ) that did not quite match onto the grid sizes that we used ( 512 ) . inside this radius the bullet density is @xmath34 with @xmath35 . a thin boundary layer with a hyperbolic tangent profile has been applied to the bullet , using a characteristic width of two zones . since the grid is cartesian , the bullet has a cylindrical form and its axis is aligned with the @xmath31-direction . the computational domain is @xmath36 $ ] , @xmath37 $ ] . at the start of each computation the bullet center is @xmath38 . each simulation begins with the bullet at rest while an ambient medium flows from left to right at speed @xmath39 , so that the sound speed in the ambient medium , @xmath40 , is unity . interaction with this `` wind '' accelerates the bullet towards the positive @xmath41-direction . in a fixed frame this would cause the bullet to be expelled from the grid on a timescale of only @xmath42 . in order to enable the simulations to follow the bullet evolution as long as possible , we have incorporated a feature that adjusts the reference frame speed at each time step so that the intersection of the bullet bow shock with the @xmath41-axis is approximately stationary on the grid . the bullet still moves slowly towards the right grid boundary as the mach number of the bow shock decreases and the stand - off distance between the shock and the bullet nose increases . but this procedure has enabled us to follow bullet evolution for at least @xmath43 and more typically @xmath44 . we have used a range of numerical resolutions . following @xcite , we have characterized them in terms of the number of zones across one bullet radius ; , @xmath45 , refers to a simulation with 50 zones across the bullet radius . since the grid is uniform the total number of zones would be @xmath46 for an @xmath45 run and @xmath47 for an @xmath48 run , for examples . to understand better the evolution of the bullet , we have included a passive , lagrangian , conserved `` tracer '' , @xmath49 , that can be termed the `` bullet fraction '' or , in the language of @xcite , the `` color '' of the fluid . this is set equal to unity for fluid initially inside the bullet @xmath50 and zero @xmath51 everywhere else . the bullet fraction is followed with a tvd advection routine identical to that utilized for mass advection in the code . this enables us to compute several useful quantities , such as the bullet mass inside the grid , @xmath52 where @xmath53 and @xmath54 represents the volume of the entire calculation domain . other related useful quantities include the mass - weighted speed of the bullet , @xmath55 and the mean height of the bullet in the @xmath56-direction , @xmath57 we will present @xmath58 as measured in the initial rest frame of the bullet . at the start @xmath59 . another useful integrated quantity is the magnetic energy enhancement inside the grid . that is conveniently normalized in terms of the magnetic energy initially inside the bullet , @xmath60 , where we set @xmath61 , and @xmath62 . then using the initial total magnetic energy inside the computational domain , @xmath63 , we have @xmath64 } , \label{emag}\ ] ] where @xmath65 in our units . as mentioned earlier we have carried out simulations with two initial magnetic field geometries ; namely a field transverse to the motion of the bullet ( henceforth identified as `` t '' models ) and parallel or aligned with the bullet motion ( henceforth identified as `` a '' models ) . table [ tbl-1 ] characterizes six mhd simulation pairs with a range of field strengths for each geometry . models t1t5 and a1a5 include 50 zones across a radius ( thus , termed @xmath45 runs ) while t6 and a6 use 100 zones ( @xmath48 runs ) . three additional , control models , n1n3 , have set the magnetic field to zero . they vary @xmath5 , but are otherwise identical to the t1t5 and a1a5 models . these purely gasdynamical cases are intended to assist in identifying mhd effects on the flows that might develop even when the field is initially weak . in addition we have explored the degree to which flow behaviors depend on the density contrast , @xmath5 , enabling us to compare the current simulations with our previous ones . all of the mhd models have assumed @xmath66 , while the gasdynamical models , n1,n2 and n3 have used @xmath66 , @xmath67 and @xmath68 , respectively . the duration of each run is indicated in table [ tbl-1 ] as @xmath69 . the appendix evaluates `` convergence '' issues associated with the grid . to summarize quickly the appended discussion , we conclude that the @xmath45 simulations are well converged by such global measures as the quantities defined in eqs . [ mbull][yav ] . the magnetic energy enhancement ( eq . [ emag ] ) is reasonably well converged in the transverse field cases and in the aligned field cases at early times . however , the topology of the aligned field makes it more sensitive to the details of eddy structures that form around the bullet as it is destroyed , and those details depend on numerical resolution . so , at late times the magnetic energy enhancement is not converged in our aligned - field simulations . in association with this , it is clear that structural details of the bullets at late times , @xmath70 , do depend on resolution , especially since the perturbations in the bullet structure that eventually lead to its destruction come out of the mismatch between the bullet geometry and the grid geometry in our simulations . those aspects should serve as reminders that calculations such as these are idealized efforts to understand the physics of clouds interacting with their environments and not intended to be used as predictors of detailed structures . that point is also made stronger by the recognition that bullet evolution , particularly after @xmath71 , depends physically on the initial bullet structure , including the bullet geometry ( see @xcite ) and the density contrast ( see discussion below ) . 1 , 2 and 3 contain images that provide a summary spanning pretty well the behaviors of the mhd models we have computed before significant mass stripped from the bullets begins to leave the grid . panels ( a ) and ( b ) in each of those figures represent transverse field cases , with t2 ( @xmath72 ) in ( a ) and t5 ( @xmath73 ) in ( b ) . panels ( c ) and ( d ) represent the analogous aligned field cases , with a2 in ( c ) and a5 in ( d ) . for each model two times are shown : @xmath74 and @xmath75 . 1 illustrates evolutionary aspects of the log gas density , fig . 2 shows the magnetic field lines and fig . 3 the vorticity . in the appendix , figs . a1 and a2 show log density and field lines at @xmath76 and @xmath77 for models t5 and a5 as well as their higher resolution representations , t6 and a6 . for comparison , 4 illustrates the log density distributions for the gasdynamical models n1n3 at @xmath71 and @xmath78 . it is convenient to begin our discussion with an outline of the evolution of bullet structures and associated magnetic fields in the cases with fields aligned to the direction of the bullet motion . those show less dramatic dynamical differences from the gasdynamical bullets discussed in @xcite , and closely resemble in some ways the aligned field simulations of shocked mhd clouds described in @xcite . several key points are clear from the figures listed above . first , in all the aligned - field cases the bullet shows signs of penetration by one or more large r - t bubbles by @xmath74 . there is little effect from the magnetic field on the bullet structure by this time . that impression extends to the strongest field cases we considered ; namely , a2 @xmath79 and a1 @xmath80 . indeed , the density distributions for a2 and a5 at this time are very similar and also hardly distinguishable from the analogous run with @xmath81 , model n1 ( see figs . 1 and 4 ) . this comparison is more quantitatively apparent in fig . 5 , which includes plots of density , @xmath82 , velocity as measured in the initial rest frame of the bullet , @xmath83 , gas pressure , @xmath84 , and magnetic pressure , @xmath85 , along a cut just above the @xmath41-axis for the same situations as in figs . 1 , 2 , and 3 . the solid lines correspond to @xmath74 , while dotted lines illustrate quantities at @xmath86 . furthermore , the vorticity structure around the bullet is practically the same for both a2 and a5 ( fig . 3 ) providing a very good indicator that the dynamics in the two models are very similar at this time ( see , , @xcite ) . qualitatively this vorticity structure is also the same as that found in three - dimensional gasdynamical simulations of shocked clouds at comparable dynamical times ( @xcite ) . the vorticity associated with flow over the bullet is negative ( clockwise ) . we see in fig . 3 that in the stronger field case , a2 , some vorticity of the opposite sign has been generated in the bullet wake in response to magnetic tension there , however . comparing at @xmath74 model a5 in fig . 1 with model n1 shown in fig . 4 , we see no detectable difference in density distributions . the field has not had any appreciable influence on the development of r - t and k - h instabilities , since it is not strong enough anywhere along the bullet boundary to suppress them directly . k - h instabilities will be suppressed by the field along the boundary according to linear theory , if the local alfvn speed exceeds roughly the velocity difference across the boundary ( @xcite ) . since the local mach number of the flow along the bullet boundary is generally less than or about unity , the criterion for the magnetic field removing k - h instabilities is roughly @xmath87 along the bullet boundary . similarly if the alfvn wave crossing time through a bubble perturbation on the bullet edge is less than the `` buoyancy rise time '' , @xmath88 where @xmath14 is the acceleration and @xmath89 is the length scale , the magnetic field will inhibit r - t instabilities . using eq . [ ub ] ( see below ) to estimate the acceleration of the bullet , we obtain a rough criterion for the magnetic field to stabilize r - t instabilities ; namely , @xmath90 . that leads in the present context again to @xmath87 , since @xmath91 . even in the case a1 with @xmath25 , the local @xmath27 along the bullet face and sides is everywhere greater than unity . in fact , we do not find at any time in aligned field models a1a6 that the magnetic field adjacent to the bullet is ever strong enough that @xmath92 ( there is one region with @xmath93 in the wake , as we will discuss below ) . however , even a relatively weak magnetic field can play a significant role in small scale flow dynamics , as pointed out by others ( , @xcite ; @xcite ) and as we shall explain further on . by @xmath74 the states of k - h and r - t instabilities are , on the other hand , rather strongly influenced by the density contrast choice , @xmath66 . the other panels in fig . 4 illustrate how , as the density contrast is increased , the penetration of the r - t bubbles into the bullet is enhanced for fixed @xmath20 . while , in the limit of large @xmath5 , the linear growth times on a given length for both k - h and r - t instabilities scale directly with @xmath22 , both growth rates include an additional term that increases with @xmath5 when it is finite . that tendency is consistent with the behaviors illustrated in fig . 4 . as a further test , we can compare run n3 ( @xmath94 ) with very similar gasdynamic simulations reported in @xcite . those earlier simulations were carried out using a ppm gasdynamic code and were designed to study dynamical feedback on bullets from cosmic - rays accelerated in shocks associated with the bullet . however , model number 1 in that paper omitted cosmic - rays . their fig . 4 illustrates density structure at @xmath74 , showing a good correspondence between the cloud structures in the two simulations , except that our current calculations show less evidence of small scale k - h structures along the bullet boundary . even when @xmath95 the mhd tvd code does suppress the smallest scale k - h instabilities in comparison with the ppm code , because it spreads the very strong contact shear layer of the bullet edge over several more zones . k - h instabilities are suppressed on scales less than the thickness of the shear layer . as a direct consequence of the penetration of the r - t bubbles , the bullet begins appreciably to expand laterally by @xmath96 . for all the aligned field cases , the mass weighted bullet height , @xmath97 , has expanded to roughly five times or more its initial value by @xmath86 ( see fig . this characteristic behavior was also noted by others with regard to gasdynamic simulations . as illustrated in fig . a1 , before @xmath98 , our bullets develop a roughly `` c''-shaped morphology on the computed half - plane in all of the aligned field cases , a1-a5 . ( including the reflected portion below the computed space , the cloud shape would be an `` e '' . ) that `` c '' continues to open up and thin itself even until the bullet moves off the grid . as the figures show bullet mass is stripped and carried into the wake , eroding the bullet body . in response to drag forces the bullet is accelerated towards a terminal velocity , @xmath8 , as measured in the original reference frame ; that is , the bullet or its fragments should come to rest in the surrounding flow . @xcite derived a simple theoretical model for the acceleration of a cloud based on ram pressure , taking into account the lateral expansion . the drag force actually comes from the difference in the total pressure across the bullet . assuming highly supersonic motion , we can estimate in the gasdynamic case from bernoulli s equation the pressure difference on the symmetry axis to be @xmath99 , where @xmath100 is the instantaneous speed of the bullet with respect to its initial reference frame and we have set @xmath21 . applied across the full bullet this leads to the usual expression for its acceleration , @xmath101 where @xmath102 is a drag coefficient that absorbs our ignorance of details of the pressure distribution and @xmath103 is the effective bullet radius as a function of time . borrowing the notion in @xcite that the bullet expansion begins only after @xmath71 , we write @xmath104 $ ] where we can term @xmath105 an expansion coefficient ( see also @xcite , @xcite ) . if @xmath106 , drag on the bullet increases with time , enhancing its acceleration . in an mhd flow we have to account for magnetic pressure in eq . [ accel ] . for aligned cases , however , the magnetic pressure does not enter into the bernoulli equation on axis and the magnetic pressure is not significant anyway . thus we expect the bullet motion to behave pretty much as in the gasdynamic case . we will comment later on modifications to eq . [ accel ] appropriate to transverse field cases . [ accel ] can be integrated to give @xmath107}}\right ) . \label{ub}\ ] ] all the aligned field cases can be fit well with a drag coefficient , @xmath108 , and an expansion coefficient , @xmath109 . there is a weak trend visible in fig . 6 for the acceleration of the bullets to be faster when the field is smaller , corresponding to a value of @xmath105 that depends inversely on @xmath110 . that just reflects the fact that stronger aligned fields do resist lateral expansion , keeping the bullet cross section somewhat smaller . we expect the magnetic fields themselves to respond differently to the motion of the bullet for the two field geometries we have used . that is apparent even by @xmath71 as shown in fig . the evolution of the aligned field is similar to that found by @xcite for shocked clouds with aligned fields . to the front of the bullet , field lines that initially pass through the bullet are swept and then `` folded '' over the top of the bullet in a configuration that is unstable to the resistive tearing mode . that instability leads to `` magnetic islands '' within and behind the bullet as seen at @xmath71 in fig . a2 c and d or at @xmath74 in fig . 2 c and d. the compact magnetic island closest to the axis coincides with the strong vortex at the rear of the bullet ( fig . magnetic flux initially formed into these islands is mostly annihilated by about @xmath98 ( or more properly `` expelled '' , since the total magnetic flux through the computational box does not change over time for the aligned field configuration ) . magnetic reconnection also takes place inside the bullet in consequence of the circulation developing after the bullet shock has exited ( see figs . 2 , 3 , and a2 ) . that significantly reduces the field strength within the bullet over time . through these reconnection events , magnetic flux is separated into two elements : that which passes around the bullet without reversal and that which passes through the bullet . the latter flux element continues to be involved with vortices around the bullet and subjected to reversals and reconnection . before the magnetic flux separation , there is a thin region of strongly concentrated field formed along the axis behind the bullet , analogous to the `` flux rope '' discussed for shocked clouds in some detail by @xcite . in our situation , as in theirs , this feature forms as a consequence of compression of field into the low pressure wake behind the bullet , followed by field - line stretching . in our case those field lines are temporarily anchored in the bullet on one end and drawn out to the rear by the expanding rarefaction wave ( see figs . a1 and a2 ) . just as for the shocked cloud case considered by @xcite , the magnetic field in this region can become locally dominant with @xmath111 . however , as fig . 2 shows clearly , the field configuration on the edge of the flux rope is susceptible to tearing mode instabilities , so that the rear flux rope disappears as part of the flux separation event . we do _ not _ see in any of our aligned field simulations that @xmath24 , nor that this flux rope plays a significant dynamical role in the evolution of the bullet . however , this feature is the dominant source of enhanced magnetic field energy before its flux is expelled ( see fig . 6 ) . the normalized magnetic energy enhancement , @xmath112 , ( eq . [ emag ] ) peaks at values between @xmath113 and @xmath68 near @xmath114 ( see fig . 6 ) . the peak value in model a1 ( @xmath25 , @xmath115 ) is about half that reported by @xcite for their shocked cloud during this phase for the same numerical resolution across the cloud and the same @xmath110 . since our post - bullet flux rope is significantly thinner than theirs , the comparison seems very good . our higher resolution , @xmath48 , weak field run , a6 , differs in @xmath112 by @xmath116% from the analogous low resolution , @xmath45 , a5 run for @xmath117 . so , through this stage the magnetic energy behavior seems well converged . for the aligned field geometry there is very little compression of the field , so from the start we should expect that field line stretching would be the primary contributor to magnetic field enhancement . along the boundary of the bullet the field remains weak before about @xmath98 , even in the @xmath25 case , because it is subject to reconnection that shortens the field lines . however , that reconnection leads directly to the flux separation mentioned , and through that the field topology above the bullet returns to something resembling its initial form ; namely , field lines pass directly around the bullet from front to back without folding . as the bullet body expands laterally , lines above it are stretched significantly ( see fig . 2 ) , but are no longer subject to reconnection . a relatively thin flux tube especially on the rear perimeter of the bullet envelops the now - distended cloud , extending into the wake . almost independently of the initial @xmath110 or the numerical resolution , the minimum local @xmath118 in this structure . after the flux separation event these field lines do not penetrate the strong vortex at the rear of the bullet . on the other hand , the flux that now passes through the bullet is drawn into the big r - t bubble and an associated vortex pair . through reconnection , however , those field lines divide into magnetic islands ( for the weaker initial field cases ) that are annihilated and flux that penetrates directly through the bullet . some flux passing through the bullet is drawn into the strong vortex at the rear and base of the bullet , leading to another region of strong magnetic flux on its perimeter . even though the magnetic pressure in these flux tubes is never dominant , the field can still play a major role in reducing the vorticity and leading the flow to become more nearly laminar and less disruptive . that dynamical behavior was shown recently in high resolution simulations of the mhd k - h instability ( @xcite ) . there it was found in k - h unstable mhd flows that even when @xmath119 the magnetic field , acting as catalyst , realigns the flow into a stable , broad and laminar shear layer . the field acts as a catalyst in the sense that kinetic energy is temporarily stored in the field so that locally the magnetic tension is at least significant if not dominant . that stored energy is released again during reconnection , but one result of the reconnection is that the velocity and magnetic fields are more closely aligned ; that is , the magnitude of the cross helicity is increased . the smoothing and spreading of the flow should significantly reduce stripping from the bullet . this conclusion is consistent with that reached by @xcite , that the magnetic field reduces the intensity of vortices around their clouds and , thus , increases the cloud survivability . the presence of the magnetic field is clearly felt by the bullets in our aligned field simulations . clouds in the stronger field cases are less distended ( figs . 1 , 2 , and 6 ) and there is an apparently stable density concentration at the top of the bullet `` c '' . examination of the forces applied there shows that the density is mostly confined by ram pressure , but that magnetic pressure and tension contribute at the 10% level , as anticipated from earlier discussion . at late times in these cases material no longer seems to be stripped from the bullets ( see fig . 6 ) , consistent with our comments above . for the stronger field cases , there is less thinning of the main bullet body , as well . magnetic energy enhancement in the aligned field models begins to rise again after @xmath120 , mostly in response to the development of flux tubes around the bullet and in the associated vortices . our simulations are not able to capture this final rise completely , because the flux tube fine structures are still resolution dependent ( see figs . a2 and a4 ) and because significant magnetic structures begin to leave the grid after @xmath121 . it appears likely , however , that the excess magnetic energy becomes at least comparable in this stage to the peak value noted earlier . @xcite similarly emphasized the lack of convergence in the magnetic field within their simulations . they also used an aligned field geometry , so the issues responsible were analogous . although the details are fairly complex , the summary of aligned - field bullet evolution is straightforward . magnetic fields are initially swept over the bullet , stretched and folded there . eventually reconnection separates the field into flux passing directly through the bullet and flux passing directly around it . except within a thin flux tube formed temporarily along the symmetry axis , there is no place that the magnetic field becomes energetically dominant ; that is , almost everywhere @xmath122 . in the meantime the disruptive , ram pressure force that applies to gasdynamical bullets ( and gasdynamical shocked clouds ) causes the initial cloud to become distended and stripped . magnetic field stretched over the top of the bullet can have a significant stabilizing influence that should prolong bullet coherence even though magnetic energy never becomes dominant . for the transverse cases where the initial field crosses the bullet s path , field lines are also swept and stretched around the bullet . however in this case the field lines do not reconnect around the bullet , except in a `` magnetotail '' along the axis , so that a region of high magnetic pressure develops on the bullet nose as shown in fig . 2 . even in the t5 case with @xmath73 , the field is strong enough that @xmath123 on the bullet nose by @xmath74 . for all the other transverse field cases we computed , @xmath87 along the nose of the bullet by this time . in models t1 and t2 , @xmath124 here and @xmath24 directly on the leading edge of the bullet . except in t5 , @xmath125 along the top and towards the rear of the bullet boundary , partly because the gas pressure is low , but mostly because field lines have been greatly stretched . that characterization of the field lines is obvious in fig . 2 or fig . a2 . even though field lines are compressed by the bow shock , it is stretching , not compression , that is important to the evolution of the magnetic field and to its eventual dynamical role . that supports the expectations expressed by @xcite , based on passive fields . one can see in fig . 6 that the normalized magnetic energy enhancement for the transverse geometry grows to more than @xmath126 for the weakest field cases and more than @xmath127 even for the strongest field cases . since the volume containing highly compressed plasma is never more than a few times the initial bullet volume , the maximum field enhancement through compression would be @xmath128 , emphasizing the importance of stretching . we note that the magnetic energy in the transverse cases seems fairly well converged through the full duration of the simulations and better than in the aligned field geometry . that makes sense , since the transverse fields are not so closely tied to vortical flows behind the bullet . we can write down a crude model for the enhancement of the magnetic field energy in the transverse cases that seems to qualitatively account for what is seen . for ideal mhd the magnetic induction equation can be expressed in a form @xmath129}{b^2 } } , \label{dbdt}\ ] ] where @xmath130 is the convective derivative . at the stagnation point on the bullet nose we can estimate @xmath131 and @xmath132 \sim b^2 u_{b0}/r$ ] , so that @xmath133 where @xmath134 is a `` fudge factor '' accounting for various details . if this field growth occurs within a `` shield '' of volume @xmath135 , we can estimate the associated magnetic energy enhancement to be @xmath136 that growth seems consistent with the results shown in fig . 6 , for @xmath137 . since the field on the bullet nose saturates with @xmath138 , we expect a maximum for @xmath139 . that is also roughly consistent with the results in fig . the early magnetic energy enhancement is similar for all the values of @xmath110 , but it saturates sooner and the dynamical influence of the field is sooner when @xmath110 is smaller . the field lines draped over the bullet are also drawn down near the symmetry axis into the bullet wake , producing a region that resembles the flux rope seen in the aligned field cases . again , this is a region of @xmath140 , but primarily because the gas pressure is very low from the initial evacuation of this region . it is always the case in our simulations that @xmath141 within this feature . in addition , the @xmath142 boundary of our grid separates field lines of opposite direction ; , there is a current sheet there , much as in the earth s magnetotail . that is also unstable to tearing mode instabilities that limit fields in this region . like the earth s magnetosphere does , the penetration of the bullet through a `` quasi - transverse '' field should produce on the bullet nose an induced electric field perpendicular to both the magnetic field and the bullet velocity . in this case that projects out of the computational plane . if the third dimension of the bullet @xmath143 , then the induced potential across the bullet would be @xmath144 . taking numbers that might be appropriate for knots in a young snr like cas a ( @xmath145 gauss , @xmath146 c , @xmath147 cm ) ( , @xcite ) , we find that @xmath148 volt is possible . supposing that reconnection within the bullet magnetotail can generate electric fields aligned with the magnetic fields , then these regions might be important sites for nonthermal particle acceleration . for the weak transverse field cases , the bullet dynamics initially resembles the aligned field and gasdynamical cases to the first approximation . the r - t bubbles seen in those other situations form here , too . in fact at @xmath74 there are no obvious differences in either bullet morphology or flow dynamics between models a5 and t5 , for example ( see figs . 1 and 3 ) . as time progresses , however , the bullet in all transverse field cases becomes significantly influenced by the field . even though the same `` c''-shaped bullet morphology is seen in the two weakest field cases , t4 and t5 , field stretched over the top and back of the `` c''-shaped cloud develops magnetic pressures significantly greater than the gas pressure , so that @xmath149 along the bullet edge . this condition protects the bullet from further disruption . in model t5 that happens quite late , so as late as @xmath86 the bullet morphology is pretty similar to a5 . however , even in model t4 ( @xmath150 ) the field becomes dynamically strong ( @xmath151 along the bullet nose just after @xmath74 ) . that case still expands laterally , but not so fast as in the cases with negligible magnetic influence ( see fig . for all the other transverse cases with @xmath152 , the maxwell stresses grow large enough to completely inhibit lateral expansion of the bullet . the bullets are then enshrouded by a strong magnetic shield . for cases involving an initially weak field , the bullet is partially disrupted , but eventually the shield protects the r - t fingers from further erosion . the bullet embedded in a stronger field is compressed , but then , as the field confines it , develops a streamlined profile and is not strongly eroded . remarkably , fig . 6 shows that the acceleration of the transverse - field bullet increases with the strength of the magnetic field . this trend is reversed from that of aligned - field bullets . there we have observed that the tendency of stronger field to resist lateral expansion reduces the evolution to a greater cross section . it was that increased cross section and the augmented drag that were modified by the field . for the transverse field models , lateral expansion can be halted entirely , yet the drag force clearly is enhanced with the stronger field . this seeming paradox is easily explained by accounting properly for the role of the magnetic pressure in the drag . whereas the aligned field did not contribute directly to the force across the bullet , the transverse field does . the mhd bernoulli equation on the symmetry axis gives the result in the highly supersonic limit that @xmath153 , where @xmath154 on the nose of the bullet in this case . as already noted , @xmath85 increases over the first few crushing times and can become comparable to @xmath155 in that region . although the expansion coefficient @xmath156 in eqs . [ accel ] and [ ub ] the drag coefficient , @xmath102 , is an effectively increasing function of time and that enhances the bullet s acceleration . we find , for example , that eq . [ ub ] gives a good fit to the motion of the strongest field case , t1 , by setting @xmath156 , replacing @xmath102 with @xmath157 , and using the same @xmath158 that we have used for aligned field models . that corresponds to an increase in drag of about 5% at the end of the simulation . thus , the picture that develops for the transverse field bullets is rather different from and considerably simpler than the ones for aligned field . magnetic field is swept around the bullet to form a protective shield . the magnetic pressure becomes comparable to the ram pressure through the bullet bow shock ; thus , the magnetic energy is enhanced by an amount approaching @xmath159 . again taking that to be roughly the volume of the bullet , we recover the observed magnetic energy enhancements with the expression @xmath160 . our bullets have a cylindrical form , so that field lines can not be swept around the sides of the bullet in the @xmath31-direction . such sweeping may reduce the magnetic field enhancement somewhat in a three - dimensional bullet . but we still expect the same qualitative behavior , since much of the field that forms into the shield comes from the field lines that penetrate into the `` skin '' of the bullet , and since irregularities in the spherical bullet would capture field lines much like the cylindrical one does . the brief summary of transverse bullet evolution is the following . even when the initial field is of modest strength measured in terms of the ratio of magnetic pressure to background plasma pressure , the stretched magnetic field effectively confines the bullet and prevents its fragmentation . further , the magnetic pressure applied to the nose of the bullet increases the rate at which the bullet is brought to rest with respect to the ambient medium ; more effectively , in fact , than the lateral expansion that accompanies the acceleration of gasdynamical bullets . we have carried out an extensive set of two - dimensional mhd simulations exploring the role of magnetic field in the dynamics of supersonic clumps of plasma . we have examined the influence of both field strength and orientation on the problem . of those two characteristics field orientation is far more important . even a very modest ambient magnetic field that crosses the path of the bullet tends to be amplified by field line stretching around the bullet until the maxwell stresses become comparable to the ram pressure associated with the bullet motion . a field that is aligned with the bullet motion , on the other hand , develops reconnection - prone topologies that shorten the stretched field and release the excess energy it contains . the field is also swept around the bullet in this geometry and can temporarily become moderately strong . however , the energy in the field is not enhanced so much for this geometry as for the transverse geometry . rather , as a consequence of reconnection , there is a transformation and relaxation of the field over a few bullet crushing times . some magnetic flux passes directly through the bullet , where it can become entrained in vortices and amplified around the vortex perimeter . the remaining flux passes directly around the bullet , where it can act to resist lateral expansion as it is stretched by that expansion . in this geometry , however , the maxwell stresses on the bullet never approach the ram pressure level . even so , a field of even moderate initial strength becomes strong enough to help realign the flow around the bullet into a smoother , more laminar form that reduces the tendency of the bullet to fragment . both of the above field geometries are highly idealized , and more generally the bullet would encounter a field at some oblique angle to its motion . in that situation , we will want to know which of the two special cases is more relevant . the most important detail should be if , as the field lines are swept over the bullet , they are folded over the top as well , thus leading to reconnection . conceptually that depends on the relative rates at which the field lines are swept past the bullet body , on the one hand , and at which the `` foot - points '' of the field lines along the bow shock move past the bullet , on the other hand . if the field - line bow - shock foot - point moves faster ( as it must for transverse field cases ) , the lines are not folded over the top , but stretched directly to the bow shock . if , however , the field - line bow - shock foot - point moves slower ( it is at rest for aligned field cases ) , then the field lines are folded over inside the bow shock on one side of the bullet and likely to reconnect . we can derive an approximate expression for the condition that the field foot - point moves downstream faster than the flow around the bullet , if we take the mach bow shock to be a simple cone with half angle @xmath161 and suppose that the flow speed around the bullet is @xmath162 . defining the motion of the foot - point as the translation of the intersection between the field line and the bow shock , this leads to the constraint @xmath163 where @xmath1 is the angle between the bullet motion and the ambient magnetic field . in the limit @xmath18 , with @xmath164 , this becomes @xmath165 so , we conclude that for supersonic bullets most field directions will lead to behaviors similar to the transverse field cases , while as the motion becomes transsonic the dividing line would be closer to 45 . except in circumstances with preferential alignments between the motion and the field , that would lead to the further conclusion that even a weak magnetic field will have a substantial impact on the evolution and dynamics of supersonic clumps of ionized , conducting gas . in young supernova remnants , polarized radio synchrotron emission indicates that there is a net radial direction to the magnetic field inside the remnants ( @xcite ) . this might suggest that bullets in young remnants would generally encounter a radial field and , hence , one nearly aligned with their motions . that view is somewhat simplistic , however , because the same radio observations also show a very small net polarization ( @xcite ) indicating that , to first order , the field is disordered . that , in turn , argues that a wide range of field orientations may be encountered by small projectiles . in either limiting field geometry and presumably those in between , field line stretching is the dominant process for magnetic field amplification . that supports the conclusions reached by @xcite and @xcite that nonthermal radio emission associated with supersonic clumps in supernova remnants , for example , is likely to be largely controlled by the generation of stretched magnetic fields around the perimeters of the clumps , rather than primarily highlighting the bow shock where field is mostly enhanced by compression . we note that the structures of the bow shock and the bullet boundary may even superficially resemble each other in observations , so that morphology alone can be misleading . the physical difference is important , however , since the bow shock is probably the site of local particle acceleration , while on the bullet perimeter one sees primarily energetic particles that come from some other site ( possibly including the bow shock , of course ) . our estimates of the field strengths expected in the two situations are very different , however . so , observational estimates of the local magnetic field ( based on equipartition , for example ) would possibly lead us to very different conclusions about the local conditions . in addition , since the magnetic pressure on the nose of the bullet may become comparable to the ram pressure and hence the total pressure behind the bow shock , the gas pressure there could be substantially lower than that in a gasdynamical bullet . that means , as well , that the temperature in the region on the nose of the bullet would be lower than that predicted in the gasdynamical case . that detail can alter expectations of the thermal emission , including x - rays and uv - ir lines . in summary , the role of a magnetic field in the evolution and in the appearance of supersonic clumps is very important , even if the magnetic field is nominally not strong in the ambient medium . the importance comes because magnetic field lines can be stretched and amplified if they become draped around the bullet perimeter . that effect seems especially strong when the field lines are `` quasi - transverse '' to the motion , a concept that depends on the speed of the bullet , but seems to include most directions . additional amplification within vortices associated with the destruction of the bullets can also occur , but seems to be less important . if the fields are quasi - transverse , then they can effectively confine the bullet and prevent its disruption . the same amplified fields may be important to emissions used to analyze the bullets . they can control the radio synchrotron emission expected and possibly become strong enough to alter the local thermodynamics of the gas and influence thermal emissions , as well . we are grateful to adam frank , byung - il jun , bob lysak and larry rudnick for fruitful discussions and to mordecai - mark mac low for helpful comments on the manuscript . at the university of minnesota this work was supported in part by nsf ( ast-9318959 ) , nasa ( nagw-2548 ) , and the minnesota supercomputer institute . at chungnam national university this work was supported in part by the basic science research institute program , korean ministry of education 1995 , project no . bsri-95 - 5408 . we have carried out several experiments to understand the influence of finite numerical resolution on our simulations . there are two issues . first , we are nominally simulating ideal mhd , but depend on the existence of a small , albeit finite , dissipation on the grid cell scale to simulate the physical viscous and resistive dissipation that certainly takes place on very small scales . the existence of the numerical dissipation is necessary , for example , to allow shocks to form and magnetic reconnection to occur . there is fairly good evidence that conservative monotonic schemes such as ours do a good job of approximately representing physical viscous and resistive dissipative processes presumed to exist on scales somewhat smaller than the grid ( , @xcite ) . for the astrophysical environments being simulated , the dissipative scales are probably very much smaller than those that can be modeled directly , however . thus , the numerical solutions , if they are complex , will generally differ to some degree from the asymptotic `` physical '' solution on the smallest scales after long periods of time . one hopes , on larger scales and over moderate periods of time , that the behaviors will be converged for important characteristics . our code is second - order accurate in smooth flows , so that the effective reynolds numbers increase with length , @xmath166 , approximately as @xmath167 where @xmath168 is the size of a computational zone ( @xcite ) . for moderately strong magnetic fields ( @xmath169 ) , @xmath170 in our code . for much weaker fields the dissipation is several times less on a given scale . in any case , we expect that dissipation would be confined to scales of only a few zones and that _ inertial _ structures formed within scales of less than @xmath171 zones would have histories that are resolution - dependent . the second , related issue has to do with the fact that the late history of a bullet depends on the nonlinear evolution of instabilities that in our simulations form off of the initial surface of the bullet . consequently that history can depend to some degree on the exact structure of the initial perturbation in a way that extends beyond the size scale of the perturbation . most of our simulations ( t1t5 and a1a5 ) have been conducted using a grid placing 50 zones across the bullet radius ( @xmath45 ) and with the bullet centered in the middle of a zone along the bottom axis @xmath172 . to evaluate the importance of the zone size on our calculations , we have carried out simulations identical to t5 and a5 except that 100 zones spanned the bullet radius ( @xmath48 ; t6 and a6 ) . in addition , we have performed a number of test runs with varying resolution using this code but setting the magnetic field strength to zero . in some of those , we have simulated the impact of a mach 10 shock on the cloud . the intent there was to repeat the resolution tests shown in @xcite . since they used cylindrical geometry , and we have used cartesian geometry , an exact comparison is not meaningful . but we conclude that our mhd code produces resolution dependencies very comparable to those they displayed as computed from their gasdynamical code . some simple illustrations make plain the limitations imposed by our grid . a1 compares the density structures computed for models t5 and t6 ( top panels ) as well as a5 and a6 ( bottom panels ) . results in each case are shown at two times , @xmath173 and @xmath174 , with the earlier time displayed above the later time . a2 displays the magnetic field structures for the same cases . @xmath45 runs , t5 and a5 , are shown on the left and @xmath48 runs , t6 and a6 , on the right . these have the weakest initial fields of any of the simulations we have run in this study . they are , thus , the cases where we would expect the greatest dependence on numerical resolution , since a strong magnetic field has a tendency to produce smoother flows and more organized field structures . at @xmath173 the paired runs agree almost exactly for both field configurations . at @xmath98 the structural agreement is qualitatively consistent , but there are clear differences . especially in the magnetic field lines shown in fig . a2 , there is additional fine structure in the higher resolution runs . also the field along the leading edge of the bullet is somewhat stronger in the higher resolution runs . both of these features are anticipated , since they come directly from the reduced dissipation scale in the @xmath48 runs . most apparent , however , is that the instabilities that begin to destroy the bullet have a somewhat different nonlinear development ; , the shapes of the bullets are beginning to differ . as it turns out , that result comes at least as much from the fact that the form of the initial perturbations on the bullet is slightly different as a consequence of the different resolutions as from the differences in dissipation scales . the reason is because our initial perturbation results from the mismatch between a circular cross section and a cartesian grid . thus , either a change in the number of zones spanning the bullet or a shift in the bullet center will alter the locations of those irregularities . to make that point clear , we compare in fig . a3 the density distributions for two @xmath45 runs at @xmath175 and @xmath176 . the one shown on the left is a5 , while the one on the right is identical to a5 _ except for one small detail _ ; namely , that the initial bullet center is shifted @xmath177 zone to the right from a zone center to a zone edge along the bottom grid boundary . the differences between these two runs are almost as great as those between the two different resolutions , and , in particular , the degree of the changes in bullet shape is comparable . this emphasizes that , once the instability becomes nonlinear , the precise structure becomes sensitive to exact details of the initial conditions . considering these comparisons , one should concentrate on more global characteristics than on specific structural features in simulations like those in this paper . in that global sense , the model comparisons are much closer . to demonstrate this , fig . a4 compares the evolution of the bullet mass , speed , and height as well as the normalized magnetic energy enhancement in the computational domain for each resolution pair ( t5,t6 and a5,a6 ) . in each pairing , these important dynamical quantities agree very well . for example , the magnetic energy enhancement in the two transverse field runs approaches @xmath178 by @xmath98 . the two runs agree within about 10% on that figure . in the aligned field pair , the enhancement is much smaller and @xmath112 peaks between 50 and 60 for the two resolutions with about @xmath179 more enhancement in the @xmath48 run . those differences reflect , on the one hand , that for the transverse field cases the magnetic field is more effectively stretched , but , on the other hand , that for the aligned field cases the field interacts much more with vortices . vortex structure and magnetic field behavior there are fairly sensitive to numerical resolution . overall , it is clear that run pairs are largely consistent with each other . slightly less stripped bullet material has left the grid in each of the high resolution runs at the last time shown , but is less than 1% of the bullet mass , in any case . ccccccc t1 ( a1 ) & @xmath45 & 1 & 9.13&10 & 0.6 & 5 @xmath22t2 ( a2 ) & @xmath45 & 4 & 18.3 & 10 & 0.6 & 8 @xmath22t3 ( a3 ) & @xmath45 & 16 & 36.5 & 10 & 0.6 & 8 @xmath22t4 ( a4 ) & @xmath45 & 64 & 73.0 & 10 & 0.6 & 8 @xmath22t5 ( a5 ) & @xmath45 & 256 & 146 & 10 & 0.6 & 8 @xmath22t6 ( a6 ) & @xmath48 & 256 & 146 & 10 & 0.6 & 4 @xmath22n1 & @xmath45 & @xmath180 & @xmath180 & 10 & 0.6 & 2 @xmath22n2 & @xmath45 & @xmath180 & @xmath180 & 40 & 1.2 & 2 @xmath22n3 & @xmath45 & @xmath180 & @xmath180 & 100 & 2.0 & 2 @xmath22 fig . 1 : : grayscale images of the log density for models t2 ( a ) , t5 ( b ) , a2 ( c ) , and a5 ( d ) at two times ( @xmath181 ) ( top and bottom , respectively ) . this shows bullet evolution for strong and weak magnetic fields ( @xmath182 ) and for both transverse and aligned magnetic field geometries . values increase from dark tones to high tones . fig . 2 : : same as fig . 1 , except showing the magnetic field lines ( contours of the magnetic flux function ) . about 40 equal - interval contours of flux function are shown in each frame . fig . 3 : : same as fig . 1 , except showing the grayscale images of the vorticity . dark tones are negative vorticity while high tones are positive . fig . 4 : : grayscale images of the log density for gasdynamical bullets ( @xmath183 ) . the three runs shown have @xmath66 , ( a ) ; @xmath184 , ( b ) ; and @xmath94 , ( c ) . otherwise , the simulations are identical to mhd @xmath45 simulations . each model is shown at @xmath71 ( top ) and @xmath74 ( bottom ) . fig . 5 : : cuts along @xmath185 ( or the third cell from the bottom axis with @xmath186 ) for models t2 ( a ) , t5 ( b ) , a2 ( c ) , and a5 ( d ) at the same time as in fig . 1 . solid lines correspond to @xmath74 , while dotted lines represent profiles at @xmath86 . shown top to bottom are : log gas density , @xmath82 , velocity as measured in the initial rest frame of the bullet , @xmath83 , log gas pressure , @xmath84 , and log magnetic pressure , @xmath85 . nd log magnetic energy enhancement , @xmath112 ( eq . [ emag ] ) . fig . 6 : : evolution of various integrated quantities . panel ( a ) presents values for models with transverse magnetic fields ( t1t5 ) , while panel ( b ) for models with aligned fields ( a1a5 ) . the line types distinguish models of different initial field strength : solid ( @xmath25 ) , long dash ( @xmath72 ) , short dash ( @xmath187 ) , dot - dash ( @xmath150 ) , and dot ( @xmath73 ) , with @xmath188 . shown top to bottom are : total bullet mass inside the grid , @xmath189 ( eq . [ mbull ] ) , mass - weighted bullet speed as measured in the initial rest frame of the bullet , @xmath58 ( eq . [ uxav ] ) , mean height of the bullet material , @xmath97 ( eq . [ yav ] ) , and log magnetic energy enhancement , @xmath112 ( eq . [ emag ] ) . fig . a1 : : grayscale images of the log density comparing @xmath45 with @xmath48 runs for both magnetic field geometries when @xmath73 . models t5 ( a ) , t6 ( b ) , a5 ( c ) and a6 ( d ) are shown at @xmath173 and @xmath77 . a2 : : magnetic field lines for the same situations as fig . a1 . about 40 lines are shown in each frame . a3 : : grayscale images of the log density showing sensitivity of bullet evolution to the initial details along the symmetry axis @xmath172 . panel ( a ) shows model a5 and panel ( b ) an identical simulation with the initial bullet center shifted along the @xmath41-axis @xmath177 zone . times shown are @xmath175 ( top ) and @xmath190 ( bottom ) . a4 : : same quantities as fig . 6 , except now comparing models with different resolution , t5 ( dotted lines ) and t6 ( solid lines ) in ( a ) and a5 ( dotted lines ) and a6 ( solid lines ) in ( b ) .

aharonov and casher showed that pauli hamiltonians in two dimensions have gapless zero modes . we study the adiabatic evolution of these modes under the slow motion of fluxons with fluxes . the positions , , of the fluxons are viewed as controls . we are interested in the holonomies associated with closed paths in the space of controls . the holonomies can sometimes be abelian , but in general are not . they can sometimes be topological , but in general are not . we analyze some of the special cases and some of the general ones . our most interesting results concern the cases where holonomy turns out to be topological which is the case when all the fluxons are subcritical , , and the number of zero modes is . if it is also non - abelian . in the special case that the fluxons carry identical fluxes the resulting anyons satisfy the burau representations of the braid group .

1403.2216

the pauli hamiltonian describes a non - relativistic electron with gyromagnetic constant @xmath6 @xmath7 @xmath8 is the vector of pauli matrices and @xmath9 acts on spinors . we use units where @xmath10 . the electric and magnetic fields are determined by the 4-potential @xmath11 : [ fields ] = , = - + a_0 in 1979 aharonov and casher @xcite observed that the pauli operator for static magnetic field in two dimensions , @xmath12 , so @xmath13 , has ( normalizable ) zero energy modes . the zero modes turn into gapped bound states . ] they are gapless ground states and their number @xmath14 , is determined by the total magnetic flux @xmath15 measured in units of quantum flux , [ zero - modes ] d= ceiling of @xmath16 , i.e. the smallest integer @xmath17 . , the total number of zero modes is 6 . since all the fluxons are critical no mode is localized on any one fluxon . if the two clusters are well separated one can choose two of the zero mode to be localized on the cluster of three fluxons and another three modes to be localized on the cluster of four fluxons . the remaining 6-th mode is supported on the two clusters no matter how far they are separated . , width=6 ] we consider a magnetic field @xmath18 localized on a finite number of disjoint fluxons labeled by @xmath19 . the magnetic flux of the @xmath20-th fluxon , @xmath21 , is localized in a region of radius @xmath22 centered at @xmath23 . we _ do not _ assume that @xmath21 is quantized or that all the fluxes @xmath21 are identical . we shall assume w.l.o.g . that @xmath24 . we say that the @xmath20-th fluxon is super - critical if @xmath25 , subcritical if @xmath3 and critical if @xmath26 . the fluxons are viewed as classical parameters and _ not _ as dynamical degrees of freedom : they _ do not _ have a wave function or an equation of motion . ( the dynamical degree of freedom is the electron wave function . ) when the @xmath20-th fluxon is super - critical it can create @xmath27 zero modes which are _ confined _ to it , in the sense that their wave function decays ( as a power law ) over a typical distance @xmath28 determined by the fluxon radius @xmath22 . more interesting are the zero modes which are bound jointly by a number of separate fluxons . we shall call these solutions _ free zero modes_. these states wave functions live in between the fluxons and have typical size determined by the inter - distance @xmath29 . when _ all _ the fluxons are subcritical , @xmath30 _ all _ the zero modes are free : the probability of finding the charge on any of the fluxons is close to zero ( as @xmath31 ) . in general , confined and free modes coexist . the confined modes behave like the charge - flux composites one encounters in the fractional quantum hall effect @xcite , except that here the charge is quantized but the flux is not whereas in the hall effect it is the flux that is quantized and the charge is not . the free modes are a different kettle of fish as the composite involves a single electron jointly bound by several fluxons . as we shall see , these modes can sometimes turn the fluxons into non - abelian anyons @xcite . these new `` topological '' objects are quite interesting . the distinction between confined and free zero modes is meaningful when the radius of the individual fluxons , @xmath22 , is the smallest length scale in the problem , @xmath32 and is sharp for point - like fluxons . the total number of free modes @xmath33 is , as we shall see , [ extended ] 0d_f= max\{0,_a _ a-1}n-1 , _a=_a - d_a we say that the number of free modes is maximal if @xmath34 . this turns out to be the case where the fluxons become non - abelian anyons . if all the fluxons are identical then to have maximal number of free modes leading to interesting representation of the braid group requires [ con ] 1 - 1 n < _ a < 1 ( the case @xmath35 leads to a trivial representation of the pure braid group . ) to study the holonomy of the zero modes we treat the fluxon coordinates , @xmath23 as ( classical ) adiabatic controls . the adiabatic theory we shall need and describe is of interest in its own right , since the weak electric fields generated by the slow motion of the fluxons are important for the adiabatic transport and since the zero modes are gapless ( see sec . [ sec : ad ] for more details ) . adiabaticity means that the characteristic time scale of the controls is large compared with the characteristic time scale of the system . we shall argue that the characteristic time scale in the case of point - like fluxons is set by their mutual distances . this means that points in control space where fluxons collide must be removed : fluxon collisions is like gap closures in gapped systems . both endow control space with an interesting topology which is sine qua non for interesting topological behaviour . the holonomies of braiding point - like fluxes are summarized in : * the berry phase associated with the confined mode on the @xmath20-th super - critical fluxon braided by the fluxon @xmath36 is the aharonov - bohm phase @xmath37 . * the berry phase for a non - degenerate free mode , ( @xmath38 ) , and two fluxons ( @xmath39 ) is _ topological _ ( path independent ) given by @xmath40 . * the berry phase for a single free mode , ( @xmath38 ) and @xmath41 fluxons is abelian but path _ dependent_. in other words , the adiabatic curvature is non trivial , see fig . [ fig : curvature ] . * for @xmath42 and maximal number of free modes , @xmath34 the holonomy is non - abelian and _ topological_. braiding anyon @xmath20 with anyon @xmath43 is associated with the monodromy matrix , _ a = e^-2i_a the eigenvalues of the holonomy matrix are @xmath44 . this is our most significant result . * if , in addition to @xmath34 , all the fluxons carry identical fluxes , then exchanging them make sense and is described by matrices that give the burau representation of the full braid group @xcite : , the fluxons are identical anyons . like ordinary anyons @xcite they have topological braiding and fusion rules . ( the fusion rules are simply flux addition . ) but , unlike ordinary anyons , they are gapless and hence fragile . * @xmath41 and @xmath45 : the holonomy is non - abelian and , in general , path dependent i.e. not topological . a key to aharonov - casher ( ac ) observation is the fact that when @xmath46 and @xmath6 the pauli hamiltonian is a prefect square @xmath47 since @xmath48 the zero modes , if any , are ground states and are the normalizable solutions of ( -i - e)=0 where @xmath49 is a two component spinor . the second key observation is special to two dimensions . it is convenient then to use complex notation and @xmath50 z = x+iy,2_z=_x -i _ y , 2a_z = a_x - ia_y , one then has ( -i - e ) = -2i ( cc 0 & _ z - iea_z + ) = q+q^ * since @xmath51 the pauli hamiltonian is super - symmetric @xcite : @xmath52 the zero - modes then lie in @xmath53 , i.e. they are either spin up states that lie in the kernel of @xmath54 , or spin down states in the kernel of @xmath55 . for the zero - modes with spin up : [ kernel1 ] = ( c _ + 0 ) , ( |_z - ie|a_z)_=0 let us first look for a solution that does not vanish anywhere , so @xmath56 is well defined . we shall call this a fundamental solution and denote it by @xmath57 . it is given by [ alpha ] |_z _ 0=ie|a_z _ z|_z _ 0=ie_z |a_z using 4_z|_z= , 4_z|a_z = ` div ` + i ` curl ` it follows that @xmath58 is a solution of poisson s equation whose source term is determined by @xmath59 [ poisson ] _ 0 = -eb + ie consequently , a unique choice of @xmath58 is made by means of the poisson kernel : ^-1(z , z)= 1 2 |z - z| by elliptic regularity @xmath58 is at least as regular as @xmath60 . in the coulomb gauge @xmath61 . consequently @xmath58 is real and the fundamental solution is positive . clearly [ alphapk ] _ 0 -_t|z| with @xmath15 the total magnetic flux . since @xmath62 is gauge invariant the fundamental solution ( in any gauge ) decays polynomially : | _ 0| |z|^-_t the fundamental solution is square integrable iff @xmath63 . similarly , the spin down fundamental solution decays at infinity if @xmath64 and is square integrable iff @xmath65 . we shall assume from now on that @xmath24 and consider only spin up zero modes . now with @xmath66 any holomorphic , [ yr7 ] _ = p(z)_0 _ ker(|_z - ie|a_z ) @xmath66 can not have poles , since this conflicts with ( local ) square integrability of @xmath67 ( and the regularity of @xmath58 ) . @xmath66 must therefore be a polynomial . @xmath67 is square integrable provided @xmath68 it follows that there are @xmath14 zero modes with @xmath14 given by eq . ( [ zero - modes ] ) . these results of aharonov and casher ( ac ) @xcite may be viewed as an example of an index theorem for non - compact manifold @xcite . we now turn to the study of static disjoint fluxons with fluxes @xmath21 localized inside discs of radius @xmath22 centered at @xmath23 . we shall denote @xmath69 and @xmath70 . an interesting and very useful feature of the ac zero modes , which follows from eq . ( [ poisson ] ) , is the ` superposition ' property : has no simple relation to the normalization of the factors @xmath71 . ] the fundamental solution for @xmath0 fluxons is the product of the single fluxon fundamental solutions : [ superposition ] _ 0=_a ( _ a)_0 in particular , in the coulomb gauge , @xmath72 consider a single fluxon with uniform @xmath18 localized in a disc of radius @xmath73 centered at the origin . in the coulomb gauge [ buniform ] eb()=\ { lr 1 & r < r + 0 & r > r . e= ( ) ^2 & r < r + 1 & r > r the fundamental solution ( in the coulomb gauge ) is , by eq . ( [ poisson ] ) , [ rnz ] _ 0= \ { lr ( -_t ) & r < r + ( r r)^_t ( -_t ) & r > r . for @xmath74 the fundamental solution is not square integrable near infinity . a sub - critical fluxon can not support zero modes . when the total flux is super - critical @xmath63 the fundamental solution is square integrable for any @xmath75 and it is spread over an area of typical diameter @xmath76 . similar conclusions hold for asymmetric @xmath77 with the same total flux @xmath15 since the asymptotics of eq . ( [ rnz ] ) remains unchanged . it is instructive to compare this result with the formal solution for single point fluxon . the fundamental solution for a delta localized magnetic field is [ point ] _ 0= r^-_t , which is _ never _ square integrable , in contrast with what we found for finite @xmath73 ; the limit @xmath78 must be taken with care . in our discussion of point - like fluxons we shall assume that @xmath73 is smaller than any other length scale in our system but is actually non - zero and can therefore serve as a small distance cutoff . this issue will however not be relevant if all fluxons are sub - critical , in which case one may safely put @xmath79 . consider @xmath0 sub - critical , point - like fluxons . by the superposition property the fundamental solution is just a product of ( translated ) solutions of the form eq . ( [ point ] ) : [ prod ] _ 0()=|-_a|^-_a the solution is locally square integrable since , by assumption , the individual fluxons are sub - critical @xmath3 and distinct @xmath80 . it is square integrable at infinity if the total flux is super - critical , @xmath81 . eq.([prod ] ) is the prototype of free zero modes . when the @xmath20-th fluxon collides with the @xmath43-th fluxons the norm of the fundamental solution @xmath82 diverges if @xmath83 . the condition @xmath84 endows the control space of point - like fluxons with a non - trivial topology . for point fluxes , a useful gauge , besides the coulomb gauge , is the _ holomorphic gauge _ which formally has @xmath85 except for cuts where it is delta like . the fundamental solution in this gauge is a holomorphic function in the cut complex plane [ holomorphic ] _ 0(z;)=_a(z-_a)^-_a,_a= _ a ( + i ) , ^n @xmath86 is analytic in @xmath87 . the cut @xmath88 runs from @xmath89 to infinity . is then well defined at infinity . this can be understood as reflecting the dirac flux quantization on compact manifolds . ] the general solution obtained by multiplying by a polynomial @xmath66 is then also holomorphic with the same cuts . [ lijs ] ( z;)=p(z)_a(z-_a)^-_a,(p ) < _ t-1 in general one may allow having both positive and negative fluxons . the solution eq . ( [ lijs ] ) corresponding to @xmath90 is typically not square integrable if some of the fluxons are super - critical . it becomes a legitimate solution only for @xmath91 where @xmath92 . the corresponding modes are then identical to those occurring in a system of ( sub - critical ) fluxons having @xmath93 ( assuming that @xmath94 ) . we call these states free zero modes . as we shall see their behaviour and holonomies are identical to those of a system with fluxes @xmath95 . since in general @xmath96 , the system will also contain another type of zero modes . in these states the electron typically sits in a small @xmath76 neighbourhood of a specific fluxon . we therefore call these states confined . in the special case of integer fluxons certain states may incorporate feature of both confined and free states ( see e.g. fig . [ fig : qubits ] ) . the main focus of this paper is on the holonomies of free states and the reader may assume , for simplicity , that all fluxons are subcritical so that @xmath97 and all states are free . for completeness we give below a brief account of the confined states . confined states localized near the @xmath20th fluxon are typically constructed by taking @xmath98 in eq . ( [ yr7 ] ) . it is convenient to take advantage of our superposition principle and write @xmath49 as a product of ( not necessarily fundamental ) solutions @xmath99 corresponding to all fluxons . a confined state at @xmath20 then takes the form ( z)=_a^j(z)_ba_b(z)_a^j(z)_ba_b(_a).[uigf]here @xmath100 are some wave - functions which depend on the detailed shape and radius @xmath22 of the confining fluxon , while for @xmath101 we have @xmath102 in holomorphic gauge . the approximate equality on the right of eq.([uigf ] ) follows from the fact that @xmath103 is sharply peaked near @xmath104 . the cases when @xmath105 or @xmath106 require a more careful analysis . we shall not delve into this here since our main interest is in the free modes . modes localized on different fluxons are clearly mutually orthogonal . some extra thought also shows that when properly normalized the overlap of free state with a confined state localized at @xmath23 scales as @xmath107 and hence vanishes in the pointlike limit . we are interested in the evolution of the zero modes when the fluxes move adiabatically . control space , parametrized by the fluxon coordinates @xmath108 is @xmath109 dimensional . since the motion of the fluxons generates electric fields we need first to construct corresponding ( time - dependent ) pauli operator , eq . ( [ pauli2 ] ) , with both @xmath110 and @xmath111 . by faraday law a moving magnetic field must be accompanied by a nonzero electric field . if the motion is adiabatic the velocity @xmath112 is small and the acceleration negligible . it follows that radiation and retardation can be neglected . the fields resulting from the motion can be obtained by lorentz transformation to the moving frame [ evb ] = - we therefore need first to determine the full pauli operator , allowing for both scalar and vector potentials , eq . ( [ pauli2 ] ) , due to the motion of fluxons . the main result of this subsection is eq . ( [ phi ] ) which we shall now derive . to determine @xmath113 associated with a moving fluxon we substitute eq . ( [ evb ] ) in the definitions of the potentials , eq . ( [ fields ] ) , -a_0=-== ( ) = ( ) - ( ) ( and on the right we used the fact that @xmath114 is a vector , not a vector field for a discussion of the general case . ] . ) this may be rearranged as [ potentials ] = ( + ) = ( + a_0 ) let the static magnetic field of the @xmath20-th fluxon be described by the vector potential @xmath115 . take @xmath116 to be the rigid transport of @xmath117 , so that @xmath118 , and choose @xmath119 so that eq . ( [ potentials ] ) is satisfied , . ] namely [ rot ] ( , t)=_a(-_a(t)),a_0(,t)=-_a _ a(-_a(t ) ) @xmath120 is the trajectory of the fluxon . this 4-potential generates the fields of a rigidly moving fluxon : [ fields2 ] ( , t)=_a(-_a(t ) ) , ( , t)=-_a(,t ) the generalization to a number of fluxons each moving along its own trajectory is obviously ) may be derived by applying a lorentz boost to the vector @xmath121 and keeping terms only up to first order in @xmath122 . ] [ phi ] = _ a=1^n_a(-_a(t ) ) , a_0=-_a=1^n_a_a(-_a(t ) ) note that @xmath113 _ is not _ necessarily in the coulomb gauge . it has the pleasant feature that a closed path in the space of controls @xmath123 is represented by a closed path of the potential , and hence a closed path of the hamiltonian . we are interested in the evolution of the zero modes due to adiabatic motion of the fluxons . more specifically , we are interested in the holonomy that describes the braiding of fluxons . this adiabatic problem has three subtle points : * gapless zero modes . the zero modes lie at the threshold of the continuous spectrum so the adiabatic evolution is not protected by a gap . one may then appeal to adiabatic theorems that cover the gapless case . for point - like fluxons the only length scale is the distance between them . ] @xcite . these theorems hold provided the space of zero modes changes smoothly . consider the collision of sub - critical point - like fluxons . the norm of the fundamental solution @xmath124 diverges when they form a super - critical fluxon . it follows that the space of zero - modes does not behave smoothly upon flux collisions . flux collisions then play the role of gap closure in gapped adiabatic evolutions . removing points of flux collisions endows the control space with a non - trivial topology . * gauge freedom : adiabatic phases are well defined ( gauge invariant ) for closed cycles of the hamiltonian @xcite . braiding of fluxons is described by a closed cycle in control space @xmath123 , and therefore also a closed cycle in the space of em _ fields _ , but not necessarily in the space of hamiltonian which depends on the potentials . to correctly compute the holonomy , one needs the em _ potentials _ to make a closed cycle . ( one _ is not _ interested in phases that come simply from a change of gauge between the initial and final hamiltonian . ) this is taken care of by the choice of gauge made in eq . ( [ phi ] ) . * parallel transport : in standard adiabatic hamiltonians @xcite the the adiabatic evolution is determined by the frozen hamiltonian . this is not the case here where the weak electric field generates the evolution . it turns out to be instructive @xcite to write the pauli evolution equation as id_t=1 2 m ( ( -i - e ) ) ^2,d_t=_t -iea_0 with @xmath125 replaced by the covariant time derivative @xmath126 . let @xmath127 denote the spectral projection on the zero modes of ( the frozen ) pauli operator . the evolution generated by [ kato ] i= ( _ p + i [ ] ) maps unitarily @xmath128 @xcite . the first term describes the action of the hamiltonian on @xmath129 and the second term guarantees that the states remain within the instantaneous spectral subspace . usually , the first term acts on @xmath129 as a c - number giving it just an overall phase and therefore in spite of being @xmath130 it is usually less important than the second which is only @xmath131 . for the case at hand , both terms act nontrivially on the zero modes space and are @xmath131 . ( [ kato ] ) reduces to i=(i [ , ] -e a_0 ) = ( i [ , ] + e_a=1^n _ a _ a ) and , we have used eqs . ( [ pauli2],[phi ] ) . in particular , if @xmath49 and @xmath132 are zero modes then @xmath133 and @xmath134 and the evolution of zero modes is governed by [ evolution1 ] i| d = e _ a d_a | _ a | = e_a| v_a a_a+ covariant derivative @xmath126 . ( here @xmath135 . ) given @xmath59 , the instantaneous fundamental solution @xmath136 is uniquely determined as in sec . [ sec : ac ] . the adiabatic evolution can be viewed as a rule for evolving the polynomial @xmath137 [ poly ] = p(z , t)_0 = p(z , t)_a(_a)_0 , p(z , t)=_j=0^d -1 p_j(t ) z^j in the rest of this section we show that the matrix elements of @xmath116 in the evolution equation ( [ evolution1 ] ) can be traded for the derivatives of the zero modes overlaps . the main results are eq . ( [ pt ] ) , or equivalently eq . ( [ gy876 t ] ) below . note first that fundamental mode @xmath71 of each individual fluxon satisfies [ kmode ] 0=(|_z - ie|a_a)(_a)_0=-(|_a+ie|a_a)(_a)_0 , _ a=__a it follows from this that @xmath138 the evolution equation then takes the form [ pt0 ] 0=| d + d_a |_a | - d|_a | |_a using the fact that the fundamental modes @xmath71 evolve by rigid motion d_0= ( d_a_a)_0=(d_a_a+d|_a|_a)_0 we finally arrive at the evolution equation for the polynomial @xmath137 : [ pt ] 0=|d p| + d_a _ a| , = p(z , t)_0,=q(z)_0 which may also be stated as : the evolution of a zero mode under the change of the controls @xmath139 is determined by the evolution of the corresponding polynomial @xmath137 . this is determined by the equations _ 0||qdp|_0+d_a_a _ 0||q p |_0=0 [ gy876 t ] when the fluxons are pointlike subcritical @xmath3 and the fundamental mode is chosen in the holomorphic gauge , as in eq . ( [ holomorphic ] ) , the sum on the right of eq . ( [ pt0 ] ) vanishes and the evolution equation simplifies to the statement that the velocity in the manifold of zero modes vanishes : [ parallel ] 0=| d , = p(z , t)_0,=q(z)_0 note that the scalar product in the holomorphic gauge is well defined even for fractional @xmath21 , independently of the way one chooses the cuts as long as this choice is done consistently . the @xmath14 dimensional space of zero modes can be naturally identified with the space of holomorphic polynomials with @xmath141 . natural coordinates are the coefficient @xmath142 in @xmath143 . let p=(p_0 , , p_d-1)^t , p^d the hilbert space metric induces a metric on @xmath144 [ metric ] ( ) _ jk= _ j | _ k , _ j = z^j_0 , j , k0 , , d-1 with the properties : * @xmath140 is a positive , hermitian , @xmath145 matrix . * @xmath140 is gauge invariant : it is independent of the choice of gauge for the ( frozen ) pauli operator . * @xmath140 is a smooth function of the fluxes , @xmath21 , provided @xmath146 . it blows up as the total flux @xmath147 reflecting the loss of one mode . * when all the fluxons are finite , @xmath148 , the metric is an everywhere smooth function of @xmath149 , the fluxon coordinates . in the limit of pointlike fluxons , @xmath32 , we can say more : * the metric has an ( approximate ) block structure : all the ( properly normalized ) confined modes are in @xmath150 blocks , and all the free modes are in a single block . ( the terms connecting different blocks scale as positive powers of @xmath151 . ) * the block of the free zero modes is given by [ bkj ] ( ) _ jk= _ j | _ k , _ j(z ; ) = z^j _ a(z-_a)^-_a , _ a=_a - max\{0,_a } * under scaling the metric of the free modes behaves like : [ scaling ] _ jk()=^k|^j ||^2(1-_t ) _ jk ( ) note that the magnitude of @xmath152 describes dilation and its phase a rigid rotation of the control space . * when two point - like fluxons @xmath20 and @xmath43 collide the metric blows up if @xmath153 . * it is natural to remove from control space ( with coordinates @xmath89 ) the points where @xmath154 . this endows the control space of point fluxons with an interesting topology . consider a path @xmath155 in the space of controls . we are interested in the evolution of @xmath156 along the path . making use of @xmath140 , the transport equation , eq . ( [ gy876 t ] ) , takes the form [ asd ] 0= d p + ( d_a _ a ) p this can be written more compactly using the dolbeault operator stands for dolbeault to be distinguished from @xmath157 and @xmath158 . ] @xmath159 @xcite ( similarly @xmath160 ) : [ yr ] 0= d p + ( ) p the transport equation may then be written in terms of a connection@xmath161 [ gp ] 0= ( d+a)p , = ^-1 ( ) the connection determines the ( adiabatic ) curvature 2-form @xmath162 by a standard formula @xcite : [ rr ] r= da+a =| ( ^-1 ) in the abelian case @xmath163 this simplifies into [ ra ] r=| the curvature vanishes , @xmath164 , when the connection @xmath165 is a pure gauge : [ puregauge ] a= ^-1_0d _ 0the connection @xmath166 of eq . ( [ gp ] ) is `` half way '' to be a pure gauge ( @xmath167 is missing ) . it has special properties which we return to below . the study of the connection for point like fluxons is easier in some cases and harder in others : it is relatively easy for the confined modes where arguments based on the aharonov - bohm effect apply . it is more difficult in the more interesting case of free zero modes . this case too splits into cases with different levels of difficulty : the abelian case is simpler than the non - abelian case , and rigid translations and rotations are easier than braiding . this section is organized so that we treat the easier and special cases first . a reader who prefers to start with the most general results may want to read the subsections below in different order . by eqs . ( [ gp],[rr ] ) the connection @xmath166 and curvature @xmath162 inherit the block structure of @xmath140 . as all the confined modes are orthogonal to the free modes ( in the limit @xmath90 ) and to each other , one concludes that the connection @xmath166 and curvature @xmath162 split into a block corresponding to all free states and a number of @xmath168 blocks corresponding to each of the confined modes . each block remains completely unaffected by other blocks and may therefore be discussed separately . braiding the pair @xmath169 one expects is finite due to the power law tails of the state and exact in the limit @xmath31 . ] a confined mode on @xmath20 to acquire aharonov - bohm phase @xmath170 . let us see how this can be understood from the machinery developed in sec . [ ch : adiabatic ] . due to the block structure of the connection , [ sec : block ] , the adiabatic evolution of a confined state has only an abelian phase factor @xmath171 . from sec . [ super ] a mode confined at fluxon @xmath20 is described ( e.g. in holomorphic gauge ) by @xmath172 where @xmath173 is independent of the @xmath174s . the metric ( @xmath168 block ) is the function _ a | _ a _ ba @xmath175 in particular , if @xmath174 encircles @xmath89 the phase is the aharonov - bohm phase @xmath176 . * under ( rigid ) translations @xmath178 of the entire fluxon configuration independent of @xmath20 but not necessarily of @xmath179 the wave function undergoes simple translation @xmath180 . this fact may be verified by using eqs . ( [ kernel1],[evolution1 ] ) . * a rigid rotation of the entire configuration is described by ( complex ) scaling @xmath181 . from eqs . ( [ scaling ] ) we find @xmath182 from eq . ( [ yr ] ) the @xmath183-th coefficient rotates independently of the others dp_k(t)=-(k+1-_t ) p_k(t ) ( t ) dt in particular a full @xmath184 rotation result in acquiring of the ( abelian ) berry phase [ berry ] 2(_t - k-1 ) as @xmath183 is an integer this phase is identical ( mod @xmath184 ) for all free states . ( this could be anticipated from the fact that changing the origin of rotation mixes different @xmath183s . ) consider @xmath0 subcritical fluxons @xmath3 . the metric @xmath140 is simply the norm of the fundamental mode = _ 0 | _ 0 , _ 0(z;)=_a(z-_a)^-_a suppose @xmath185 , holding @xmath186 fixed . the adiabatic curvature in the remaining @xmath187 coordinate is given by eq . ( [ ra ] ) [ curva ] r = da=|_u_u ( u ) , ( u)= there is no reason for @xmath188 to be a harmonic function . in appendix [ 3flux ] we show that @xmath140 and @xmath189 can be evaluated in terms of hyper - geometric functions . indeed as one can see from fig . [ fig : curvature ] for three half fluxes the curvature does not vanish . and @xmath190 . it is symmetric under @xmath191 . ( [ fig : r ] ) and appendix [ 3flux ] for more details on how the figure was drawn.,height=4 ] when @xmath39 and @xmath163 braiding is topological by a special reason that we shall discuss in sec . [ sec : square ] . however , for @xmath41 and @xmath163 , the braiding of two fluxons @xmath169 , keeping all the other fluxons fixed is , in general , path dependent . i.e. @xmath192 . the diagonal blue lines delineate regions without zero modes , with one zero mode and with 2 zero modes . in the triangle marked by @xmath193 the charge is localized on @xmath194 fluxons and berry s phase is @xmath195 as one would expect from the aharonov - bohm effect . the triangle marked @xmath196 describes a zero mode confined on 3 and the phase is @xmath197 . in the triangle marked by @xmath198 the state is free and the berry phase is @xmath199.,height=6 ] for the sake of simplicity of the notation we shall write @xmath200 and @xmath201 in this section . in appendix [ sec : integrals ] we show that the metric @xmath140 for the free zero modes factorizes into a holomorphic and anti - holomorphic factors . by eq . ( [ gexplicit ] ) : [ kjdc ] ( ; ) = ^ * ( ;) ( ) ( ;) where : * @xmath202 is an @xmath203 holomorphic ( matrix ) function whose matrix elements are given in eq . ( [ paj ] ) , as [ paj1 ] _ ak ( ) = _ _ 0^_a d _ k ( ) , a1, ,n , k0 , , d-1 @xmath202 involves a choice , which we call a _ gauge choice _ , and is hidden in the freedom of @xmath204 . alternatively one could add arbitrary integration constant @xmath205 to the r.h.s . * @xmath206 is an @xmath207 hermitian matrix which is independent of the controls @xmath208 . its explicit form is given in appendix [ sec : integrals ] . it has rank @xmath209 with its kernel generated by the vector @xmath210 . note that this guarantees that the metric @xmath211 is not affected by adding arbitrary integration constant to eq.([paj1 ] ) . substituting the expression ( [ kjdc ] ) for the metric into eq . ( [ yr ] ) leads to the transport equation _ dn d(p)=0 [ hjs ] these are @xmath14 equations for the evolution of the @xmath14 coefficients @xmath212 . the solutions of these equations are , in general , path dependent . if one could cancel out the left factor @xmath213 , it would follow that the holonomy is topological . in general however this can not be done since @xmath213 is not an invertible ( or even a square ) matrix . the holonomy is , in general , not topological . when the number of free zero modes is maximal @xmath214 the holonomy is topological . the simplest example of this kind is @xmath39 and @xmath38 . the berry phase associated with taking one fluxon around the other is _ topological _ and is given by eq . ( [ berry ] ) with @xmath215 @xmath216 one way to show that this is the case is by using the ` gauge ' freedom to choose @xmath217 in eq.([paj1 ] ) so that @xmath218 for all @xmath219 . by throwing its last row we can then view @xmath220 as a square @xmath145 matrix which we denote @xmath221 . we may then write for the metric [ square1 ] _ jk = _ a , b=1^d |_aj g_ab _ bk = ^ * _ _ _ where @xmath222 are all square @xmath223d matrices . repeating the arguments of the previous subsection eq . ( [ hjs ] ) now takes the form ^ * _ _ ( _ p ) = 0 since @xmath211 is positive , all of its ( @xmath145 ) factors are invertible . the equations of parallel transport therefore reduces into @xmath14 conservation laws and @xmath165 is a pure gauge : [ ksg ] d ( _ p)=0 , = ^-1 = ^-1_d _ it follows that the holonomy of adiabatic transport is determined by the monodromy of the ( multivalued function ) @xmath221 . the gauge choice @xmath217 has the disadvantage of treating @xmath224 on different footing than the other @xmath89s . moreover , in braiding operations in which the @xmath0th fluxon is more than a spectator , any dependence of @xmath204 on @xmath224 can lead to extra complication . for this reason we shall prefer in the next section to fix @xmath204 to be a constant independent of @xmath149 . one may avoid the ` gauge ' choice @xmath217 and rewrite equation ( [ ksg ] ) in a gauge independent way as [ dgij ] d ( p ) _ n where @xmath225 is the generator of @xmath226 and the complex coefficient on the r.h.s depends on the arbitrary integration constant chosen in the definition of @xmath220 . we have seen that when @xmath34 there is no curvature . hence , if the unitary holonomy operators are non trivial , and non - abelian , then fluxon braiding can be viewed as ( non abelian ) topological unitary operations on the manifold of zero modes . we start by computing the monodromy of braiding of distinct fluxons . in the special case that the fluxons carry identical fluxes , they may be viewed as identical anyons . in particular , when the fluxes are identical , it obeys the braiding rules of burau representation of the braid group . ( 300,160)(0,0 ) ( 30,40)(1,0)200 ( 30,40 ) ( 80,90)(1,0)150 ( 80,90 ) ( 47,140)(1,0)183 ( 47,140 ) ( 25,25)@xmath227 ( 125,45)@xmath228 ( 220,65)@xmath229 ( 75,75)@xmath230 ( 135,95)@xmath231 ( 220,105)@xmath232 ( 42,125)@xmath233 ( 92,145)@xmath234 ( 220,155)@xmath235 ( 220,20)@xmath236 ( 7,10)(1,0)240 ( 7,10)(0,1)150 ( 230,-4)@xmath16 ( -9,149)@xmath237 we start by computing the monodromy of @xmath238 . in subsection [ uyufc ] we shall relate it to the holonomy of the adiabatic evolution . the components of the matrix @xmath220 are by eq . ( [ paj1 ] ) , [ ppppsi ] _ aj()=__0^_a d ^j _ b=1^n ( -_b)^-_b since @xmath219 is an integer the factor @xmath239 in the integrand has no interesting effect on the monodromy and we can ignore the index @xmath219 ( and henceforth drop it ) without risk . choosing integration paths from @xmath204 to @xmath240 which do not cross any of the cuts shown in fig . [ fig : cuts ] leads to a standard definition of @xmath241 . upon cyclically moving the fluxon positions , we keep fixed @xmath204 independent of @xmath149 . if we do otherwise the monodromy would get extra contribution from possible movement of @xmath204 . ] @xmath89 these paths are deformed as seen in figs . [ fig : braidb1],[fig : braida2 ] into paths which typically do cross the cuts . this leads to another branch @xmath242 of the multivalued function . the monodromy is an @xmath207 matrix @xmath243 : [ m ] = , = ( _ 1, ,_n)^t it is useful to collect properties of @xmath243 before one actually computes it , as they provide tests on the computations . * since the fluxons may be moved backwards , the monodromies @xmath243 must be invertible and generate a group . * since @xmath211 ( of eq . ( [ kjdc ] ) ) must not be affected by the monodromy , @xmath243 and @xmath244 must satisfy a consistency condition [ unitary ] = ^ * ( this may be viewed as a unitarity condition ) . * adding an integration constant ( or equivalently changing @xmath204 ) in eq . ( [ ppppsi ] ) corresponds to @xmath245 and @xmath246 . consistency with eq.([m ] ) thus requires @xmath247 with its ( red dashed ) branch cut @xmath88 going counter clock - wise around fluxon @xmath248 with its ( red dashed ) branch cut @xmath249 . as @xmath20 encircles @xmath43 the integration path ( blue ) from @xmath204 to @xmath174 loops around the branch point of the cut @xmath88 . the integration from @xmath20 to @xmath43 in the old and new paths are related by a factor @xmath250.,width=13 ] a proper definition of the monodromy of @xmath251 requires choosing some definite convention for placements of the cuts , see fig . [ fig : cuts ] . we shall take the cut @xmath88 to run from @xmath89 to @xmath252 , and we order them in such a way that as one goes counter - clockwise along a big circle ( near infinity ) the cuts @xmath88 are traversed successively according to their @xmath20-indexing . let us now compute the monodromy as the flux @xmath253 encircles an adjacent flux @xmath254 counter - clockwise . the integration path associated with @xmath255 loops around the branch point @xmath20 as shown in fig . [ fig : braidb1 ] . the deformed path gives @xmath256 by similar considerations as @xmath20 loops around @xmath43 , the integration path associated with @xmath257 `` stitches '' @xmath43 as shown in fig . [ fig : braida2 ] . it follows that @xmath258 to the fluxon at @xmath89 . as the @xmath20-th fluxons encircle b counter - clockwise it `` stitches '' @xmath43 . the old and new values of @xmath257 differ by integrations from @xmath20 to @xmath43 along two sides of the two cuts . , width=13 ] all other components of @xmath251 remain unaffected . the @xmath259 nontrivial block of the monodromy matrix is therefore ( _ a,_b)= ( cc 1-_a+_a_b & _ a(1-_b ) + 1-_a & _ a ) , = _ a_b the monodromy matrix is not symmetric in @xmath260 due to our convention for ordering the cuts counter clockwise . @xmath261 is related to @xmath262 by @xmath263 by eq.([unitary ] ) the spectrum of @xmath264 should lie on the unit circle . is spanned by @xmath265 known to be an eigenvector of @xmath264 . ] indeed : [ evm ] eigenvalues ( ) = \{1,_a _ b } in the special case where all fluxons are identical ( having the same @xmath21 ) it makes sense to consider also a permutation of two adjacent fluxons . this leads to the burau representation of the braid group @xcite . indeed , inspecting fig . [ fig : ex ] we see @xmath266 the monodromy matrix has the single nontrivial @xmath267 block [ ex ] ( ) = ( cc 1- & + 1 & 0 ) , = -with eigenvalues @xmath268 . note that when @xmath269 this reduces to the standard representation of the symmetric group . this may be understood as due to the fact that in this limit the free zero - modes turn into confined modes which move with the fluxons . the case @xmath270 ( i.e. @xmath271 ) does not occur since it is inconsistent with the assumption @xmath34 , see eq . ( [ con ] ) . the ( non - abelian ) holonomy @xmath272 for a closed path @xmath273 and base point @xmath149 , acts unitarily on the @xmath34 dimensional space of ( free ) zero modes at @xmath149 : @xmath274 one may write @xmath275 or equivalently @xmath276 . since the basis @xmath277 is not orthonormal , the matrix @xmath278 is not unitary . instead it satisfies @xmath279 . this is consistent with unitarity of the holonomy operator @xmath280 the previous sections make it clear that on the @xmath34 dimensional space of free zero modes , the @xmath281 matrix @xmath282 is closely related to the @xmath207 monodromy matrix @xmath264 . the exact relation is however complicated by the gauge freedom of fixing @xmath204 . we would like in this section to state this relation more precisely . for simplicity , consider first only braidings which do not involve the @xmath0-th fluxon . using the gauge choice @xmath217 one writes the conservation law , eq . ( [ ksg ] ) , taking @xmath283 around a closed path ( based at @xmath149 ) : _ p= _ p= _ _ p ( @xmath284 is the @xmath145 matrix obtained from @xmath243 by deleting its last row and column . ) the last identity used the definition of @xmath285 eq . ( [ m ] ) . hence [ u ] p_i=_j u_ijp_j , =( _ ) ^-1 ^-1 _ _ the derivation given above was special to the case where @xmath224 was a spectator . below we give an analysis of the general case . the monodromy matrices @xmath264 acting on @xmath286 preserve the vector @xmath287 i.e. satisfy @xmath288 . it follows that @xmath264 defines a linear transformation @xmath289 on the quotient space @xmath290 . we shall show that the holonomy @xmath272 for a closed path @xmath273 and base point @xmath149 is obtained from @xmath289 by a similarity transformation . the hermitian matrix @xmath291 satisfies @xmath292 . therefore the hermitian form it defines on @xmath286 , projects to a hermitian form @xmath293 on @xmath290 . moreover since @xmath291 has @xmath14 positive eigenvalues the form @xmath293 must give a ( non degenerate ) inner product on @xmath294 . ( [ unitary ] ) shows that @xmath289 are unitary relative to this inner product on @xmath294 . by eq.([dgij ] ) and the definition of the monodromy @xmath264 one has @xmath295 recall that @xmath238 is an @xmath296 matrix i.e. a map @xmath297 . denoting by @xmath298 the corresponding map into @xmath290 we conclude @xmath299 as @xmath298 is clearly invertible ( as follows e.g. from @xmath300 ) , we see that @xmath301 in particular the eigenvalues of the holonomy @xmath302 of fluxon braiding are related to the eigenvalues of the monodromy @xmath264 by [ mu ] eigenvalues ( ) = eigenvalues()\{1}=eigenvalues ( ^-1)\{1 } it follows from the results of the previous sections that when the fluxon @xmath20 goes around the fluxon @xmath43 , the holonomy matrix eigenvalues are [ main2 ] eigenvalues ( ) = \{1 , |_a|_b } by considering the @xmath207 matrix @xmath303 obtained by adding @xmath265 as an extra column to @xmath304 and defining @xmath305 we can obtain the simple relation @xmath306 as well as @xmath307 . the set @xmath308 is however an over - spanning set rather than a basis as it satisfies @xmath309 . in this section we show that the metric @xmath140 for the free modes of point - like fluxons factorizes into a product of a holomorphic and anti - holomorphic factors . since we are interested only in the free states , we will , for notational simplicity , assume all fluxons are subcritical . if this is not the case one should replace @xmath21 by its fractional part . let @xmath310 denote the primitive integral of @xmath311 : [ f ] _ j(z;,_0)=__0^z d _ j ( ) , _ j()=^j_a ( -_a)^-_a , j=0, ,d-1 we shall refer to the choice of @xmath204 as a choice of a gauge . for @xmath312 mathematica can evaluate @xmath313 of eq . ( [ f ] ) in terms of known special functions . in general , when @xmath314 the integral form is the best we can do . since @xmath315 one of the two integrations in eq . ( [ bkj ] ) for the metric is for free [ g ] _ jk = i 2 d _ kd|_j = i 2 _ a _ _ a _ k |_j d|z and we have used the generalized stokes theorem . the remaining contour integrals encircle the cuts @xmath88 running from @xmath89 to @xmath252 ( see fig . [ fig : cuts ] ) . the value of @xmath316 above and below the cut are related by _ -=_a ( _ j)_+_a = e^-2i_a to see how @xmath317 behaves across the cut @xmath88 write _ k(z;,_0)= _ ak ( ) + _ k(z;,_a ) , where [ paj ] _ ak ( ) = _ k ( _ a;,_0)= _ _ 0^_a d ^k _ 0 ( ) , the first term is a finite ] constant ( independent of @xmath318 ) . the second term inherits the @xmath250 discontinuity of @xmath311 . it follows that @xmath319 is continuous across the cut @xmath88 and does not contribute to the integral in eq . ( [ g ] ) . the metric reduces to @xmath320 we denote by @xmath321 the value attained by @xmath322 as @xmath318 tends to infinity in the region between the cuts @xmath88 and @xmath323 . ( see fig . [ fig : cuts ] . ) the limit is well defined at infinity provided @xmath324 , which is what we need for the metric . rewrite eq . ( [ eq : g ] ) as a matrix equation [ eq : g2 ] = ( _ -)^ * _ 1 , ( _ 1)_ab= i 2 _ ab(1-|_a ) @xmath325 is independent of @xmath149 . the @xmath0-tuples @xmath326 and @xmath327 are linearly dependent [ yjf ] _ j(_a-1)-_aj=_a ( _ j(_a)-_aj),a1 , , n where @xmath328 . this comes from integrating @xmath329 along @xmath88 . ( [ yjf ] ) too may be written as a matrix equation [ yjf2 ] _ 2 _ = _ 1 , ( _ 2)_ab = i 2 ( _ ab_a-_a , b+1 ) it follows from eq . ( [ eq : g2 ] ) and eq . ( [ yjf2 ] ) that the @xmath145 matrix @xmath140 can be factored as [ gexplicit ] ( ; ) = ^ * ( ;) ( ) ( ;) , = ( _ 1^*(_2 ^ -1)^*- ) _ 1 where : * @xmath202 is an @xmath203 holomorphic ( matrix ) function whose matrix elements are given in eq . ( [ paj ] ) . * @xmath206 is an @xmath207 matrix which is independent of the controls @xmath208 . * since @xmath211 is a @xmath145 positive matrix , @xmath244 must be an hermitian matrix . it must have at least @xmath14 positive eigenvalues and the image of @xmath220 must lie in the positive cone " of @xmath244 . in fact one may show that @xmath244 has exactly @xmath14 positive eigenvalues . * the definition of @xmath310 as a primitive integral in eq . ( [ f ] ) allows addition of an arbitrary integration constant ( possibly @xmath219-dependent ) corresponding to a free choice of @xmath204 . change of this choice will change the columns of @xmath220 by constant columns : [ psigauge ] = ( + ccc c_1 & & c_d + & & + c_1 & & c_d + ) since changing @xmath204 must not affect the metric , it follows that the kernel of @xmath244 contains the vector @xmath330 . it is in fact spanned by it . * one convenient ` gauge ' choice is @xmath217 which makes the last row of @xmath331 vanish . as a result eq . ( [ gexplicit ] ) takes the form @xmath332 where @xmath333 is @xmath334 and @xmath335 is @xmath281 . * an explicit expression for @xmath244 is [ ggn ] _ ab= + -((_t-_a ) ) & a = b + ( _ b ) & a < b + the values for @xmath336 may be deduced from hermiticity condition @xmath337 . alternatively the same value may be found from the relation @xmath338 . * when all the fluxes are identical @xmath244 is a tplitz matrix , i.e. constant along the diagonals , _ ab= g_a - b , a , b1, ,n explicitly , if each fluxon carries @xmath339 , then for , @xmath340 , [ gn ] g_0=- ( ) ( ( n-1))(n ) , g_k= e^i ( n-2k ) * away from the threshold for appearance of a new zero - mode , @xmath341 , the elements of @xmath244 are well defined and free of singularities . ( as are the elements of @xmath342 . ) for three fluxons one can find explicit expressions for the metric @xmath140 and the curvature . exploiting translation rotation and dilatation symmetries allow us to fix the location of two fluxons at will . we shall therefore assume the three fluxon are located at @xmath343 . choosing @xmath345 leads in the case @xmath163 to the following [ ucg ] = ( c 0 + u^-_3(1-_1 ) ( 1-_2 ) ( 2-_1-_2 ) _ 2_1(1-_1 , _ 3 ; 2-_1-_2 ; 1u ) + u^1-_1-_3(1-_1 ) ( 1-_3 ) ( 2-_1-_3 ) _ 2_1(1-_1 , _ 2 ; 2-_1-_3;u ) + ) http://functions.wolfram.com/hypergeometricfunctions/hypergeometric2f1/[@xmath346 is a hypergeometric function . for three identical fluxes @xmath339 this reduces into [ psi ] = ( 1-)^2 ( 2 - 2 ) ( c 0 + u^- f(1 u ) + u^1 - 2 f(u ) + ) , f ( u)=_2_1(1-,;2 - 2 ; u ) in particular in the special case @xmath347 it becomes [ hk]= ( cc 0 + 2k(1u ) + 2k(u ) + ) with http://www.wolframalpha.com/input/?i=elliptic+k[@xmath348 the complete elliptic integral of the first kind . using eq . ( [ gn ] ) for @xmath349 , one then finds a simple formula for the metric [ fig : r ] ( u)=8 re ( k(u)k(1-u)^ * ) the associated curvature is plotted in fig . ( [ fig : curvature ] ) . since @xmath220 is defined only up to addition of an arbitrary ( @xmath282-dependent ) multiple of @xmath265 , one may write down various alternative expressions to eq . ( [ ucg ] ) . using the following @xmath350 with @xmath349 the @xmath351 matrix given in eq . ( [ ggn ] ) , leads to expressing the @xmath150 metric @xmath140 as a combination of two squares @xmath352 for @xmath353 this is always positive . one can also get explicit formulas for the non - abelian case . @xmath220 is @xmath355 : [ psig ] = u^-_3 ( cc 0 & 0 + a & b + u^1-_1c & u^2-_1d + ) where @xmath356 when all three fluxes are identical @xmath339 this becomes [ psi2 ] = ( 1-)^22(2 - 2 ) ( cc 0 & 0 + 2u^-f(1 u ) & u^-g(1 u ) + 2u^1 - 2 f(u ) & u^2 - 2 g(u ) + ) where @xmath357 @xmath349 is again @xmath351 given by eq . ( [ ggn ] ) and @xmath140 is @xmath259 . in this appendix we give another ( more abstract ) construction of the connection described in sec . [ sec : connection1 ] and sec . [ sec : square ] corresponding to the free zero modes around point - like fluxons . in particular it shows that in general one may embed our @xmath33-dimensional bundle into a flat @xmath209 bundle . let @xmath294 be the fixed @xmath358-dimensional complex vector space @xmath359 where @xmath225 . since @xmath360 , the @xmath207 hermitian matrix @xmath291 defines a pseudo ( hermitian ) metric on @xmath294 . let @xmath361 be the space of possible positions of @xmath0 fluxons . consider the trivial bundle @xmath362 . for each @xmath219 the vector function @xmath363 defines a ( multivalued ) section of @xmath364 . we shall denote this section by @xmath313 as well although it is actually an equivalence class under quotienting by @xmath365 . at each point @xmath366 the vectors @xmath367 generate a @xmath14-dimensional subspace @xmath368 of @xmath294 . these spaces make up together a @xmath14-dimensional sub - bundle @xmath369 of @xmath364 . the restriction of @xmath291 to @xmath369 is a positive definite hermitian metric . this follows from the fact that @xmath370 is the hilbert space metric on our pauli zero - modes . in particular it follows that @xmath371 the @xmath291-orthogonal complement of @xmath369 is well defined and hence also the @xmath291-orthogonal projection @xmath372 . in fact one may write explicitly @xmath373 where @xmath374 is the inverse of the matrix @xmath375 . as @xmath364 is trivial it is natural to use the trivial connection @xmath376 on it . the projection @xmath372 then defines a connection @xmath377 on @xmath369 . consider a general section @xmath378 of @xmath369 . using the fact that @xmath379 we find that the covariant derivative is given by : @xmath380 the equation @xmath381 for parallel transport thus takes the form @xmath382 which is exactly identical to the transport equation eq . ( [ yr ] ) . consider adiabatically turning one of the flux tubes around itself once . to find the holonomy of zero energy bound states we first need to find the electric and magnetic fields generated by adiabatic rotation at angular rate @xmath383 . to find these , we need a model of a fluxon . consider the following simple model of fluxon , shown in fig . [ box ] : two concentric thin cylinders of radius @xmath73 with charge @xmath384 ( per unit length ) , and charge density @xmath385 , rotating at constant angular velocity @xmath386 . since the overall charge vanishes and the fields are time independent , there is no electric field . the magnetic field is static and it satisfies , ( recall @xmath387 ) @xmath388 leading to a jump in the boundary conditions @xmath389 assuming @xmath390 we then have [ flux ] b ( ) = 4 q & | |<r + 0&| |>r it follows that the flux , per eq . ( [ zero - modes ] ) , is _ t = 2 e q r^2 and charge density @xmath391 . the red cylinder rotates clockwise and the blue counter - clockwise with the same angular velocity . rotating the fluxon causes the red cylinder to rotate faster and the blue cylinder slower . this creates a voltage difference between the inside and outside of the fluxon . , width=6 ] consider what happens when one adiabatically rotates the whole arrangement by @xmath392 so the two cylinders rotate at different angular velocities . to figure out the addition of angular velocities @xmath393 and @xmath394 , let @xmath395 if @xmath396 then addition gives @xmath397 . but if we allow @xmath398 , the rule follows from additivity of the rapidity @xmath399 . one finds ( assuming @xmath383 small ) @xmath400 if this was all that happened , rotating the fluxon would have no effect on the fields ( to order @xmath383 ) . however , this is not all . relativistic lorentz contraction implies that the geometry of the cylinders must change . the perimeter of the cylinders should contract by the usual rule . as the embedding space remains euclidean the radius needs to adjust to accommodate the contraction . for a cylinder of finite width this would inevitably lead to nontrivial internal stresses , but in the zero width limit we consider here this issue can be ignored . thus if @xmath73 denotes the radius for the cylinder rotating with @xmath393 then the contraction of the radii is given by @xmath401 it follows that , to first order in @xmath383 @xmath402 hence @xmath403 this imply that in the annulus between the two cylinders there is a radial electric field and hence a potential difference between the inside and the outside of the fluxon : [ 2 ] v= 2 q 4 q r=4q r^2 = 2 e_t integrating over the time needed to complete one full rotation gives @xmath404 consider a charged particle having wave function @xmath49 in the presence of the fluxon . the above suggests that fluxon rotation would induce a phase on the part of the wave function which is inside the fluxon . if the evolution is adiabatic this relative phase can be translated into the overall phase @xmath405 @xmath406 is the ( fraction ) of charge inside the fluxon . the phase depends on the total charge inside the fluxon but is independent of how it is distributed there . it is instructive to contrast this result with what one expects from the aharonov - bohm effect . a ( classical , localized ) magnetic flux encircling a localized ( quantum ) charge @xmath406 gives half this phases . not only is the factor 2 intriguing but , even more importantly , the aharonov bohm argument implies that the phase should depend also on the distribution of charge inside the fluxons . in the discussion of moving fluxons , section [ 7tuygd ] , it was assumed for simplicity that @xmath114 stands for a vector rather than a vector field . as a result self rotations of fluxons were not permitted only the position @xmath89 of the fluxons centers were rotated . ] . it is in fact possible to generalize parts of our arguments to arbitrary vector fields . the aim of this appendix is to explain this . it is worthwhile to note that the following does not even require the introduction of a riemann metric as the argument are completely independent of it . by a slowly moving fluxon we shall mean a fluxon whose magnetic field @xmath18 is being dragged along the vector field @xmath112 . more precisely , this is described mathematically by writing @xmath407 where @xmath408 stands for the lie derivative along @xmath112 . in two dimensions @xmath18 is a scalar density and using explicit form of the lie derivative gives @xmath409 the first term describes moving along @xmath112 while the second makes @xmath18 behave as a density in cases where the flow defined by @xmath112 does not preserve volume . in particular this relation guarantees that the total flux is unchanged . in case of translations or rotations the second term vanishes anyway . the corresponding vector potential @xmath59 is of course not determined uniquely , but it is most convenient to assume it is dragged in a similar way . this leads to @xmath410 since @xmath59 is a covariant vector the explicit form in vector notation is @xmath411 the second term represents the required rotation of the vectorial components of @xmath59 . this relation may also be expressed as @xmath412 substituting this into eq.([fields ] ) for the electric field we find @xmath413 as in section [ 7tuygd ] it follows that the relation @xmath414 is consistent with the choice @xmath415 . this holds generally regardless of whether @xmath112 is a rigid motion or a deformation , and of whether it is constant or time dependent . it does not even matter here whether space is flat or curved . ( this may however matter when one considers the spinor @xmath49 . )

two dimensional lattice models such as the quantum double models , which includes the toric code , can be constructed from transfer matrices of lattice gauge theories with discrete gauge groups . these transfer matrices are built out of local operators acting on links , vertices and plaquettes and are parametrized by the center of the gauge group algebra and its dual . for general choices of these parameters the transfer matrix contains operators acting on links which can also be thought of as perturbations to the quantum double model driving it out of its topological phase towards a paramagnetic phase . these perturbations can be thought of as magnetic fields added to the system which destroy the exact solvability of the quantum double model . we modify these transfer matrices with perturbations and extract exactly solvable models which remain in a quantum phase , thus nullifying the effect of the perturbation . the algebra of the modified vertex and plaquette operators now obey a deformed version of the quantum double algebra . the abelian cases are shown to be in the quantum double phase whereas the non - abelian phases are shown to be in a modified phase of the corresponding quantum double phase . this is shown by working with the groups and for the abelian and non - abelian cases respectively . the quantum phases are determined by studying the excitations of these systems . the fusion rules and the statistics of these anyons indicate the quantum phases of these models . the implementation of these models can possibly improve the use of quantum double models for fault tolerant quantum computation . we then construct theories which arise from transfer matrices that are not the transfer matrices of lattice gauge theories . in particular we show that for the case this contains the double semion model . more generally for other discrete groups these transfer matrices contain the twisted quantum double models . these transfer matrices can be thought of as being obtained by introducing extra parameters into the transfer matrix of lattice gauge theories . these parameters are central elements belonging to the tensor products of the algebra and its dual and are associated to vertices and volumes of the three dimensional lattice . as in the case of the lattice gauge theories we construct the operators creating the excitations in this case and study their braiding and fusion properties .

1408.2501

the concept of topological order @xcite was initiated in the 80 s with the discovery of the fractional quantum hall effect and high temperature superconductivity @xcite . since then it has also been seen in short range resonating valence bond states @xcite and in quantum spin liquids @xcite . once it s usefulness was realized in solid state quantum computation @xcite several exactly solvable models have been constructed achieving this . the classic example emerged when kitaev wrote down the toric code hamiltonian in two dimensions @xcite . these systems are examples of lattice models which host anyons @xcite as part of their low energy excitations . they can also be thought of as particular phases of the @xmath2 lattice gauge theory which host these deconfined excitations with anyonic statistics . they posses ground states with degeneracies which are topological invariants . this degeneracy is stable up to the addition of weak perturbations . this feature makes this model a probable candidate for fault tolerant quantum computation @xcite . these models have been extended to other discrete groups and involutary hopf algebras as well @xcite . in these cases they can be thought of as arising from lattice gauge theories based on these discrete groups or involutary hopf algebras @xcite . earlier works showing the existence of anyons in two dimensional discrete gauge theories can be found in @xcite . discrete gauge theories emerge in these models when a continuous gauge group breaks down to one via spontaneous symmetry breaking @xcite . these systems are usually perturbed by adding qudit terms to act on the edges of the lattice which carry the gauge degrees of freedom . they drive the system out of the topological phase @xcite . the resulting models are rendered unsolvable analytically and are thus subject to study using numerical methods . however by considering restricted plaquette and vertex operators they can be made solvable . such studies were carried out in @xcite . this resulted in condensed phases of the quantum double model . these works were crucial in understanding the stability of these topological phase represented by the quantum double model . it is thus an important problem to find exactly solvable models which remain in topological phases in the presence of these perturbations . in this spirit we introduce exactly solvable models which are constructed by looking at possible hamiltonians that can be generated using the transfer matrices of such systems . we show exactly solvable models which include the single qudit perturbations . the cases of abelian and non - abelian groups are studied separately . it is shown that in the abelian case the system remains in the topological phase corresponding to the quantum double model . the situation turns out different for the non - abelian cases . we find that the model remains in a topological phase but it is in a modified version with respect to the corresponding topological phase of the quantum double model . these are seen by studying the examples of the group algebras of @xmath0 and @xmath1 denoted by @xmath3 and @xmath4 respectively . we then go beyond the transfer matrices of lattice gauge theories by introducing more parameters in the transfer matrix of lattice gauge theories to find other topological phases . this transfer matrix is made up of two - qudit operators apart from the usual operators making up the transfer matrix of lattice gauge theories . we work with the @xmath5 case to show how one can obtain the double semion phase from such considerations . for more general groups these transfer matrices contain the twisted quantum double models as defined in @xcite . such models were also defined in @xcite while considering trivial global symmetry groups . the paper is organized as follows . section 2 describes the transfer matrices of two dimensional lattice gauge theories . the local operators used to build these transfer matrices and the parameters used to study phase transitions between different phases is explained here . hamiltonians of two dimensional lattice models are obtained from these transfer matrices by taking their logarithms . this can be done in different ways resulting in several models . these include exactly solvable models and models which can not be solved analytically . the models of interest are presented in section 3 . their excitations along with their braiding statistics and fusion rules are studied in section 4 . section 5 shows the construction of the double semion model from the transfer matrix picture . section 6 makes up our concluding remarks . the system is defined on a two dimensional lattice @xmath6 with gauge degrees of freedom located on the links of this lattice . the gauge degrees of freedom belong to the group algebra of a gauge group @xmath7 , denoted by @xmath8 , with basis elements @xmath9 and multiplication rule given by @xmath10 . thus the states on the links are given by linear combinations of @xmath11 . in the case of @xmath5 this is nothing but the familiar two state system given by @xmath12 and @xmath13 . the system is evolved in time resulting in a three dimensional manifold @xmath14 which is more precisely given by @xmath15 $ ] where @xmath16 $ ] is the unit time interval . @xmath14 is also discretized with elements of the algebra @xmath8 living on the links . this can be used to construct partition functions for this system by associating weights to links and faces of the closed triangulated 3-manifold @xmath14 in the spirit of state - sum models @xcite . however when @xmath14 is not closed the tensors associated to the weights have non contracted indices at the ends of the unit time interval giving us transfer matrices . such a procedure was used to construct transfer matrices of lattice gauge theories in @xcite . in @xcite the partition function , and hence the transfer matrix , was built out of structure constants of the algebra @xmath17 , which in this case is given by @xmath8 , as the weights associated to links and plaquettes . apart from these structure constants the partition function is also parametrized by a pair of elements @xmath18 belonging to the center of the algebra @xmath17 and @xmath19 belonging to the center of the dual algebra @xmath20 . @xmath21 and @xmath22 denote space and time directions . thus the transfer matrix in our case is given by @xmath23 . we do not go into the details of constructing this transfer matrix in this paper . the details can be found in @xcite . for the purposes of this paper we will start with the most general transfer matrix for @xmath24 with particular choices of @xmath7 and show how we can construct various models of physical interest from them . for a general group algebra based on a group @xmath7 the transfer matrix is written as [ tg ] u((g ) , z_s , z_t , z_s^ * , z_t^ * ) & = & _ p ( _ c _ cb_p^c ) + & & _ j(_gg b_gt_j^g ) _ i ( _ r a_rx_i^r ) _ v a_v where @xmath25 , @xmath26 , @xmath27 and @xmath28 denotes plaquettes , links and vertices respectively . the operators in eq.([tg ] ) are given by [ operatorsg ] a_v & = & _ gg_g l_i_1^gr_i_2^g^-1l_i_3^gr_i_4^g^-1 + [ operatorsbg ] b_p^c & = & _ \{g_i}(g_1 g_2 g_3 g_4,c ) t_j_1^g_1t_j_2^g_2t_j_3^g_3 ^ -1t_j_4^g_4 ^ -1 + [ operatorsxg ] x_i^r & = & _ gg_r(g ) l_i^g where @xmath29 is the character of the group element @xmath30 in the irr @xmath31 , @xmath32 is the number of elements with non - zero character in the irr @xmath31 , @xmath33 $ ] is the set of conjugacy classes in @xmath7 , and the operators @xmath34 act on the states @xmath35 on the links as follows t_i^g|k & = & _ g , k|k ; t_i^g^-1|k= _ g^-1 , k|k , + l_i^g|k & = & |gk ; r_i^g^-1|k= |kg^-1 . the operator in eq.([operatorsxg ] ) can also be defined using the @xmath36 operators instead of the @xmath37 operators . however this does not matter as it can be shown that they are the same . this is because the elements of a given conjugacy class have the same coefficients . so the orientation of the lattice does not matter for the definition of these operators . however in the operator @xmath38 the coefficients of the elements in a given conjugacy class are not the same and hence the orientation of the lattice does matter in this case . the convention is that if the orientation of the link matches with the orientation of the plaquette then we use @xmath39 otherwise we use @xmath40 . the parameters on the two sides of eq.([tg ] ) are related by the following z_s & = & _ c_c_gc_g + z_t & = & _ r a_r _ gg_r(g ) _ g + z_s^ * & = & _ ggb_g_g + z_t^ * & = & _ gg _ g_g where @xmath41 is the basis of the algebra dual to @xmath8 that is it is dual to the basis of the algebra given by @xmath42 . these are the most general form for the central elements of a group algebra and its dual . in this paper we are not using @xmath43 directly . instead , we describe the models in terms of @xmath44 and @xmath45 . the parametrization by algebra and dual algebra elements is important in the context of @xcite but it will not play a major role here . the action of the operators in eq.([operatorsg]-[operatorsxg ] ) is shown in fig.([lattop ] ) . and plaquette operator @xmath46 are shown along with the orientations of the lattice . ] note from the form of the terms in the transfer matrix in eq.([tg ] ) that the operators on the plaquettes @xmath25 , are the same as the plaquette operators of the quantum double model @xcite . however the operators on the vertices @xmath28 are not the same as the star operators of the quantum double models . in order to obtain precisely the star operators of the quantum double case we need to choose @xmath47 to be z_t^ * = _ r g_r_gg_r(g)_g . this results in the following vertex operator a_v = _ r g _ r a_v^r with a_v^r = _ gg_r(g ) l_i_1^gr_i_2^g^-1l_i_3^gr_i_4^g^-1 . the operators @xmath48 commute with @xmath46 . when the irr @xmath31 is the identity representation we obtain the star operator projecting to the vacuum sector . this is the operator which appears in the hamiltonian of the quantum double model @xcite . note that these subtleties do not matter for the abelian models . they only make a difference in the non - abelian instances . the operators acting on the links @xmath27 , @xmath26 in eq.([tg ] ) can be thought of as magnetic field terms . such terms were considered in @xcite to induce condensations of quasiparticle excitations leading to phase transitions while remaining in the exactly solvable regime . we now write down the transfer matrices for the examples we consider in this paper . the hamiltonian is given by [ sidesg ] h = _ v a_v^g + _ c_c_p b_p^c where @xmath46 is the projector to a conjugacy class @xmath81 of @xmath7 and is given by b_p^ c = _ gc b_p^g and @xmath82 is a real parameter . the modified vertex operator @xmath83 is labeled by an element @xmath84 . it is given by a_v^g & = & l_i_1^gr_i_2^g^-1l_i_3^gr_i_4^g^-1 + & + & l_i_1^g^-1r_i_2^gl_i_3^g^-1r_i_4^gwith the complex parameters @xmath85 and @xmath86 satisfying a_h_1a_h_1^ * & = & a_h_2a_h_2^ * , h_1 , h_2 g h_1h_2 + b_h_1b_h_1^ * & = & b_h_2b_h_2^ * , h_1 , h_2 g h_1h_2 + b_gu & = & b_u , ug . the choice of parameters depend on the orientation of the @xmath87 lattice . the oriented lattice is shown in fig([lattop ] ) . every time the arrow goes into the vertex we use the @xmath86 set of parameters . the @xmath85 set of parameters is used when the arrow goes away from the vertex . with this definition the model can be defined on a @xmath87 lattice with an arbitrary triangulation . for these choice of parameters the vertex operators commute for adjacent vertices . the commutation between the modified vertex operator @xmath83 and the plaquette operator @xmath46 projecting to the conjugacy class @xmath81 is non - trivial . for this proof we need not consider the @xmath88 as it trivially commutes with the plaquette operator . the proof for the remaining part of the vertex operator goes as follows . consider the plaquette @xmath25 and the vertex @xmath28 shown in oriented square lattice of fig([proof ] ) . ) and eq.([rhs ] ) . ] they share two links labeled @xmath27 and @xmath26 in the fig([proof ] ) . according to the orientation of the links shown in the fig([proof ] ) we have [ lhs ] a_v^gb_p^c & = & ( l_i^gl_j^g r_j_1^g^-1r_j_2^g^-1 ) + & = & ( r_j_1^g^-1r_j_2^g^-1 ) and b_p^ca_v^g[rhs ] & = & ( l_i^gl_j^g r_j_1^g^-1r_j_2^g^-1 ) + & = & ( r_j_1^g^-1r_j_2^g^-1 ) + & = & ( r_j_1^g^-1r_j_2^g^-1 ) where we have used the relation t^h_il^g_i = l^g_it^g^-1h_i . thus the shifted summation in eq.([rhs ] ) only shuffles the terms in the projector . this shows that the two summations in eq.([lhs ] ) and eq.([rhs ] ) are the same implying @xmath83 commutes with @xmath46 . we can use a similar proof to show that the other three vertex operators commute with the plaquette operator . thus the hamiltonian in eq.([sidesg ] ) is exactly solvable . note that we have not used the full projectors for the modified vertex operators . this can be simply understood by looking at the identity term in the expression for the projectors . the identity term coupled with the perturbations is like adding magnetic fields to the system . this will take us out of the exactly solvable regime we are interested in and also we will no longer remain in a topological phase . thus we avoid full projectors for the modified vertex operators . we can write down a similar model where we modify the plaquette operator with the @xmath89 operators defined in eq.([operatorsxg ] ) . in the examples to follow we will write down these models for the @xmath3 case alone . the corresponding models for the non - abelian cases are more cumbersome and so we do not show them here . before we write down the specific examples let us look at the deformed quantum double algebra . consider the product of two deformed vertex operators @xmath90 and @xmath91 . it is given by a_v^g_1a_v^g_2 = a_v^g_1g_2 m(g_1 , g_2 ) , with m(g_1 , g_2 ) = . we can now check associativity of the product by comparing @xmath92 and @xmath93 which requires [ mc]m(g_1g_2 , g_3)m^g_3(g_1 , g_2 ) = m(g_1 , g_2g_3)m(g_2 , g_3 ) , which is seen to be true by using the expression for @xmath94 and from m^g_3(g_1 , g_2 ) = . the quantum double algebra given by a^g_vb_p^h = b^ghg^-1_pa_v^g , is unchanged but the product in the algebra is changed to ( a_v^g_1b_p^h_1)(a_v^g_2b_p^h_2 ) = _ h_2 , g_2 ^ -1h_1g_2a_v^g_1g_2m(g_1 , g_2 ) b_p^h_2 . this algebra is similar to quasi quantum doubles as seen in @xcite where the deformation occurs through a 2-cocycle . such algebras also have a non - coassociative coproduct with an associator given by a 3-cocycle related to the 2-cocycle by a slant product @xcite . the condition in eq.([mc ] ) is similar to the one obtained in studying obstructions to implementations of global symmetries on physical systems as used in analysis of symmetry protected topological phases @xcite . note that we have not studied the co - algebra structure of these operators here . it is not necessary for what is to follow and we stop with the above crucial remarks . we now look at the simple example of @xmath49 to illustrate the form of the transfer matrix defined in eq.([tg ] ) . [ tz2 ] u((_2 ) , z_s , z_t , z_s^ * , z_t^ * ) & = & _ p ( _ 1 b_p^1 + _ -1 b_p^-1)_j ( b_1 t_j^1 + b_-1 t_j^-1 ) + & & _ i ( a_1 x_i^1 + a_-1 x_i^-1 ) _ v(_1 a_v + _ -1 a_v^-1 ) where @xmath28 , @xmath25 , @xmath27 and @xmath26 label vertices , plaquettes and links respectively . the operators @xmath50 , @xmath51 , @xmath52 and @xmath53 are all projectors and are given by a_v^1 & = & + b_p^1 & = & + x_i^1 & = & + t_j^1 & = & . these are the operators appearing in the toric code @xcite system and their action is shown in the figure([lattop ] ) . the parameters on both sides of eq.([tz2 ] ) are related in the following way z_s & = & _ 1 _ 1 + _ -1_-1 + z_t & = & ( ) _ 1 + ( ) _ -1 + z_s^ * & = & b_1 _ 1 + b_-1 _ -1 + z_t^ * & = & ( ) _ 1 + ( ) _ -1 where @xmath54 and @xmath55 are the basis elements of @xmath5 and its dual respectively . the hamiltonian in this case is given by [ sides2v ] h = _ v a_v + _ p(_1 b_p^1 + _ -1b_p^-1 ) with @xmath95 given by a_v & = & where the complex parameters @xmath96 have the same modulus . it is easy to check that the model is exactly solvable as the terms commute with each other . to compute the spectrum of the hamiltonian we need to find the eigenvalues of the modified vertex operator @xmath95 and the plaquette operators . each plaquette operator is a sum of two projectors and so the eigenvalues of that operator depends on the magnitude of @xmath97 . the eigenvlaues of @xmath95 are given by @xmath98 . the dual of the above model in eq.([sides2v ] ) is given by the following hamiltonian [ sides2 ] h = _ v(_1 a_v^1 + _ -1a_v^-1 ) + _ p b_p where b_p = ( b_p^1 - b_p^-1)_i_kp , k=1 ^ 4(a_1 x_i_k^1 + a_-1x_i_k^-1 ) + h.c . with @xmath99 being the perimeter of the plaquette @xmath25 . this model can be defined on any two dimensional lattice for an arbitrary triangulation . on the square lattice this plaquette operator continues to be a four spin interaction term as in the usual quantum double model . it is easy to check that for two adjacent plaquettes @xmath100 and @xmath101 , these operators commute and that @xmath102 . this operator clearly commutes with the vertex operators as can be easily seen from the expression for @xmath103 . thus the hamiltonian in eq.([sides2 ] ) continues to be an exactly solvable model . the hamiltonian in eq.([sides2 ] ) is got by the following choice of the parameters in the transfer matrix of eq.([tz2 ] ) : z_s^ * & = & _ 1 + _ -1 + z_t^ * & = & ( ) _ 1 + ( ) _ -1 . the values of the other two parameters @xmath104 and @xmath105 can be found but after a cumbersome calculation involving ten variables . we do not show this computation here but just remark that it is possible to find these parameters as well . though this computation may seem irrelevant it is an important calculation to show that the semion model can not be obtained from the transfer matrix in eq.([tz2 ] ) . we will make more comments about this in the remarks section . in general it is important to consider the orientation of the two dimensional triangulated lattice . there are orientations for the links and the plaquettes of this lattice . these are shown in fig([lattop ] ) . for the @xmath5 case the orientations are not important as the inverse of the group elements in this case is the same as the group element itself . the ground states @xmath131 for the hamiltonian in eq.([sides2 ] ) is given by the following conditions a_v^1|g= |g , a_v^-1|g= 0 , b_p|g= -|g ; v , p.to validate these conditions we assume the relative magnitudes of the parameters in the hamiltonian in eq.([sides2 ] ) according to the conditions stated . for the @xmath5 case we have the following operators create the flux ( violations of the plaquette operator conditions ) and charge ( violation of the vertex operator conditions ) . [ excite2 ] v_^*^f & = & _ i^ * _ i^x + v_^c & = & _ i _ i^z_j(n_1x_j^1 + n_-1x_j^-1 ) where @xmath132 and @xmath133 are strings in the direct and dual lattice respectively . the parameters @xmath134 and @xmath135 satisfy @xmath136 . this condition makes the operator creating the flux independent of @xmath137 . the dyonic excitations is created by applying both these strings on the dual and direct lattice respectively . the statistics of these excitations of these particles can easily be found out using the operators in eq.([excite2 ] ) . the self statistics of fluxes are trivial as the operator creating the flux is the same as in the toric code case . hence they remain bosonic . the self statistics of charges is also seen to be bosonic as ^z_i(n_1 x_i^1 + n_-1x_i^-1)^z_i(n_1 x_i^1 + n_-1x_i^-1 ) = 1 . the mutual statistics between the charges and the fluxes is clearly fermionic as can be trivially seen from the expressions of the operators creating the charges in eq.([excite2 ] ) . for convenience we now label the particles in this model as 1 , @xmath138 , @xmath139 and @xmath140 . 1 is the vacuum , @xmath138 is the charge , @xmath139 is the flux and @xmath140 is the dyonic excitation . the fusion rules can be obtained from the algebra of the orators creating them in eq.([excite2 ] ) , and they are found to be [ fusion2 ] & = & mm= = 1 + m & = & and the remaining fusion rules involving the fusion of the vacuum are all trivial . thus we find the fusion rules to be the same as the ones for the @xmath5 anyon model . the elements of the group @xmath0 are denoted by @xmath56 with @xmath57 , the elements of the group algebra @xmath3 and its dual are denoted by @xmath58 and @xmath59 respectively . in this case the transfer matrix is written as [ tzn ] u((_n ) , z_s , z_t , z_s^ * , z_t^ * ) & = & _ p(_k=0^n-1_^k b_p^^k)_j(_k=0^n-1b_^kt_j^^k ) + & & _ i(_k=0^n-1a_^kx_i^^k)_v(_k=0^n-1_^k a_v^^k ) where the index @xmath60 labels the irr of @xmath0 . the operators @xmath61 , @xmath62 , @xmath63 and @xmath64 are given by a_v^^k & = & _ l=0^n-1_^k(^l)x_i_1^lx_i_2 ^ -lx_i_3^lx_i_4 ^ -l + b_p^^k & = & _ l=0^n-1_^k(^l)z_j_1^lz_j_2^lz_j_3 ^ -lz_j_4 ^ -l + x_i^^k & = & _ l=0^n-1_^k(^l)x_i^l + t_j^^k & = & _ l=0^n-1 _ ^k(^l)z_j^l where @xmath65 and @xmath66 are single qudit operators generating @xmath0 and are defined by [ genxn ] x_i |^k & = & [ genzn ] z_j|^k & = & ^k|^k . @xmath67 is the character of the element @xmath56 in the irr labeled by @xmath60 . the parameters on the two sides of eq.([tzn ] ) are related in the following way z_s & = & _ k=0^n-1_^k_^k + z_t & = & _ k=0^n-1(_l=0^n-1_^k(^l)a_^l)_^k + z_s^ * & = & _ k=0^n-1b_^k_^k + z_t^ * & = & _ k=0^n-1(_l=0^n-1_^k(^l)_^l)_^k . the hamiltonian in this case is given by [ sidesnv ] h = _ v ( a_v + h.c . ) + _ p ( _ k=0^n-1_^kb_p^^k ) with @xmath95 given by a_v & = & & & _ i_k , kp , k=(1,2 ) , k=(3,4)(_j=0^n-1a_^jt_i_k^^j)(_j=0^n-1b_^jt_i_k^^j ) where the parameters @xmath106 and @xmath107 satisfy @xmath108 for @xmath109 and @xmath110 , @xmath111 for @xmath112 . these operators commute for adjacent vertices when the parameters satisfy these conditions . it is also easy to check that @xmath113 and hence it has the nth roots of unity as its eigenvalues . as the hermitian conjugate is added to @xmath95 the eigenvalues are real . the dual version of the model in eq.([sidesnv ] ) is given by the hamiltonian [ sidesn ] h = _ v(_k=0^n-1 _ ^ka_v^^k ) + _ p ( b_p+h.c . ) where @xmath103 is given by b_p & = & ( _ k=0^n-1 ^n - k b_p^^k ) + & & _ i_k , kp , k=(1,2 ) , k=(3,4 ) ( _ j=0^n-1 a_^jx_i_k^^j)(_j=0^n-1 b_^jx_i_k^^j ) with the solutions for the @xmath114 s given by the relations @xmath115 . the modulus of these parameters are the same as in their dual version . these operators commute for adjacent plaquettes and @xmath116 . for the @xmath3 case , the hamiltonian is given by eq.([sidesn ] ) . the conditions for the ground states are similar to the @xmath141 case . again we assume the parameters to be such that these conditions hold . as in the @xmath141 case the fluxes do not change but the operators creating the charges have to be modified . there are @xmath142 charges and we denote them by @xmath143 with @xmath144 . the operators creating them are given by [ chargen]v_^e^n - k = z^n - k(_j=0^n-1m_(n - k)^jx^^j ) where @xmath132 is a direct triangle @xcite and the parameters @xmath145 satisfy [ recurn]a_^j - km_(n - k)^j = m_(n - k)^j - n+1a_^j for @xmath146 . at this point there seems to be one free parameter among the @xmath145 s for a given charge @xmath143 . the remaining are found in terms of this single `` free '' parameter using the recursion relations in eq.([recurn ] ) . however this is not there once we consider the fusion rules . they consistently fix all the parameters @xmath145 in the string . the string operators for the fluxes are the same as in the @xmath3 quantum double model and are given by [ fluxn ] v_^*^f^k = x_j^k where @xmath133 is a dual triangle and @xmath26 is the label for this edge . it is easy to check using the string operators in eq.([chargen ] ) and eq.([fluxn ] ) that the statistics of these particles are the same as in a @xmath0 anyon model . the charges and fluxes have bosonic self - statistics . the mutual statistics are not trivial and they are given by e^n - k_1f^k_2 = ^(n - k_1)k_2f^k_2e^n - k_1 . the fusion rules of the fluxes do not change . the fusion rules of the charges are the same when we constrain the free parameters of the parameters @xmath145 . this constrain is given by m_(n - k_1)^j - k_2m_(n - k_2)^j = m_(n-(k_1+k_2))^j for @xmath147 , @xmath148 belong to @xmath149 and @xmath146 . this relation comes when we impose the @xmath0 fusion rules for the charges for the operators creating the charges in eq.([chargen ] ) which is e^n - k_1e^n - k_2 = e^n-(k_1+k_2 ) . these fix all the @xmath145 parameters unambiguously . as we have shown the existence of all the charges and fluxes of the @xmath3 quantum double model exist in our model they are in the same phase . a similar construction can be carried out when the vertex operator is modified and the flux operator is left unchanged as given in eq.([sidesnv ] ) . this model is also in the same phase as the @xmath3 quantum double phase . the group @xmath68 . it has three conjugacy classes , @xmath69= \left([1 ] , [ r ] , [ \tau]\right)$ ] , where @xmath70=\{1\}$ ] , @xmath71 = \{r , r^2\}$ ] and @xmath72 = \{\tau , \tau r , \tau r^2\}$ ] . we denote the group elements by @xmath30 , the elements of the group algebra and its dual by @xmath73 and @xmath74 respectively . the transfer matrix is given by [ ts3 ] u((s_3 ) , z_s , z_t , z_s^ * , z_t^ * ) & = & _ p ( _ c _ cb_p^c ) + & & _ j(_gs_3 b_gt_j^g ) _ i ( _ r a_rx_i^r ) _ v a_v with [ operatorss3]a_v & = & _ gs_3_g l_i_1^gr_i_2^g^-1l_i_3^gr_i_4^g^-1 + b_p^c & = & _ \{g_i}(g_1 g_2 g_3 g_4,c ) t_j_1^g_1t_j_2^g_2t_j_3^g_3 ^ -1t_j_4^g_4 ^ -1 + [ operatorsxs3 ] x_i^r & = & _ gs_3_r(g ) l_i^g . the operators @xmath75 are projectors to the various elements @xmath76 . they are given by the following relations t^1 & = & + a + z_11 + ( z^2)_11 + t^r & = & + a + ^2z_11 + ( z^2)_11 + t^r^2 & = & + a + z_11 + ^2 ( z^2)_11 + t^ & = & - a + z_22 + ( z^2)_22 + t^r & = & - a + ^2z_22 + ( z^2)_22 + t^r^2 & = & - a + z_22 + ^2 ( z^2)_22 where & = & ( cc 1 & 0 + 0 & 1 ) ; a = ( cc 1 & 0 + 0 & -1 ) + z_11 & = & ( cc z & 0 + 0 & 0 ) ; ( z^2)_11 = ( cc z^2 & 0 + 0 & 0 ) + z_22 & = & ( cc 0 & 0 + 0 & z ) ; ( z^2)_22 = ( cc 0 & 0 + 0 & z^2 ) where each of the matrices is divided in blocks of three by three matrices . 1 denotes the three by three identity and @xmath77 is the generator of @xmath78 defined in eq.([genzn ] ) . the parameters on the two sides of eq.([ts3 ] ) are related by z_s & = & _ [ 1]_1 + _ [ r](_r + _ r^2 ) + _ [ ] ( _ + _ r + _ r^2 ) + z_t & = & _ rs_3 a_r _ gs_3_r(g ) _ g + z_s^ * & = & _ gs_3b_g_g + z_t^ * & = & _ gs_3 _ g_g . the transfer matrix in eq.([tg ] ) can be used to obtain the hamiltonians through u((g ) , z_s , z_t , z_s^ * , z_t^ * ) = e^-h where @xmath79 is the hamiltonian . in the next section we will see that there are several ways of grouping terms in the transfer matrices while taking their logarithms which result in different hamiltonians for a given set of parameters . the group @xmath1 has six elements given by @xmath117 with @xmath118 and @xmath119 . we choose @xmath30 to belong to a normal subgroup @xmath120 of @xmath1 . the only normal subgroup of @xmath1 is given by @xmath121 . thus the modified vertex operator has either @xmath122 or @xmath123 . there are three conjugacy classes for @xmath1 given by @xmath69 = \left(\{1\ } , \{r , r^2\ } , \{\tau , \tau r , \tau r^2\}\right)$ ] . the hamiltonian is given by [ sides3v ] h = _ v ( a_v^r + h.c . ) + _ p_c_c(_kcb_p^k ) with @xmath124 given by a_v^r = l^r_i_1(_hs_3 a_ht^h)r^r^2_i_2(_hs_3 b_ht_i_2^h)l^r_i_3(_hs_3 a_ht_i_3^h)r^r^2_i_4(_hs_3 b_ht_i_4^h ) with the parameters satisfying the following b_r & = & b_1 ; b_r^2 = b_1 + b_r^2 & = & b _ ; b_r = b _ and they have the same modulus . the eigenvalues of @xmath124 are found by computing its cube as @xmath125 . this gives the eigenvalues as @xmath126 and @xmath127 . for the chosen parameters the model is solvable as all the terms commute with each other . we can define exactly solvable models by choosing @xmath30 to be either @xmath128 , @xmath129 , or @xmath130 . these models are in another phase which we do not discuss here . in the next section we find out the quantum phases in which these models live in by studying their excitations . the quantum phase of this case is not in the phase of the corresponding quantum double model . this is because some of the particles in the quantum double phase condense and some others do not exist . we will write down which ones condense and which ones do not exist for the @xmath1 example in what follows . we will work with the vertex version of the model given by eq.([sides3v ] ) . the @xmath1 anyon model consists of eight excitations including the vacuum @xcite . they are given by @xmath150 1 is the vacuum , @xmath151 and @xmath152 are the abelian and non - abelian charges respectively . @xmath153 and @xmath154 are the non - abelian fluxes in the model . @xmath155 , @xmath156 and @xmath157 are the dyonic excitations . in the notation of @xcite these excitations are given by @xmath158 we will switch to this notation when we discuss the fusion rules in this model . the operators creating these excitations , called ribbon operators , can be found in @xcite . we do not write all these operators here . we will only write down operators which will be modified in our model . the abelian charge @xmath151 condenses as the ribbon operator creating this excitation commutes with both the plaquette and the modified vertex operators . the ribbon operator creating the non - abelian charge @xmath152 is not modified . the non - abelian flux @xmath153 is created by the following modified operators . [ modfl3 ] _ ; id(r)^(1,1)(1,1 ) & = & f _ ; id(r)^(1,1)(1,1)(_hs_3 n_ht^h ) + _ ; id(r)^(2,1)(2,1 ) & = & f _ ; id(r)^(2,1)(2,1)(_hs_3 m_ht^h ) where the parameters are given by n_r & = & n_1 ; n_r^2 = n_1 + n_r & = & n _ ; n_r^2 = n_and m_r^2 & = & m_1 ; m_r = m_1 + m_r^2 & = & m _ ; m_r = m_. at this point @xmath134 , @xmath159 , @xmath160 and @xmath161 seem to be free parameters . however the ribbon operators should not depend on these free particles as this will indefinitely increase the number of particles . we will see that the fusion rules fix these parameters unambiguously . the non - abelian flux @xmath154 does not exist in this case as if we attempt to modify the corresponding ribbon operator to commute with the modified vertex operator we will be forced to constrain the parameters @xmath85 which eventually makes the model trivial in the sense they remove the additional qudit terms on the sides . hence this changes the model and the phase . so we conclude that in this phase this flux excitation does not exist or they create states which can not be expanded in the basis of the excitations of the condensed model . this is not surprising as even in the @xmath4 quantum double model we can write operators which after acting on the ground states create states that can not be expanded in the basis of the eight excitations of the @xmath1 anyon model . one such operator is given by the following matrix o = ( cc 0 & z + 0 & 0 ) . it can be easily checked that this matrix does not commute with both the plaquette and vertex operators of the quantum double model of @xmath4 . it can also be checked that this matrix can not be expanded in terms of the matrices creating the eight excitations of the @xmath4 quantum double model . the operator creating the flux excitation @xmath154 faces a similar situation in our model . the non - existence of this flux also means that the dyonic excitation @xmath157 does not exist in this model . on the other hand the dyonic excitations given by @xmath155 and @xmath156 do exist in this model and they are created by the modified operators creating flux excitations and the unchanged charge operators . we now fix the remaining `` free '' parameters @xmath134 , @xmath159 , @xmath160 , @xmath161 using appropriate fusion rules . the fusion rules of the @xmath1 anyon model can be found in @xcite . as we do not have the excitations denoted by @xmath162 and @xmath163 we can ignore their fusion rules . they are decoupled from the fusion rules of the other particles . the fusion rule which will be useful to fix the remaining parameters is given by j^xj^x = 1 + a + j^x.the operators creating the excitation @xmath164 are not modified in our model and hence we can not use this channel to fix the remaining parameters . we use the other two channels instead . so we have _ ; id(r)^(1,1)(1,1 ) _ ; id(r)^(2,1)(2,1 ) = 1.this gives us the conditions m_1n_1 = ; m_n_= . the second channel is obtained by _ ; id(r)^(1,1)(1,1 ) _ ; id(r)^(1,1)(1,1 ) = _ ; id(r)^(2,1)(2,1 ) which gives us the conditions m_1 = n_1 ^ 2 ; m_= n_^2.these conditions unambiguously fix all the parameters in the ribbon operators of eq.([modfl3 ] ) . thus we have seen that this model is in a quantum phase different from the @xmath4 quantum double model . the hamiltonians can be got from the transfer matrix @xmath80 by taking its logarithm . there are several ways to do this and in general we can obtain many complicated models with several spins interacting . however for simplicity we will only consider four spin couplings . unless specified we will work only on the square lattice . however the models we define are not sensitive to the details of the lattice in the sense that the phase they describe is the same irrespective of the choice of the lattice . the first two operators on the vertices and plaquettes are four spin interaction terms by construction . the operators on the links @xmath27 , @xmath26 are not interaction terms between spins but magnetic field terms or perturbations @xcite as noted previously . these terms do not commute with the vertex and plaquette operators in general making the procedure for taking logarithms in their presence cumbersome . we now write down exactly solvable models with vertex operators modified by the perturbations and which continue to remain in a topological phase . this model is first written for a general group algebra @xmath8 and then we discuss the abelian and non - abelian cases separately . we find out the quantum phase for the three examples considered . the transfer matrices considered so far were made up of two kinds of operators , the plaquette and vertex operators , given by @xmath46 and @xmath165 , which coupled four qudits and the single qudit operators , given by @xmath166 and @xmath167 . such operators arise in a transfer matrix parametrized by the centers of the algebra @xmath17 and its dual @xmath20 . we denoted these transfer matrices as @xmath168 where @xmath17 was taken to be the group algebra @xmath8 . the parameters @xmath104 , @xmath105 , @xmath169 and @xmath47 are the centers of the algebra and its dual respectively . in the pictorial language of @xcite these parameters were associated to the spacelike plaquettes , timelike plaquettes , spacelike links and timelike links respectively . these are the operators which result in the transfer matrix of a lattice gauge theory and we studied the quantum phases we can produce using the hamiltonians from these transfer matrices . in particular we obtained the quantum double phase of abelian gauge groups from non - trivial hamiltonians constructed by modifying these transfer matrices . for example in the @xmath2 case we obtained the toric code phase . however it is not possible to obtain the double semion phase from this modified transfer matrix . in order to obtain this we need to consider more general transfer matrices made up of other kinds of operators . such operators can be obtained by considering more general transfer matrices parametrized by the centers of not just the algebra @xmath17 but by two copies of the algebra @xmath170 and its dual which result in operators coupling two qudits . a transfer matrix of this sort can be formally written as @xmath171 where @xmath172 and @xmath173 are the centers of @xmath170 and its dual respectively . in the language of the pictorial formalism introduced in @xcite , these parameters are associated to the volumes and the vertices of the triangulated three dimensional lattice respectively . we will not go into the details of this pictorial formalism in this paper and only show the hamiltonians that we can obtain from such a transfer matrix . it is possible to build such transfer matrices made up of the usual four qudit plaquette and vertex operators as well as two and three qudit operators which are combinations of @xmath166 and @xmath167 , acting on different links , coupled together . such considerations will lead to both the toric code phase and the double semion phase as we shall see now . we note that such transfer matrices are not the transfer matrices of the usual lattice gauge theory based on discrete groups . we only work with the case of @xmath49 in what follows . we will work on a lattice where there are vertices that have valency three . the modified vertex operator acting on this vertex is shown in fig([vertex ] ) . in general the lattice is oriented but as we are working with the @xmath2 case in this section we ignore the orientation . the operators coupling two links together are given by @xmath174 , @xmath175 and @xmath176 using these three two link operators we can write down the modified vertex operator as [ z2d ] a_v^-1 & = & x_l_1x_l_2x_l_3 + & & ( a_1 t_l_1 ^ 1t_l_2 ^ 1t_l_3 ^ 1 + a_2 t_l_1 ^ 1t_l_2 ^ 1t_l_3 ^ -1 + a_3 t_l_1 ^ 1t_l_2 ^ -1t_l_3 ^ 1 + a_4 t_l_1 ^ 1t_l_2 ^ -1t_l_3 ^ -1 . a_5 t_l_1 ^ -1t_l_2 ^ 1t_l_3 ^ 1 + a_6 t_l_1 ^ -1t_l_2 ^ 1t_l_3 ^ -1 + a_7 t_l_1 ^ -1t_l_2 ^ -1t_l_3 ^ 1 + a_8 t_l_1 ^ -1t_l_2 ^ -1t_l_3 ^ -1 ) . the relation between @xmath177 s and @xmath114 s are given by a_1 & = & b_11b_21b_31 , a_2 = b_11b_22b_33 , a_3 = b_12b_23b_31 , a_4 = b_12b_24b_33 + a_5 & = & b_13b_21b_32 , a_6 = b_13b_22b_34 , a_7 = b_14b_23b_32 , a_8 = b_14b_24b_34 . for adjacent vertex operators to commute the parameters have to satisfy the following relation = = = .thus there are four free parameters given by @xmath178 , @xmath179 , @xmath180 and @xmath181 . in order to have @xmath182 we further restrict them with the relation @xmath183 . thus there are just three free parameters now . the plaquette operator is unchanged and is that of the toric code . we now obtain the excitations in this model . the operator creating the vertex excitations that is the charges is unchanged with respect to the usual toric code . the change is seen in the operators creating the fluxes . they have to be modified in order for them to commute with the modified vertex operator given in eq.([z2d ] ) . consider the following operator creating the flux on the two plaquettes adjacent to link @xmath184 as shown in the fig([triangle ] ) . and the two plaquettes adjacent to it . ] [ f2d ] f_l_1 & = & x_l_1 + & & ( n_1 t_l_1 ^ 1t_l_2 ^ 1t_l_3 ^ 1 + n_2 t_l_1 ^ 1t_l_2 ^ 1t_l_3 ^ -1 + n_3 t_l_1 ^ 1t_l_2 ^ -1t_l_3 ^ 1 + n_4 t_l_1 ^ 1t_l_2 ^ -1t_l_3 ^ -1 . n_5 t_l_1 ^ -1t_l_2 ^ 1t_l_3 ^ 1 + n_6 t_l_1 ^ -1t_l_2 ^ 1t_l_3 ^ -1 + n_7 t_l_1 ^ -1t_l_2 ^ -1t_l_3 ^ 1 + n_8 t_l_1 ^ -1t_l_2 ^ -1t_l_3 ^ -1 ) + & & ( n_1 t_l_1 ^ 1t_l_2^1t_l_3^1 + n_2 t_l_1 ^ 1t_l_2^1t_l_3^-1 + n_3 t_l_1 ^ 1t_l_2^-1t_l_3^1 + n_4 t_l_1 ^ 1t_l_2^-1t_l_3^-1 . n_5 t_l_1 ^ -1t_l_2^1t_l_3^1 + n_6 t_l_1 ^ -1t_l_2^1t_l_3^-1 + n_7 t_l_1 ^ -1t_l_2^-1t_l_3^1 + n_8 t_l_1 ^ -1t_l_2^-1t_l_3^-1 ) . for this operator to commute with the vertices @xmath185 and @xmath186 we have the following relations n_5 = n_1 k , n_6 = n_2 k , n_7 = n_2 k , n_8 = n_1 k , with @xmath187 . the relation between @xmath134 and @xmath188 is fixed by the condition that @xmath189 commutes with the vertex operators associated to @xmath28 and @xmath190 . this is given by n_1 ^ 2 = n_2 ^ 2.we now study the two cases separately . when @xmath191 we have the following form for the operator @xmath189 given by [ f1 ] f_l_1 = n_1 ^ 2 x_l_1 where @xmath192 and @xmath77 are pauli matrices acting on the link @xmath184 . this string operator can not create the semion phase and hence remains in the toric code phase . this can be seen by computing the fusion rules and the braiding statistics . the fusion rule for this operator is given by f_l_1f_l_1 = n_1 ^ 4k^2.by choosing @xmath193 we obtain the fusion rules for the toric code phase . let us work on the hexagonal lattice as shown in fig([braiding ] ) for computing the braiding rules . we consider two sets of fluxes created by the operators @xmath194 and @xmath195 respectively . the braiding is carried out as shown in fig([braiding ] ) and is given by [ gb ] f_f_ & = & x_12x_13x_43x_23x_13x_14 + & & nt^23_0,2,3 nt^13_0,1,3 nt^23 , 13_2 , 3 , 4 nt^13 , 14_2 , 1 , 4 nt^13 , 14_2 , 1 , 4 nt^13_1 , 3 , v nt^14_1 , 4 , v + & & nt_1 , 3 , v^ nt_1 , 2 , v^ nt_2 , 1 , 4 nt_2 , 3 , 4 nt_v^ , 1 , 3 nt_v^ , 4 , 3 where @xmath196 is the part of @xmath195 that intersects @xmath194 . [ bg ] f_ f _ & = & x_23x_13x_14x_12x_13x_43 + & & nt^13_1,3 , v^ nt^12_1,2 , v^ nt^12 , 13_2 , 1 , 4 nt^13 , 34_2 , 3 , 4 nt^13_v^ , 1 , 3 nt^34_v^ , 4 , 3 + & & nt_0 , 2 , 3 nt_0 , 1 , 3 nt_2 , 3 , 4 nt_2 , 1 , 4 nt_1 , 3 , v nt_1 , 4 , v. the notation requires explanation . the vertices are numbered as shown in fig([braiding ] ) . with this numbering @xmath197 , for example , denotes the @xmath192 pauli matrix acting on the link between vertices numbered 1 and 2 . we also have nt_0 , 2 , 3 nt_0 , 1 , 3 & = & + & & . and we have nt^23_0,2,3 nt^13_0,1,3 & = & + & & . it is easy to see from the expression for the operator creating the flux in eq.([f1 ] ) that the two equations eq.([gb ] ) and eq.([bg ] ) are the same . hence the flux excitation is bosonic . for the case when @xmath198 we have the following operator that creates the flux excitation . [ f2 ] f_l_1 = n_1 ^ 2 x_l_1z_l_2z_l_3z_l_2z_l_3. the fusion rule for this operator is the same as the fusion rule of the operator in eq.([f1 ] ) . it is easy to check using the expression for the flux operator in eq.([f2 ] ) in eq.([bg ] ) and eq.([gb ] ) that the flux operator is bosonic in this case as well . in order to obtain the double semion phase we need to use the following form for the modified vertex operator [ asem ] a_v = _ l_1 , l_2 , l_3\{1 , -1 } x_l_1x_l_2x_l_3 ( l_2 ^ -1l_1 ^ -1 , -l_2 , -1)(-l_2 , -1 , l_3)(-l_1 , -1 , l_3)^-1t_l_1^l_1 t_l_2^l_2 t_l_3^l_3where @xmath199 is an element of the cohomology group @xmath200 the cohomology group @xmath200 is given by @xmath2 . hence there are two cocycles one of which is the trivial cocycle . it is easy to see that for the trivial cocycle the operator @xmath165 in eq.([asem ] ) reduces to the operator of the toric code . for the non - trivial cocycle @xmath201 , the rest are equal to the identity element . the operator in eq.([asem ] ) is precisely the same as the one used to define the twisted quantum double model in @xcite for the @xmath5 case . the two - link operators that are used to obtain the modified vertex operator in eq.([asem ] ) are given by @xmath202,\ ] ] @xmath203\ ] ] and @xmath204.\ ] ] the expressions for the modified flux operator on a single link @xmath184 is given by f_l_1 = x_l_1where @xmath205 and @xmath206 are functions of the variables on the links @xmath184 , @xmath207 , @xmath208 , @xmath209 and @xmath210 . the action of this string is shown in fig([triangle ] ) . using the convention of @xcite we order the vertices as follows @xmath211 . the operator creating the flux , @xmath189 has to commute with @xmath165 , @xmath212 , @xmath213 and @xmath214 . this imposes the following conditions on the coefficients @xmath205 and @xmath206 & = & + & = & + & = & 1 . using these conditions it is easy to check that the fusion and braiding rules for the operator creating the fluxes is that of the double semion phase . we have used modified transfer matrices of 3d lattice gauge theories with discrete gauge groups to find exactly solvable quantum models in two dimensions . the abelian cases were studied with the example of the @xmath0 groups and were shown to generate a quantum phase identical to the quantum double phase based on the group algebra of these abelian groups . this was seen by studying the excitations of these systems . the excitations were the same as the @xmath0 anyon models . the only change was in the operator creating these excitations . in the case when the vertex operators were modified the flux excitations were created by new operators whereas in the case where the plaquette operators were modified the operators creating the charge excitations were modified . thus these models present another example of a system hosting @xmath0 anyons on the two dimensional lattice . there have been other examples of exactly solvable lattice models which host the @xmath0 anyons @xcite . the important point about the models presented in this paper is that they involve terms which are usually added as perturbations on the links to induce quantum phase transitions of the @xmath0 quantum double models . the models remain solvable in spite of including these terms . the only difference is that the modified vertex and plaquette operators obey a deformed version of the quantum double algebra . it is interesting question to find out if a suitable choice of a representation category can generate these models using the string - net formalism @xcite . the story changes when the discrete groups are non - abelian . the example of the dihedral group @xmath1 was studied and the phase was found to be different from that of the corresponding quantum double phase . in particular the phase was found to be a modified version of the @xmath1 quantum double phase . the abelian charge was no longer an excitation of the system instead it condensed to the ground state . the flux excitations corresponding to the @xmath72 $ ] conjugacy class is not existent that is can not be expanded in the set of excitations of the system . similar models were studied in @xcite where they add the magnetic terms to the links just like perturbations . the plaquette operators were considered to be projectors to a normal subgroup of the full gauge group @xmath7 . the resulting phases in @xcite are condensed versions of the @xmath1 quantum double phase . the construction of the transfer matrices for two dimensional lattice gauge theories was shown in @xcite . in particular it was shown to contain the quantum double models for any involutory hopf algebra . the models in @xcite were found by trivializing the @xmath105 and @xmath169 parameters . this implies there were no perturbation terms included in the transfer matrices considered in @xcite . the models considered in this paper can be considered as an extension of the formalism presented in @xcite where we make non - trivial choices for the parameters @xmath105 and @xmath169 . condensed phases of abelian quantum double models were exhibited in @xcite . the models considered in this paper can be regarded as another way to obtain condensed phases along lines similar to @xcite . the models considered in this paper included only four spin couplings . we could include more couplings which will result in more complicated models . for example if we couple a plaquette with the legs of the plaquette we can obtain models similar in spirit to the double semion model @xcite . however we can not use the transfer matrix in eq.([tz2 ] ) to obtain the double semion model . the reason being that there are no choices of parameters @xmath104 , @xmath105 which will result in this model . this calls for the construction of more general transfer matrices which can lead to these kinds of models . in section 5 we showed how we can go about finding the double semion phase by including two - qudit operators in the transfer matrix of lattice the gauge theory . we only worked out the case of @xmath5 in this section . for other groups this reduces to the twisted quantum double models constructed in @xcite . these models can be thought of as arising due to the introduction of new parameters to the vertices which we denoted as @xmath173 . hence it is natural to use @xmath172 to obtain the models dual to the twisted quantum double models . we will explore these in future works . it is interesting to note that the transfer matrix approach provides a variety of lattice gauge models including topological and non - topological ones . this approach may very well be appropriate for studying quantum phase transitions in these lattice models . pp , mjbf and pts thank fapesp for support of this work . jpij thanks cnpq for support during this work . d. v. else , c. nayak , _ classifying symmetry - protected topological phases through the anomalous action of the symmetry on the edge _ , phys . b 90 , 235137 ( 2014 ) and arxiv:1409.5436 [ cond-mat.str-el ] . m. koch - janusz , m. levin , a. stern , _ exactly soluble lattice models for non - abelian states of matter in 2 dimensions _ b 87 , 104406 ( 2013 ) and arxiv:1212.0593 [ cond-mat.str-el ] . phys . rev . b 88 , 115133 ( 2013 ) and arxiv:1306.2789 [ cond-mat.str-el ] .

pickands dependence functions characterize bivariate extreme value copulas . in this paper , we study the class of polynomial pickands functions . we provide a solution for the characterization of such polynomials of degree at most , , and show that these can be parameterized by a vector in belonging to the intersection of two ellipsoids . we also study the class of bernstein approximations of order of pickands functions which are shown to be ( polynomial ) pickands functions and parameterized by a vector in belonging to a polytope . we give necessary and sufficient conditions for which a polynomial pickands function is in fact a bernstein approximation of some pickands function . approximation results of pickands dependence functions by polynomials are given . finally , inferential methodology is discussed and comparisons based on simulated data are provided . ./style / arxiv - general.cfg

1601.03906

bivariate extreme value copulas are characterized by the pickands dependence functions , these are functions @xmath3 \to{\mathbb{r}}$ ] which satisfy the conditions : 1 . _ boundary conditions _ : @xmath4 , @xmath5 $ ] . convexity condition _ : @xmath6 is convex . in the presence of the _ convexity condition _ , the above _ boundary conditions _ are simply saying that the lines @xmath7 and @xmath8 , @xmath9 $ ] , are both _ support lines _ of @xmath6 at @xmath10 and @xmath11 respectively , and it follows that they can be equivalently replaced by 1 . _ endpoint conditions _ : @xmath12 and @xmath13 , @xmath14 ( see , e.g. , roberts and varberg @xcite , page 14 , problem k ) . note also that in the case where @xmath6 is twice differentiable the _ convexity condition _ can be replaced by @xmath15 , for all @xmath5 $ ] . let @xmath16 be the space of pickands dependence functions . for @xmath17 , the copula is @xmath18 and @xmath19 has a density if @xmath20 is absolutely continuous on @xmath21 $ ] . bivariate extreme value copulas may also be parameterized by the so - called spectral measure @xmath22 , that is , a positive measure on @xmath21 $ ] such that @xmath23}w h(\mathrm{d}w)=1=\int_{[0,1]}(1-w ) h(\mathrm{d}w)$ ] . let @xmath24 denote the space of spectral measures . the one - to - one correspondence between @xmath16 and @xmath24 , is given by @xmath25}\bigl\{(1-t)w\bigr\ } \vee\bigl \{t(1-w)\bigr\ } h(\mathrm{d}w),\qquad t\in[0,1],\ ] ] and @xmath26)=1+a'_+(w)$ ] , for all @xmath27 , where @xmath28 is the right derivative of @xmath6 , see beirlant _ _ @xcite . in the past literature , parametric models for the function @xmath6 have been studied for instance in tawn @xcite , coles and tawn @xcite , joe _ et al . _ @xcite , ledford and tawn @xcite and dupuis and tawn @xcite . the modeling of @xmath22 has also been considered , as it offers some advantages , especially when extensions to the multivariate case are desired . inference within parametric families of spectral measures can be found in boldi and davison @xcite , coles and tawn @xcite , coles and tawn @xcite , de haan _ et al . _ @xcite , einmahl _ et al . _ @xcite , joe _ et al . _ @xcite , ledford and tawn @xcite and smith @xcite . polynomial splines models have also been proposed by guillotte _ _ @xcite and fougres _ et al . _ @xcite . in view of the infinite - dimensional nature of the space of pickands functions , inference is often done nonparametrically for maximum flexibility , see for instance capra _ @xcite , hall and tajvidi @xcite , fils - villetard _ et al . _ @xcite and recently bcher _ et al . _ @xcite , to cite only a few . however , the nonparametric estimators do not usually satisfy the properties of pickands functions for finite samples and are often modified to do so . this can be cumbersome , see , for instance , fils - villetard _ et al . _ @xcite and the references therein . one alternative to the nonparametric approaches is based on having a series of nested models indexed by @xmath29 , @xmath1 . each model is parametric but @xmath29 is not bounded . within each model , a genuine estimator of the pickands dependence function is proposed . in this spirit , we consider modeling the pickands function @xmath6 using polynomials on @xmath21 $ ] . the latter are natural candidates here , the quadratic case being known as the symmetric mixed model , while the cubic case corresponds to the asymmetric mixed model , see beirlant _ et al . _ @xcite , pages 308 and 309 , and the references therein . an attempt at characterizing the space of pickands polynomials for higher degrees was made in klppelberg and may @xcite , page 1472 , theorem 2.5 . it is stated there that a polynomial ( in the power basis ) @xmath30 , @xmath31 $ ] , is a pickands dependence function if and only if its coefficients satisfy the four following conditions : @xmath32 it turns out that these conditions fail to be sufficient for @xmath33 . in fact , a counterexample dates back to beirlant _ et al . _ @xcite , page 308 : the polynomial @xmath34 satisfies the above conditions but is not convex . we will return to this example in section [ sect : characterization ] . first , section [ sect : bernstein ] is a short preliminary section giving the essential definitions and general mathematical results concerning the bernstein basis used throughout the paper . the reason for this choice of basis here is , essentially , its ability to reflect geometric and algebraic particularities of this problem on the coefficients . this will be reinforced , with an example and pointers to the text in the concluding remarks of section [ sect : conclusion ] . in section [ sect : characterization ] , the aim is to characterize , for each polynomial having degree at most @xmath29 , @xmath1 , the domain in which the coefficients of the polynomial ( in the bernstein basis ) must belong . we are looking for a constructive solution leading ultimately to a parameterization , which could be directly applicable in the search of the maximum likelihood estimate , for instance . we approach the problem by exploiting the one - to - one correspondence between the pickands function and the related spectral measure mentioned above . for polynomials , this boils down to the link between the pickands function and its second derivative ( a nonnegative polynomial ) . the solution therefore exploits the characterization of nonnegative polynomials given by lukcs , see karlin and shapley @xcite . we subsequently refer to this solution as being the solution to the _ full model_. in section [ sect : approximation ] , the focus is on characterizing , for each degree @xmath1 , the space of polynomials that can be obtained from a bernstein approximation of a pickands dependence function . we call this model the _ submodel_. we try to give an answer to the question : what is the gap between the _ full model _ and the _ submodel _ ? now in section [ sect : quality ] , we use tools from approximation theory and probability for obtaining accurate bounds for measuring the closeness between the space of pickands dependence functions and the one obtained from the _ submodel_. note that bernstein approximations of copulas have appeared in the past literature and some of their properties have been studied in sancetta and satchell @xcite and sancetta and satchell @xcite . finally , in section [ sect : simulations ] , we present the results of a simulation comparing the maximum likelihood estimator from the full model with the one from the submodel and a version of the popular nonparametric `` cfg '' estimator in capra _ our estimators should be easily implemented , and could become part of standard extreme value statistics packages , such as the evd package in r. the existing packages offer only polynomial models of degree at most three , this may be explained by the absence of a successful characterization for higher degrees in the previous literature . let @xmath35 be the space of polynomials on the interval @xmath21 $ ] , with degree at most @xmath29 , @xmath1 . we shall represent polynomials in @xmath35 using the commonly called bernstein basis . for self - containedness of the exposition , we begin with a short ( preliminary ) section describing some of the key properties of this basis that will serve in the following developments . the connection between this basis and the binomial distribution provides a very powerful tool in some of the proofs . we are also going to exploit the binomial identities given below . henceforth , let @xmath36 be a @xmath37 random variable , @xmath38 , where the value of @xmath39 $ ] will always be clearly indicated by the context . the members of the bernstein basis of degree @xmath29 , @xmath1 , are the polynomials @xmath40,\end{aligned}\ ] ] for @xmath41 . the coefficients of a polynomial @xmath42 in the bernstein basis of degree @xmath29 , will be denoted by @xmath43 , and so for @xmath44 we have @xmath45,\qquad x\in[0,1].\ ] ] let @xmath46 be the forward difference operator . here , when applied to a function @xmath47 of two arguments : @xmath48 , it is understood that @xmath46 operates on the first argument : @xmath49 and @xmath50 , for all @xmath51 . here are some basic results related to the bernstein basis . [ thm : bern_prop ] for @xmath41 , we have a. @xmath52 , for @xmath53 , with the convention @xmath54 for @xmath55 , b. for @xmath56 , @xmath57 , c. @xmath58 $ ] . the following are direct but useful consequences of the above proposition . properties ( i ) and ( ii ) ( obtained using the summation by parts formula ) show how derivatives of polynomials represented in the bernstein basis act on the coefficients . [ thm : bern_calc ] let @xmath59 , with @xmath60 , @xmath61 , its first and second derivative respectively . a. @xmath62 , @xmath63 , @xmath53 , so that @xmath64,\ ] ] b. @xmath65 , @xmath66 , @xmath67 , so that @xmath68,\ ] ] c. for @xmath69 and @xmath70 , and @xmath5 $ ] , @xmath71 in section [ sect : approximation ] , we will construct a submodel based on bernstein approximations of pickands functions . in general , the @xmath29th - order bernstein approximation of any function @xmath72\to { \mathbb{r}}$ ] is given by @xmath73 , \qquad t \in[0,1].\ ] ] it transforms @xmath47 into the polynomial @xmath74 , with coefficients equal to the function values on the uniformly spaced grid @xmath75 . notice that @xmath76 and @xmath77 . however , @xmath78 does not necessarily interpolate @xmath47 on the grid . as a first implication of proposition [ thm : bern_calc ] above , we get that the convexity of @xmath47 implies that of @xmath78 . we will make extensive use of the following identities . let @xmath79 be independent @xmath80 random variables , @xmath5 $ ] , and let @xmath81 be a function . here @xmath82 , and @xmath83 \nonumber \\[-8pt]\\[-8pt ] & = & t{\mathrm{e}}_t\bigl\{f(s_{m-1}+1)\bigr\}+(1-t ) { \mathrm{e}}_t\bigl\{f(s_{m-1})\bigr\},\nonumber \\ \label{eq : bin_expecsfs } { \mathrm{e}}_t\bigl\{s_{m}f(s_{m } ) \bigr\}&=&m{\mathrm{e}}_t\bigl\{x_{m}f(s_{m-1}+x_{m } ) \bigr\ } = m{\mathrm{e}}_t\bigl[{\mathrm{e}}_t\bigl \{x_{m}f(s_{m-1}+x_{m})\mid s_{m-1}\bigr\ } \bigr ] \nonumber \\[-8pt]\\[-8pt ] & = & mt{\mathrm{e}}_t\bigl\{f(s_{m-1}+1)\bigr\},\nonumber\end{aligned}\ ] ] so that from ( [ eq : bin_expecfs ] ) and ( [ eq : bin_expecsfs ] ) , we have @xmath84 . these identities are special cases of more general identities developed for the exponential families , see hudson @xcite . the problem of characterizing the polynomial pickands functions is solved in section [ sect : polpick ] below . to do so , we will take further advantage of the one - to - one correspondence between the spectral measure @xmath22 and the related pickands function @xmath6 . in fact , under absolute continuity of @xmath20 , we will show that the problem of modeling @xmath6 ( or @xmath22 ) boils down to modeling some nonnegative function @xmath85 on @xmath86 satisfying the condition @xmath87 although this is ( briefly ) mentioned in beirlant _ et al . _ @xcite , page 269 , we provide the details in section [ sect : representation ] because the use of this special function @xmath85 is a key element in the paper . note that this framework is slightly more general then what is really needed for obtaining the characterization later on , because we will be working with polynomials rather than merely absolutely continuous functions . however , it is effortless to do so here . let @xmath88 , let @xmath22 on @xmath21 $ ] be the spectral measure associated with @xmath6 and let @xmath89 be the lebesgue measure on @xmath21 $ ] . assume that @xmath20 is absolutely continuous on @xmath21 $ ] , then , in particular , the copula @xmath19 in equation ( [ eq : ca ] ) has a density with respect to the lebesgue measure on @xmath21 ^ 2 $ ] . in this case , we derive a convenient integral representation for @xmath90 , @xmath5 $ ] , in terms of the radon nikodm derivative @xmath85 of the restriction of @xmath22 on @xmath86 . note that @xmath22 may still have point masses at 0 and 1 , which can be seen by @xmath91 and @xmath92 . under the above regularity condition on @xmath20 , we will show that the knowledge ( almost everywhere ) of @xmath93 alone is enough to evaluate @xmath6 , and similarly , the knowledge of the radon nikodm derivative @xmath85 of the restriction of @xmath22 on @xmath86 is enough to know the measure @xmath22 . in particular , we have @xmath94 almost everywhere on @xmath86 . [ thm : representation ] a. let @xmath95 , let @xmath22 be the spectral measure associated with @xmath6 . assume that @xmath20 is absolutely continuous and let @xmath94 almost everywhere on @xmath86 . we have @xmath96 h(w ) \,\mathrm { d}w,\qquad t\in[0,1],\ ] ] and the function @xmath85 satisfies the condition ( [ function_h ] ) . let @xmath89 be the lebesgue measure and @xmath97 denote a dirac measure at @xmath98 , the spectral measure is given by @xmath99 , with @xmath100 b. conversely , let @xmath101 be a nonnegative function satisfying condition ( [ function_h ] ) . the measure @xmath22 given by ( [ h_representation ] ) is a spectral measure , its corresponding pickands function @xmath6 is given by ( [ a_representation ] ) , and we have @xmath102 almost everywhere on @xmath86 . \(i ) integration by parts yields @xmath103,\ ] ] and @xmath104,\ ] ] and by combining these two equations , we obtain @xmath105.\end{aligned}\ ] ] letting @xmath11 in ( [ 0tot ] ) and @xmath10 in ( [ tto1 ] ) shows that the condition ( [ function_h ] ) is satisfied : @xmath106 from the one - to - one correspondence between pickands functions and spectral measures , the only thing left to show is that if @xmath22 is the measure given by ( [ h_representation ] ) , then it is the spectral measure associated to @xmath6 , that is , equation ( [ bijection ] ) holds : @xmath107 h(w ) \,\mathrm{d}w \nonumber \\ & = & ( 1-t ) \biggl(1-\int_0^tw h(w ) \ , \mathrm{d}w \biggr)+t \biggl(1-\int_t^1(1-w ) h(w ) \,\mathrm{d}w \biggr ) \nonumber \\ & = & ( 1-t ) \biggl(h_1+\int_t^1w h(w ) \,\mathrm{d}w \biggr)+t \biggl(h_0+\int_0^t(1-w ) h(w ) \,\mathrm{d}w \biggr ) \\ & = & ( 1-t)h_1+th_0+\int_0 ^ 1 \bigl[\bigl\{(1-t)w\bigr\}\vee\bigl\{t(1-w)\bigr\}\bigr ] h(w ) \ , \mathrm{d}w \nonumber \\ & = & \int_{[0,1]}\bigl\{(1-t)w\bigr\ } \vee\bigl\{t(1-w)\bigr\ } h(\mathrm{d}w ) , \qquad t\in[0,1],\nonumber\end{aligned}\ ] ] as claimed . \(ii ) for the converse , if @xmath108 satisfies the condition ( [ function_h ] ) , then the measure @xmath22 given by ( [ h_representation ] ) is a spectral measure : it is clearly positive , and @xmath109 } w h(\mathrm{d}w)=h_1+\int _ 0 ^ 1wh(w)\,\mathrm{d}w=1 = h_0+\int _ 0 ^ 1(1-w)h(w)\,\mathrm{d}w = \int _ { [ 0,1 ] } ( 1-w ) h(\mathrm{d}w ) . \ ] ] let @xmath6 be its corresponding pickands function given by equation ( [ bijection ] ) . the above equalities that lead to ( [ identity ] ) show that @xmath6 is also given by ( [ a_representation ] ) . finally , a direct calculation shows that @xmath102 almost everywhere on @xmath86 . the space @xmath110 corresponds to polynomial pickands functions with degree at most @xmath29 , @xmath111 . the reader should refer to section [ sect : bernstein ] for notations used in the following . it is immediate that @xmath112 , and @xmath113 if and only if @xmath114 , for all @xmath5 $ ] . the nontrivial cases start with pickands polynomials of degree at least two . we will characterize @xmath115 , for all @xmath67 . note that in the bernstein basis , the _ endpoint conditions _ are quite simple to verify because they are directly related to the first and last two coefficients only . [ thm : pol_constraints ] let @xmath116 , @xmath1 . we have @xmath117 if and only if the two following conditions are verified : 1 . _ endpoint conditions _ : @xmath118 and @xmath119 2 . _ convexity condition _ : @xmath15 , @xmath5 $ ] . we need to verify the _ endpoint conditions _ , that is , @xmath120 , @xmath121 . in the bernstein basis , by proposition [ thm : bern_calc ] , we have @xmath122 and @xmath123 it is therefore the _ convexity condition _ above that is more delicate to express in terms of the coefficients of @xmath6 . however , the spaces @xmath124 and @xmath125 are still easy to characterize , as can be seen in the following example . [ ex : a2a3 ] we have @xmath126 if and only if @xmath127 and @xmath128 . to see this , let @xmath129 . the polynomial @xmath6 satisfies the _ endpoint conditions _ of proposition [ thm : pol_constraints ] if and only if @xmath127 and @xmath130 . moreover @xmath131 , for all @xmath5 $ ] , and by proposition [ thm : bern_calc ] , @xmath132 . therefore , @xmath6 is convex if and only if @xmath133 . we have @xmath134 if and only if @xmath135 , and the couple @xmath136 belongs to the polytope derived from the four linear inequalities : @xmath137 and @xmath138 to show this , let @xmath139 . the polynomial @xmath6 satisfies the _ endpoint conditions _ of proposition [ thm : pol_constraints ] if and only if @xmath135 , and @xmath140 . also , since @xmath141 , @xmath5 $ ] , we have @xmath142 , in @xmath21 $ ] if and only if @xmath143 . by evaluating @xmath144 for @xmath145 , we get that @xmath6 is convex if and only if @xmath146 . for @xmath147 , finding @xmath148 explicitly ( in example [ ex : a2a3 ] above ) is easy mostly because @xmath149 is nonnegative if and only if @xmath150 . when @xmath117 with @xmath151 things become more complicated , as the following example illustrates . [ ex : counterexpol ] consider the polynomial @xmath34 , which served as the counterexample provided by beirlant _ et al . _ @xcite discussed in the introduction . here , @xmath152 , so @xmath153 but @xmath154\neq[0,1]$ ] , and therefore @xmath155 . essentially , it all boils down to finding the coefficients making @xmath93 nonnegative , and the problem is now reoriented towards the characterization of such @xmath93 . we will essentially be using the _ integral representation _ theorem [ thm : representation ] and a characterization of the nonnegative polynomials on @xmath21 $ ] due to lukcs . the main result is given in theorem [ thm : mainthm ] , while a parametric representation of the polynomial pickands functions and the geometric shape of the corresponding parameter space is provided via theorem [ thm : parameterization ] . to get to these results , we first need two technical lemmas , namely lemma [ thm : htoa ] and lemma [ thm : pandq ] . because of the new orientation of the problem as mentioned above , it is convenient to introduce new spaces @xmath156 , @xmath1 . we say that a polynomial @xmath85 belongs to @xmath157 , @xmath1 , when the following conditions are satisfied : 1 . _ endpoint derivatives conditions : _ @xmath158 2 . _ nonnegativity condition : _ @xmath159 , @xmath5 $ ] . these are the same conditions as those given at the very beginning of this section when restricted to polynomials , in particular , the _ endpoint derivatives conditions _ ( [ eq : boundcondh ] ) come directly from the condition ( [ function_h ] ) , by using proposition [ thm : bern_prop ] . when a polynomial @xmath6 on @xmath21 $ ] has an integral reprentation ( [ a_representation ] ) , the condition ( [ function_h ] ) corresponds to @xmath160 , hence the name _ endpoint derivatives conditions_. from the expression ( [ a_representation ] ) we get the following result ( notice the resemblance between the polynomial @xmath6 and its coefficients in the bernstein basis ) . [ thm : htoa ] let @xmath1 . a. if @xmath161 and @xmath162 h(w ) \,\mathrm { d}w,\qquad t\in[0,1],\ ] ] then @xmath116 and its @xmath163th coefficient , @xmath164 , in the bernstein basis is @xmath165\\[-8pt ] & & \quad = 1-\frac{1}{m+1}\sum_{j=0}^m \biggl [ \biggl\ { \biggl(1-\frac{k}{m+2 } \biggr)\frac{j+1}{m+2 } \biggr\ } \wedge\biggl\{\frac{k}{m+2 } \biggl(1-\frac{j+1}{m+2 } \biggr ) \biggr\ } \biggr]c(j , m;h).\nonumber\end{aligned}\ ] ] b. if @xmath116 and @xmath94 , then @xmath166 and @xmath167 \(i ) we have the equality @xmath168 h(w ) \,\mathrm{d}w\nonumber\\[-8pt]\\[-8pt ] & & \quad = \int _ 0^tw h(w ) \,\mathrm{d}w+ t\int _ t^1 h(w ) \,\mathrm{d}w- t\int _ 0 ^ 1w h(w ) \,\mathrm{d}w.\nonumber\end{aligned}\ ] ] using proposition [ thm : bern_calc](iii ) and the binomial identities ( [ eq : bin_expecsfs ] ) , we obtain : @xmath169 and putting the three pieces together shows that ( [ eq : inteq ] ) equals @xmath170c(j , m;h ) \biggr),\ ] ] which in turn equals @xmath171 , and this gives ( [ eq : htoa ] ) . ( ii ) this is directly obtained from proposition [ thm : bern_calc ] . [ ex : a4 ] we can show directly that @xmath172 is nonnegative on @xmath21 $ ] if and only if @xmath173 indeed , if @xmath172 is nonnegative on @xmath21 $ ] , then necessarily @xmath174 and @xmath175 . since the latter conditions need to be satisfied , lets assume them and look at the remaining coefficient @xmath176 . if @xmath177 , then @xmath85 is nonnegative on @xmath21 $ ] . if however @xmath178 , then we have @xmath179 and @xmath180 . by expressing @xmath181,\ ] ] we see that @xmath85 is convex and attains its minimum on @xmath86 . the result follows from the sign of the minimum . the above conditions on @xmath182 and @xmath183 , together with the _ endpoint derivatives conditions _ and a direct application of lemma [ thm : htoa ] gives the characterization of @xmath184 . notice that , so far ( for @xmath185 ) , we have characterized @xmath115 using only elementary mathematics . for higher degrees however , we use a finer result by lukcs , concerning the characterization of nonnegative polynomials , which can be found in for instance karlin and shapley @xcite , szeg @xcite and plya and szeg @xcite . it says that a polynomial @xmath186 of degree at most @xmath29 , is nonnegative if and only if there are polynomials @xmath187 and @xmath188 with @xmath189 and @xmath190 , such that @xmath191 for all @xmath5 $ ] . the following lemma exploits lukcs result and provides expressions for the coefficients , in the bernstein basis , of nonnegative @xmath161 , @xmath53 , using the hypergeometric distribution . here @xmath192 if @xmath193 [ thm : pandq ] a polynomial @xmath161 , @xmath53 , is nonnegative on @xmath21 $ ] if and only if there exist polynomials @xmath187 and @xmath188 such that , the coefficients of @xmath85 are related to the coefficients of @xmath187 and @xmath188 ( in the bernstein basis ) in the following way : ( for notational simplicity here , let @xmath194 , @xmath195 and @xmath196 denote the coefficients ) a. when @xmath29 is even , @xmath197 , @xmath198 and @xmath199\\[-8pt ] & & \hphantom{{\mathrm{e}}\biggl\{}{}+ \frac{k(m - k)}{m(m-1)}q \biggl(y_2,\frac{m-2}{2 } \biggr)q \biggl(k - y_2 - 1 , \frac{m-2}{2 } \biggr ) \biggr\},\nonumber\end{aligned}\ ] ] with @xmath200 , @xmath201 , and @xmath202 , @xmath203 , b. when @xmath29 is odd , @xmath204 and @xmath205\\[-8pt ] & & \hphantom{\frac{1}{m}{\mathrm{e}}\biggl\{}{}+ ( m - k)q \biggl(y_2,\frac{m-1}{2 } \biggr)q \biggl(k - y_2 , \frac { m-1}{2 } \biggr ) \biggr\},\nonumber\end{aligned}\ ] ] with @xmath206 , @xmath207 , and @xmath208 , @xmath209 . the proof is a bit technical , and can be skipped without affecting the readability of what follows . see the . by putting everything together , we get our main theorem . [ thm : mainthm ] a. for @xmath53 , we have @xmath210 if and only if @xmath85 satisfies the endpoint derivatives conditions ( [ eq : boundcondh ] ) and there exist polynomials @xmath187 and @xmath188 such that the coefficients of @xmath85 ( in the bernstein basis ) are given by ( [ eq : even ] ) when @xmath29 is even or by ( [ eq : odd ] ) when @xmath29 is odd . b. if @xmath210 , @xmath1 , then @xmath6 given by ( [ eq : pola ] ) is in @xmath148 . lemma [ thm : htoa ] gives the coefficients of @xmath6 in terms of the ones of @xmath85 . c. if @xmath117 , @xmath1 , then @xmath211 . lemma [ thm : htoa ] gives the coefficients of @xmath85 in terms of the ones of @xmath6 . a parameterization of @xmath157 ( and therefore of the polynomial pickands functions ) can therefore be made via the coefficients of the polynomials @xmath187 and @xmath188 above . the following theorem describes the corresponding parameter space . [ thm : parameterization ] let @xmath1 , for polynomials @xmath212 and when @xmath53 @xmath213 , let @xmath214 be the concatenation of the coefficients of @xmath187 and the ones of @xmath188 in the bernstein basis , that is , @xmath215 let @xmath216 be constructed using polynomials @xmath187 and @xmath188 in formula ( [ eq : h ] ) . when @xmath217 , the coefficients of @xmath218 are given in lemma [ thm : pandq ] . the parameter space @xmath219 is given by @xmath220 where @xmath221 and @xmath222 are two ellipsoids in @xmath2 . for every fixed value of @xmath5 $ ] , the function @xmath223 is a positive semidefinite quadratic form . in fact , using the bernstein basis and ( [ eq : h ] ) , we find that @xmath224 is given by @xmath225 note that @xmath218 is the zero polynomial if and only if @xmath226 , and so if @xmath227 , then the function @xmath228 is positive except perhaps at a finite number of points ( roots ) . now let @xmath229 be the polynomial obtained from formula ( [ eq : htoa ] ) with @xmath85 being replaced by @xmath218 . since @xmath230 if @xmath227 , it follows that @xmath231 and @xmath232 are both positive definite quadratic forms . finally , theorem [ thm : mainthm ] says that @xmath233 if and only if the _ endpoint derivatives conditions _ are verified : @xmath234 and so the sets @xmath235 are therefore both ellipsoids in @xmath2 and @xmath236 . let @xmath237 be the set of bernstein s approximations of pickands functions . the second part of the following lemma gives a useful necessary and sufficient geometric condition on the coefficients ( in the bernstein basis ) for a polynomial @xmath238 to belong to @xmath239 . the first part is used for this characterization and is also useful in the next propositions . [ thm : piecewisepick ] for any polynomial @xmath240 , @xmath241 , represented in the bernstein basis by @xmath242 , @xmath5 $ ] , let @xmath243 be the piecewise linear function interpolating the points @xmath244 , @xmath245.\ ] ] a. if @xmath246 , @xmath247 , @xmath67 , then @xmath243 is convex , b. ( characterization ) @xmath248 if and only if @xmath249 . \(i ) consider the step function @xmath250 , @xmath5 $ ] . geometrically , @xmath251 corresponds to the right derivative of @xmath243 on [ 0,1 ) . now , @xmath246 , @xmath247 , implies that @xmath252 , @xmath209 , is nondecreasing , and therefore @xmath251 is nondecreasing on @xmath21 $ ] . since @xmath253 , for all @xmath254 $ ] , it follows that @xmath243 is convex . ( ii ) we have @xmath255 . if @xmath249 , then @xmath256 and therefore @xmath257 . for the converse , if @xmath258 , then there is some @xmath259 such that @xmath260 . for @xmath261 , the fact that @xmath249 follows trivially . for @xmath67 , the convexity of @xmath262 implies @xmath263 and ( i ) above shows that @xmath243 is convex . also , @xmath264 and finally @xmath265 and @xmath266 this shows that @xmath249 . the following proposition says that bernstein approximations of pickands functions are themselves pickands functions . [ thm : bernsteinapproxpick ] we have @xmath267 , for every @xmath53 . moreover , @xmath268 for @xmath269 , while @xmath270 . let @xmath271 , @xmath5 $ ] . for the first statement , let @xmath95 and @xmath272 , @xmath53 . we have @xmath273 , and @xmath274 is convex , see section [ sect : bernsteinapprox ] . using jensen s inequality , we obtain @xmath275,\ ] ] and this shows that the _ boundary conditions _ of a pickands function are verified and so @xmath276 . for the second statement , recall that @xmath277 . if @xmath278 for either @xmath279 or @xmath280 , then the arguments presented in example [ ex : a2a3 ] can serve to verify that the piecewise linear function @xmath243 interpolating the points @xmath244 belongs to @xmath16 . lemma [ thm : piecewisepick ] then implies that @xmath281 , when @xmath282 . the fact that @xmath243 belongs to @xmath16 is not necessarily true for @xmath33 as the following shows : take for instance @xmath283.\ ] ] it can be easily verified that @xmath284 , and we have @xmath285 , with @xmath286 , for @xmath287 and @xmath288 . suppose that @xmath289 , for some @xmath290 . then @xmath291 . this implies that @xmath292 , for all @xmath5 $ ] , which in turn implies that @xmath293 , for all @xmath5 $ ] . therefore , @xmath294 . the above result says that @xmath270 , but we can say even more than this : @xmath295 for every @xmath296 . this follows as a direct application of _ the gap _ theorem coming up ( notice that for @xmath6 given by in ( [ eq : a4 + ] ) we have @xmath297 ) . for @xmath1 , we call @xmath115 the _ full model _ and @xmath298 the _ submodel_. in order to give a parameterization of the submodel , let @xmath299 the next theorem shows that the ( parametric ) space of the coefficients @xmath300 is a polytope by showing that @xmath301 , @xmath1 . [ thm : h+ ] we have @xmath302 , @xmath1 , if and only if @xmath303 and @xmath304 let @xmath305 . if @xmath302 then @xmath210 and , by definition of @xmath157 , the conditions @xmath306 hold . when @xmath302 , there exists @xmath307 such that @xmath308 . therefore , by lemma [ thm : htoa ] and convexity of @xmath6 , @xmath309 conversely , suppose that a polynomial @xmath186 satisfies ( [ eq : boundary ] ) and ( [ eq : positivity ] ) , then @xmath210 , and from theorem [ thm : representation ] , there exists an @xmath117 such that @xmath94 . now let @xmath310\to\mathbb{r}$ ] be the piecewise linear interpolation of @xmath311 . the _ endpoint conditions _ on the coefficients of @xmath6 given by proposition [ thm : pol_constraints ] directly imply that @xmath243 satisfies the _ endpoint conditions _ of a pickands function . the convexity of @xmath243 is obtained by the nonnegativity of the coefficients of @xmath85 : lemma [ thm : htoa ] gives @xmath312 , @xmath201 , and the first part of lemma [ thm : piecewisepick ] gives the convexity of @xmath243 . therefore , @xmath249 , and by the second part of theorem [ thm : bernsteinapproxpick ] , @xmath313 , which means that @xmath314 . a useful property of the submodel @xmath298 , @xmath217 , is that it is nested . we show this in the next theorem , and we find the gap between the full model and the submodel . the proof relies on the lorentz degree of a positive polynomial on @xmath86 , see for instance powers and reznick @xcite . in the bernstein basis , it is easy to see that @xmath59 is nonnegative if @xmath315 , @xmath201 . this sufficient condition is not a necessary condition , see karlin and shapley @xcite , szeg @xcite and plya and szeg @xcite . it may happen that for some @xmath316 , @xmath317 the question is when does this happen ? for the ( interesting ) case : @xmath318 , the necessary and sufficient condition for having @xmath319 for some @xmath320 is that @xmath187 be positive on @xmath86 . this is known in the literature as bernstein s theorem , and it motivates the definition of the lorentz degree of a polynomial @xmath59 which is positive on @xmath86 : @xmath321 to illustrate this notion , here is a very instructive example that will lead us naturally to _ the gap _ theorem . the symmetric polynomials in @xmath322 have the following form : @xmath323 , \mbox { with } \alpha\in(0,1 ] \mbox { and } \beta\in[-1,2]\setminus \{0\}.\end{aligned}\ ] ] as can be seen in figure [ fig : asym ] below , @xmath324 ( which is the absolute value of the first derivative at the endpoints ) sets the triangle in which the curves belong , and then @xmath325 controls their curvature . and @xmath184 in the symmetric case . here @xmath326 with @xmath327 $ ] , @xmath5 $ ] . in both plots , the three curves correspond to @xmath328 and 2 respectively ( from bottom to top ) . the shaded regions hold the curves in @xmath329 . ( a ) @xmath330 . ( b ) @xmath331 . ] let @xmath332 , and lets find the values of @xmath67 , such that @xmath333 . the smallest such @xmath29 , if it exists , is the lorentz degree of @xmath334 , and as we will see , it depends on @xmath325 here . the function @xmath335 is given , for @xmath5 $ ] , by @xmath336,\qquad m\geq2.\end{aligned}\ ] ] using newton s binomial formula , we get @xmath337 fix @xmath338 $ ] . from ( [ all_coef ] ) , we see that when @xmath339 , we have @xmath340 , for @xmath341 , and this holds for every @xmath67 . therefore , the lorentz degree in this case is @xmath342 , and @xmath343 if and only if @xmath67 . next , when @xmath344 the smallest coefficient occurs when @xmath345 , and we get @xmath346\\[-8pt ] & & \quad = 2\alpha \biggl\{(1+\beta)-6\beta\frac{\lfloor m/2\rfloor(m-\lfloor m/2\rfloor)}{m(m-1 ) } \biggr\},\qquad m\geq2.\nonumber\end{aligned}\ ] ] by looking at the sign of ( [ min_coef ] ) and solving the inequality for @xmath29 we obtain @xmath347 if and only if @xmath348 . this tells us that the lorentz degree in this case is @xmath349 and also that @xmath343 if and only if @xmath350 . the lorentz degree increases without bound as @xmath325 increases to 2 . finally when @xmath351 , we get @xmath352 so @xmath353 or equivalently @xmath354 . as the following theorem shows , this is true because @xmath355 , that is , @xmath356 is nonnegative but fails to be positive on @xmath86 . finally , for this example , we note that @xmath357 if and only if @xmath358\times[-1,0)\cup(0,1/2]$ ] , see figure [ fig : asym ] . [ thm : gap ] for all @xmath53 , we have @xmath359 . moreover , for @xmath360 we have @xmath361 if and only if @xmath362 and @xmath363 for some @xmath364 . first , since @xmath365 , for @xmath269 , we already have @xmath366 . let @xmath367 , for some @xmath53 . we want to show that @xmath368 . since @xmath369 we just need to show that @xmath370 , @xmath371 . using the binomial identities from section [ sect : binident ] , we have @xmath372 leading to the identity @xmath373 from theorem [ thm : h+ ] it follows that @xmath374 , which proves the first assertion . now let @xmath360 . to prove the second assertion , we show that we have @xmath375 if and only if @xmath376 or @xmath377 for all @xmath364 . first , assume that @xmath375 . if @xmath378 , then @xmath376 . otherwise , @xmath379 , so @xmath362 and this implies that @xmath380 and @xmath381 for all @xmath364 . it follows that @xmath382 , for all @xmath364 . conversely , if @xmath376 then @xmath378 , while if @xmath85 is positive on @xmath86 then @xmath383 , where @xmath384 is the lorentz degree of @xmath85 . we are now concerned with the flexibility of the submodel @xmath298 consisting of the bernstein approximations of pickands functions . more precisely , we provide an answer to how well a pickands function @xmath95 can be approached in the space @xmath298 . we also work out the range @xmath385 , as @xmath29 varies , of some dependence measures @xmath386 on @xmath16 . [ thm : bernsteinapprox ] for @xmath53 , @xmath95 , @xmath5 $ ] , we have @xmath387 where @xmath388 moreover , when @xmath389 , with @xmath390 , @xmath5 $ ] , we get the finer approximation @xmath391 where @xmath392 , for @xmath393 , while @xmath394 for @xmath364 , with equalities in the last two inequalities when @xmath395 . the inequality @xmath396 follows from jensen s inequality @xmath397.\ ] ] the approximation formula ( [ eq : bernapproxa ] ) is obtained as follows . first , the convexity and the fact that @xmath398 , @xmath5 $ ] imply that @xmath399 , for all @xmath400 $ ] . from that , for @xmath5 $ ] , @xmath401 the last two equalities can be found in , for instance , johnson @xcite . in particular , the last equality is obtained using stirling s formula , @xmath402 , ( @xmath403 ) . here we should point out that the cauchy inequality gives the weaker result @xmath404 although the rate of convergence remains of the same order . to show ( [ eq : bernapproxv ] ) , making use of the binomial identities ( once again ) , we have , for @xmath5 $ ] , @xmath405 thus , @xmath406 the latter inequality being an equality at @xmath395 . finally , @xmath407 , and the rate of convergence follows again by stirling s formula . dependence measures for bivariate extremes have been studied in the literature , see for instance tawn @xcite and weissman @xcite . in particular , for @xmath95 the following measures were proposed @xmath408 , \qquad u\sim\mathcal{u}(0,1).\ ] ] let @xmath95 and consider the regions @xmath409 , @xmath410 , given by @xmath411 since @xmath95 we have @xmath412 and @xmath413 because , @xmath414 in particular , @xmath415,\qquad \tau_2\bigl \{b_m(a , \cdot)\bigr\}={\mathrm{e}}\bigl[4\bigl\{1-a(u_m / m)\bigr \}\bigr],\ ] ] where @xmath416 . note that since @xmath417 and @xmath418 converge in distribution to @xmath419 and @xmath420 respectively , and since @xmath6 is continuous and bounded , we obtain that @xmath421 , as @xmath422 , @xmath423 . the next proposition gives the range of these measures for the bernstein approximations . [ thm : depmeasures ] let @xmath390 , @xmath5 $ ] ( the pickands function corresponding to complete monotone dependence ) . for @xmath53 , @xmath424 $ ] , @xmath425 , with @xmath426 first , if @xmath427 , @xmath428 , then @xmath429 , so that @xmath430 , @xmath431 , and this implies that @xmath432 $ ] . the reverse inclusion follows from the convexity of the space @xmath433 , the fact that @xmath434 is a pickands function and the linearity ( under convex combinations ) of the functionals @xmath435 and @xmath436.=1 we now compare the maximum likelihood estimator from the full model and the one from the submodel through simulated data . in both cases , the maximum likelihood estimator ( mle ) is computed numerically . for a fixed @xmath111 , when using the full model , the estimator is the polynomial @xmath437 , where @xmath438 is the solution of the nonlinear constrained maximization problem @xmath439 where @xmath440 is given in ( [ eq : thetam ] ) , for @xmath53 , @xmath441 $ ] , @xmath442 is the data and the likelihood @xmath443 is obtained by computing the coefficients of @xmath85 via lemma [ thm : pandq ] , by deriving its associated pickands function @xmath6 via lemma [ thm : htoa ] , and then by using the mixed partial derivative of ( [ eq : ca ] ) . the estimator for the bernstein approximations submodel is @xmath444 , the problem to solve is @xmath445 where @xmath446 is the polytope given in theorem [ thm : h+ ] and @xmath447 is such that @xmath448 . again the likelihood @xmath449 is obtained by lemma [ thm : htoa ] , and ( [ eq : ca ] ) . both problems have solutions by continuity of the likelihood over their ( compact ) feasible regions @xmath440 and @xmath446 , respectively . these are easily solved by using the matlab global search algorithm , for instance . we consider the three following models for the simulation : the asymmetric logistic model @xmath450^{\alpha } , \qquad t\in[0,1],\ ] ] with @xmath338 $ ] , @xmath451 , the symmetric mixed model @xmath452,\ ] ] @xmath453 $ ] , and the polynomial @xmath6 obtained via lemma [ thm : htoa ] with @xmath454 , @xmath455 and @xmath456 , leading to @xmath457.\ ] ] while it is clear from the characterization of @xmath458 given after the proof of lemma [ thm : htoa ] that @xmath459 , it turns out that it has lorentz degree @xmath460 , where @xmath94 , so that @xmath461 . the first two models have also been considered for simulations studies in fils - villetard _ et al . _ @xcite and bcher _ et al . _ @xcite . here we take the parameter values @xmath462 , @xmath463 , @xmath464 in the first model , and @xmath465 in the second model . note that for this particular choice of asymmetric logistic model , @xmath466 , and for the symmetric mixed model , @xmath467 . a general algorithm for drawing independent couples from the copulas associated to these models is provided by ghoudi _ _ @xcite . here we draw @xmath468 samples of sizes @xmath469 from the copulas corresponding to each of the above three models . for every sample , the maximum likelihood estimate is obtained using the full model by solving ( [ eq : mlef ] ) and the submodel by solving ( [ eq : mles ] ) both with @xmath470 . this gives polynomial estimates of degree at most 7 , that is @xmath471 for the full model and @xmath472 for the submodel . for comparisons , we also consider the popular @xmath473 estimator from capra _ et al . the latter can lead to estimates which are not genuine pickands functions because they do not satisfy either the _ boundary conditions _ or the _ convexity condition_. a modification which leads to pickands functions as estimates is @xmath474 , where @xmath390 , @xmath9 $ ] . another good estimator , according to genest and segers @xcite , is the optimally corrected cfg estimator @xmath475 proposed initially by segers @xcite . the estimator that we use for the comparisons is @xmath476 we show 95% point - wise confidence intervals for model ( [ eq : polfull ] ) in figure [ fig : band ] . ) . on part ( a ) are the results using the maximum likelihood estimator from the full model computed via ( [ eq : mlef ] ) , in part ( b ) using the maximum likelihood estimator from the submodel computed via ( [ eq : mles ] ) and on part ( c ) using the optimal cfg estimator ( [ eq : cfgopt ] ) . here , @xmath470 for the polynomials . ( a ) model ( [ eq : polfull ] ) , full mle @xmath437 . ( b ) model ( [ eq : polfull ] ) , sub mle @xmath444 . ( c ) model ( [ eq : cfgopt ] ) , @xmath477 . ] to compare the performance of the various estimators in the cases considered here , we look at estimates of the mean squared error @xmath478,\ ] ] where @xmath479 is an estimator and @xmath6 is the true pickands function . we also look at the variance and bias separately . these results are plotted in figure [ fig : mse ] . for the polynomials . ( a ) model ( [ eq : alog ] ) . ( b ) model ( [ eq : mix ] ) . ( c ) model ( [ eq : polfull ] ) . ( d ) model ( [ eq : alog ] ) . ( e ) model ( [ eq : mix ] ) . ( f ) model ( [ eq : polfull ] ) . ] in the simulation results , the mle of the submodel ( uniformly ) outperforms the optimal cfg estimator ( [ eq : cfgopt ] ) in terms of mean squared error and also variance . the mle of the full model has uniformly smaller variance than ( [ eq : cfgopt ] ) . the bias of all three estimators is much smaller in the symmetric case than in the other two cases . when @xmath470 , the results do not clearly indicate an overall winner between the two maximum likelihood estimators for sample sizes @xmath469 . this seems to change , however ( according to further simulations we have done ) , when @xmath29 is greater with respect to the data size @xmath480 . this will be noticed in the following . we run the entire simulation again , using the same data set , but this time with @xmath481 . this gives polynomial estimates of degree at most 10 , that is @xmath482 for the full model and @xmath483 for the submodel . the results are plotted in figure [ fig : mse8 ] . for the polynomials . ( a ) model ( [ eq : alog ] ) . ( b ) model ( [ eq : mix ] ) . ( c ) model ( [ eq : polfull ] ) . ( d ) model ( [ eq : alog ] ) . ( e ) model ( [ eq : mix ] ) . ( f ) model ( [ eq : polfull ] ) . ] we can observe that the performance of the estimator from the submodel is roughly the same as in the case @xmath470 , but the performance of the estimator from the full model has somehow worsened in the two nonsymmetric cases in part due to an increase of bias . as a final numerical investigation , we compare the performance of the mle from the full model and the one from the submodel , this time , when the sample size @xmath480 increases and @xmath470 is held fixed . here , in every case we have tried , the mean squared error of the mle of the submodel was uniformly smaller than that of the full model when the polynomial degree @xmath29 was large in comparison to the sample size @xmath480 . then , when @xmath480 increases and @xmath29 is held fixed , the two estimators mean squared error curves are close and cross eachother . figure [ fig : mseasymp ] gives a typical illustration . ) and @xmath470 for the polynomials . ( a ) @xmath484 . ( b ) @xmath485 . ( c ) @xmath486 . ] first , the results obtained in section [ sect : characterization ] could be developed in any other polynomial basis , and we took the popularity of the power basis into consideration . however , it appears that some of the results have an easy interpretation when the coefficients are expressed in the bernstein basis but seem meaningless when the coefficients are expressed in the power basis . for instance , the link between @xmath117 and @xmath94 , together with the _ endpoint derivatives conditions _ can be written @xmath487h(u)\bigr),\end{aligned}\ ] ] with @xmath5 $ ] , and @xmath488 with @xmath420 . when we express these expressions in terms of the coefficients in the bernstein basis , see lemma [ thm : htoa ] and inequalities ( [ eq : boundcondh ] ) respectively , we get a striking similarity @xmath489c\bigl((m+2)u_m-1,m;h \bigr ) \bigr),\end{aligned}\ ] ] with @xmath490 , @xmath491 , and @xmath492 with @xmath493 . now let @xmath494.\ ] ] if we do the same exercise in the power basis this time , we obtain @xmath495 and @xmath496 these are definitely difficult to grasp . secondly , in view of the model constructed in section [ sect : approximation ] , the bernstein basis is clearly the right basis for approximating continuous functions on @xmath21 $ ] by polynomials , think of the weierstrass theorem for instance . it turns out that the bernstein basis is very appealing for other reasons as well . for example , the geometry of the pickands functions is directly reflected in the coefficients , this is made clear by lemma [ thm : piecewisepick ] . it is also shown that @xmath365 for @xmath269 , but @xmath497 . working with the bernstein basis throughout made this finding easier . the convexity condition on the full model in section [ sect : characterization ] is obtained via intermediate polynomials ( @xmath187 and @xmath188 ) , so that the coefficients of the resulting pickands function @xmath6 are parameterized by those of @xmath187 and @xmath188 . in this parameterization , there is an identifiability problem with the parameters . to see this , in ( [ eq : h ] ) , take for example the four couples @xmath498 , @xmath499 , @xmath500 and @xmath501 , these will all produce the same function @xmath85 . however , for inferential purposes , the parameters of interest remain the coefficients of @xmath85 ( or @xmath6 ) , not the ones of @xmath187 and @xmath188 . the submodel is not concerned with this issue at all , the parameters of the submodel are the coefficients . the simplicity of this model , its polytopal parameter space , the gap theorem and its flexibility makes it more appealing , for us at least , than the full model in section [ sect : characterization ] . a practical consequence of the simplicity of the submodel is in finding the maximum likelihood estimator ; it is more than twice as fast , numerically , than for the complete model . finally , we mention that although the likelihood is not concave ( and rather complicated ) , local maxima could indeed make difficult the search for a global maximum . however , using a state - of - the - art matlab global search algorithm , we have not encountered problematic situations . proof of lemma [ thm : pandq ] consider @xmath502 , a sequence of independent @xmath80 random variables . let @xmath503 , be the sequence of the cumulative sums , @xmath504 . suppose that @xmath29 is even , @xmath505 , @xmath197 , @xmath198 . we can write , by equation ( [ eq : h ] ) , @xmath506.\ ] ] now , @xmath507,\end{aligned}\ ] ] with @xmath508 , @xmath509 . similarly , @xmath510\\[-8pt ] & = & { \mathrm{e}}_t\bigl[{\mathrm{e}}\bigl\{q\bigl(y , ( m-2)/2\bigr)q \bigl(s_{m-2}-y , ( m-2)/2\bigr)\mid s_{m-2 } \bigr\}\bigr],\nonumber\end{aligned}\ ] ] with @xmath511 , @xmath512 . from ( [ eq : q2 ] ) and the binomial identities , we get @xmath513,\ ] ] with @xmath514 , @xmath203 . now , when @xmath29 is odd , @xmath204 , and we can write @xmath515.\ ] ] here , @xmath516\\[-8pt ] & = & { \mathrm{e}}_t \bigl[{\mathrm{e}}\bigl\{p\bigl(y,(m-1)/2\bigr)p \bigl(s_{m-1}-y , ( m-1)/2\bigr)\mid s_{m-1 } \bigr\ } \bigr],\nonumber\end{aligned}\ ] ] with @xmath517 , @xmath518 . moreover , from ( [ eq : p2 ] ) and the binomial identities , @xmath519,\ ] ] with @xmath520 , @xmath207 . similarly , @xmath521,\end{aligned}\ ] ] with @xmath517 , @xmath518 . finally , again from the binomial identities , @xmath522,\ ] ] with @xmath523 , @xmath209 . we are particularly grateful to the two referees and especially to the associate editor , the comments and suggestions have led to a very much improved version of this paper . we wish to thank libasse coly for bringing the problem to our attention and johan segers ( universit catholique de louvain ) for great advice and interesting discussions . we also acknowledge the financial support of the natural sciences and engineering research council of canada ( nserc ) .

we introduce techniques to analyze unitary operations in terms of _ quadratic form expansions _ , a form similar to a sum over paths in the computational basis when the phase contributed by each path is described by a quadratic form over . we show how to relate such a form to an entangled resource akin to that of the one - way measurement model of quantum computing . using this , we describe various conditions under which it is possible to efficiently implement a unitary operation , either when provided a quadratic form expansion for as input , or by finding a quadratic form expansion for from other input data .

0801.2461

in the one - way measurement model @xcite , quantum states are transformed using single - qubit measurements on an entangled state , which is prepared from an input state by performing controlled-@xmath2 operations on pairs of qubits , including the input system and ancillas prepared in the @xmath3 state . this model lends itself to ways of analyzing quantum computation which do not naturally arise in the circuit model , _ e.g. _ with respect to depth complexity @xcite and discrete structures underlying unitary operations @xcite . in this article , we present another result of this variety , by introducing _ quadratic form expansions_. [ dfn : quadformexpan ] let @xmath4 be a set of @xmath5 elements , and @xmath6 be ( possibly intersecting ) subsets . for a binary string @xmath7 , let @xmath8 and @xmath9 be the restriction of @xmath10 to those bit - positions indexed by elements of @xmath11 and @xmath12 , respectively . then a _ quadratic form expansion _ is a matrix - valued expression of the form @xmath13 @xmath14 , where @xmath15 is a real - valued quadratic form on @xmath10 , and @xmath16 . quadratic form expansions bear a formal similarity to a representation of a propagator of a quantum system in terms of a sum over paths . for a unitary @xmath1 given as in , the outer product @xmath17 essentially specifies a particular coefficient , in the row indexed by the substring @xmath8 and the column indexed by @xmath9 : the amplitude of the transition between these standard basis states is proportional to a sum of complex units specified by @xmath8 , @xmath9 , and the auxiliary variables @xmath18 . representations of unitary transformations as sums over paths is a well - developed subject in theoretical physics ( see e.g. @xcite ) ; and a representation of unitaries as a sum over paths was used in @xcite to provide a simple proof of @xmath19 . ; comments made in section vi of that paper essentially anticipate quadratic form expansions with discretized coefficients . we describe how their techniques provide a means of constructing quadratic form expansions from circuits in appendix [ apx : sumpaths ] . ] however , there are also examples of quadratic form expansions which arise without explicitly seeking to represent unitaries in terms of path integrals : the quantum fourier transform over @xmath20 can readily be expressed in such a form , and quadratic form expansions for clifford group operations are implicit in the work of dehaene and de moor @xcite , as we will describe in section [ sec : synthclifford ] . given such an expression for a unitary @xmath1 , we show how to obtain a decomposition of @xmath1 in terms of operations similar to those used in the one - way measurement model . using this connection , we demonstrate techniques involving quadratic form expansions to efficiently implement a unitary operator , when the coefficients of the quadratic form satisfies certain constraints related to `` generalized flows '' ( or _ gflows _ ) @xcite or clifford group operations . in particular , we exhibit an @xmath21 algorithm to obtain a reduced _ measurement pattern _ ( an algorithm in the one - way model ) for clifford group operations from a description of how they transform the pauli group , based on the results of @xcite . we can formulate the one - way measurement model as a way of transforming quantum states in the following way . given a state @xmath22 on a set of qubits @xmath11 ( the _ input system _ ) , we embed @xmath11 in a larger system @xmath4 , where the qubits of @xmath23 are prepared in the @xmath24 state . we then perform entangling operations on the qubits of @xmath4 , by performing controlled-@xmath2 ( denoted @xmath25 ) operations on some sets of pairs of qubits . ( these operations are symmetric and commute with each other , and so we may characterize the entangling stage by a simple graph @xmath26 whose vertices are the qubits of @xmath4 : we call this the _ entanglement graph _ of the procedure . ) we then measure each of the qubits of @xmath4 in some sequence , except for some set of qubits @xmath27 ( the _ output subsystem _ ) which will support a final quantum state . we may represent the measurement result for each qubit @xmath28 by a bit @xmath29 which indexes the orthonormal basis states of the measurement . the measurement basis for each qubit may depend on the results of previous measurements , but without loss of generality may be expressed in terms of a `` default '' basis which is used when all preceding measurements yield the result @xmath30 . depending on the measurement results , a final pauli operator may be applied to the qubits in the output subsystem @xmath12 . , @xmath11 , and @xmath12 for these sets of ( labels for ) qubits as for the sets in definition [ dfn : quadformexpan ] will become apparent in the next section . ] in the original formulation of the one - way measurement model , the measurement bases were described by some axis of the bloch sphere lying on the plane , which is sufficient for universal quantum computation . it is also easy to prove that restricting this to states which are an angle @xmath31 from the axis is sufficient for approximately universal quantum computation @xcite . while it is reasonable to extend beyond this for choices of measurement bases @xcite , we will only need to consider measurement bases from the plane . consider a unitary @xmath1 given by a quadratic form expansion as in , where the quadratic form @xmath15 is given by @xmath32 for some angles @xmath33 , and where the sum includes terms for @xmath34 . note that @xmath35 can be expressed as an expectation value @xmath36 , where @xmath37 is a @xmath38-local diagonal operator : @xmath39 \;\;\;+\;\;\ ; \sum_{v \in v } \theta_{vv } { \left\lvert1\right\rangle\mspace{-1.5mu}}{\mspace{-1.5mu}\left\langle1\right\rvert}_v \;\;.\end{aligned}\ ] ] then we may decompose @xmath1 as follows : @xmath40 } { \mathrm e}^{i h}\!\!\ : { \left [ \sum_{{\mathbf{x } } \in { \left\{0,1\right\}}^v } \!\!{\left\lvert{\mathbf{x}}\right\rangle\mspace{-1.5mu } } { \mspace{-1.5mu}\left\langle{\mathbf{x}}_i\right\rvert } \right ] } \notag\\[1ex]\propto&\;\;\ ; r_o \ ; { \mathrm e}^{ih } \ , p_i \;,\end{aligned}\ ] ] where @xmath41 is a unitary embedding which introduces fresh ancillas ( indexed by @xmath42 ) initialized to the @xmath3 state , and @xmath43 is a map projecting onto the @xmath3 state for all qubits in @xmath44 ( tracing those qubits out afterwards ) . equation is a _ phase map decomposition _ @xcite for @xmath1 : that is , it expresses @xmath1 in terms of a process of postselecting observables , as follows . decompose @xmath37 into terms @xmath45 , @xmath46 , and @xmath47 , where @xmath45 consists of the @xmath48-local terms on the qubits of @xmath12 , @xmath46 consists of the @xmath48-local term on the remaining qubits , and @xmath47 contains the remaining terms from . we then have @xmath49 . note that @xmath50 and @xmath51 are simply single - qubit @xmath2 rotations applied to the elements of @xmath12 and @xmath52 respectively , where in each case the qubits @xmath28 in those sets are rotated by an angle @xmath53 . then , the composite map @xmath54 projects each the state of each qubit @xmath55 onto the vector @xmath56 for each @xmath55 . we then have @xmath57 , which is a decomposition of @xmath1 into the preparation of some number of @xmath3 states , followed by a diagonal unitary operator consisting of two - qubit ( fractional ) controlled-@xmath2 operations , followed by post - selection of states on the bloch equator for @xmath55 , and ( unconditionally applied ) single - qubit @xmath2 rotations on the remaining qubits . if @xmath58 for all distinct @xmath59 and for @xmath60 , the above describes precisely the action of a measurement - based computation in which the qubits @xmath55 are measured in the eigenbases of observables of the form @xmath61 , in the special case where all measurements result in the @xmath62 eigenstate ( which we may label with the bit @xmath63 ) . if we are able to extend the above into a complete measurement algorithm , with defined behavior when not all measurements yield a specific outcome , we obtain a measurement - based algorithm for @xmath1 : we discuss this problem in the next section . conversely , from every measurement based algorithm , we may obtain a quadratic form expansion : [ thm : quadformexpansionsuniversal ] every unitary operator on @xmath5 qubits may be expressed by a quadratic form expansion with @xmath64 , and where the quadratic form has coefficients @xmath58 for all cross - terms @xmath65 and @xmath66 for all terms @xmath67 . furthermore , any unitary can be approximated to arbitrary precision by such an expansion where we further require @xmath68 . * proof . * from @xcite ( and using the notation of that article ) , the measurement pattern @xmath69 performs the unitary transformation @xmath70 $ ] for @xmath71 , from the state space of a qubit @xmath72 to that of a `` fresh '' qubit @xmath28 . these operations generate @xmath73 , and generate a group dense in @xmath73 if we restrict to @xmath74 , by @xcite . for any @xmath5 qubit unitary @xmath1 , there exists a measurement pattern composed of such patterns together with two - qubit controlled-@xmath2 operations ( which we denote @xmath25 ) which implements @xmath1 . let @xmath26 be the entanglement graph of this pattern , and @xmath11 and @xmath12 be the qubits defining the input space and output space ( respectively ) of the measurement pattern . by @xcite , in this measurement pattern , the probability of every measurement resulting in the @xmath62 eigenvalue ( i.e. @xmath63 for all @xmath55 ) is non - zero . then , @xmath75 , where @xmath76 \;\;\;-\;\;\ ; \sum_{v \in o{^{\textsf{c } } } } \alpha_v { \left\lvert1\right\rangle\mspace{-1.5mu}}{\mspace{-1.5mu}\left\langle1\right\rvert}_v \;\;. \end{aligned}\ ] ] by , this yields a quadratic form expansion for @xmath1 , with @xmath77 for a quadratic form expansion approximating @xmath1 , it is sufficient to consider measurement patterns approximating @xmath1 using angles @xmath78 . as we remarked above , the connection from quadratic form expansions to phase map decompositions may allow us to obtain an implementation for @xmath1 , provided we can determine how to adapt measurements in case the measurements for qubits @xmath55 do not all yield the result @xmath63 . in a measurement pattern performing @xmath79 measurements , the computation may follow any of @xmath80 branches , corresponding to the different combinations of measurement results . let us call the branch in which every measurement produces the result @xmath63 the _ positive branch _ of the measurement pattern . eigenvalues of their respective observables @xmath81 . ] without loss of generality , we may restrict our attention to patterns where no classical feed - forward is required in the positive branch : then , the positive branch of a measurement pattern is characterized by the _ geometry _ @xmath82 of the pattern ( where @xmath26 is the entanglement graph of the measurement algorithm , and @xmath83 are the sets of qubits defining the input / output space of the pattern ) , and the angles @xmath84 defining the measurements to be performed . to extend the description of the positive branch of a measurement algorithm into a _ complete _ measurement algorithm performing a unitary is the subject of the following problem : for input data @xmath85 , describing a unitary embedding @xmath1 as _ the positive branch _ of a measurement pattern with geometry @xmath82 and performing measurements @xmath86 , determine if there a measurement pattern @xmath87 with geometry @xmath82 which performs the transformation @xmath1 . this problem is open , and seems to be difficult in general . we may attempt to make the problem easier by considering a more restricted problem : for an input geometry @xmath82 , determine if there exist measurement patterns @xmath88 parameterized by a choice @xmath86 of measurement angles , each with geometry @xmath82 , such that the pattern @xmath88 performs a unitary embedding for all @xmath86 . gmpi addresses , in an _ angle - independent _ manner , the subject of the structure of measurement patterns which perform unitary transformations a special case of the gmpi which has been solved are those geometries @xmath82 which have a `` generalized flow '' ( or _ gflow _ ) , which are the `` _ _ yes _ _ '' instances of gmpi such that the patterns @xmath88 yield maximally random outcomes on all of their measurements @xcite . the following is the definition of gflows in @xcite , for measurements restricted to the plane : [ dfn : gflow ] given a geometry @xmath82 for a measurement pattern , a _ gflow _ is a pair @xmath89 , where @xmath90 is a function from @xmath52 to subsets of @xmath91 and @xmath92 is a partial order , such that the following conditions hold for all @xmath72 and @xmath28 in the graph @xmath26 : @xmath93 where @xmath94 is the set of vertices adjacent to an odd number of elements of @xmath95 . here , @xmath96 essentially represents , for two qubits @xmath72 and @xmath28 , that @xmath28 is measured no earlier than @xmath72 ; a gflow then specifies an ordering in which the qubits are to be measured ( with the function @xmath90 providing a description of how to adapt later measurements ) . mhalla and perdrix @xcite present an algorithm which determines if a geometry has a gflow in this sense in polynomial time , which in turn yields a polynomial time solution to the gmpi for that case . as a result , any instance of the mpi where the geometry @xmath82 has a gflow can be efficiently solved . a different special case of the measurement pattern interpolation problem which has been solved is that where the measurement angles are restricted to multiples of @xmath97 ( or slightly more generally , where the measurement observables are pauli operations ) . in this case , as noted in @xcite , no measurement adaptations are necessary , and the corrections can be determined via the stabilizer formalism @xcite . in the following section , we apply these solutions to special cases of the mpi to describe how to synthesize implementations for a unitary @xmath1 given by a quadratic form expansion . in order to apply the partial solutions to the mpi described above , it will be useful to define the following : [ dfn : quadraticformgeometry ] for a quadratic form expansion @xmath98 the _ geometry induced by the quadratic form _ is a triple @xmath82 , where @xmath26 is a _ weighted _ graph with vertex - set @xmath4 , edge - set @xmath99 , and edge - weights @xmath100 . because we can require @xmath101 for all @xmath59 , we may without loss of generality restrict @xmath26 to have edge - weights @xmath102 . we will assume that this holds for the remainder of the article , and speak of edges being either of _ unit weight _ or _ fractional weight_. in this section , we consider the problem of synthesizing an efficient implementation of unitaries @xmath1 in terms of the geometry induced by a quadratic form expansion for @xmath1 by reduction to the solved cases of the measurement pattern interpolation problem discussed in the previous section . consider a geometry @xmath82 induced by a quadratic form expansion for a unitary embedding @xmath1 , where @xmath26 has only edges of unit weight : then @xmath82 is also a geometry for a measurement pattern . to obtain a measurement pattern for @xmath1 , it suffices to find a gflow for @xmath82 : in that case , by theorem 2 of @xcite , for any choice of measurement angles @xmath84 , we may consider the pattern @xmath103}\ ! { \left [ \prod_{u \sim v } e_{uv } \right]}\ ! { \left [ \prod_{u \in i{^{\textsf{c } } } } n_u \right]}\end{aligned}\ ] ] where the left - hand product may be ordered right - to - left in any linear extension of the order @xmath92 , and @xmath104 denotes the adjacency relation of @xmath26 . this pattern thus steers the reduced state after every measurement to the state which would occur if the result had been the @xmath62 eigenvalue . every branch of the pattern then performs the same operation as the positive branch , and so the pattern implements a unitary operation @xmath1 . to obtain a pattern in standard form ( with corrections only on output qubits ) , it is sufficient to propagate the corrections to the left , absorbing them into the measurement bases . in @xcite , an @xmath105 algorithm is provided to determine whether or not a geometry @xmath82 has a gflow where every qubit is to be measured in the plane ( and obtain one in the case that one exists ) , where @xmath106 . the measurement pattern of can be constructed in time @xmath107 by first producing a pattern where corrections undo byproduct operations after each measurement , commuting these corrections to the end , and simplifying ; the resulting pattern will have @xmath108 operations each with complexity @xmath108 . thus : for a unitary embedding @xmath1 given as a quadratic form expansion with geometry @xmath82 with unit edge - weights , there is an @xmath105 algorithm which either determines that @xmath82 has no gflow , or constructs a measurement pattern consisting of @xmath107 operations operations by the solovay - kitaev theorem @xcite , where @xmath109 is the precision of the coefficients of @xmath15 . ] implementing @xmath1 ( using measurement angles of arbitrary precision ) , where @xmath106 . a geometry @xmath82 which has fractional edges lies , at first glance , outside of the domain of the measurement interpolation problems described above . however , given a quadratic form expansion with such a geometry , we may still be able to synthesize a circuit for a unitary @xmath1 represented by that expansion by considering _ flows _ , which correspond to gflows where the function @xmath90 maps each vertex @xmath55 to a singleton set : we may say @xmath110 is a flow if and only if @xmath111 is a gflow , where @xmath112 . geometries which have flows are a solvable special case of the gmpi , where the resulting measurement patterns are very `` circuit - like '' . specifically , the positive branch of a measurement pattern whose geometry has a flow can be represented by a circuit with the following characteristics @xcite : * edges of the form @xmath113 for @xmath55 correspond to @xmath114 gates on some wire , separating two wire segments which we label @xmath28 and @xmath115 ; * edges @xmath116 for @xmath117 and @xmath118 correspond to @xmath25 operations acting on the wire segments labelled by @xmath72 and @xmath28 ; * wires whose initial segments are labelled by vertices of @xmath11 accept arbitrary input states , while those labelled by vertices @xmath119 take input @xmath3 . in the above formulation , the edges of the form @xmath113 can be interpreted as implementing single - qubit teleportation , in which case a fully entangling unitary is important in order to transfer the information of the `` source '' qubit to the `` target '' qubit upon measurement . however , considering the analysis of @xcite , it is not important that the edges of the second kind above be fully entangling operations : using such edges to represent fractional powers of @xmath25 will also yield unitary circuits . this motivates the following definition : suppose @xmath82 is a geometry of a quadratic form expansion for a unitary transformation @xmath1 . we may say that @xmath120 is a _ fractional - edge flow _ for @xmath82 if it is a flow for that geometry , and for all @xmath121 with @xmath122 , we have @xmath123 and @xmath124 . if @xmath82 has a fractional - edge flow , we may synthesize a circuit from a quadratic form expansion for @xmath1 using the description above , where edges @xmath125 of fractional weight correspond to @xmath126 gates on the wire segments labelled by @xmath127 and @xmath128 rather than simple @xmath25 gates . we will make use the following easily verified lemma to consider how to compose / decompose quadratic form expansions : [ lemma : quadformexpancomposition ] let @xmath129 , @xmath130 be matrices given by quadratic form expansions @xmath131 in the following , @xmath132 , and sums are over @xmath133 . if @xmath134 , then @xmath135 if @xmath136 and @xmath137 are disjoint , then @xmath138 where @xmath139 and @xmath140 . we prove the circuit construction given by inducting on the number of edges of fractional weight . for the base case , if @xmath82 has no fractional - weight edges at all , we may synthesize a circuit for @xmath1 as above , as it corresponds to a normal measurement pattern with a flow , and so falls under the analysis of @xcite . we may then induct for geometries with fractional edge - weights if we can show we can decompose the geometry into ones with fewer fractional edge - weights . for any arbitrary fractional edge @xmath121 and each each @xmath141 , we may define @xmath142 to be the maximal vertex @xmath143 in the ordering @xmath92 subject to @xmath144 being in the orbit of @xmath28 under @xmath145 ( that is , @xmath146 for some @xmath147 ) , such that at least one of @xmath148 or @xmath149 holds . for a set @xmath150 , let @xmath151 $ ] represent the subgraph of @xmath26 induced by @xmath95 ( i.e. by deleting all vertices in @xmath26 not in @xmath95 ) . then , define the following subgraphs of @xmath26 , and corresponding geometries : * let @xmath137 be the set of vertices @xmath142 for each @xmath152 : it is easy to show that @xmath153 . let @xmath154 $ ] , and let @xmath155 . * let @xmath136 be the set of vertices @xmath156 such that @xmath96 for some @xmath157 ; let @xmath158 { \smallsetminus}{\left\{uv \,\big|\ , u , v \in v_2\right\}}$ ] ; and let @xmath159 . * let @xmath160 be the set of vertices @xmath156 such that @xmath161 for some @xmath157 ; let @xmath162 { \smallsetminus}{\left\{uv \,\big|\ , u , v \in v_2\right\}}$ ] ; and let @xmath163 . this decomposes the geometry @xmath82 into three geometries with fractional - edge flows , as illustrated in figure [ fig : decompose ] . ( 1350,460)(0,0 ) ( 0,40 ) ( 20,0 ) illustration of the decomposion of a quadratic form expansion about an edge @xmath125 , expressed in terms of geometries . @xmath137 is a set of maximal vertices under the constraint of being bounded from above , by the vertices @xmath127 and @xmath128 , in a partial order @xmath92 associated with a fractional - edge flow . arrows represent the action of the corresponding fractional - edge flow function , @xmath145 . , title="fig : " ] ( 50,12)[c]@xmath11 ( 365,-23)[c]@xmath12 ( 140,12)[c]@xmath136 ( 280,-23)[c]@xmath160 ( 127,299)(15,30)(40,30)(60,0 ) ( 135,330)[r]@xmath137 ( 190,390)[c]@xmath82 ( 480,205)[c]@xmath164 ( 540,40 ) ( 20,40 ) illustration of the decomposion of a quadratic form expansion about an edge @xmath125 , expressed in terms of geometries . @xmath137 is a set of maximal vertices under the constraint of being bounded from above , by the vertices @xmath127 and @xmath128 , in a partial order @xmath92 associated with a fractional - edge flow . arrows represent the action of the corresponding fractional - edge flow function , @xmath145 . , title="fig : " ] ( 50,12)[c]@xmath11 ( 215,12)[c]@xmath137 ( 130,370)[c]@xmath165 ( 850,210)[c]@xmath166 ( 900,40 ) ( 20,40 ) illustration of the decomposion of a quadratic form expansion about an edge @xmath125 , expressed in terms of geometries . @xmath137 is a set of maximal vertices under the constraint of being bounded from above , by the vertices @xmath127 and @xmath128 , in a partial order @xmath92 associated with a fractional - edge flow . arrows represent the action of the corresponding fractional - edge flow function , @xmath145 . , title="fig : " ] ( 50,12)[c]@xmath137 ( 50,370)[c]@xmath167 ( 1050,210)[c]@xmath166 ( 1100,40 ) ( 20,0 ) illustration of the decomposion of a quadratic form expansion about an edge @xmath125 , expressed in terms of geometries . @xmath137 is a set of maximal vertices under the constraint of being bounded from above , by the vertices @xmath127 and @xmath128 , in a partial order @xmath92 associated with a fractional - edge flow . arrows represent the action of the corresponding fractional - edge flow function , @xmath145 . , title="fig : " ] ( 50,12)[c]@xmath137 ( 205,-23)[c]@xmath12 ( 120,370)[c]@xmath168 let @xmath169 be a quadratic form on @xmath170 consisting of the terms @xmath65 of @xmath15 for @xmath171 or @xmath172 , but not both ; @xmath173 be a quadratic form on @xmath174 consisting of the terms @xmath65 of @xmath15 for _ distinct _ @xmath175 ; and similarly let @xmath176 be defined on @xmath177 , and consist of the remaining terms of @xmath15 . then @xmath169 , @xmath173 , and @xmath176 define quadratic form expansions for some operations @xmath129 , @xmath130 , and @xmath178 ( respectively ) with geometries @xmath179 , @xmath180 , and @xmath181 ( respectively ) . * @xmath130 in particular will be a product of operations @xmath182 for distinct @xmath183 , as it is a quadratic form expansion whose input and output indices coincide . then @xmath130 can be represented as a circuit with a wire for each @xmath184 , with fractional controlled-@xmath2 gates @xmath182 for each edge @xmath116 . * both @xmath179 and @xmath181 have fractional - edge flows , but fewer fractional edges than @xmath82 . by induction , @xmath129 and @xmath178 are also unitary embeddings , and have circuits with wire - segments connected by @xmath185 gates ( where @xmath186 are the coefficients of the terms @xmath67 in each quadratic form ) and possibly fractional @xmath25 gates ( as in the case for @xmath130 ) . * in the circuits described above , the terminal wire - segments for @xmath129 and ( a subset of ) the initial wire - segments for @xmath178 have the same labels as the wires for @xmath130 . the composite circuit for @xmath187 can then use these labels to arrive at a unified labelling of its wire - segments . because @xmath188 for all @xmath189 by construction , the composite operation @xmath187 can differ from @xmath1 by at most a scalar factor by lemma [ lemma : quadformexpancomposition ] ; so the circuit obtained implements the operation @xmath1 . in @xcite , an @xmath190 algorithm is provided to determine whether or not a geometry @xmath82 has a flow , and obtain one if it exists , where @xmath106 and @xmath191 . for each edge @xmath192 , we may check whether one of @xmath193 or @xmath194 and @xmath195 $ ] holds : if all edges satisfy this constraint , the circuit described above is well - defined . by iterating through the vertices of @xmath196 in an arbitrary linear extension of @xmath92 , we may construct the circuit described above can be constructed in time @xmath197 , and the size of the resulting circuit will also be @xmath197 , where @xmath198 . by an extremal result @xcite , any geometry with a flow has @xmath199 : thus , the total running time of this algorithm is @xmath190 . in the case @xmath200 , a flow function @xmath145 is unique if it exists , by @xcite ; so in this case , if @xmath82 has a flow but there is an edge @xmath113 of fractional weight , there is no fractional - weight flow for @xmath82 . we then have : for a unitary transformation @xmath1 given as a quadratic form expansion with geometry @xmath82 , there is an @xmath190 algorithm which either determines that @xmath82 has no fractional - edge flow , or constructs a circuit consisting of @xmath190 operations gates and fractional @xmath25 gates of arbitrary precision . a corresponding circuit using a finite elementary gate set may be of size @xmath201 by the solovay - kitaev theorem @xcite , where @xmath109 is the precision of the coefficients of @xmath15 . ] implementing @xmath1 , where @xmath106 and @xmath191 . the fourier transform over @xmath20 is given by the matrix formula @xmath202 \big [ \sum\limits_{j=0}^{n-1 } 2^j y_j \big]/2^n } } { \left\lvert{\mathbf{y}}\right\rangle\mspace{-1.5mu } } { \mspace{-1.5mu}\left\langle{\mathbf{x}}\right\rvert } , \end{aligned}\ ] ] which is a quadratic form expansion ; its quadratic form can be given by @xmath203 this has a fractional - edge flow for all @xmath5 . figure [ fig : qftsynth ] illustrates this geometry for @xmath204 , and the circuit ( due to @xcite ) which may be synthesized from it . if a quadratic form expansion has a geometry whose edges all have unit weight , and its other coefficients are multiples of @xmath97 , then it corresponds to the positive branch of a measurement pattern which measures only @xmath205 or @xmath206 observables . a measurement pattern of this sort , if it performs a unitary operation , performs a clifford group operation in particular . an algorithm of aaronson and gottesman @xcite can produce a circuit of size @xmath207 in classical deterministic time @xmath21 for a clifford group operation @xmath1 acting on @xmath5 qubits , from a description of how @xmath1 transforms pauli operators by conjugation . by converting the circuit into a measurement - based algorithm , and performing the graph transformations of @xcite to remove auxiliary qubits , we may obtain a pattern of at most @xmath208 qubits qubits are described as having minimal patterns for are described as requiring at most @xmath209 qubits ; however , this only holds up to local clifford operations on the output qubits . ] in time @xmath210 . building on the results of @xcite , we show how to classically compute such a minimal pattern in time @xmath21 by solving the mpi for a quadratic form expansion for @xmath1 . for the sake of completeness , we outline the relevant results of @xcite . define the following notation for bit - flip and phase - flip operators on a qubit @xmath211 out of a collection @xmath212 : @xmath213 let @xmath214 represent the vector of the diagonal elements of any square boolean matrix @xmath215 ; and let @xmath216 m\right ) \in { \mathbb z}_2^{\ , 2n}$ ] for a @xmath217 matrix @xmath215 over @xmath218 . then , we may represent an @xmath5 qubit unitary @xmath1 by a @xmath217 boolean matrix @xmath219 and a vector @xmath220 , whose coefficients are jointly given by @xmath221}\end{aligned}\ ] ] for each @xmath222 . ( note that the factor of @xmath223 is only necessary to ensure that the image of @xmath224 is hermitian , and does not serve as a constraint on the value of @xmath219 as a matrix . ) we will call an ordered pair @xmath225 a _ leuven tableau _ for a clifford group element @xmath1 if it satisfies ( [ eqn : leuventableau]).$ ] is similar to a _ destabilizer tableau _ as defined in @xcite . ] provided a leuven tableau @xmath225 for a clifford group operation @xmath1 , @xcite provides a matrix formula for @xmath1 which we may obtain for @xmath1 , as follows . decompose @xmath219 as a block matrix @xmath226 $ ] with @xmath227 blocks , and then find invertible matrices @xmath228 over @xmath218 such that @xmath229 $ ] for some @xmath230 ( using _ e.g. _ the decomposition algorithm of @xcite to obtain @xmath231 and @xmath232 in terms of elementary row operations ) . then , define the matrices @xmath233 \,\tilde{e}_{21}\ , & \,\tilde{e}_{22}\ , \end{matrix } \right ] \;=&\;\ ; \tilde{r}_1{^\top}e \tilde{r}_2 \ , , & r_1 \;=&\;\ ; \tilde{r}_1 \ ; , & r_2 \;=&\;\ ; \left[\begin{matrix } \,\tilde{e}_{11}{^{-1}}\ , & 0 \\[1ex ] \;0\ ; & \;{\mathbbm{1}}_r\ ; \end{matrix}\right]{^\top}\!\ ! \tilde{r}_2{^\top}\;,\end{aligned}\ ] ] where @xmath234 is taken to be a block of size @xmath235 . we may then obtain the block matrices @xmath236 e_{21 } & e_{22 } & f_{21 } & f_{22 } \\[1ex ] 0 & 0 & h_{11 } & h_{12 } \\[1ex ] 0 & { \mathbbm{1}}_r & h_{21 } & h_{22 } \end{smallmatrix}\right ] \;\;=&\;\ ; \left[\begin{smallmatrix } r_1{^\top } & 0 \\ 0 & r_1{^{-1}}\end{smallmatrix}\right ] c \left[\begin{smallmatrix } r_2{^\top } & 0 \\ 0 & r_2{^{-1}}\end{smallmatrix}\right ] \,,\end{aligned}\ ] ] and use these to construct the @xmath227 boolean matrices @xmath237 \;e_{12}{^\top}\ ; & \;e_{22}\ ; \end{matrix}\right]\ , , & m_{bc } \;=&\;\ ; \left[\begin{matrix } \;0\ ; & \;h_{21}{^\top}\ ; \\[1ex ] \;h_{21}\ ; & \;h_{22}\ ; \end{matrix}\right]\,.\end{aligned}\ ] ] next , define @xmath238 where @xmath239 is the strictly lower - triangular part of a square matrix @xmath215 ( with all other coefficients set to @xmath30 ) . finally , define @xmath240 $ ] and @xmath241 for the sake of brevity , and let and @xmath242 may be obtained by repeated application of theorem 2 of @xcite . ] @xmath243 { \mathbf{h } } \;+\ ; { \mathop{\operator@font diag}\nolimits}\big ( { \left [ r_2{^{-1}}\pi_r \right ] } \!l_{br } { \left [ r_2{^{-1}}\pi_r \right]}{^\top}\big ) \ ; , \\[0.5ex ] \begin{split } { \mathbf{h}}_{bc } \;\;=&\;\ ; \left[\begin{matrix}\ ; 0 \,&\ , r_2{^{\,-\!\top}}\,\end{matrix}\right ] { \mathbf{h } } \;+\ ; r_2{^{\,-\!\top}}{\mathop{\operator@font diag}\nolimits}\big ( r_2{^\top}\big [ l_{bc } \,+\ , \pi_r m_{bc } \\[-0.5ex]&\mspace{100mu } \,+\ , { \left ( \pi_r{^\bot}+ \pi_r m_{bc } \right ) } l_{br } { \left ( \pi_r{^\bot}+ m_{bc } \pi_r \right ) } \big ] r_2 \big ) \;. \end{split}\end{aligned}\ ] ] then theorem 6 of @xcite states that the unitary operation @xmath1 for the clifford operation characterized by @xmath244 is given by the matrix formula @xmath245 & ( -i)^{{\left ( { \mathbf{d}}_{br}{^\top}{\mathbf{x}}_{br } \,+\ , { \mathbf{d}}_{bc}{^\top}{\mathbf{x}}_{bc } \right ) } } \ ; { \left\lvertr_1\big . { \mathbf{x}}_{br}\right\rangle\mspace{-1.5mu}}{\mspace{-1.5mu}\left\langler_2{^{-1}}{\mathbf{x}}_{bc } + { \mathbf{t}}\right\rvert } \,\bigg ] \ , , \end{split}\end{aligned}\ ] ] where @xmath246 $ ] and @xmath247 $ ] are @xmath5 bit boolean vectors . the formula in ( [ eqn : cliffordformula ] ) shows strong similarities to a quadratic form expansion . in particular , consider disjoint sets of indices @xmath248 , @xmath249 , and @xmath250 , with @xmath251 and @xmath252 . let @xmath253 , @xmath254 , and @xmath255 , and define the following notation for @xmath7 : @xmath256 \;=\ ; \left[\begin{matrix } { \mathbf{x}}_{v_b } \\ { \mathbf{x}}_{v_c } \end{matrix}\right ] \in { \left\{0,1\right\}}^{i } \ ; , & { \mathbf{x}}_o \;=&\ ; \left[\begin{matrix } { \mathbf{x}}_b \\ { \mathbf{x}}_r \end{matrix}\right ] \;=\ ; \left[\begin{matrix } { \mathbf{x}}_{v_b } \\ { \mathbf{x}}_{v_r } \end{matrix}\right ] \in { \left\{0,1\right\}}^{o } \;,\end{aligned}\ ] ] @xmath257 then , ( [ eqn : cliffordformula ] ) is equivalent to @xmath258 ( 1470,400)(0,0 ) ( 40,0 ) ( 0,40 ) illustration of geometries arising from quadratic form expansions yielding the same matrix . on the left is a geometry whose inputs and output intersect ; on the right is a geometry from an equivalent quadratic form expansion , constructed so that the input and output indices are disjoint.,title="fig : " ] ( 30,20)[c]@xmath11 ( 190,20)[c]@xmath12 ( 110,370)[c]@xmath259 ( 335,190)[c]@xmath260 ( 400,0 ) ( 0,40 ) illustration of geometries arising from quadratic form expansions yielding the same matrix . on the left is a geometry whose inputs and output intersect ; on the right is a geometry from an equivalent quadratic form expansion , constructed so that the input and output indices are disjoint.,title="fig : " ] ( 50,20)[c]@xmath11 ( 210,20)[c]@xmath261 ( 50,370)[c]@xmath248 ( 210,370)[c]@xmath262 ( 725,270)[c]@xmath166 ( 800,0 ) ( 0,41.5 ) illustration of geometries arising from quadratic form expansions yielding the same matrix . on the left is a geometry whose inputs and output intersect ; on the right is a geometry from an equivalent quadratic form expansion , constructed so that the input and output indices are disjoint.,title="fig : " ] ( 30,370)[c]@xmath248 ( 110,370)[c]@xmath263 ( 190,370)[c]@xmath262 ( 1100,190)[c]@xmath164 ( 1180,0 ) ( 0,40 ) illustration of geometries arising from quadratic form expansions yielding the same matrix . on the left is a geometry whose inputs and output intersect ; on the right is a geometry from an equivalent quadratic form expansion , constructed so that the input and output indices are disjoint.,title="fig : " ] ( 50,20)[c]@xmath11 ( 210,20)[c]@xmath261 which is essentially a quadratic form expansion sandwiched between two networks of controlled - not and @xmath205 gates . to obtain a simple quadratic form expansion , we would like to perform a change of variables on @xmath8 and @xmath9 ; but this can not be done as @xmath11 and @xmath12 intersect at @xmath248 , and the changes of variables do not necessarily respect the partitioning of @xmath11 and @xmath12 with respect to this intersection . however , we may add auxiliary variables in order to produce an expansion with disjoint input and output indices . note that @xmath264 where @xmath265 is the kronecker delta . let @xmath266 and @xmath262 be disjoint copies of @xmath248 , and set @xmath267 and @xmath268 . writing @xmath269 and @xmath270 for the restriction of @xmath271 to @xmath266 and @xmath262 , we then define @xmath256 \in { \left\{0,1\right\}}^{\ , i } \ ; , & { \mathbf{x}}_{o ' } \;=&\ ; \left[\begin{matrix } { \mathbf{x}}_{b ' } \\ { \mathbf{x}}_r \end{matrix}\right ] \in { \left\{0,1\right\}}^{\ , o ' } \;,\end{aligned}\ ] ] @xmath272 0 \end{smallmatrix}\right]\ ! { \mathbf{x}}_a \;+\ ; \pi { \mathbf{x}}_{o'}{^\top}\!\left[\begin{smallmatrix } { \mathbbm{1}}_{n { \text{--}}r } \\[0.5ex ] 0 \end{smallmatrix}\right]\ ! { \mathbf{x}}_a \notag\\&\mspace{150mu } \;-\ ; \frac{\pi}{2 } \big ( { \mathbf{d}}_{br}{^\top}{\mathbf{x}}_{o ' } \,+\ , { \mathbf{d}}_{bc}{^\top}{\mathbf{x}}_i \big ) \;. \ ] ] note that the difference between the expressions for @xmath273 and @xmath15 is essentially that all instances of @xmath9 have been replaced with @xmath274 ( which is independent from @xmath8 ) , and the presence of the terms involving @xmath269 . ( this manipulation is illustrated in figure [ fig : idxsplit ] as a transformation of geometries . ) we therefore have @xmath275 substituting the final expression of ( [ eqn : quadformtransform ] ) into ( [ eqn : cliffordalmostquadformexpan ] ) and performing the appropriate change of variables , we have @xmath276 note that the quadratic form of the expansion in ( [ eqn : cliffordquadformexpan ] ) has only angles @xmath277 which are multiples of @xmath97 , with @xmath278 for @xmath279 . this then represents the positive branch of a one - way measurement pattern on the geometry @xmath280 of the quadratic form expansion of [ eqn : cliffordquadformexpan ] , using only @xmath205 or @xmath206 basis measurements , and having only @xmath281 auxiliary vertices . we can augment this to a measurement pattern by applying the techniques of the stabilizer formalism @xcite to the stabilizer code generated by the operators @xmath282 for @xmath283 ( where again @xmath104 is the adjacency relation of @xmath26 ) , as follows . to obtain the final correction , we do classical pre - processing simulating the evolution of the _ state space _ when we perform one measurement at a time . for each measured qubit @xmath72 , there is an associated correction @xmath284 which we may perform immediately after the measurement if we obtain the result @xmath285 . we store for each qubit @xmath28 two boolean formulas @xmath286 and @xmath287 , representing the @xmath205 and @xmath2 components of the accumulated corrections to be performed on @xmath28 . when @xmath28 is measured , the pending @xmath205 corrections will affect the result of any @xmath206 measurement , and the pending @xmath2 corrections will affect the result of any @xmath205 or @xmath206 measurement , in each case by exchanging the significance of the two measurement outcomes . just prior to the ( simulated ) measurement of @xmath288 , let @xmath289 if @xmath28 is to be measured with an @xmath205 observable , and @xmath290 if @xmath28 is to be measured with a @xmath206 observable . thus , upon measuring @xmath28 , the following operations are accumulated into the corrections which must be performed : * for every qubit @xmath291 where @xmath284 acts with an @xmath205 or @xmath206 operation , we must add @xmath292 to @xmath293 ; * for every qubit @xmath291 where @xmath284 acts with a @xmath206 or @xmath2 operation , we must add @xmath292 to @xmath294 . this accounts for the accumulated corrections due to the measurement of @xmath28 and every preceding measurement which affects it . by simulating measurement for all of the qubits in @xmath52 in this way , we obtain boolean formulae for the corrections on @xmath12 in terms of the results of the measurements : the correction to be performed for some @xmath295 is @xmath296 , for @xmath293 and @xmath294 constructed after all of the ( simulated ) measurements . to obtain @xmath293 and @xmath294 for all @xmath295 in this way takes time @xmath107 . it is easy to show that the resulting measurement pattern is irreducible by the techniques of @xcite , by the following argument . let @xmath297 denote the set of auxiliary vertices corresponding to the bit positions of @xmath269 : note that in the measurement pattern , these are all to be measured with the observable @xmath205 , and are adjacent only to the input / output variables @xmath8 and @xmath9 . for @xmath298 or cross - term @xmath65 for @xmath299 before the change of variables in ) , and the change of variables itself does not introduce any . ] to eliminate a vertex @xmath298 using the methods of @xcite on the geometry induced by the quadratic form expansion , we would have to identify an output variable @xmath300 adjacent to @xmath301 , and apply the graph transformation in ( * ? ? ? * proposition 1 ) . this would result in a geometry where @xmath302 has the former neighbors of @xmath28 in @xmath26 ( and in particular is not adjacent to any more removable vertices ) , and where a local clifford ( which is not a pauli operator ) must be applied to @xmath302 after the entangling procedure . because @xmath302 is not adjacent to any other auxiliary qubit after this transformation , the local clifford can not be undone or made into a pauli operator by e.g. another vertex removal ; then , except by extending the computational model to allow for corrections which are local clifford operations , performing the local clifford can only be done by introducing an auxiliary qubit ( or rather , a new output qubit following @xmath302 , making the latter an auxiliary qubit ) . thus : for an @xmath5-qubit clifford group operation @xmath1 given in the form of a leuven tableau , there is an @xmath21 algorithm which produces a minimal one - way measurement pattern for @xmath1 . the ability to obtain a quadratic form expansion representing a reduced measurement pattern yields a more efficient algorithm to find totally reduced clifford patterns than from using existing techniques to obtain one via the circuit model . the quadratic form of ( [ eqn : cliffordquadformexpan ] ) can be found from a leuven tableau @xmath244 in time @xmath21 , which is dominated by the time required to compute @xmath303 and @xmath304 . to contrast , an approximately optimal quantum circuit for a clifford group operation ( _ i.e. _ consisting of @xmath207 gates ) can be found from a leuven tableau in time @xmath21 by transforming it into a destabilizer tableau , and then applying the algorithm of @xcite . to obtain a measurement pattern from such a circuit by composing the patterns for each gate , removing vertices opportunistically ( with each removal taking time @xmath107 ) , requires time @xmath210 . thus , making use of quadratic form expansions provides us with a faster algorithm to obtain reduced measurement patterns for clifford group operations . we have introduced quadratic form expansions , and developed techniques which suggest that they may be useful for synthesizing efficient implementations for unitary operations . we described conditions under which implementations may be efficiently found for unitaries specified by quadratic form expansions ; and we showed how quadratic form expansions leads to more efficient algorithms for obtaining reduced patterns for clifford operations in the one way measurement model . in the introduction , we mentioned that quadratic form expansions are similar in form to a sum - over - paths representation of unitary operations , which is a well - developed subject in theoretical physics . this raises the question of whether the techniques developed here are useful e.g. for developing algorithms to simulate physical systems . it is not known whether the solved cases of the measurement pattern interpolation problem correspond to _ natural _ ( in the more literal sense ) unitaries expressed as sums over paths : this question , and how to extend the solved cases of the mpi to include propagators for interesting physical systems , remain open . 000 r. raussendorf and h. briegel . _ a one - way quantum computer_. physical review letters 86 ( 5188 ) , 2001 . r. raussendorf and h. briegel . _ computational model underlying the one - way quantum computer_. quantum information & computation , vol 2 # 6 ( 443 ) 2002 . a. broadbent and e. kashefi . _ parallelizing quantum circuits_. arxiv:0704.1736 , 2007 . r. p. feynmann , a. r. hibbs . _ quantum mechanics and path integrals_. mcgraw - hill , new york , 1965 . l. s. schulman . _ techniques and application of path integration_. wiley - interscience , new york , 1981 . v. danos and e. kashefi . _ determinism in the one - way model_. physical review a 74 ( 052310 ) , 2006 . arxiv : quant - ph/0506062 . d. e. browne and h. j. briegel . _ one - way quantum computation a tutorial introduction_. arxiv : quant - ph/0603226 ( 2006 ) . d. e. browne , e. kashefi , m. mhalla , and s. perdrix . _ generalized flow and determinism in measurement - based quantum computation_. new j. physics vol . 9 ( 250 ) , 2007 . arxiv : quant - ph/0702212 c. m. dawson , h. l. haselgrove , a. p. hines , d. mortimer , m. a. nielsen , and t. j. osborne . _ quantum computing and polynomial equations over _ . quantum information & computation vol 5 # 2 ( 102 ) , 2004 . arxiv : quant - ph/0408129 n. de beaudrap , v. danos , and e. kashefi . _ phase map decompositions for unitaries_. arxiv : quant - ph/0603266 , 2006 . v. danos , e. kashefi , and p. panangaden . _ the measurement calculus . _ j. acm vol . 54 , 8 ( 2007 ) . arxiv : quant - ph/0412135 v. danos , e. kashefi , and p. panangaden . _ parsimonious and robust realizations of unitary maps in the one - way model_. physical review a vol 72 ( 064301 ) , 2005 . arxiv : quant - ph/0411071 m. mhalla and s. perdrix . finding optimal flows efficiently . arxiv:0709.2670 , 2007 . a. kitaev , a. shen , and m. vylalyi . _ classical and quantum computation_. graduate texts in mathematics , vol 47 , american mathematical society , providence ri , 2002 . n. de beaudrap and m. pei . _ an extremal result for geometries in the one - way measurement model_. to appear in quantum information and computation , vol . 8 # 5 ( 430 ) , 2008 . arxiv : quant - ph/0702229 d. gottesman . . phd thesis , caltech , 1997 . arxiv : quant - ph/9705052 . j. dehaene and b. de moor . _ clifford group , stabilizer states , and linear and quadratic operations over _ . physical review a vol 68 ( 042318 ) , 2003 . arxiv : quant - ph/0304125 m. hein , j. eisert , and h. j. briegel . _ multi - party entanglement in graph states . _ physical review a vol 69 ( 62311 ) , 2004 . arxiv : quant - ph/0307130 s. aaronson and d. gottesman . _ improved simulation of stabilizer circuits_. physical review a vol 70 ( 052328 ) , 2004 . arxiv : quant - ph/0406196 n. de beaudrap . finding flows in the one - way measurement model . arxiv : quant - ph/0611284 , 2006 . a. fowler , s. devitt , and l. hollenberg . _ implementation of shor s algorithm on a linear nearest neighbor qubit array_. quantum information & computation vol 4 # 4 ( 237 ) , 2004 . k. n. patel , i. l. markov , and j. p hayes . efficient synthesis of linear reversible circuits . to appear in quantum information & computation vol 8 , 2008 . arxiv : quant - ph/0302002 let @xmath82 be the geometry of a quadratic form expansion , as defined on page . in the special case when @xmath82 has a fractional - edge flow as defined in section [ sec : circuitsviaflows ] , the quadratic form expansion corresponds exactly to a sum over paths as described in @xcite , for the elementary gate set of @xmath37 , @xmath305 , and @xmath306 , where @xmath307 ( _ i.e. _ admitting arbitrary @xmath2 rotations and fractional controlled-@xmath2 gates ) . in order to demonstrate the sense in which quadratic form expansions are sums over paths in this case , and because it represents a reasonably simple algorithm for converting quantum circuits into quadratic form expansions , we now present an alternate proof of theorem [ thm : quadformexpansionsuniversal ] based on the techniques of @xcite . that any quadratic form expansion with geometry with a fractional - edge flow can be constructed in this way follows by reversing the construction below . * proof of theorem [ thm : quadformexpansionsuniversal ] . * consider a quantum circuit implementing @xmath1 exactly , using the operations @xmath37 , @xmath306 , and @xmath305 . enumerate the wires of the circuit from @xmath48 to @xmath308 , and for each wire @xmath309 , introduce a _ path label _ @xmath310 for the input end of the wire , corresponding to an input bit @xmath311 . we set @xmath312 . divide each wire into _ segments _ , bounded on each end by either a hadamard gate , the input terminal of the wire , or the output terminal . we label the wire segments with path variables : for the segments at the inputs , we apply the labels @xmath310 for @xmath313 , and we introduce new path variables to label the remaining wire segments . computational paths in the circuit are then described by setting all of the the path variables @xmath314 collectively to some particular binary string in @xmath315 . the phase contribution of each paths , governing how they interfere to produce an output state for any given input state , is described by a function @xmath316 depending the gates of the circuit as follows : for every hadamard gate on a single wire , with a path variable @xmath317 labelling the segment preceding the hadamard and a path variable @xmath310 labelling the segment following the hadamard , we add a term @xmath318 . for every @xmath306 operation between two wires , with a path variable @xmath317 labelling the segment of one wire and @xmath310 labelling the segment of the other wire in which the @xmath306 operation is performed , we add a term @xmath319 . for every @xmath305 operation on a wire segment labelled with a path variable @xmath310 , we add a term @xmath320 . ( because the path variable @xmath310 ranges over @xmath321 , the extra power of @xmath38 has no effect . ) in particular , the function @xmath316 is a quadratic form , where without loss of generality the coefficients may be constrained to @xmath322 . the phase of a given path , described by a bit - string @xmath323 , is then given by @xmath324 . each path also has an associated amplitude of @xmath325 , where @xmath326 is the number of hadamard gates in the circuit . to be simply a polynomial over @xmath0 , in terms of the descriptions used in section vi of @xcite , one may consider @xmath327 to be a polynomial over the ring @xmath328 . if we restrict to @xmath329 , we may simplify this to the finite ring @xmath330 by multiplying all of the coefficients by @xmath331 , and using it to describe powers of @xmath332 rather than of @xmath333 . ] let @xmath12 be the set of indices @xmath334 such that some wire is labelled by the path - variable @xmath310 at its output end . then , the initial points of computational paths are described by bit - vectors @xmath335 , and the terminal points of paths are described by @xmath336 . the coefficients @xmath337 can then be given as the sum of the contributions of all paths beginning at @xmath338 and ending at @xmath339 : @xmath340 which is an expression of the coefficients of @xmath1 as a quadratic form expansion . to obtain a proof of theorem [ thm : quadformexpansionsuniversal ] , it is sufficient to note that without loss of generality we may restrict ourselves to using @xmath306 gates only for @xmath341 to implement @xmath1 exactly ; and that to implement @xmath1 to arbitrary precision , it suffices to use @xmath305 gates where @xmath211 is restricted to multiples of @xmath342 .

this paper is the third in a series in which we present deep chandra acis - s imaging spectroscopy of the seyfert 1 galaxy ngc 4151 , devoted to study its complex circum - nuclear x - ray emission . emission features in the soft x - ray spectrum of the bright extended emission ( erg s ) at pc ( 2 ) are consistent with blended brighter ovii , oviii , and neix lines seen in the _ chandra _ hetgs and xmm - newton rgs spectra below 2 kev . we construct emission line images of these features and find good morphological correlations with the narrow line region clouds mapped in [ oiii] . self - consistent photoionization models provide good descriptions of the spectra of the large scale emission , as well as resolved structures , supporting the dominant role of nuclear photoionization , although displacement of optical and x - ray features implies a more complex medium . collisionally ionized emission is estimated to be% of the extended emission . presence of both low and high ionization spectral components and extended emission in the x - ray image perpendicular to the bicone indicates leakage of nuclear ionization , likely filtered through warm absorbers , instead of being blocked by a continuous obscuring torus . the ratios of [ oiii]/soft x - ray flux are approximately constant ( ) for the 1.5 kpc radius spanned by these measurements , indicating similar relative contributions from the low- and high - ionization gas phases at different radial distances from the nucleus . if the [ oiii ] and x - ray emission arise from a single photoionized medium , this further implies an outflow with a wind - like density profile . using spatially resolved x - ray features , we estimate that the mass outflow rate in ngc 4151 is yr at 130 pc and the kinematic power of the ionized outflow is erg s , approximately of the bolometric luminosity of the active nucleus in ngc 4151 .

1103.1913

the seyfert 1 galaxy ngc 4151 is well known to have an extended narrow line region ( enlr ) in the optical ( e.g. , perez et al . 1989 ; robinson et al . 1994 ; crenshaw et al . 2000 ) , indicating interaction between the active galactic nucleus ( agn ) and the host galaxy ( unger et al . thanks to its proximity ( @xmath9 mpc , @xmath10 pc ; mundell et al . 1999 ) and extensive coverage at all wavelengths ( ulrich 2000 and references therein ) , ngc 4151 thus represents an excellent laboratory to study observationally nuclear feedback . presence of extended emission in the soft x - rays is also well established ( elvis et al . 1983 ; morse et al . 1995 ; ogle et al . 2000 ; yang et al . 2001 ; wang et al . 2010a,2011a ) . the nature of the kpc - scale extended x - ray emission remains uncertain ; suggestions include collisionally ionized plasma , photoionized gas , a hybrid of both , or electron - scattered nuclear emission ( elvis et al . 1983 , 1990 ; heckman & balick 1983 ; morse et al . 1995 ; schulz & komossa 1993 ; ogle et al . 2000 ; komossa 2001 ; yang et al . 2001 ) , although electron - scattering is less favored , because the required scattering column and spectral variability do not match the observations ( weaver et al . 1994 ; yang et al . 2001 ) . chandra _ imaging studies of ngc 4151 ( ogle et al . 2000 ; yang et al . 2001 ; gonzalez martin 2008 ; wang et al . 2009 , 2010a ) found good morphological agreements between the extended x - ray emission and the optical forbidden line emission ( in particular [ oiii ] @xmath35007 ) . such a close spatial correspondence between the soft x - rays and the [ oiii ] emission appears to be common in nearby seyfert 2 galaxies , leading bianchi et al . ( 2006 ) to propose that optical and x - ray features arise from a single photoionized medium . high spectral resolution grating observations of ngc 4151 with _ chandra _ and _ xmm_-newton further show that the soft x - ray emission is due to both blended emission lines and radiative recombination continua ( rrc ) from he - like and h - like transitions of carbon , oxygen , neon , and nitrogen ( ogle et al . 2000 ; schurch et al . 2004 ; armentrout et al . the x - ray line ratio diagnostics and electron temperatures of the rrc features support the conclusion that most of the gas is photoionized ( @xmath11 k ) by the agn ( schurch et al . 2004 ; armentrout et al . 2007 ) , although photoexcitation by the agn continuum may also play an important role ( kinkhabwalla et al . 2003 , ogle et al . 2003 ) . however , these grating studies of ngc 4151 can not provide the spatial distribution of each spectral feature . the extracted spectra consist of both nuclear emission and extended emission from various physical scales . therefore , based on the high resolution spectra alone , one can not exclude that the photoionized emission may be associated with the bright narrow - line region ( nlr ) gas clouds close to the nucleus ( @xmath12 pc ) , while a significant contribution from collisionally ionized gas may be present at larger radii ( e.g. , ngc 1365 , wang et al . 2009 , guainazzi et al . 2009 ) . although yang et al . ( 2001 ) presented acis imaging data , the limited signal to noise ratio ( @xmath13 ) of these data prevented detailed high spatial resolution comparisons of optical [ oiii ] and x - ray morphology , or the spectral analysis of x - ray emission from the spatially resolved features . the x - ray emission mechanisms in this prototypical seyfert 1 galaxy deserve further investigation with high quality x - ray data that allow spatially resolved spectral analysis of the circum - nuclear region . this is the aim of the study presented in this paper , the sixth and last in our series of studies on the nuclear and circum - nuclear region of ngc 4151 using deep _ chandra _ acis and hrc observations ( pi : fabbiano ) . using the hrc data and image deconvolution techniques , wang et al . ( 2009 ) resolved the nuclear emission on spatial scale of @xmath530 pc and made comparisons with _ hst _ observations . wang et al . ( 2010a ) presented the x - ray spectral analysis of the ngc 4151 nuclear emission . wang et al . ( 2010b ) reported the discovery of faint soft x - ray emission extending 2 kpc out from the active nucleus and filling in the cavity of the h i material . in paper i of the present series ( wang et al . 2011a ) , based on in - depth analysis of the acis data set , we have examined the kpc scale x - ray morphology and the relations between x - ray absorption features and the cold ism in the host galaxy . in paper ii ( wang et al . 2011b ) we have reported on strong evidence of jet cloud interaction in the nuclear region ( @xmath14 pc ) and compared these findings to our previous hrc study . here we focus on the comparison between the x - ray line emission of the ionization cone and the ionized gas traced by the optical line emission , and we report on the spectral analysis of spatially resolved x - ray features . the paper is organized as follows . in 2 , we first briefly describe the data reduction , present the x - ray emission line morphologies ( 2.1 ) and the spectral modeling of all extended emission within a @xmath15-radius ( @xmath16 kpc ) and the spatially resolved emission ( 2.2 ) . the results are discussed in 3 and summarized in 4 . ngc 4151 was observed by _ chandra _ for a total of 180 ks ( after screening for high background intervals ) with the spectroscopic array of the advanced ccd imaging spectrometer ( acis - s ; garmire et al . 2003 ) in 1/8 sub - array mode during march 27 - 29 , 2008 . we have presented the details of our acis observations of ngc 4151 and the data reduction in paper i. the data were analyzed following the standard procedures using ciao ( version 4.2 ) with the caldb 4.2.1 provided by the _ chandra _ x - ray center ( cxc ) . subpixel event repositioning and subpixel binning techniques ( paper i and references therein ) have been applied to the acis images to improve the spatial resolution . the complexity in the data analysis caused by the bright nuclear emission was described in papers i and ii . we established that for the @xmath17 x - ray emission ( @xmath18 kev ) , photon pile - up is most severe in the inner @xmath19 , mild at @xmath20 , and not an issue at @xmath21 . _ chandra _ psf simulations were performed to provide an estimate of the expected contamination from the nuclear emission in an extended feature . we have taken into account this information in the following analysis . figure [ define ] ( top ) shows the 0.32 kev acis image ( @xmath50.5 per pixel ) . bright structured soft x - ray emission along the northeast ( ne ) southwest ( sw ) direction is apparent , together with fainter , less extended emission along the northwest ( nw ) southeast ( se ) direction . the inner @xmath22 nuclear region , where we identified interactions between the radio jet and the optical clouds , has been discussed in paper ii . here we focus on the kpc - scale extended emission . we first extracted counts from every @xmath23 sectors to create azimuthal surface brightness profiles for the extended x - ray emission ( @xmath24 ) , which provides a clear view of its angular distribution , and to identify sectors that contain the bright outflows ( figure [ define ] , bottom ) . this plot was used to guide further spectral extraction ( see [ spec_analysis ] ) . the extended emission along the nw se direction is interesting as this is the direction of the putative obscuring torus . figure [ contour ] demonstrates the clear elongation perpendicular to the bicone ( in particular the nw sector ) , using the 0.32 kev acis image smoothed with the ciao tool @xmath25 . we extracted the 0.32 kev radial profiles of the nw sector , the bright sw cone , and the faint emission in between ( the `` control region '' ) , shown in figure [ 3profile ] . the nw emission is on average 5 times fainter than the bright sw cone , but brighter than the control region at @xmath26 ( @xmath27 significance ) . beyond @xmath28 , it becomes indistinguishable with the faint emission . the implication on the nuclear obscuring structure will be discussed later in [ evidence ] . a quick examination of the spectrum of the extended emission extracted from a circular region between a radius of @xmath29 and @xmath15 shows the clear presence of strong blended emission lines ( figure [ conti ] ) , corresponding to the emission lines ( @xmath302 kev ) seen in hetg observations ( ogle et al . these lines can not be uniquely resolved with the spectral resolution of the acis imaging observation . most notably the blended lines appear as three strong emission features approximately centered at 0.57 kev ( ovii ) , 0.68 kev ( oviii ) , 0.91 kev ( neix ) and a few weak lines between 12 kev ( mg xi , si k@xmath31 , sixiii ; figure [ conti ] ) . the blended emission features seen in the acis spectrum and the corresponding hetg lines are summarized in table 1 ( see 2.2.1 ) . following paper ii , we created adaptively smoothed images of the ngc 4151 circum - nuclear region using the subpixel resolution data . we extracted images from the merged data in three spectral bands dominated by emission lines below 2 kev : 0.30.7 kev ( `` soft band '' containing the ovii and oviii emission ) , 0.71.0 kev ( `` medium band '' containing the neix emission ) , and 12 kev ( `` hard band '' ) . figure [ 3color]a presents a false color composite image of the central @xmath32 ( @xmath53 kpc on a side ) , circum - nuclear region of ngc 4151 , where the soft , medium , and hard band smoothed images are shown in red , green , and blue , respectively . figure [ 3color]b zooms in to the central 1 kpc - radius region , emphasizing the bicone emission . for visualization of the faint features , the nuclear region ( see paper ii ) is saturated and excluded from this figure . the x - ray biconical morphology has been described in paper i , and here we focus on the spectral differences . the extended bicone emission ( @xmath33 ) appears rich in 0.31 kev emission ( red and green ) . similar color emission ( mixed red / green ) is also seen perpendicular to the bicone direction with a hint of harder ( blue ) emission . clumps of 0.30.7 kev ( red ; ovii and oviii ) emission are also seen , some particularly prominent features are seen in the sw cone @xmath34 and @xmath35 from the nucleus . to further highlight the regions where the strong emission lines arise , we also extracted the x - ray emission in narrow energy intervals ( 0.530.63 kev ; 0.630.73 kev ; 0.850.95 kev ) to create line strength images . this is a reasonable approach since the line emission dominates over the weak underlying continuum ( schurch et al . 2004 ) in these narrow bands . the resulting images are shown in figure [ lines ] . the position of the nucleus is indicated with a black cross . ovii and oviii line emission is prevalent in the extended regions . neix emission , while extended along the general p.a . of the large scale extended emission , is bright in the central @xmath36 but becomes fainter quickly at @xmath37 compared to the ovii emission . we examined similar emission line maps for the weak lines in the 12 kev range ( nex , neix rrc , mgxi ) and found that they are highly concentrated within the inner @xmath36 region . this is better demonstrated in figure [ lines]d , where the radial profiles of the line emissions are shown . spectra and instrument responses were generated using ciao version 4.2 and analyzed with xspec version 12.6 ( arnaud 1996 ) . background spectra were taken from a source - free region from the same chip . spectra were grouped to have at least 20 counts per energy bin to allow @xmath38 fitting . unless otherwise noted , we restricted our modeling to photon energies above 0.3 kev , where the acis calibration is good , and below 2 kev , above which the nuclear continuum dominates @xcite . the 90% confidence interval for a single interesting parameter is reported for all fitting results . we first characterize the x - ray spectrum of bright extended emission extracted from the @xmath39 circular region , exploring a phenomenological spectral fitting approach that was broadly adopted in previous studies at ccd resolution ( yang et al . 2001 ; see also ogle et al . 2000 , smith & wilson 2001 ) . the `` continuum '' in the acis spectrum ( a combination of true underlying continuum and highly blended weak emission lines ) was described by a smooth bremsstrahlung emission . we emphasize that the bremsstrahlung emission here should not be assigned any physical origin , whereas the derived line emission should be considered as a lower limit in such an approach . repeating the technique used to fit the spectrum of the nucleus ( wang et al . 2010 ) , narrow emission lines were then added to the model . hetg and _ xmm_-newton rgs spectra ( ogle et al . 2000 ; schurch et al . 2004 ; armentrout et al . 2007 ) were used as a guide for the identification of the soft x - ray lines . the x - ray spectrum of the nucleus of ngc 4151 is notoriously complex and variable . the hard x - ray continuum is dominated by a power law component suppressed by a partially covering absorber and a compton reflection component ( weaver et al . 1994 ; schurch et al . 2004 ; de rosa et al . 2007 ; wang et al . 2010b ; lubinski et al . 2010 ) , and the absorption consists of both a mildly - ionized `` warm absorber '' and a cold absorber @xcite . the psf scattered contribution from the nuclear emission is taken into account in the spectral fitting , using the nuclear component fixed at the best fit nuclear spectral model and normalization @xcite scaled to the expected fraction from psf simulations . the extraction region contains 4.7% of the 0.32 kev nuclear emission due to psf scattering , corresponding to @xmath525% of the extracted counts . this can be taken as the upper limit of contamination to the extended line emission , as the line emission dominates the nuclear spectrum . the identified emission line blends are summarized in table [ tab : extended_flux ] and the spectrum shown in figure [ 4reg2 ] ( see also figure [ conti ] ) . the absorption column required by the fit is consistent with the line - of - sight galactic column towards ngc 4151 ( @xmath40 @xmath41 ; murphy et al . the x - ray luminosity of the extended emission ( between 130 pc and 2 kpc from the nucleus ) corrected for the galactic absorption is @xmath42 erg s@xmath1 , compared with @xmath43 erg s@xmath1 for the unabsorbed nuclear source ( wang et al . 2010b ) . however , such a simplified model does not provide constraints on the emission mechanism , and more physically meaningful spectral models are needed . fitting the data with combinations of absorbed optically - thin thermal emission with solar abundances ( @xmath44 model ; smith et al . 2001 ) was attempted but gave poor results ( reduced @xmath45 ) . the abundances were then allowed to vary from solar abundance in the apec model using the variable abundance apec model ( vapec ) , in which @xmath46 , @xmath47 , @xmath48 , and @xmath49 were left free while other elements were fixed at solar abundance . when @xmath50 is allowed to vary , the fit generally improves but still is far from satisfactory ( @xmath51 ) . moreover , the required abundances become unphysically low ( @xmath52 ) as the fit attempts to reproduce a smooth continuum . to have a self - consistent model , we have made use of the cloudy photoionization modeling code ( ferland et al . although the resolution of the acis ccd spectrum prevents us from performing line - based diagnostics , e.g. deriving line ratios of the helium - like triplets , spectral fitting with photoionization models still provides useful constraints about the photoionization status ( as in bianchi et al . 2006 , 2010 ; gonzalez - martin et al . 2010 ) . using cloudy version c08.00 , which enables a cloudy / xspec interface ( porter et al . 2006 ) , we modelled the @xmath53 kev x - ray spectrum assuming an open plane - parallel geometry ( `` slab '' ) . the dimensionless ionization parameter ( osterbrock & ferland 2006 ) is defined as @xmath54 , where @xmath55 is the hydrogen number density , @xmath56 is the distance to the inner face of a model slab , @xmath57 is the speed of light , and @xmath58 is the emitting rate of hydrogen ionizing photons ( s@xmath1 ) by the ionizing source . we assumed the broken power - law used by kraemer et al . ( 2005 ) for the agn continuum ( see also armentrout et al . 2007 ) . we varied @xmath59 and @xmath60 , the column density of the slab to create spectral models covering a grid of parameters ( @xmath61 , @xmath62 @xmath41 ) , which were fed to xspec . armentrout et al . ( 2007 ) successfully modeled the xmm - newton / rgs soft x - ray spectrum of ngc 4151 with three photoionization components : a high ionization component ( @xmath63 , @xmath64 ) , a medium ionization component ( @xmath65 , @xmath64 ) , and a low ionization component ( @xmath66 , @xmath67 ) . in our initial attempt , we included all these components , with the ionization parameters and the column densities fixed at these values and the normalizations set free . in addition , the photoionized emission was absorbed with a line - of - sight column density , @xmath68 . the fit to the acis spectrum was poor ( @xmath69 ) . this is not a total surprise , given that these best fit parameters , are optimized for the brightest photoionized material close to the nucleus ( 10@xmath701 pc ; armentrout et al . 2007 ) that dominates the soft x - ray spectrum of ngc 4151 , whereas the emission of interest here arise on the @xmath71130 pc1.5 kpc scale . when the ( @xmath59 , @xmath60 ) parameters for the photoionized components were set free to vary , we were able to obtain a statistically satisfactory fit ( @xmath72 ) , reproducing the observed acis spectrum of the large scale extended x - ray emission well ( figure [ phot3 ] ) . the line - of - sight column density required by the fit is @xmath73 @xmath41 , consistent with the galactic column towards ngc 4151 @xcite . two components , a high ionization phase ( @xmath74 , @xmath75 ) and a low ionization phase ( @xmath76 , @xmath77 ) , are required to produce both the hydrogen - like and helium - like neon and oxygen line emissions . the cloudy model predicts gas pressures of @xmath78 dynes @xmath41 and @xmath79 dynes @xmath41 for the high @xmath59 component and low @xmath59 component , respectively . if they are collocated , additional pressure perhaps from a hot intercloud medium in a multiphase ism ( elvis et al . 1983 , ogle et al . 2000)is needed to reach pressure equilibrium . note that the @xmath76 component closely resembles the `` low '' model component in armentrout et al . ( 2007 ) and the `` d+ea '' x - ray absorption component in kraemer et al . ( 2005 ) . these model components still left small residuals at @xmath50.5 kev ( the nvii ly@xmath31 ) and 1.02 kev ( the nex ly@xmath31 ) . adding in a third photoionized component can further improve the fit ( best fit parameters @xmath80 , @xmath81 ) , but the improvement appears not significant when an @xmath82-test is performed ( at @xmath83 significance level ) . thus the two - component model is used as a template for the following analysis of subregions . given the satisfactory spectral fits with photoionized components and lack of clear signatures of collisional ionization in the enlr , how much thermal emission from hot gas may be present in the circum - nuclear region of ngc 4151 ? we obtain a constraint on the possible presence of the hot ism by invoking a collisionally ionized component in the best fit photoionization model . after adding a thermal emission component ( @xmath44 ) to the photoionized emission , the fit is improved with a @xmath84 kev ( at @xmath85 significant level by f - test ) and the residual at 0.91.1 kev is mostly accounted for . this thermal component may contribute @xmath86 of the total soft x - ray emission , @xmath87 erg s@xmath1 . as a cautionary note , although @xmath44 component improves the fit statistically , it remains possible that the residuals are due to photoexcitation of fe l - shell and the enhanced resonance lines ( e.g. , ogle et al . 2003 ; kinkhabwala et al . 2003 ) that are inadequately accounted for with the cloudy photoionization model components . this will be explored with the high resolution hetg spectrum in future work . we will discuss the possible presence of the thermal component further . we emphasize that , because of likely complex geometrical dependencies and absorption between these components in the kpc scale emission , this modeling may still be over - simplistic to simultaneously reproduce the observed features in the broad band . to perform spectral analysis of the extended x - ray emission , the large scale region was divided into four sectors with position angles ( p.a . ) as outlined in figure [ define ] . extraction regions are termed the `` sw '' , `` nw '' , `` se '' and `` ne '' regions respectively ( table 2 ; see figure [ define ] ) . the bisectors of ne and sw correspond to the approximate p.a . of the enlrs ( e.g. , robinson et al . 1994 ) . they are further divided into subregions along the radial direction ( e.g. , sw 1 , 2 , 3 , and 4 ; ne 1 , 2 , and 3 ) , with @xmath51000 counts ( 0.32 kev ) in the fainter outer regions ( sw4 and ne3 ) . the nw and se sectors cover regions ( @xmath88 ) perpendicular to the bicone , which is the direction of the large scale weak bar ( asif et al . 1998 ) . corresponding to the variations in color , spectral differences are clearly visible when the acis spectra across sectors are compared ( figure [ compare4 ] ) . the fitting results are summarized in table 2 , and figures [ sw ] , [ ne ] , and [ se - nw ] show the corresponding spectra and the fits . we find that : 1 . except for the outer - most regions ( ne3 and sw4 ) , a high ionization component is present in all regions ( @xmath891.26 ) . lower ionization components are present along the bicone direction ( ne sw ) , with @xmath90 values ranging between -0.5 and 0 . recall that the fit to the integrated spectrum , which includes emission from all the subregions , finds the @xmath91 and @xmath92 components . 2 . in the ne3 and sw4 regions , a very low ionization component ( @xmath93 ) and a high column density ( @xmath94 ) are required by the fit to improve the residuals in the mgi k@xmath31 and sii k@xmath31 . an f - test indicates that this fluorescent component significantly improve the fit at @xmath95 confidence . the second photoionized component in the nw se regions , perpendicular to the ionization cone , shows a lower ionization ( @xmath96 ) . we also extracted spectra of the x - ray emission in the bicone regions that are prominent in 0.30.7 kev emission ( or `` ovii '' rich , visually red in figure [ 3color ] ) and in 0.71 kev emission ( or `` neix '' rich , visually green in figure [ 3color ] ) . the spectral fits are shown in figure [ red - green]a , b respectively . whereas the fitted ionization parameter for the `` neix '' rich emission is close to the average value , the `` ovii '' rich emission is generally of lower ionization , likely due to higher number density . this is consistent with the appearance of clumpy material in the optical . kiloparsec - scale extended x - ray emission is ubiquitously observed in nearby obscured seyfert galaxies with enlrs ( bianchi et al . 2006 ; guainazzi et al . 2009 ; bianchi et al . 2010 ; gonzalez - martin et al . 2010 ) , but the ionization mechanism for this emission has been a subject of active debate . for ngc 4151 , yang et al . ( 2001 ) proposed a two phase model for the extended x - ray emission , including a photoionized warm gas and a hotter collisionally ionized gas ( see also komossa 2001 ) . the relatively low electron temperatures ( @xmath97 k ) inferred from the radiative recombination continua ( rrc ; see liedahl 1999 for review ) and the line ratios from the he - like triplets strongly support the presence of a photoionized component ( ogle et al . 2000 ; schurch et al . 2004 ; kraemer et al . 2005 ) . however , does the observed soft spectrum entirely consist of photoionized emission , or does collisional ionization become more important than the photoionization mechanism beyond the nuclear region ( armentrout et al . 2007 ) ? several of our results point to photoionization as the dominant mechanism . * _ emission line gas morphologies _ first of all , the morphological agreement between soft x - ray and optical [ oiii ] emission is very strong . in particular , there is a good spatial correlation between [ oiii ] line emission and the three strong x - ray emission lines ( figures [ lines]a - c ) . there is clearly an overall morphological coincidence with the ionized gas bicone traced by [ oiii ] emission , as noted in previous x - ray studies ( e.g. , ogle et al . 2000 , yang et al . 2001 ) . a good agreement between the peaks of optical and x - ray line emissions in the inner @xmath98 is found . * _ [ oiii]-to - soft x - ray ratio of the _ hst_-resolved clouds _ bianchi et al . ( 2006 ) studied a sample of eight seyfert 2 galaxies , and suggested that the same kpc - scale gas photoionized by the agn continuum can simultaneously produce the observed x - ray and [ oiii ] emission . our hrc and acis studies of the inner 2 ( @xmath99 pc ) emission in ngc 4151 ( wang et al . 2009 and paper ii ) have shown that , for most of the clouds close to the nucleus , the ratios of [ oiii ] to soft x - ray emission are consistent with the typical ratios in bianchi et al . ( 2006 ) , the exception being clouds collisionally ionized by the radio jet . + our subpixel - processed acis images of ngc 4151 enable us to directly compare the extended x - ray emission and those of the resolved [ oiii ] clouds ( figure [ lines ] ) . the main cloud features are labeled in figure [ oiii_cloud ] . using the calibrated continuum - subtracted hst / wfpc2 f502n image ( kaiser et al . 2000 ) , we measured the [ oiii ] fluxes for the clouds and listed them in table [ fluxratio ] . acis spectra were extracted from the same regions and fitted with the continuum plus emission line model to obtain their 0.52 kev x - ray fluxes . figure [ ratio ] shows the [ oiii ] to soft x - ray ratio for the distinct cloud features at various radii to the nucleus ( @xmath5150 pc1500 pc ) . the ratios listed in table [ fluxratio ] are consistent with a constant [ oiii]/x - ray flux ratio of @xmath515 for the 1.5 kpc range spanned by these measurements . the radially constant [ oiii]/x - ray ratios indicate that the relative contributions of the low- and high - ionization gas phases at different radii are similar , as pointed out by our referee . if the [ oiii ] and x - ray emission arise from a single photoionized medium , as discussed in bianchi et al . ( 2006 ) , this further implies that the ionization parameter does not vary significantly to a large radial distance and the density is approximately decreasing as @xmath100 , as expected for a freely expanding nuclear wind based on mass conservation ( kraemer et al . 2000 ; bianchi et al . 2006 ) . * _ spectral fitting _ the spectral analysis shows that , a model with just two photoionized components of high- and low - ionization parameters can describe well the observed x - ray spectra in various regions across the circum - nuclear area , whereas in the spectral fits that invoke only collisionally ionized components strong emission line residuals persist . this supports the dominant role of nuclear photoionization . * _ kinematics of the ionized gas _ the full velocity field of the nlr for ngc 4151 was mapped by kaiser et al . ( 2000 ) , which supports a hollow biconical outflow model for these clouds . at distances exceeding 4 ( @xmath5260 pc ) from the nucleus , kaiser et al . ( 2000 ) find that the gas cloud kinematics are consistent with circular rotation in a gravitational field , identical to the host ism . therefore , except for a few clouds in the nuclear region that appear disturbed by the impact of the radio jet ( * ? ? ? * ; * ? ? ? * see also paper ii ) , there is no evidence for strong kinematic shocks or any strong shocks in the outflow , which is consistent with the gas being photoionized by the nuclear radiation . based on all the above considerations , we conclude that photoionization by the nucleus is most likely the dominant ionization mechanism in the enlr of ngc 4151 , confirming the suggestions of previous studies ( ogle et al . 2000 ; schurch et al . 2004 ; kraemer et al . 2005 ; armentrout et al . 2007 ; wang et al . it is reassuring to establish a role for nuclear photoionization in the extended nlrs that is consistent with the grating studies . however , there is also evidence that points to a more complex emission mechanism than a single photoionized medium ( e.g. , bianchi et al . we discuss these results and their implications below in sections 3.2 through 3.4 . in [ sec - large_scale ] , we estimated the extent of thermal emission from hot gas ( @xmath101 erg s@xmath1 ) that could be present in the circum - nuclear region of ngc 4151 . chandra _ grating study of ngc 4151 by ogle et al . ( 2000 ) provided the first direct evidence of x - ray line emission from a collisionally ionized hot plasma . armentrout et al . ( 2007 ) also noted that the poor fit to the ovii line profile could be due to a non - photoionized component . wang et al . ( 2011b ) further show that a @xmath102 kev thermal emission component is present in the inner @xmath36 region , which likely originates from jet - cloud interaction . although star formation activity in the circum - nuclear region is at a low level ( asif et al . 2005 ; dumas et al . 2010 ) , hot gas may originate from the ngc 4151 galactic disk , since its temperature ( @xmath103 kev ) and luminosity ( @xmath104 ) are comparable to the disk diffuse emission in some star forming galaxies , @xmath102 kev , @xmath105 erg s@xmath1 ( e.g. , ngc 253 ; strickland et al . 2002 ) . another plausible alternative is that the hot plasma is the intercloud medium uniformly filling in the outflow ( i.e. , filling factor @xmath106 ) , providing pressure confinement for the nlr clouds ( elvis et al . 1983 ; cf . krongold et al . 2007 for warm absorber outflows ) . the volume emission measure for the thermal component is @xmath107 cm@xmath70 , derived from the normalization after fitting an additional thermal component ( apec ) to the photoionization models of the bright biconical emission ( 2.2.1 ) . assuming the hollow bi - conical nlr geometry ( crenshaw et al . 2000 ; das et al . 2005 ) , we obtain @xmath109 cm@xmath70 and so an average density of @xmath110 cm@xmath70 . thus the thermal pressure of the @xmath111 k gas is @xmath112 dyne @xmath41 . comparing with the thermal pressure of @xmath11 k and @xmath113 cm@xmath70 nlr clouds ( robinson et al . 1994 ) gives @xmath114 dyne @xmath41 . thus pressure equilibrium between the hot and cool phase is highly likely , confirming the findings on pressure confinement in ogle et al . ( 2000 ) . figure [ lines]a clearly shows that one distinct ovii emission blob `` a '' ( at @xmath115 , @xmath5390 pc sw from the nucleus ) does not align with a bright [ oiii ] emitting cloud , designated as `` k1 '' in robinson et al . ( 1994 ) and offset by @xmath116 to the se of `` a '' ( see yang et al . we find this is also the case for a more distant x - ray blob `` b '' ( @xmath117 sw ) where the closest [ oiii ] emission cloud is 2 to the sw . similar displacements between the x - ray enhancement and the [ oiii ] blob are also clearly found in the oviii line emission ( figure [ lines]b ) . note that the _ chandra _ astrometry is accurate to 0.3(paper i ) , therefore the mis - alignment between features at this spatial scale is significant and indicates that the origin of the optical / x - ray emission needs further investigation . the radial location of x - ray blob``a '' lies at the boundary between nlr and enlr identified in the optical ( robinson et al . 1994 ) , where the density drops rapidly from @xmath118 cm@xmath70 to @xmath119 cm@xmath70 @xcite . we can estimate the ionization parameter of k1 . from the nuclear spectrum @xcite , we obtain an ionizing luminosity @xmath120 erg s@xmath1 . assuming a number density of 200 cm@xmath70 ( penston et al . 1990 ; robinson et al . 1994 ) and a radial distance of @xmath121 cm to the nucleus , we obtain an ionization parameter @xmath122 . this is lower than , but in agreement with , the value of @xmath123 derived from the [ oiii]/[oii ] ratio in robinson et al . ( 1994 ) , supporting the photoionization assumption . since the x - ray blob and the [ oiii ] cloud are located at the same radial distance , their spatial offset suggests that the x - ray blob is in a higher ionization phase than the [ oiii]-bright cloud . a likely explanation is that the x - ray blob has a lower density in the clump . if their densities are similar , the difference might arise from a higher ionizing flux received at the surface of the x - ray blob , perhaps due to less screening than in the direction towards the [ oiii ] cloud . our spectral fitting results show that a range of ionization states are needed to obtain good fits , in agreement with the multiple ionization parameters required in the grating spectra ( e.g. , armentrout et al . 2007 ) . a high ionization ( @xmath124 ) component is found in the spectral fitting for all regions surrounding the nucleus , including the nw and se sectors perpendicular to the ionization cone ( table 2 ) . a low ionization ( @xmath125 ) component is also present in the nw se direction that has lower @xmath59 than found in the cone direction ( ne sw ) . this strongly indicates that there is no continuous absorbing `` torus '' blocking all the ionizing photons . this hypothesis is supported by the agreement between the observed h@xmath126 line flux in the `` torus '' direction and the value predicted from x - ray - irradiation by the active nucleus ( paper i ) , and by the measurement of hi absorption in the nuclear region , where high levels of clumping on the smallest scales is found ( mundell et al . we note that a `` patchy torus '' or clumpy wind arising from the accretion disk in which the clouds are optically thick ( e.g. , * ? ? ? * ) , may allow leakage of nuclear continuum for such a high @xmath59 component . however , it can not reproduce the @xmath125 component in the directions where the continuum is blocked . extended emission along the `` torus '' direction is clearly present in the large scale x - ray images shown in figures [ contour ] ( indicated by the arrows ) and [ 3color]a . the base of this emission is at the location of the nucleus , thus may indicate ionization by the leaked photons in this direction . moreover , we found an intriguing coincidence between radial regions of enhanced x - ray surface brightness and the p.a . of outflowing optical clouds identified by the _ hst _ study ( das et al . they pointed out that these faint clouds flowing close to the plane of the torus would need to be ionized by leaked nuclear radiation , which is attested by our x - ray data . the low @xmath59 suggests a lower ionizing flux received by some clouds , likely due to nuclear continuum filtered by `` warm absorbers '' ( e.g. , * ? ? ? * ) or the cone walls ( e.g. , * ? ? ? evaluating the difference between the ionization parameters in the direction of torus and in the bicone , we estimate that approximately 1% of the total ionizing flux is seen by the nw se clouds . previous work on the nlr kinematics of ngc 4151 have established that the bi - cone geometry in ngc 4151 results in one side of the conical outflow exiting the galactic plane at a small angle , partly intersecting the host disk ( e.g. , robinson et al . 1994 ; crenshaw et al . 2000 ; das et al . 2005 ; storchi - bergmann et al . 2009 ) . thus part of the host ism is likely illuminated by the agn radiation . our x - ray work finds that multiple phases of photoionized gas are present in the enlr of ngc 4151 . the fitting results for the `` ovii '' rich ( red in figure [ 3color ] ) and `` neix '' rich ( green in figure [ 3color ] ) regions indicate that there is a range of ionizations , with a lower ionization phase associated with the clumpy clouds and the higher ionization with the diffuse outflow . there is a lowest ionization ( @xmath127 ) component at larger radii in the sw and ne sectors . this component may come from high density material in the host disk plane , perhaps the nuclear spirals , illuminated by the nuclear radiation . we further note that interactions between the biconical outflow and the host galaxy ism , similar to the radio jet cloud interaction in the nuclear region , may contribute to the heating of any hot gas present in the host disk . currently there is no evidence in the gas kinematics that supports any strong interaction ( kaiser et al . however , as we discuss below , there is some morphological evidence that , at the largest radii , the outflow runs into a clump in the inhomogeneous ism and perhaps results in local shock heating of the gas . figure [ morph ] gives a large scale ( 1 arcmin@xmath1281 arcmin ) view of the key ism components in the circum - nuclear region of ngc 4151 , illustrating the directions of the kinematic major axis of the host galaxy ( green solid line ; p.a.@xmath522@xmath129 , pedlar et al . 1992 ; mundell et al . 1999 ) , the large scale `` weak fat bar '' ( green dotted line ; p.a.@xmath5130@xmath129 ; mundell & shone 1999 ) , the enlr bicone ( cyan line ; p.a.@xmath565@xmath129 , evans et al . 1993 ) , and the distribution of hi ( mundell et al . 1999 ) and co ( dumas et al . 2010 ) gas , respectively . in paper i we noted that the ne soft x - ray emission reaches as far as 12 ( 780 pc ) from the nucleus , but that the surface brightness then decreases significantly by a factor of 5 . comparing the continuum - subtracted h@xmath31 emission ( knapen et al . 2004 ) with the extended soft ( 0.31 kev ) x - ray emission ( figure [ fig8]a ) , the ionized gas traced by bright h@xmath31 emission is mainly located in the @xmath130 biconical region along the ne sw direction centered on the nucleus , which also closely follows the soft x - ray emission . an arc - like feature in the h@xmath31 image ( @xmath131 ) is clearly present at the @xmath132 location ( figure [ morph ] ) at the edge of the x - ray emission , and the northern co gas lane is also found here ( figure [ fig8]b , and dumas et al . 2010 ) . figure [ fig8]b provides a multiwavelength , enlarged view of this region , showing the spatial relations between the soft x - ray emission , the h@xmath31 , the co gas lanes , and the hi distribution @xcite . the termination of the ne x - ray emission appears to be closely associated with the presence of the dense cold neutral material ( traced by the co and hi emission ) in the galactic disk . no similar feature is seen to the sw of the nucleus . the origin of the h@xmath31 arc ( figure [ fig8]a ) at the terminus of the ne x - ray emission is intriguing . it is spatially very close to the northern co gas lane and is one of the dust arcs identified in the optical @xmath133 color map @xcite , which are gaseous compressions driven by the bar potential @xcite . in this context , the h@xmath31 arc may be a more clumpy continuation of the dense co arc and photoionized by the agn radiation . the rate of ionizing photons from the agn is approximately @xmath134 s@xmath1 ( kraemer et al . 2005 ; c.f . @xmath120 erg s@xmath1 , wang et al . 2010b ) . assuming that the arc has a line - of - sight depth that is comparable to its projected length ( @xmath51 kpc ) , the arc s intercepted ionizing photon rate is @xmath135 s@xmath1 at its distance to the nucleus ( 970 pc ) , a covering factor of almost @xmath136 . this photon budget should be treated as an upper limit as some fraction of the ionization photons will be absorbed or scattered on the way to the arc . if we assume that the gas is in photoionization equilibrium , for the case b effective recombination rate of h@xmath31 @xcite , a rate of @xmath137 s@xmath1 ionizing photons s@xmath1 is needed to produce the observed h@xmath31 luminosity of the arc ( @xmath138 erg s@xmath1 ) , which is only 1% of the maximal rate of available ionizing radiation in the direction of the arc ( for a covering factor @xmath139 ) . the scenario where part of the co gas lane is photoionized by the agn and produces the h@xmath31 arc seems plausible with a modest covering factor of @xmath140 , implying that the arc has a filamentary structure with a thickness of a few pc . on the other hand , we consider the possibility that the arc could be a bow shock feature from interaction between the biconical outflow and dense gas in the host disk piled in the co gas lanes by the large scale stellar bar . assuming that the x - ray emission is due to shock heating as the outflow encounters the dense co lane , and that the shock is strong and adiabatic , we estimate the shock velocity @xmath141 , adopting a postshock temperature @xmath142 ( lehnert et al . 1999 ) , where @xmath143 is the mass per particle ( co and hi ) and @xmath144 the boltzmann constant . for the @xmath145 kev x - ray temperature measured at the arc , the required @xmath141 is @xmath5150 km s@xmath1 , which is much higher than the local sound speed @xmath146 km s@xmath1 in the @xmath147 k h@xmath31-emitting gas . although there is no direct measurement at the exact position of the h@xmath31 arc , the measured outflow velocity in [ oiii ] at @xmath148 ne of the nucleus is fully consistent with this value of @xmath141 , e.g. , cloud#28 which has @xmath149 km s@xmath1 ( kaiser et al . 2000 ) . for two of the brighter , spatially resolved [ oiii]-emitting clouds seen in the @xmath150 images ( clouds # 1 and # 2 in figure [ oiii_cloud ] ; see also figure 4 in kaiser et al . 2000 ) , we were able to constrain the number density for the x - ray emitting gas with cloudy modeling of their x - ray spectra when an inner radius @xmath151 pc to the nucleus and an ionizing photon rate @xmath152 are set . the best - fit density is @xmath153 cm@xmath70 . we use this number density to estimate the mass loss rate and the kinematic power of the hot phase outflow , which are key quantities that gauge the impact of agn outflow on the host galaxy ism . following barbosa et al . ( 2009 ) and @xcite , we calculate the mass loss rate using @xmath154 = @xmath155 , where @xmath156 is the radial velocity ( taken as the approximate outflow velocity ) , @xmath157 the proton mass , @xmath158 the gas filling factor , and @xmath159 the area in the cross section of the outflow . for a radial distance of @xmath160 cm ( 130 pc ) and the hollow cone geometry of das et al . ( 2005 ) , we obtain @xmath161 @xmath162 . adopting @xmath163 @xcite and @xmath164 km s@xmath1 ( crenshaw et al . 2000 ; storchi - bergmann et al . 2010 ) , we obtain a mass outflow rate @xmath165 yr@xmath1 and a kinetic luminosity for the outflow @xmath166 ergs s@xmath1 , or @xmath167 of the bolometric luminosity of the agn ( @xmath168erg s@xmath1 ; kaspi et al . 2005 ) . the @xmath154 derived from the hot phase outflow is comparable to the value derived from nir studies of ionized gas in ngc 4151 @xcite , which estimated mass outflow rate @xmath1691.2 @xmath6 yr@xmath1 along each cone . the @xmath154 value is over 10 times higher than the outflow rate derived from @xmath170 and optical spectra ( 0.16 @xmath6 yr@xmath1 ; crenshaw & kraemer 2007 ) , but the latter is measured much closer to the nucleus ( at 0.1 pc ) . the optical emitting gas likely has a much smaller filling factor than the x - ray emitting gas , which explains the difference in the measurements . if our study measured an outer part of the same outflow , it could have accelerated to a higher velocity or have loaded ism between 0.1 pc and 100 pc , hence a higher @xmath154 . how does @xmath154 compare to the accretion rate of the smbh in ngc 4151 ? the bolometric luminosity of ngc 4151 , @xmath171 erg s@xmath1 ( kaspi et al . 2005 ) , is about 1.2% of its eddington luminosity , which is @xmath172 erg s@xmath1 for the black hole mass @xmath173m@xmath174 of ngc 4151 ( bentz et al . 2006 ) . adopting @xmath175 and @xmath176 for the efficiency of the accretion disk ( shakura & sunyaev 1973 ) , this implies that the central engine of ngc 4151 is accreting at @xmath177 yr@xmath1 . since the measured outflow rate is 160 times the accretion rate necessary to feed the active nucleus , there must be either significant entrainment of the host galaxy s ism while the nuclear outflow is expanding , or part of the gas fueling the nucleus is ejected before radiating , or a high fraction ( over 99% ) of the x - ray gas does not take part in the outflow and is illuminated by the agn `` in situ '' . however , the latter seems unlikely as the grating spectra strongly indicate that the emission lines have outflow velocities of 200300 km s@xmath1 ( e.g. , armentrout et al . 2007 ) . significant mass loading of the outflow by the host ism may be most plausible and common , since previous estimated nlr outflow rates in seyferts ( @xmath50.110 @xmath6 yr@xmath1 ) generally exceed tenfold the accretion rate ( veilleux et al . 2005 ) . nevertheless , even with such a high @xmath154 in ngc 4151 , the kinematic power @xmath178 suggests that only 0.3% of the available accretion power is extracted to drive the outflow , similar to the findings of @xcite and holt et al . ( 2006 , 2011 ) . for a sample of seyfert nlr outflows ( ngc 4151 not included ) studied in the [ siii]@xmath179 emission , barbosa et al . ( 2009 ) reported a lower @xmath180 . as pointed out in mathur et al . ( 2009 ) , this poses a problem to the majority of quasar feedback models which require a higher fraction of the accretion power of the black hole ( @xmath55% of @xmath181 to unity ) to be thermally coupled to the host ism . our @xmath182 is in rough agreement with a two - stage feedback model recently proposed by hopkins & elvis ( 2010 ) , which requires only 0.5% of the radiated energy to drive the initial hot outflow to efficiently shut down star formation in the host . in this paper we present spectral analysis and emission line images from deep _ chandra _ observation of ngc 4151 , aiming to resolve and characterise the x - ray emission in the nlr . the findings are summarized as follows : * the soft x - ray emission line images of ngc 4151 ( ovii , oviii , and neix ) are clearly extended and show remarkable morphological coincidence with the biconical nlr mapped by the [ oiii ] emission , which supports a common emission mechanism for the hot- and cool - phase of the nlr gas . * extended emission in the x - ray image is detected along the nw se sectors , which is the direction of a putative torus . this may explain the faint rogue clouds identified in previous hst studies that require ionization in this direction , indicating leakage of nuclear ionization instead of full blocking by a continuous obscuring torus . * spectral models involving smooth continua ( a bremsstrahlung plus a power law ) with emission lines provide good descriptions of the spectra . the emission lines can not be uniquely identified with the present spectral resolution , but are consistent with the brighter lines seen in the _ chandra _ hetgs and xmm - newton rgs spectra below 2 kev . the absorption corrected x - ray luminosity of the extended emission between @xmath183 pc and @xmath184 kpc is @xmath42 erg s@xmath1 . * photoionization models successfully reproduce the soft x - ray emission , supporting the dominant role of nuclear photoionization . there are considerable variations in ionization states across the circum - nuclear region . a high ionization ( @xmath185 ) component is present in most regions . a low ionization ( @xmath92 ) component is present along the bicone direction ( ne sw ) , whereas a lower ionization ( @xmath125 ) component is found in the nw se direction , which is consistent with filtered nuclear emission by warm absorbers instead of a continuous absorbing torus . the lowest ionization component ( @xmath127 ) appears to be associated with the dense gas in the host plane . a thermal component may still be present at @xmath86 of the total soft emission , perhaps related to hot ism in the galactic disk or shocks associated with the outflow . * the measured ratios of [ oiii]/soft x - ray flux are consistent with a constant ratio of @xmath515 for the 1.5 kpc radius spanned by these measurements . this suggests a similar relative contributions from the low- and high - ionization gas phases at different radii . if the [ oiii ] and x - ray emission arise from a single photoionized medium , this further implies an outflow with a wind - like density profile ( @xmath186 ) . * the estimated mass outflow rate in ngc 4151 is @xmath52@xmath6 yr@xmath1 at 130 pc and the kinematic power of the ionized outflow is @xmath187 erg s@xmath1 , 0.3% @xmath181 . this value is significantly lower than the expected efficiency in the majority of quasar feedback models , but comparable to the two - stage model described in @xcite . placing all our findings in the context of previous studies , we obtain a comprehensive view of the ngc 4151 circum - nuclear region at various spatial scales , from the inner - most @xmath188 pc ( as illustrated in the schematic drawing in figure [ cartoon]a ) to as far as @xmath189 kpc ( figure [ cartoon]b ) . the key points are recapped here . photoionization by the nucleus is important at all scales . in the nuclear region , except for a few clouds that show interaction with the radio jet ( wang et al . 2009 , 2011b ) , most of the x - ray nlr clouds are consistent with being part of a photoionized biconical outflow , with a @xmath190 density profile expected for a nuclear wind . in the enlr of ngc 4151 , photoionization is still the dominant ionization mechanism for the observed x - ray emission ( e.g. , ogle et al . 2000 ; kraemer et al . 2005 ) , but a wide range of ionization states is present . this is likely related to the nature of the enlr . the bi - cone geometry indicates that one side of the conical outflow exits the galactic plane at a small angle , partly intersecting the host disk ( e.g. , crenshaw et al . 2000 ; das et al . 2005 ; storchi - bergmann et al . 2010 ) . optical studies ( e.g. , robinson et al . 1994 ; kaiser et al . 2000 ) have identified such a boundary between nlr and enlr at @xmath191 ( sw ) , and the ionized gas kinematics become consistent with the rotation velocity of the galactic disk at @xmath192 . thus the origin of the enlr is best explained as the inhomogeneous ism in the galactic plane ionized by the agn . part of the high density nuclear spirals illuminated in the path of the bicone can produce the curvy shape of x - ray enhancement and the associated low ionization . the presence of ionized emission perpendicular to the bicone indicates leaked nuclear emission along the putative torus direction , likely filtered nuclear emission by warm absorbers . in the large 3 kpc - scale cavity of the h i material , we find faint soft diffuse x - ray emission that provides evidence for agn host interaction in ngc 4151 ( wang et al . 2010a ) , originated from either hot gas heated by the nuclear outflow or photoionized gas from a recent nuclear outburst . as a concluding remark , our findings in ngc 4151 demonstrate that abundant information can be extracted from x - ray spectral imaging studies when supplemented with multiwavelength data . many valuable new radio , optical and ir observations and modeling of other nearby seyfert galaxies are emerging ( crenshaw et al . 2010 ; riffel et al . 2010 ; fisher et al . 2011 , schnorr mller et al . 2011 ) , and progress in the x - rays are being made to further our understanding of agn feeding and feedback processes in these galaxies . we thank the anonymous referees for providing us with detailed and constructive comments that have improved the clarity of this manuscript . this work is supported by nasa grant go8 - 9101x and go1 - 12009x . we acknowledge support from the cxc , which is operated by the smithsonian astrophysical observatory ( sao ) for and on behalf of nasa under contract nas8 - 03060 . cgm acknowledges financial support from the royal society and research councils u.k . gd was supported by dfg grants sch 536/4 - 1 and sch 536/4 - 2 as part of spp 1177 . j. w. thanks t. kallman , g. ferland , s. bianchi , a. marinucci , and s. chakravorty for advice on photoionization modeling . this research has made use of data obtained from the _ chandra _ data archive , and software provided by the cxc in the application packages ciao and sherpa . some of the data presented in this paper were obtained from the multimission archive at the space telescope science institute ( mast ) . stsci is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 . cccccc ovii & & 0.57 & ... & @xmath193 & 13% + ... & ovii f & 0.5617 & 0.5611 & @xmath194 & + ... & ovii i & 0.5691 & 0.5687 & @xmath195 & + ... & ovii r & 0.5745 & 0.5740 & @xmath196 & + + oviii & & 0.66 & ... & @xmath197 & 13% + ... & oviii ly@xmath31 & 0.6541 & 0.6537 & @xmath198 & + ... & nvii rrc & 0.67 & ... & @xmath199 & + + ovii & & 0.73 & ... & @xmath200 & 12% + ... & ovii rrc & 0.74 & ... & @xmath201 & + ... & oviii ly@xmath202 & 0.7749 & 0.7747 & @xmath203 & + + oviii & & 0.82 & ... & @xmath204 & 2% + ... & oviii ly@xmath205 & 0.8175 & 0.8171 & @xmath206 & + ... & oviii rrc & 0.86 & ... & @xmath206 & + + neix & & 0.91 & ... & @xmath207 & 21% + ... & neix f & 0.9064 & 0.9052 & @xmath208 & + ... & neix i & 0.9172 & 0.9151 & @xmath209 & + ... & neix r & 0.9229 & 0.9221 & @xmath210 & + + nex & & 1.03 & ... & @xmath204 & 17% + ... & nex ly@xmath31 & 1.0224 & 1.0219 & @xmath210 & + ... & neix @xmath211 & 1.0749 & 1.0738 & @xmath212 & + + nex & & 1.20 & ... & @xmath213 & 12% + ... & neix rrc & 1.20 & ... & @xmath214 & + ... & nex ly@xmath202 & 1.2118 & 1.2110 & @xmath215 & + + mgxi & & 1.30 & ... & @xmath213 & 9% + ... & mgxi f & 1.3542 & 1.3313 & @xmath216 & + ... & mgxi r & 1.3525 & 1.3524 & @xmath214 & + + sii & & 1.75 & ... & @xmath217 & 9% + ... & sii k@xmath31 & 1.7413 & 1.7391 & @xmath207 & + ... & mgxii ly@xmath202 & 1.7472 & 1.7450 & @xmath218 & + ne 1 & 2432 & 0.04@xmath219 & 0.86 & 19.7 & -0.22 & 19.7 & @xmath220 & 63/65 & @xmath221 + .... 2 & 1814 & 0.07@xmath222 & 1.05 & 19.8 & 0.06 & 20.1 & @xmath223 & 60/53 & @xmath224 + .... 3 & 913 & 0.03@xmath225 & 0.49 & 19.9 & -1.87 & 23.5 & @xmath226 & 26/29 & @xmath227 + sw 1 & 2096 & 0.02 & 0.83 & 20.0 & -0.38 & 19.4 & @xmath228 & 60/57 & @xmath229 + .... 2 & 1667 & 0.02 & 1.06 & 20.0 & -0.42 & 19.8 & @xmath230 & 58/45 & @xmath231 + .... 3 & 1560 & 0.02 & 1.00 & 19.6 & -0.50 & 19.6 & @xmath232 & 54/46 & @xmath233 + .... 4 & 966 & 0.02 & 0.34 & 19.0 & -1.28 & 23.5 & @xmath234 & 34/30 & @xmath235 + nw & 1050 & 0.22@xmath219 & 1.03 & 19.2 & -0.98 & 19.5 & @xmath236 & 35/35 & @xmath237 + se & 1081 & 0.16@xmath238 & 1.26 & 20.3 & -1.16 & 19.9 & @xmath239 & 28/33 & @xmath240 + ovii - rich & 2626 & 0.02@xmath219 & 0.14 & 20.3 & -1.22 & 20.3 & @xmath223 & 60/56 & @xmath241 + neix - rich & 2141 & 0.07@xmath219 & 0.6 & 19.6 & -0.21 & 20.4 & @xmath242 & 50/56 & @xmath243 + cccccc 1 & 2.6 & 169 & 29.8@xmath244 & 25.7@xmath245 & 11.6 + 2 & 2.7 & 175 & 21.1@xmath246 & 15.2@xmath247 & 13.8 + 3 & 3.2 & 208 & 2.6@xmath248 & 2.4@xmath249 & 10.7 + 4 & 4.8 & 312 & 7.9@xmath250 & 5.4@xmath251 & 14.5 + 5 & 3.5 & 228 & 1.2@xmath248 & 1.6@xmath252 & 7.5 + 6 & 4.7 & 306 & 1.9@xmath253 & 4.7@xmath254 & 4.1 + 7 & 6.1 & 397 & 10.2@xmath255 & 6.6@xmath256 & 15.2 + 8 & 10.7 & 696 & 3.9@xmath257 & 4.2@xmath258 & 9.6 + 9 & 17.9 & 1164 & 0.9@xmath259 & 0.2@xmath260 & 43.4 + 10 & 17.7 & 1151 & 1.8@xmath259 & 0.9@xmath261 & 20.3 +

we study the internal response of a galaxy to an unbound encounter and present a survey of orbital parameters covering typical encounters in different galactic environments . overall , we conclude that relatively weak encounters by low - mass interloping galaxies can cause observable distortions in the primaries . the resulting asymmetries may persist long after the interloper is evident . we focus our attention on the production of structure in dark halos and in cluster ellipticals . any distortion produced in a dark halo can distort the embedded stellar disk , possibly leading to the formation of lopsided and warped disks . we show that distant encounters with pericenters in the outer regions of a halo can excite strong and persistent features in the inner regions . features excited in an elliptical are directly observable and we predict that asymmetries in the morphologies of these systems can be produced by relatively small perturbers . for example , a fly - by on an orbit with pericenter equal to the half - mass radius of the primary system and velocity of 200 km / s ( a value typical for groups ) can result in a significant dipole distortion for perturbers with mass as small as 5% of primary s mass . we use these detailed results to predict the distribution of the parameter defined by abraham et al . ( sensitive to lopsidedness ) and the shift between the center of mass of the primary system and the position of the peak of density for a range of environments . we find that high - density , low - velocity dispersion environments are more likely to host galaxies with significant asymmetries . our distribution for the parameter is in good agreement with the range spanned by the observed values for local galaxy clusters and for distant galaxies in the medium deep survey and in the hubble deep field . assuming that primordial galaxies were located in dense environments with previrialized low velocity dispersions , our conclusions are consistent with the observational results showing a systematic trend for galaxies at larger redshifts to be more asymmetric than local galaxies . finally , we propose a generalized asymmetry parameter which provides detailed information on the radial structure of the asymmetry produced by the mechanism explored in our work .

astro-ph9908099

galaxy interactions are likely to play a key role in determining the morphology and the structural properties of galaxies and in driving evolution at all epochs and in widely - varying environments . the importance of environmental conditions on the properties of populations of galaxies has been recognized for a long time ( hubble & humason 1931 ) and observational studies have now firmly established relationships between the relative abundance of galaxies of different morphological types and the density of the environment ( see e.g. dressler 1980 , dressler et al . 1997 ) , their location in clusters ( oemler 1974 ) as well as the evolution of the morphology of galaxies in clusters at different epochs ( butcher & oemler 1978 , 1984 ) . in particular , the ratio of early to late - type galaxies has been shown to be an increasing function of density and a decreasing function of the distance from the center of a cluster . in addition , hst observations show that the relative number of spirals to elliptical and s0 galaxies tend to increase in distant clusters , pointing to a time evolution of the morphology of galaxies from spirals to ellipticals ( see e.g. dressler et al . 1994a , 1994b , couch et al . 1998 ) . altogether , these results suggest that environmental disturbances due to interactions with other galaxies and with a cluster tidal field can lead to dramatic changes in the structure of galaxies . simulations have explored strong encounters and mergers . our recent work suggests that relatively weak encounters can give rise to significant asymmetries as well . to study weaker galaxy interactions and quantify their importance , we compute the response of a spherical stellar system to the perturbation induced by another system during a fly - by , focusing our attention on the morphology and the persistence of the distortions produced during the encounter . the range of orbital parameters considered encompasses the typical environmental conditions for field galaxies , for galaxies in groups and in clusters . our perturbative solution for these distortions provides quantitative and qualitative predictions in low to moderate amplitude that are difficult to reach by simulation . this approach nicely reveals the underlying dynamics without being masked or swamped by n - body fluctuation noise . on the other hand , numerical feasibility restricts the complexity of possible geometries and the perturbative assumption limits its validity for large amplitudes . our investigation is relevant for galactic dark halos or for spheroidal galaxies over a range of environments . we explore the possibility that the halo distortions produced during a fly - by could in turn lead to a significant and observable distortion in an embedded stellar disk ( see e.g. weinberg 1995 , 1998b , murali & tremaine 1998 , murali 1998 ) . in particular the deformations produced in a dark halo during an encounter could be one of the culprits for distorted morphologies like those observed in lopsided and warped galactic disks ( see e.g. richter & sancisi 1994 , zaritsky & rix 1997 , rudnick & rix 1998 , swaters et al . 1998 , haynes et al . 1998 , kornreich , haynes & lovelace 1998 ) . as recently shown by weinberg ( 1995 , 1998b ) , this could be the case for the milky way in which the deformations induced in the dark halo by the magellanic clouds could be responsible for the observed warp in the galactic disk . in a recent investigation , reshetnikov & combes ( 1998 , 1999 ) have found that about 40% of a sample of 540 galaxies have warped planes and their analysis revealed that warped galaxies tend to be located in denser environments supporting a scenario in which interactions would play an important role in the formation of warps . distortions of spheroidal galaxies are directly observable . our results show that in low velocity dispersion and dense environments , where low - velocity close encounters are more likely to occur , even low - mass perturbers can lead to significant asymmetries in the primary galaxy . recent morphological studies of both local and very distant galaxies have yielded comparisons of the frequencies of systems with distorted morphologies in different environments and at different distances ( see e.g. zepf & whitmore 1993 , mendes de oliveira & hickson 1994 , abraham et al . 1996a,1996b , van den bergh et al . 1996 , naim , ratnatunga & griffiths 1997 , brinchmann et al . 1998 , marleau & simard 1998 , conselice & bershady 1998 , conselice & gallagher 1999 ) . on the theoretical side , the efforts have been largely concentrated on numerical @xmath2-body simulations aimed at investigating the fate and the evolution of merging galaxies ( see e.g. barnes & hernquist 1992 , barnes 1998 and references therein ) and the distortions and the morphological alterations produced in galaxies in clusters and groups by the ensemble of `` tidal shocks '' with other members ( moore et al . 1996 , 1998 ) . this mechanism can drive the evolution of spiral galaxies into spheroidal systems and it is a strong candidate explanation for the difference in the relative number of spiral to elliptical galaxies in clusters at different redshifts . the plan of the paper is the following . in 2 we outline the method adopted for this investigation while the mathematical details are described in the appendix . in 3 we show the results obtained for initial conditions typical of galactic dark halos and their dependence on the orbital parameters of the perturber . in 4 initial conditions relevant for spheroidal galaxies and a survey of orbital parameters of the perturber typical of different environments are considered . overall , our results quantify the relationship between the interaction rate for the environment and the probability of observing irregular morphologies , aiding interpretation of any trend in the fraction of asymmetric galaxies at different redshifts . this latter aspect is of particular interest as data on the morphology of distant galaxies are now available from hst observations in the hubble deep field and in the medium deep survey projects ( abraham et al . 1996a , 1996b ) . in order to facilitate the comparison of our theoretical results with observational data , we will quantify the distortions produced by means of the asymmetry parameter @xmath0 ( abraham et al . 1996a ) which has been determined for a number of local and distant galaxies . we will also introduce a generalized asymmetry parameter @xmath1 which will provide an important information on the radial structure of the asymmetry produced by the mechanism we have considered and which will allow to test the theory presented here . the main conclusions are summarized and discussed in 5 . this section outlines the perturbative solution of the collisionless boltzmann equation used to predict the dynamical response of galaxies to interactions . additional mathematical detail is presented in the appendix . the linear perturbation theory used here has been applied in the past both to investigate the collisionless stability of stellar disks ( kalnajs 1977 ) and of spherical stellar systems ( polyachenko & shukhman 1981 , palmer & papaloizou 1987 , bertin & pegoraro 1989 , saha 1991 , weinberg 1991 , bertin et al . 1994 ; see also palmer 1994 and references therein ) and to study the response of a galaxy to perturbations induced by satellite companions ( weinberg 1989 , 1995 , 1998b ) . this method is numerically intensive but it avoids the problems of n - body simulations due to the finite number of particles which can introduce spurious noise effects ( see weinberg 1998a ) and it guarantees the resolution of resonances that lead to the excitation of patterns in the primary system . with too few particles , for example , noise can cause orbital energies and angular momenta to drift on a time scale shorter than the pattern speed of the global response of interest . this eliminates the possibility of observing this structure in the simulation . no doubt , real galaxies are not smooth and inherent fluctuations may be important to their evolution . however , many millions of n - body particles are required to approach the estimates of natural amplitudes ( nelson & tremaine 1996 , weinberg 1998a ) . we compute the perturbation induced on a spherical stellar system by a point mass perturber with mass @xmath3 on a rectilinear trajectory with pericenter @xmath4 . the response of the primary system will be determined by the simultaneous solution of the linearized collisionless boltzmann and poisson s equations , + f_1 h_0- f_0 h_1=0 [ veq ] , ^2 _ 1=4g_1 , where the subscript @xmath5 denotes the equilibrium quantities , subscript @xmath6 denotes the first order perturbation of the distribution function @xmath7 , the hamiltonian @xmath8 , and the potential @xmath9 . the boltzmann equation has been written in terms of action - angle variables @xmath10 . the perturbed potential , @xmath11 , is the sum of the tidal potential due to the perturber , @xmath12 , and the gravitational potential of the response to the perturbation , @xmath13 , and @xmath14 . the coupled boltzmann - poisson equations lead to an integrodifferential equation for the potential . this can be solved by performing a fourier transform in the angle variables , a laplace transform in the time variable and expanding the perturbed quantities in terms of spherical harmonics and of a bi - orthogonal density - potential basis functions for the angular and the radial parts respectively . the integrodifferential equation is then reduced to an algebraic equation for the vector of coefficients of the expansion of the response potential and density , * a * , = * r*(*a*+*b * ) [ mateq ] . the matrix * r * contains the time dependence of the perturbation and the equilibrium properties of the perturbed system and * b * is the vector of the coefficients of the expansion of the external perturbing potential . it is clear from equation ( [ mateq ] ) that the self - consistent response of the perturbed system is determined both by the perturbation applied by the external perturber and by the reaction of the system to its own response to this perturbation . in general , an additional contribution to the total response comes from excitation of the discrete damped modes of the primary system besides the effects of the perturber and the self - gravity of the distorted primary . the differences in the mathematical structure of these two contributions are illustrated in the appendix . the transient response due to the discrete damped modes can have a strong effect on the amplitude of the response ( cf . 3 ) and it is particularly important when the modes are weakly damped with damping timescales longer than the fly - by timescale . in this case , the discrete damped modes produce long - lived features which completely determine the amplitude and the structure of the response long after the pericenter passage of the perturber . once the above algebraic equation has been solved , the calculation of response potential and density is straightforward @xmath15\end{aligned}\ ] ] @xmath16\end{aligned}\ ] ] where @xmath17 and @xmath18 are the bi - orthogonal density - potential basis function and @xmath19 are the spherical harmonics . features excited in the halo can give rise to visible distortions in an embedded disk even though dark halo perturbations are not directly observable . while several other processes could be responsible for the observed distorted morphologies of stellar disks ( see e.g. sellwood & merritt 1994 , levine & sparke 1998 , bertin & mark 1980 , nelson & tremaine 1995 ) , the role played by a distorted galactic dark halo in exciting peculiar features in disks has been advanced for the case of an interaction of a galaxy with a satellite companion by weinberg ( 1995,1998b ) and applied to the case of the milky way - magellanic cloud system . halo deformations leading to morphological and kinematical lopsidedness has been invoked by swaters et al . ( 1998 ) in an analysis of two kinematically lopsided spiral galaxies ( ddo 9 , ngc 4395 ) . the ubiquity of magellanic - class dwarf companions of normal spirals ( e.g. zaritsky & white 1994 and zaritsky et al . 1997 ) suggests that such interactions may be important sources of structure . unbound encounters can have similar effect as we will illustrate here . our standard dark halo model is an isotropic king model with a dimensionless central potential @xmath20 ( concentration , to core radius , @xmath21 @xmath22 ) a total mass @xmath23 and a total radius @xmath24 kpc . these values are appropriate for the galactic halo ( see e.g. kochanek 1996 ) and together with a standard exponential disk results in a fair fit to the observed rotation curve . given the primary model , the perturber s velocity @xmath25 and pericenter @xmath4 determine the structure of the response to the perturbation . the amplitude of the response scales with the mass of the perturber . table 1 summarizes the values of @xmath25 and @xmath4 explored here along with the corresponding values of the ratio of the pericenter to the half - mass radius of the primary system , @xmath26 , and the ratio of the characteristic frequency of the motion of the perturber to the circular frequency at the edge of the primary system , @xmath27 . all the cases considered here correspond to fly - bys in which part of the orbit passes through the halo ; external encounters are likely to produce weaker distortions and will be considered later in 3.3 . the values of @xmath25 considered are chosen to cover a typical range of relative encounter velocities in the field , groups and clusters . in order to clearly assess the importance of weakly damped modes , we consider the evolution of the response both with and without their effect for all the cases listed in table 1 . + + + @xmath25 ( km / s ) & @xmath4 ( kpc ) & @xmath26 & @xmath28 + 200&53.6&1.0&6.45 + 500&53.6&1.0&16.1 + 1000&53.6&1.0&32.3 + 200&107.2&2.0&3.23 + 500&107.2&2.0&8.1 + 1000&107.2&2.0&16.1 + for cases that include the weakly damped modes , we must first determine their natural frequencies . we used the method adopted in weinberg ( 1994 ) : the determinant of the dispersion matrix @xmath29 ( see appendix for its definition ) is evaluated on a grid of values of @xmath30 in the upper ( @xmath31 ) complex plane where the determinant is defined and then rational functions are used to perform the analytical continuation to the lower complex plane where the zeros of the determinant corresponding to damped modes are located . rational function fits of the elements of the dispersion matrix have been adopted also for the calculation of the residue of the inverse of the dispersion matrix necessary to include the effects of damped modes in the response coefficients ( see eq . the characteristic damping times for the most weakly damped modes , @xmath32 , are @xmath33 and @xmath34 for the dipole and the quadrupole mode respectively , where @xmath35 is the dynamical time of the primary system defined at the half - mass radius , @xmath36 . figure 1 shows the evolution of the energy in the perturbation , @xmath37 , as a function of @xmath38 ( fig.1a - b ) and as a function of the position angle of the perturber @xmath39 ( fig.1c - d ) without the effects of the damped modes . we normalize to the total background energy of the primary system @xmath40 ( @xmath41 for a king @xmath42 model ) . for a given value of the pericenter , the peak of the response occurs after the closest approach . the faster the perturber , the larger the value of the position angle and the more distant the perturber location when the response of the primary system reaches its peak . the strength of response increases as the velocity of the perturber decreases . in figure 2 we have plotted the maximum value of @xmath37 as a function of the velocity of the perturber for the two values of @xmath4 investigated . the response amplitude is large when the characteristic angular frequency , @xmath43 , is in low - order resonance with primary orbital frequencies . figure 3 and figure 4 show the same plots shown in figure 1 and figure 2 but with the effects of the discrete weakly damped modes taken into account . the maximum values of @xmath37 are about two orders of magnitude larger than those reached when the effects of damped modes are not included while the dependence of the peak of @xmath37 on the velocity of the perturber and on the pericenter of its orbit are similar to each other in the two cases . _ the inclusion of the damped modes is critical to the persistence and the strength of the perturbation_. a simple empirical function of the form @xmath44 can be used to obtain a satisfactory fit of the curves shown in figure 2 and figure 4 . these relations can be used to predict ensemble properties ( see 3.4 for an example ) . table 2 summarizes the values of the parameters @xmath45 and @xmath46 for the best fits for all the cases investigated and shown in figure 2 and figure 4 . + + + parameters&@xmath45 & @xmath46 + @xmath47 , no - damped&0.005&1.41 + @xmath48 , no - damped&0.002&1.33 + @xmath47 , with damped & 1.0&1.96 + @xmath48 , with damped & 0.5&1.9 + + + + the scalings obtained will not extend to the very low - velocity regime where they would lead to a divergence of the energy associated to the response . in the limit of very slow encounters we expect the energy of the response to increase more slowly and eventually to converge to a constant value ( see murali & tremaine 1998 for an investigation of the response of galactic halos to adiabatic perturbations ) . figure 5 illustrates the differences between the response with and without the damped modes by separately showing the energy in the dipole and quadrupole components for @xmath49 km / s and @xmath47 . as expected , the dipole response is significantly stronger than that of quadrupole . the inclusion of the damped modes leads to a much longer and stronger dipole response , while producing no significant difference in the quadrupole response . such long - lived features may explain the numerous cases of galaxies with distorted morphologies but without any apparent interacting system in their vicinities ( e.g. richter & sancisi 1994 ) . for example , for a system with mass and radius equal to those of our fiducial system ( @xmath50 , @xmath51 kpc ) , the damping time of the dipole mode ( @xmath52 yrs ) is longer the hubble time and any perturbation excited in a dark halo during a fly - by as well as its effect on an embedded disk will persist well after the encounter . contour plots in figures 6 and 7 describe the density perturbation in the orbital plane for the dipole ( upper panels ) and quadrupole ( lower panels ) responses without and with the damped modes , respectively . these cases have @xmath53 km / s and @xmath47 and show the response of the primary galaxy at three different points of the perturber trajectory . a lag between the orientation of the wake and the position angle of the perturber is evident for all velocities . this lag is a consequence of angular momentum transfer between the wake and the perturber and tends to zero only when there is no resonance affecting the response of the primary system . physically , resonant momentum exchange viewed at the orbit level has a corresponding view as a global response : the global response must lag the perturber in order to apply the torque responsible for the exchange . figures 8a and 8b show the ratio , @xmath54 , of the radial position of the peak of the density of the perturbation , @xmath55 , to the half - mass radius , @xmath56 , as a function of the position angle of the perturber without the effects of weakly damped modes for @xmath53 km / s and @xmath57 ( fig.8a ) and for @xmath47 and @xmath58 km / s ( fig.8b ) . the evolution of @xmath54 is only weakly dependent on the velocity of the perturber and on the pericenter of the orbit of the perturber . the perturbation is efficiently transmitted to the inner parts of the primary system and the peak of the density of the perturbation is located at @xmath59 or in terms of the core radius , @xmath60 , @xmath61 ( for almost the entire duration of the fly - by ) . the weak dependence of the response on perturber parameters is due to the dominant effect of the lowest - order point modes on the structure of the dispersion matrix elements . for an analogy , consider the similarity of bell s ring under a variety of strikes . the strong oscillations of @xmath54 in the final phases of the fly - by are due to numerical difficulty in determining the density peak as the response weakens . figures 9a and 9b show the evolution of the maximum value of the density of the perturbation @xmath62 ( normalized to the central density of the equilibrium model ) for the same cases shown in figures 8a and 8b . the maximum response obtains for slower fly - bys ( cf . fig . 1 ) and , for a given value of @xmath25 , @xmath62 increases as @xmath4 decreases . figures 8c and 8d and 9c and 9d show the same plots as in figures 8a - b and figures 9a - b for fly - bys including the weakly damped modes . the position of the peak of the density is similar to that obtained in the case without the weakly damped modes but there are evident differences in the time evolution of @xmath54 particularly at the beginning and at the end of the fly - by due to the persistence of the damped modes . the response is significantly stronger when the effects of the damped mode are included ( compare fig.9a - b with fig.9c - d ) figure 10 shows the energy in the perturbation for the first ten basis functions and indicates the distribution of energy with spatial scale . plots shown in figure 10 are for the same cases and phases shown in figure 6 and figure 7 . figure 10 confirms what is already qualitatively evident in figure 6 and figure 7 : the energy in the dipole response is preferentially deposited on large scales and the quadrupole response energy deposited on relatively smaller scales . the typical scale represented by each index @xmath63 can be seen from the plots of figure 11 which show the density basis functions for @xmath64 . external encounters , those with trajectories everywhere outside the primary , are more frequent in groups and clusters but less dramatic . for external encounters , the stronger dipole response corresponds to a simple shift of the center of mass of the system leaving the weaker quadrupole as dominant aspherical contribution . here we briefly describe the features of externally excited responses . table 3 summarizes the characteristics of the fly - bys we have investigated . an external encounter has no intrinsic length scale and the frequency parameter @xmath28 alone now determines the structure of the response . the table reports the velocity of each fly - by assuming the pericenter to be equal to twice the total radius of the primary system ( 400 kpc ) and the maximum values of the ratio of the energy in the perturbation to the total energy of the equilibrium model . this values scales as shown for other values . + @xmath28 & @xmath65 ( km / s ) & @xmath66 + 0.1&23&@xmath67 + 0.5&115&@xmath67 + 1.0&231&@xmath68 + 2.5&578&@xmath69 + 5.0&1150&@xmath70 + + + + one sees that the effects of external encounters with less massive perturbers are weak and they can not produce a significant deformation in the primary system . on the other hand , if the perturber is more massive than the perturbed system even external fly - bys can be very important in affecting the structure of a galaxy . for example , in order to have @xmath71 for an encounter with @xmath72 the ratio of the mass of the perturber to that of the perturbed must be @xmath73 . these encounters are relevant for spiral galaxies in clusters where they can undergo frequent external encounters with more massive giant ellipticals . a recent investigation by rubin et al . ( 1999 ) has shown that about half of a sample of 89 spirals in the virgo cluster exhibit signs of kinematic disturbances in their rotation curves ; according to that investigation disturbed galaxies are preferentially on radial orbits likely to cross the central regions of the clusters where such encounters are likely to occur . the five panels of figure 12 show the contour plots of the density of the response when @xmath74 reaches its maximum value . as described in 3.2 , the response reaches its maximum strength at different phases of the orbit of the perturber depending on the value of @xmath28 : the larger the value of @xmath28 , the more distant the perturber is from the pericenter of its orbit when the response is maximum . as the characteristic frequency of the fly - by decreases and the perturbation tends to the adiabatic regime , individual resonances cease to be relevant in determining the structure of the response and the orientation of the wake aligns with the position of the perturber . finally , we note that some features are present at small radii and these may cause visible distortions in an embedded disk . we have investigated the response of a galactic dark halo to the perturbation induced by external and internal fly - bys . external fly - bys are effective when the mass of the perturber is larger than that of the primary system considered . for example , perturbations excited in spirals during external fly - bys with more massive giant ellipticals can play an important role in altering their structure and kinematics . similarly , external encounters will produce significant structure in dwarfs in the neighborhood of normal spirals . we have shown that the perturbation excited during an internal fly - by encounter can be efficiently transmitted to the inner regions of the halo where it can affect the structure of an embedded stellar disk . the halo distortions can be significant and persistent for typical dwarf interlopers . this may lead to distorted morphologies like those observed in warped and lopsided disks . the survey over different velocities and pericenter distances provides a quantitative measure of the expected importance of the process investigated in different environments . for an overall indicator of the fraction of objects distorted as a result of fly - by encounters , we have calculated the mean value of the peak relative distortion for different environments , ( 0.1m / m)^2e_p / e_back = f(v ) v^2 dvf(v ) v^2 dv where @xmath75 is the distribution of the relative velocities of encounters which we take equal to a gaussian with dispersion @xmath76 . figure 13 shows @xmath77 as a function of @xmath76 for the four scaling laws listed in table 2 . generally , low velocity dispersion compact groups with number densities @xmath78 and velocity dispersions @xmath79 km / s are dominated by low - velocity close encounters which lead to the strongest response in the perturbed galaxies ; cluster of galaxies which are characterized by higher velocity dispersions ( @xmath80 km / s ) and lower number densities ( @xmath81 ) encounters are expected to produce less pronounced effects . the effects of damped modes on the strength and the persistence of the features excited by the perturber are significant . in particular the persistence of the deformations due to the long - lived dipole mode may explain the origin of the numerous lopsided disks in field galaxies which have no apparent companion ( see e.g. richter & sancisi 1994 , zaritsky & rix 1997 ) . we have separated the response due to damped modes in figure 13 to highlight their importance , however , true physical response includes both . damped modes may be artificially suppressed in n - body simulations with small particle number owing to fluctuation - induced orbital diffusion . in the next section , we will focus our attention on elliptical galaxies modeled as high - concentration king models for which the effects of discrete modes have not been considered because they have much shorter damping times ( weinberg 1993 ) and thus are not as relevant as they are for dark halos . + + + @xmath25 ( km / s ) & @xmath4 ( kpc ) & @xmath26 & @xmath28 + 200&7.0&0.5&17.9 + 500&7.0&0.5&44.7 + 1000&7.0&0.5&89.5 + 200&14.0&1.0&8.9 + 500&14.0&1.0&22.4 + 1000&14.0&1.0&44.7 + 200&42.0&3.0&3.0 + 500&42.0&3.0&7.5 + 1000&42.0&3.0&14.9 + 200&97.8&7.0&1.28 + 500&97.8&7.0&3.2 + 1000&97.8&7.0&6.4 + we will model the primary system by an isotropic king model with @xmath82 which has a concentration @xmath83 . we adopt typical mass , total radius and half - mass radius values for cluster elliptical galaxies : @xmath84 and @xmath85 kpc , @xmath86 kpc . table 4 summarizes the properties of our model grid ( see table 1 for parameter definitions ) . the overall approach and procedure otherwise follows that described in 3 with similar physical interpretation . 4.1 briefly presents the response diagnostics developed in 3 , highlighting the differences between responses in the low- and high - concentration primaries . 4.2 explores observational diagnostics of the fly - by induced asymmetries and proposes a generalization of one of the standard observational asymmetry parameters to test the dynamical mechanism explored here . figure 14 shows the time evolution of the perturbation energy . as expected ( cf . 3 ) , the innermost fly - bys induce the strongest response and , for a given value of the pericenter , the response becomes weaker as the velocity of the perturber increases . the time evolution of @xmath37 is characterized by several peaks and a non - negligible amplitude even as @xmath87 tends to @xmath88 and the perturber moves away from the primary galaxy . the persistence of these features well past the closest passage of the perturber may explain the origin of distorted morphology in galaxies without any close visible companion . figure 15 shows the evolution of @xmath54 . the response is manifested well inside the primary galaxy and it affects the very inner regions of the system @xmath89 ( @xmath90 ) even when the pericenter of the perturber is as large as @xmath91 or @xmath92 . the value of @xmath55 is determined by the dominant resonances determining the structure of the response of the primary system . the larger dynamic range in primary profile enables excitation of high - order resonances between the characteristic frequency of the motion of the perturber and the frequencies of the orbital motion at smaller radii in the primary system . the rectilinear trajectory considered here provides continuum spectrum of frequencies that peaks at approximately @xmath43 . we found that @xmath93 is roughly constant for all the fly - bys investigated . this agrees with the results obtained by weinberg ( 1994 ) in his analysis of the structure of weakly damped modes in isotropic king models . this implies that more concentrated models are more efficient in transmitting to the inner regions the perturbation excited during a fly - by ( compare fig.8 with fig.15 ) . figure 16 shows the density distortions for a fly - by with @xmath53 and @xmath47 at three different phases of the orbit of the perturber . as in previous cases , the maximum response occurs well inside the primary system and the perturbation is efficiently transmitted to the inner regions of the system far inside the orbital pericenter . figure 17 shows plots of @xmath94 for the same case and phases shown in figure 16 describing the distribution of energy at different scales . similar to the results in 3 , the dipole response is dominated by large scale features while in the quadrupole response there is a non negligible contribution to the total energy of the perturbation from features at smaller scales . we employ an established asymmetry diagnostic to compare our predictions to observations and suggest a straightforward generalization which will help discriminate the response mechanism considered here from others . we will focus our attention on an asymmetry parameter @xmath0 for image data , defined as a [ asym ] where @xmath95 and @xmath96 are the brightness at a given point @xmath97 of the original image and in the image after a 180 degree rotation about the position of the maximum density of the primary system , respectively . for our calculation we use the density of the system on the orbital plane and thus our theoretical values are not affected by any projection effect and will provide a measure of the real asymmetry of the plane . observational values , on the other hand , are determined using the projected density and part of the underlying asymmetry of the orbital plane is likely to be washed out . in order to evaluate the effect of using the projected density on the orbital plane instead of the real plane density , we have repeated some of the calculations : we have found that , for a given value of @xmath98 , the values of @xmath0 obtained using the projected density are smaller than those obtained using the real plane density by a constant factor independent of the perturber orbital parameters and approximately equal to 3.8 ; this implies that , using the projected density , all the values of @xmath0 ( as well as those of the parameter @xmath99 introduced later in this section ) and the numerical parameters in the scaling laws of the amplitude of @xmath0 with the perturber s orbital parameters discussed in this section are not altered but they are produced by a perturber with a mass equal to @xmath100 ( instead of @xmath101 adopted in the rest of this section ) . the @xmath0 parameter was first introduced by abraham et al . ( 1996a ) to estimate the asymmetry observed in galaxies in the medium deep survey and in the hubble deep field and it has been recently adopted in several investigations on the morphology of galaxies in different environments ( e.g. van den bergh et al . 1996 , naim , ratnatunga & griffiths 1997 , marleau & simard 1998 , conselice & bershady 1998 , conselice & gallagher 1999 ) . the values of @xmath0 reported in these investigations range from @xmath5 for the most symmetric systems to @xmath102 for those showing the strongest asymmetries . a comparison of the distribution of values of @xmath0 for a sample of local galaxies , galaxies in the medium deep survey and in the hubble deep field was reported by abraham et al . ( 1996a , 1996b ) . their results show that the distribution of @xmath0 for galaxies in the hdf is significantly skewed toward larger values of @xmath0 compared to the distribution for galaxies in the mds which in turn is skewed to larger values of @xmath0 relative to the distribution of @xmath0 for a sample of local galaxies . this trend is likely to be due to the higher interaction and merging rates for distant galaxies in the hdf and the mds . in principle , a high value of @xmath0 could also result from an asymmetric distribution of star forming regions , however a recent investigation by conselice & bershady ( 1998 ) shows that the colors of many asymmetric objects in the hdf are not blue enough to support this hypothesis . in addition , we have calculated the displacement , @xmath99 , between the position of the peak of the density of the primary system and the position of its center of mass . the estimate of @xmath99 is of particular interest since many observations have provided evidence of off - center galactic nuclei and of significant displacement of isophotal centers ( see e.g. miller & smith 1992 for a short overview of several observational analyses ) . in particular , naim et al . ( 1997 ) have adopted distances between the centers of different isophotal curves as one of the diagnostic parameters characterizing the degree of peculiarity of galaxies in a sample of moderate - redshift systems and mendes de oliveira & hickson ( 1994 ) have investigated the morphology of early - type galaxies in compact groups and used the evidence of non - concentric isophotes as a diagnostic to identify perturbed galaxies . figure 18 shows the time evolution of the asymmetry parameter @xmath0 defined above ( eq . [ asym ] ) . although several well - documented processes contribute to the observed distorted morphologies , the majority of the values in the distribution of @xmath0 obtained for galaxies in the mds and in the hdf by abraham et al . ( 1996 ) ( see their fig . 2 ) fall well in the range of values of @xmath0 we have obtained in our analysis . mergers and strong interactions with more massive systems are likely to be responsible for the production of the extreme values of @xmath0 ( @xmath103 ) observed in some galaxies . we point out that the observational data refer to all the morphological types including spiral galaxies to which our results do not directly apply . figure 19 shows the time evolution of @xmath99 . + + + id.&@xmath104 & @xmath105&@xmath106 & @xmath107 + @xmath108&0.24&0.6&@xmath109&0.6 + @xmath47&0.19&0.5&@xmath110&0.5 + @xmath111&0.11&0.4&@xmath112&0.4 + @xmath113&0.03&0.1&@xmath114&0.0 + + + + as expected , the trend with @xmath25 and @xmath4 is similar to that obtained for the asymmetry parameter @xmath0 , with slow and close encounters being those capable of producing the largest values of @xmath99 . even fly - bys with large values of @xmath4 are capable of affecting the inner regions of the primary galaxies and lead to a significant displacement of the position of the density peak from the center of mass of the system . the dependence of both asymmetry parameters on @xmath25 and @xmath4 is summarized in figure 20 which shows the maximum values taken by @xmath0 and @xmath99 as a function of @xmath26 for the three values of @xmath25 considered . we have fit the scaling of @xmath115 and @xmath116 with the relative velocity of the encounter for different values of @xmath4 with a power - law @xmath117 and we report the values obtained for the cases investigated in table 5 . these fits are not valid in the limit of adiabatic encounters where the strength of the perturbation and of the diagnostic parameters tend to a constant value as described previously ( see 3.1 ) . fly - by interactions described here produce structure on well defined scales ( cf . fig . 16 ) , however , the sum in equation ( [ asym ] ) washes out the signature of our predicted response by averaging over regions with little distortion . to improve sensitivity , we propose an _ incomplete _ form of the asymmetry parameter @xmath1 which restricts the sum in equation ( [ asym ] ) to pixels enclosed by a circle of radius @xmath118 from the position of the peak of density @xmath55 . figure 21 shows the plot of @xmath119 ( calculated for @xmath120 ) for the case with @xmath49 km / s and @xmath108 . the appearance of @xmath119 is qualitatively similar for the other cases investigated . the profile of @xmath0 at any given time has a maximum at @xmath121 with a subsequent decrease which is steeper near the absolute maximum of @xmath119 and shallower far from this maximum . the value of @xmath0 always tends to a constant for large values of @xmath118 . in figure 22a we compare the radial profile of @xmath0 for the three fly - bys with @xmath47 and @xmath122 km / s . the three curves show the run of @xmath1 at peak and each curve is normalized to the maximum value of @xmath1 , @xmath123 . the radial location of @xmath123 and the shape of @xmath1 do not depend significantly on the velocity of the perturber in these half - mass units . figure 22b compares @xmath124 for all the fly - bys with @xmath49 km / s and with the different values of @xmath26 investigated . the location of the peak of @xmath124 depends very slightly on @xmath26 while there is a marked difference between the slope of @xmath124 after the peak for the fly - bys with small pericenters ( @xmath125 ) and those with larger values of @xmath26 @xmath126 . a distinct peak in the radial profile of @xmath0 is present only at some phases of the encounter and this could be difficult to capture in observations . on the other hand the decrease at small values of @xmath118 is a feature common to all the phases of the orbit of the perturber and it should be an observable signature . in this section , we have studied the response of a high - concentration spherical stellar system to the perturbation induced by a fly - by . we represent the outer profiles of cluster ellipticals by a high - concentration king model ; we choose a king profile rather than a core - free model for numerical convenience . a deformation in the structure of the primary galaxy is directly observable and we explored quantitative measures of the asymmetry produced during the fly - by encounter . in the end , we focussed our attention on the asymmetry parameter @xmath0 , introduced by abraham et al . ( 1996a , 1996b ) and measured for a number of local galaxies and in galaxies in the mds and in the hdf fields , and on the shift @xmath99 , between the position of the peak of the density of the system and its center of mass . we present the time evolution , the scaling with the perturber orbit and the radial dependence of the asymmetry to facilitate estimates for a wide variety of environments . the values of @xmath0 caused by fly - bys range from @xmath127 to @xmath128 and this same range of values of @xmath0 is found in observations of numerous local galaxies and distant galaxies in the mds and in the hdf fields ( abraham et al . 1996a , 1996b ) . extreme values of @xmath0 observed for some galaxies in the hdf field ( @xmath129 ) are probably caused by mergers . the quantities @xmath0 and @xmath99 reach their maximum values when the perturber is close to the pericenter of its orbit . as evident in figure 18 , the distortions may be long - lived ; in some cases the perturber can be quite distant from the primary system and therefore we can expect some non - negligible values of @xmath0 and @xmath99 even without any close companion in the vicinity of the primary . our results show that the values of @xmath0 and @xmath99 will be determined by both the distribution of relative encounter velocities and the density of the environment . in short , richer groups and clusters with lower velocity dispersions will lead to high values of @xmath0 and @xmath99 . figure 18 shows that disturbances persist until @xmath130 or roughly @xmath131 years after the pericenter passage . in order to provide a quantitative indication of the dependence of the response on the environmental conditions , we have calculated the probability that a galaxy has an encounter in the time interval @xmath132 such that @xmath133 assuming that the perturber has a mass @xmath134 ( or @xmath135 if the projected density is used for the calculation of @xmath0 ) . for our fiducial system @xmath136 yrs and @xmath137 , the probability @xmath138 of having such an encounter is given by p = np^2(v)vwhere @xmath63 is the galaxy number density of the environment and @xmath139 is the pericenter distance for which @xmath140 during an encounter with velocity @xmath25 . the function @xmath139 has been determined using the parametrization reported in table 5 for the values of @xmath25 available from our numerical investigation and then fitting these points with a function of the form @xmath141 . we obtain p(n , v)=10 ^ -3n(^-3 ) ( v(^-1)/200)^1.44 and describe this function in figure 23 . this quantifies our finding that galaxies in environments characterized by high density and low - velocity dispersion , like compact groups , will be more likely to show signs of past or recent interactions in agreement with the results of a number of investigations ( see e.g. zepf & whitmore 1993 , mendes de oliveira & hickson 1994 ) . these observational studies argue that the fraction of distorted morphologies in compact groups is larger than the fraction observed in clusters ( which have similar number densities but higher velocity dispersions ) and in sample of field galaxies ( which are likely to suffer low - velocity encounters but in a low - density environment ) . assuming that young galaxies were located in dense environments and characterized by lower , previrialized velocity dispersions , our conclusions are consistent with the results of abraham et al . ( 1996b ) who have shown that the distribution of @xmath0 for galaxies in the hdf is skewed toward high values compared to that for galaxies in the mds which , in turn , are more asymmetric than local field galaxies . an additional useful diagnostic is the ratio of the perturber to the primary masses necessary to produce a given value of @xmath142 . this is easily obtained from the fits discussed in 4.3 and is given by ( mm)_a=0.1 k_a a ( v ( ) 200)^_a , where the values of @xmath104 and @xmath105 depend on @xmath26 ( cf . table 5 ) . the four frames in figure 24 show the contour plots of @xmath143 in the plane @xmath144 ( as we discussed at the beginning of this section , if the projected density is used for the calculation of @xmath0 the values of @xmath98 are larger than those reported in fig . 24 by a factor 3.8 ) . any environment producing low - velocity , close encounters can lead to significant asymmetries ( @xmath145 ) in the primary system even by low - mass perturbers ( @xmath146 ) . for example , consider a system with a mass and a radius equal to those of our fiducial system ( @xmath147 , @xmath148 kpc ) . an encounter with a perturber with @xmath149 with relative velocity @xmath53 km / s leads to @xmath140 if @xmath47 ( @xmath150 kpc for our fiducial system ) . low - mass interlopers with pericenters close to the primary center could lead to significant asymmetries in cluster ellipticals as well . fly - by encounters yield the following signature in our radially - dependent asymmetry parameter , @xmath1 : 1 ) a decrease at small radii , 2 ) a peak around @xmath151 and 3 ) a convergence to a constant value at large radii . although we have not predicted the signature of @xmath1 for scenarios other than fly - bys , we believe an observational investigation would provide a test of the mechanism proposed here and would be a general probe of asymmetric structure . the computation of @xmath1 is relatively straightforward with the data already available . we have explored the structure and the persistence of the features in dark halos and cluster ellipticals excited by an unbound ( or _ fly - by _ ) interaction over a range of different galactic environments . the main conclusions of our analysis are as follows : 1 . fly - by encounters which penetrate the outer regions of a galaxy can lead to asymmetric features well inside the half - mass radius . the distortion of a dark halo may affect its disk ( e.g. weinberg 1998b ) . 2 . in dark halos the lowest - order damped modes yield strong features which persist well after the perturber s passage and could account for distorted morphologies in galaxies without obvious close companions . such damped modes are a natural part of the collisionless response but are difficult to resolve in n - body simulations with modest particle number ( @xmath152 ) due to noise . the strength of the response depends both on the velocity of the perturber and on the pericenter of its orbit ; low - velocity , close encounters produce the strongest perturbations . 4 . completely external fly - by encounters will produce a strong response if the perturber is more massive than the primary . for example such encounters can produce significant effects on spiral galaxies in clusters where they are likely to interact with more massive giant ellipticals and on dwarfs in the vicinity of normal galaxies . a proposed generalization of the asymmetry parameter @xmath0 ( abraham et al . 1996a ) , provides a diagnostic of the radial structure of the asymmetry and a test of the mechanism explored here in particular . our predicted range of @xmath0 is similar to that observed for the majority of galaxies in the medium deep survey , in the hubble deep field and for local galaxies . we quantify the range of high - density , low - velocity dispersion environments expected to produce significant distortions by fly - by encounters . indeed , compact groups have been shown to have a significant fraction of distorted galaxies . additional discussion is provided in 3.4 for dark halos and in 4.5 for cluster ellipticals . future work will explore the properties of the response for cuspy galaxies ; the magnitude of the resonant coupling to the inner galaxy in core - free profiles would nicely complement the present work . we plan to further investigate the effects of the halo distortions on an embedded stellar and gaseous disk , by extending our current study to explore the response of the disk and to quantify the induced deformations . this will permit direct comparisons with the observed distortions in the morphology of disks as well as to study possible effects of these distortions on their kinematics . in particular , the effects of even small deformations in the morphologies of disks could significantly affect their kinematics possibly giving rise to the observed scatter in the tully - fisher relation as suggested by franx & de zeeuw ( 1992 ) . we thank eric linder for comments on the manuscript . this work described here was supported in part by nsf ast-9529328 . abraham r.g.,van den bergh s. , glazebrook k. , ellis r.s . , santiago b.x . , surma p. , & griffiths r.e . 1996a , apjs , 107,1 + abraham r.g . , tanvir n.r . , santiago b.x . , ellis r.s . , glazebrook k. , & van den bergh s. 1996b , mnras , 279 , l47 + barnes , j. 1998 , in galaxies : interactions and induced star formation : saas - fee advanced course 26 lecture notes 1996 swiss society for astrophysics and astronomy ) + barnes , j. , & hernquist , l. 1992 , ara&a , 30 , 705 + bertin , g. , & mark , j.w.k . , 1980 , a&a , 1980 , 88 , 289 + bertin g. , & pegoraro f. , 1989 , in esa , proceedings of international school and workshop on plasma astrophysics , volume 1 p.329 + bertin g. , pegoraro f. , rubini f. , & vesperini , e. 1994 , apj , 434 , 94 + brinchmann j. et al . 1998 , apj , 499 , 112 + butcher h , & oemler , a. 1978 , apj , 219 , 18 + butcher h , & oemler , a. 1984 , apj , 285 , 426 + couch w.j . , barger a.j . smail i. , ellis r.s . , & sharples r.m . 1998 , apj , 497 , 188 + conselice c.j . , & gallagher iii , j.s . 1999 , aj , 117 , 75 + conselice c.j . , & bershady m. 1998 , preprint , astro - ph 9812299 + dressler a. 1980 , apj , 236 , 351 + dressler a. , oemler a. , butcher h. , & gunn j.e . 1994a , apj , 430 , 107 + dressler a. , oemler a. , sparks w.b . , & lucas r.a . 1994b , apj , 435 , l23 + dressler et al . 1997 , apj , 490 , 577 + edmonds a.r . 1960 , angular momentum in quantum mechanics , princeton university press , princeton , new jersey + franx m. , & de zeeuw t. 1992 , apj , 392 , l47 + haynes m.p . , hogg d.e . , maddalena r.j . , roberts m.s . , & van zee l. 1998 , aj , 115 , 62 + hubble e. , & humason m.l . 1931 , apj , 74 , 43 + kalnajs a.j . 1977 , aj , 212 , 637 + kochanek c.s . 1996 , apj , 457 , 228 + kornreich d.a . , haynes m.p , & lovelace r.v . 1998 , aj , 116 , 2154 + levine s.e . , & sparke l.s . 1998 , apj , 496 , l13 + marleau f.r . , & simard l. 1998 , apj , 507 , 585 + mendes de oliveira c. , & hickson p. 1994 , apj , 427 , 684 + miller r.h . , & smith b.f . 1992 , apj , 393 , 508 + moore b. , katz n. , lake g. , dressler a. , oemler a. , 1996 , nature , 379 , 613 + moore b. , lake g. , & katz n. 1998 , apj , 495 , 139 + murali , c. 1998 , preprint , astro - ph 9811223 + murali , c. , & tremaine s. 1998 , mnras , 296 , 749 + naim a. , ratnatunga u. , & griffiths r.e . , apj , 476 , 510 + nelson , r.w . , & tremaine s. , 1995 , mnras , 275 , 897 + nelson , r.w . , & tremaine s. , 1996 , in lahav o.,terlevich e. , terlevich r. , eds , proceedings of the 36th herstmonceux conference , gravitational dynamics , cambridge univ . press , cambridge , p.73 + oemler a. 1974 , apj , 194 , 1 + palmer p.l . , & papaloizou j. 1987 , mnras , 224 , 1043 + palmer p.l . 1994 , stability of collisionless stellar systems , kluwer academic publishers , dordrecht + polyachenko v.l . , & shukhman i. 1981 , sov . , 25,533 + reshetnikov v. , & combes f. 1998 , a&a , 337 , 9 + reshetnikov v. , & combes f. 1999 , preprint , astro - ph9906063 + richter o. , & sancisi r. 1994 , a&a , 290 , l9 + rubin v. , waterman a.h . , & kenney j.d.p . , preprint , astro - ph 9904050 + rudnick g. , & rix h.w . 1998 , aj , 116 , 1163 + saha p. 1991 , mnras , 248 , 494 + sellwood j. , & merritt d. 1994 , apj , 425 , 530 + swaters r.a . , schoenmakers r.h.m . , sancisi r. , & van albada t.s . 1998 , preprint , astro - ph 9811424 + tremaine s. , & weinberg m.d . 1984 , mnras , 209 , 729 + van den bergh s. , abraham r.g . , ellis r.s . , tanvir n.r . , santiago b.x . , & glazebrook k. 1996 , aj , 112 , 359 + weinberg , m.d . 1989 , mnras , 239 , 549 + weinberg , m.d . 1994 , apj , 421 , 481 + weinberg , m.d . 1995 , apj , 455 , l31 + weinberg , m.d . 1998a , mnras , 297 , 101 + weinberg , m.d . 1998b , mnras , 299 , 499 + weinberg , m.d . 1999 , aj , 117 , 629 + zaritsky d. , & white s.d.m . 1994 , 435 , 599 + zaritsky d. , & rix h.w . 1997 , apj , 477 , 118 + zaritsky d. , smith r. , frenk c. , & white s.d.m . 1997 , 478 , 39 + zepf s.e . , & whitmore b.c . 1993 , apj , 418,72 + our results are based on an explicit solution for the response of a spherical stellar system to the perturbation induced by a perturber with mass @xmath3 on rectilinear trajectory with pericenter @xmath4 . specifically , we solve the coupled linearized collisionless boltzmann equation for the distribution function of the primary stellar system and poisson s equation ( cf . 2 ) @xmath154 where the subscript @xmath5 denotes the equilibrium quantities and the subscript @xmath6 denotes the first order perturbation of the distribution function @xmath7 , and hamiltonian @xmath8 . the perturbed potential , @xmath11 , is the sum of the tidal potential due to the perturber , @xmath155 , and the gravitational potential of the response to the perturbation , @xmath13 . after performing the laplace transform in time , the fourier transform in the angle variables and solving for @xmath156 , equation ( [ veq ] ) becomes _ 1=i f_0_p,+_resp , s+i , [ lftr ] where we denote the laplace transform of a quantity @xmath157 by @xmath158 and the fourier transform of a quantity @xmath159 by @xmath160 with @xmath161 indicating the vector of integer indices for the discrete fourier expansion in the angle variables . the frequencies associated to the angle variables are denoted by @xmath162 . following the approach adopted in weinberg ( 1989 ) , we will use a potential - density biorthogonal basis , @xmath163 to expand the perturbed quantities @xmath11 and @xmath164 . the pairs @xmath165 , @xmath166 are solutions of the poisson equation . the basis we have adopted is obtained by numerical solution of the associated sturm - liouville problem in which the lowest order basis functions are tailored to the background equilibrium model following weinberg ( 1999 ) . the solution to the coupled equation , which determines the response of the halo to the perturbation , then reduces to the solution of an algebraic integral equation for the coefficients of the expansion . the subsections below sketch the main steps in the derivation of the algebraic equation for the coefficients of the expansion from equation ( [ lftr ] ) ( see e.g. weinberg 1989 for additional details ) . [ [ expansion - of - the - perturbation - potential - in - action - angle - variables ] ] expansion of the perturbation potential in action - angle variables ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ we begin by expanding the perturbing potential , @xmath155 , in terms of spherical harmonics @xmath19 ( where @xmath167 and @xmath168 denotes the angular coordinates inside the primary stellar system ) . up to quadrupole order , the potential of a point mass perturber can be written as @xmath169 \label{eq3}\end{aligned}\ ] ] where @xmath170 is the position angle of the perturber defined as the angle between the axis from the center of the system to the pericentric position of the perturber and the axis from the center to the position at time @xmath171 . we define the usual quantities @xmath172 , @xmath173 where @xmath118 and @xmath174 are the radial coordinate inside the stellar system and the distance between the center of the system and the perturber , respectively . following tremaine & weinberg ( 1984 ) , the radial part of the perturbing potential is expanded in terms of the basis function @xmath175 and each resulting term is expressed in the form @xmath176 as a fourier series in the angle variables @xmath177 . we finally obtain : where x_l_2_3^_1=12_-^dw_1u__3^l(r)e^-i[_1w_1+_2(w_2- ) ] and v_l_2_3()=r^l__2_3()y_l_3(/2,0)i^_3-_2 . the angle @xmath179 is the inclination of the orbital plane to the equatorial plane and @xmath180 are the components of the rotation matrices ( see e.g. edmonds 1960 ) . similarly , we can express the perturbing and the response potential in the form = _ l , v_l_2_3()_j a^j_l_3(t)x_l_2_3^_1,je^i*w*. for the perturbing potential , each coefficient @xmath181 can be expressed as the product of a time - independent coefficient , @xmath182 , and a function of time depending only on @xmath183 and @xmath184 , @xmath185 . the functional form of @xmath185 can be easily derived for any value of @xmath183 and @xmath184 using equation ( [ eq3 ] ) . for example , for @xmath186 , @xmath187 and @xmath188 , one finds @xmath189 equation ( [ lftr ] ) now takes the form @xmath190 . \label{lftr1}\end{aligned}\ ] ] for self - consistency , the response density , which is obtained by integrating equation ( [ lftr1 ] ) over the velocity coordinates , @xmath191 , must be equal to the density @xmath192 determined by the poisson s equation with @xmath193 . imposing the condition ^p=^resp , [ cons ] multiplying both sides of equation([cons ] ) by @xmath194 and integrating over the spatial coordinates , we obtain the following equation for the coefficients ^i_l_3(s)=_jm^ij_l_3(s)[^j_l_3(s)+b^j_l_3(s)][alapl ] . equivalently , this may be written as a matrix equation : _ l_3(s)=*d*_l_3 ^ -1(s)*m*_l_3(s)[*b*_l_3(s)][alapl1 ] where the elements of the matrix @xmath195 are defined as m^ij_l_3(s)__1,_2v^*_l-_2-_3v_l_2_3x_l-_2-_3^_1,i * x_l_2_3^_1,j and _ l_3-*m*_l_3 . equations ( [ alapl ] ) or ( [ alapl1 ] ) completely determines the response to the perturbation induced by the perturber . for @xmath196 ( no external perturbation ) , one obtains the following nonlinear eigenvalue problem ^i_l_3(s)=_jm^ij_l_3(s)[^j_l_3(s ) ] . [ al ] the eigenvalues of this problem are the frequencies of the point modes of the stellar system : frequencies with @xmath197 correspond to unstable growing modes while for stable damped modes @xmath198 . solution of equation ( [ alapl ] ) describes the evolution in frequency space . in order to get the time evolution of the perturbation we need to perform the inverse laplace transform of equation ( [ alapl ] ) this yields the following expression for the coefficients @xmath199 @xmath200b^j_{l\g_3 } , \label{tco}\end{aligned}\ ] ] where , to simplify the notation , we have defined ^kj v^*_l-_2-_3v_l_2_3x_l-_2-_3^_1,i * x_l_2_3^_1,j . equation ( [ tco ] ) does not take into account the zeros of det @xmath29 which are the point modes of the primary system . since we will consider only systems known to be stable , there are no modes with @xmath197 , while modes with @xmath198 will be damped and die away with time . nevertheless , as shown in weinberg ( 1994 ) , for king models some modes can be very weakly damped and persist long after their excitation . when the effects of damped modes are included , equation ( [ tco ] ) is valid for @xmath201 while for finite time it has an additional term for each such mode . altogether we find @xmath202b^j_{l\g_3}. \label{dm}\end{aligned}\ ] ] where @xmath203 denote the frequencies of the damped modes . numerical evaluation of equation ( [ dm ] ) requires truncating the infinite sum over @xmath204 ( the sums over the indices @xmath205 and @xmath184 range from @xmath206 to @xmath183 ; see tremaine & weinberg 1984 ) . after checking the convergence of the solution , the final value adopted is @xmath207 for the low - concentration king model studied in 3 and @xmath208 for the high - concentration king model considered in 4 . similarly , the sum over the radial basis must be truncated . since the the adopted biorthogonal functions are tailored to the equilibrium model @xmath209 is sufficient for the convergence . all the integrals have been calculated by romberg s method with the number of points chosen so to guarantee a maximum error of @xmath210 .

using the ads / cft correspondence , we calculate the transport coefficients of a strongly interacting system with a non - abelian su(2 ) global symmetry near a second order phase transition . from the behavior of the poles in the green s functions near the phase transition , we determine analytically the speed of second sound , the conductivity , and diffusion constants . we discuss similarities and differences between this and other systems with vector order parameters such as-wave superconductors and liquid helium- . pupt-2289 + christopher p. herzog and silviu s. pufu department of physics , princeton university + princeton , nj 08544 , usa [email protected] , [email protected] +

0902.0409

a new and interesting holographic perspective on the physics of superfluids and superconductors was provided by refs . these papers , which rely on the ads / cft correspondence @xcite , provide a dual description of the superconducting phase transition as the instability of a charged black hole to develop scalar hair . recalling that the ads / cft correspondence maps a strongly interacting field theory to a classical gravity description , this new perspective holds promise for deepening our understanding of superconductivity in strongly interacting regimes where bcs theory @xcite is inadequate . this paper takes as its starting point the holographic model proposed in refs . @xcite . while in refs . @xcite the black hole develops scalar hair at the phase transition , in refs . @xcite the black hole develops non - abelian hair . more precisely , refs . @xcite begin with gravity plus an abelian gauge field and charged scalar , while refs . @xcite omit the scalar and promote the abelian field to a non - abelian su(2 ) gauge field . recall that the ads / cft dictionary maps gauge fields on the gravity side of the duality to global symmetries in the field theory we are intentionally vague about the distinction between a superfluid and superconductor . as emphasized in ref . @xcite , technically , the field theory dual to the black hole construction undergoes a superfluid phase transition , i.e.spontaneous symmetry breaking of a global symmetry . to interpret this transition as a superconducting phase transition , the global symmetry must be weakly gauged . for many questions about superconductivity , the distinction is irrelevant , and from the current - current two point functions we calculate below , we can extract meaningful conductivities . this su(2 ) system has many desirable features compared to the scalar system . the scalar system appears to be less universal ; one must specify a potential for the scalar , or at the very least a mass term . in comparison , up to the strength of the gravitational and gauge couplings , the form of the lagrangian for the su(2 ) system is completely specified by gauge invariance . also , it appears more straightforward to embed the su(2 ) system in string theory or other uv completions . ( see however ref . @xcite for recent progress with the scalar system . ) this embedding can move the su(2 ) system out of the toy model realm and give us a microscopic lagrangian for the field theory @xcite discuss one possible embedding where the su(2 ) is a global flavor symmetry group for @xmath2 super yang - mills broken to @xmath3 by the addition of two hypermultiplets . super yang mills and string theory in the curved background @xmath4 . this string theory can be approximated at low energies by a gauged supergravity in @xmath5 . the su(4 ) gauge field , which maps to the su(4 ) global r - symmetry in the field theory , has an su(2 ) subgroup . unfortunately , supersymmetry constrains the relationship between the gauge field coupling and the gravitational coupling in this model , and the gauge field coupling is too weak for a superconducting phase transition to occur @xcite . ] our reason for choosing this su(2 ) system is very simple : we can get analytic results near the phase transition . the differential equations that describe these holographic systems are nonlinear , and analytic solutions do not appear to be available in most cases . the studies mentioned above make extensive use of numerics to see the phase transition , to calculate the conductivities and critical exponents . analytic results , for example the low temperature approximation of the conductivity in ref . @xcite , are scarce . our starting point is the remarkable observation in ref . @xcite that the zero mode for the phase transition for an su(2 ) gauge field in @xmath5 has a simple analytic form . from this zero mode , we are able to extract a long list of properties near the phase transition : 1 . the speed of second sound near the phase transition . 2 . that the phase transition is second order . 3 . the conductivity and in particular the residue of the pole in the imaginary part of the conductivity . 4 . the system satisfies a london type equation that implies a meissner effect . 5 . a large selection of current - current green s functions in the hydrodynamic limit , and that they satisfy the appropriate non - abelian ward identities . to clarify the title of the paper , recall that in a two component fluid , there are typically two propagating collective modes . the first mode corresponds to ordinary sound in which the two components move in phase . the second mode corresponds to second sound in which the two components move out of phase . typically , ordinary sound can be produced by pressure oscillations while second sound couples much more strongly to temperature oscillations @xcite . the order parameter for the phase transition in our su(2 ) model is the set of non - abelian global su(2 ) currents , @xmath6 . as pioneered in ref . @xcite , we introduce by hand a chemical potential in the third isospin direction which induces a charge density , @xmath7 , that breaks both the global su(2 ) symmetry to a u(1 ) sugroup and also lorentz invariance . there is a superconducting phase transition at a critical temperature @xmath8 , below which a current develops orthogonal to the third isospin direction that completely breaks the residual u(1 ) symmetry and also breaks the remaining rotational symmetry of the system to u(1 ) . for convenience , we take this current to be in the direction @xmath9 , leaving a rotational symmetry in the @xmath10-plane . the fact that rotational symmetry is broken in the superconducting phase makes the physics of this model rich and complicated . our model appears to be a holographic realization of the type of scenario described from a formal perturbative field theoretic point of view in ref . transport coefficients such as the speed of second sound and conductivities will depend on which direction we decide to look . such a breaking of rotational invariance is not unheard of in real world materials . to pick a particularly simple example , a ferromagnet will break rotational symmetry when the spins align . ref . @xcite emphasized a possible connection of this su(2 ) model to a @xmath0-wave superconductor , where the order parameter for the phase transition is a vector . of the real world materials that we considered , superfluid liquid helium-@xmath1 perhaps comes closest in approximating the physics of our model . liquid helium-@xmath1 at very low temperatures is a @xmath0-wave superfluid . two fermionic helium-@xmath1 atoms pair up to form a loosely bound bosonic molecule with weak interaction between the orbital and spin degrees of freedom of the electrons @xcite . the orbital and spin angular momenta are both equal to one , and the order parameter is often written @xmath11 where @xmath12 indexes the spin angular momentum and @xmath13 the orbital angular momentum , in surprisingly close analogy with our @xmath14 . there are many stable phases of superfluid helium-@xmath1 , depending on the pressure , temperature , and applied magnetic field . the a phases are known to break rotational symmetry . despite plausible similarities between the symmetries of our model and various real world materials , there is one crucial difference . while the order parameters for these real world materials may have vector or tensor structure , they are not currents , and the signature of the phase transition is not the production of a persistent current . in contrast , our model has @xmath15 . we begin in section 2 with a discussion of the su(2 ) model and the probe limit . we choose to work in a limit in which gravity is weak and the non - abelian field does not back react on the metric . thus , at heart , in this paper we will be solving the classical su(2 ) non - abelian yang - mills equations in a fixed background spacetime , that of a schwarzschild black hole in @xmath5 . in section 3 , we find a solution to the yang - mills equations near the phase transition . this power series solution in the order parameter and superfluid velocities allows us to demonstrate that the phase transition is second order and to calculate the speed of second sound from thermodynamic identities . in section 4 , we make some formal remarks about the current - current correlation functions for our model . we discuss the ward identities that these green s functions satisfy and some of their discrete symmetries . we also review how to calculate these two - point functions using the ads / cft correspondence . in section 5 , through a study of fluctuations about our solution near the phase transition , we extract the current - current correlation functions in the hydrodynamic limit . from the location of the poles , we independently confirm the speed of second sound calculated in section 3 . we are also able to calculate various damping coefficients and see explicitly that the green s functions satisfy the non - abelian ward identities . in the last part of section 5 , we consider the @xmath16 and @xmath17 limits . from these limits we extract conductivities and also demonstrate that the system obeys a type of london equation . consider the following gravitational action for a non - abelian gauge field @xmath18 with a cosmological constant @xmath19 : s = d^d+1 x ( r - 2 ) - d^d+1 x f^a_a b f^a ab . [ action ] our gauge field can be re - expressed in terms of a connection @xmath20 as follows : f_ab^a = _ a a_b^a - _ b a_a^a + f^a_bc a_a^b a_b^c , where @xmath21 are the structure constants for our lie algebra @xmath22 with generators @xmath23 such that @xmath24 = i { f_{ab}}^c t_c$ ] . we will take @xmath25 where @xmath26 , @xmath27 are the pauli spin matrices , and the structure constants are @xmath28 . indices @xmath29 are raised and lowered with the kronecker delta @xmath30 . the capital indices @xmath31 are raised and lowered with the five dimensional space time metric @xmath32 . we will also shortly introduce greek indices @xmath33 which will be raised and lowered with the four dimensional minkowski tensor @xmath34 . ] the equations of motion for the gauge field that follow from this action ( [ action ] ) are @xmath35 which can be expanded as _ a f^a ab + f^a_bc a_a^b f^c ab = 0 . [ gaugefieldeom ] einstein s equations can be written r_ab + ( - r ) g_ab = ( 2 f^a_c a f^ac_b - f^a_cd f^a cd g_ab ) . a solution to these equations in the case of a negative cosmological constant , @xmath36 , is a @xmath37-dimensional reissner - nordstrm black hole with anti - de sitter space asymptotics .. ] the only non - zero component of the vector potential can be taken to be is meant to evoke a charge density . the actual charge density @xmath38 . ] a_t^3 = + ^d-2 . [ atansat ] thus we are using only a u(1 ) subgroup of the full su(2 ) gauge symmetry ; this black hole solution requires only an abelian gauge symmetry . the line element for this black hole solution has the form ds^2 = 1 u^2 where @xmath39 and the warp factor is f ( ) = 1 + q^2 ( ) ^2d-2 - ( 1+q^2 ) ( ) ^d , with the charge @xmath40 being defined as q^2 ^2 _ the horizon is located at @xmath41 , and the hawking temperature is t_h = . one subtle issue to be addressed is the boundary conditions for @xmath42 at the horizon @xmath41 and at the boundary @xmath43 of our asymptotically ads space . at the horizon of the black hole , we must work in a local coordinate patch for the gauge potential such that @xmath42 has a well defined norm , @xmath44 . given the form of @xmath45 , we actually require that @xmath46 . our gauge potential ( [ atansat ] ) is well defined globally , at both the horizon and the boundary , provided = - /_h^d-2 . the boundary value of the gauge potential , @xmath47 , is interpreted via the ads / cft dictionary as a chemical potential in the dual field theory . as such , @xmath48 is an external parameter of the field theory , and we should restrict our class of gauge transformations to those which do not affect @xmath48 . for example , the class of abelian gauge transformations @xmath49 where @xmath50 is ruled out by this restriction . @xcite made the observation that below a critical temperature ( or alternately above a critical chemical potential ) , this charged black hole undergoes a second order phase transition . the component of the gauge field @xmath51 develops a profile which not only spontaneously breaks the remaining @xmath52 of the @xmath53 symmetry but also breaks rotational invariance . these authors considered the case @xmath54 and interpreted the dual 2 + 1 dimensional field theory as a @xmath0-wave thin film superconductor . more recently basu et al . @xcite looked at the @xmath55 case where they interpreted the dual 3 + 1 dimensional field theory as a pion superfluid . in this paper , we shall make two simplifying assumptions . the first is to take the probe limit , as was done in @xcite . in this limit , @xmath56 and the gauge field does not back react on the metric . the metric remains that of an uncharged black hole in anti - de sitter space with warp factor f(u ) = 1 - ( ) ^d . next , we restrict to the @xmath55 case because of the observation made in ref . @xcite that the zero mode inducing the phase transition has an analytic form . given this form , we are able to compute analytically many properties of the field theory close to the phase transition including green s functions , conductivities , diffusion constants , and the speed of second sound . we show explicitly that the phase transition is second order . we specialize to @xmath5 where we can construct an analytic solution to the gauge field equations of motion close to the phase transition . we give the solution as the first few terms in a power series in three small parameters : the order parameter @xmath57 and chemical - potential - like objects we call superfluid velocities , @xmath58 and @xmath59 . velocity is a bit of a misnomer here as the objects @xmath60 and @xmath61 , like the chemical potential @xmath48 , have mass dimension one . the name is motivated by their canonical conjugacy to the currents @xmath62 and @xmath63 . from the ads / cft dictionary , the currents @xmath64 and external field strength @xmath65 in the field theory can be determined from the small @xmath66 expansion of the bulk gauge field @xmath67 : [ adictionary ] a_^a = _ ^a + g^2 j_^a u^2 + . we begin with small steps and construct the solution in the limit @xmath68 . in the probe approximation , the equations of motion for the gauge field take the form _ t a^3_t = a^3_t _ x a^1_x = - a^1_x , [ a3t0eqanda1x0eq ] where we have defined the linear second order differential operators _ t _ ^2 - _ _ x = _ y _ ^2 + ( - ) _ . [ cald ] to keep the equations simple in what follows , we choose to put the horizon of the black hole at @xmath69 . to restore units , dimensionful quantities such as the chemical potential @xmath48 , frequencies @xmath70 , and wave - vectors @xmath71 should be replaced with the dimensionless combinations @xmath72 , @xmath73 , and @xmath74 , respectively . as pointed out by ref . @xcite , when @xmath75 there is an analytic solution to the second equation of ( [ a3t0eqanda1x0eq ] ) that is regular at the horizon , of the form [ zeromode ] a_x^1 = . from eq . ( [ adictionary ] ) , the meaning of @xmath76 in the dual field theory is , up to normalization , that of an expectation value for the non - abelian current @xmath77 . the existence of the solution indicates that the superfluid phase transition occurs when @xmath78 . where @xmath71 is a positive integer . we discuss these higher zero modes in appendix [ app : higher_zero_modes ] . as the higher zero modes have higher free energy , they should not affect the phase diagram of the system . ] given this zero mode , we look for a general solution to eqs . ( [ a3t0eqanda1x0eq ] ) as a series expansion in @xmath76 : @xmath79 the solution describes the system for @xmath80 . our strategy will be to fix the expectation value of @xmath81 but to allow the chemical potential to be corrected order by order : @xmath82 . thus , in solving the differential equations , we require the boundary condition that the @xmath83 term in @xmath84 vanish while @xmath85 is allowed to be nonzero . the differential equation governing @xmath86 is _ t _ 1 = , which has the solution _ 1 = ( 1-^2 ) _ 1 + ( 5 ^2 - ) . we applied the boundary condition that @xmath86 vanish at the horizon . also , @xmath87 corresponds to a shift of the chemical potential by @xmath88 . the value of @xmath87 is constrained by the solution for @xmath89 , as we now see . the differential equation for @xmath89 is w_1 - w_1 + w_1 = - _ 1 . we require the boundary conditions that @xmath89 be regular at the horizon and vanish at the boundary ( @xmath43 ) . these conditions leave us with the solution w_1 = + + , and the constraint _ 1 = . the term in @xmath90 proportional to @xmath91 is just the zero mode , and , consistent with our strategy , we set @xmath92 . for the free energy calculation we perform below , we also need the next order corrections , @xmath93 and @xmath94 . the expressions are too cumbersome to repeat here . the structure and boundary conditions are analogous to the case of @xmath95 and @xmath90 considered above . the near boundary expansion of our solution takes the form @xmath96 where @xmath97 these expansions match well with numerical solutions that we found close to the transition temperature . in this section , we generalize the background above to allow for the possibility of a superfluid flow . in terms of the bulk solution , this generalization requires turning on a constant value of @xmath98 and @xmath99 at the boundary corresponding to a non - zero superfluid velocity @xmath100 . the differential equations describing this background are a modification of eqs . ( [ a3t0eqanda1x0eq ] ) : @xmath101 where we set @xmath102 . the repeated covariant @xmath103 indices on the left hand side are not to be summed over . as before , we solve this system in a small @xmath76 expansion , but we also add another small expansion parameter @xmath104 . there is a non - uniformity in the limit @xmath105 and @xmath106 , and we find two branches of solutions for small values of the superfluid velocity . in the case where @xmath107 , we find @xmath108 in the case @xmath109 , we find @xmath110 these solutions can be used to compute the speed of second sound perpendicular and parallel to the order parameter @xmath51 . in a two component fluid , there are typically two propagating collective modes , ordinary and second sound . in our probe approximation , we see only the superfluid component , and the single collective motion available to us we call second sound . from our holographic perspective , ordinary sound would involve fluctuations of the metric so it is suppressed in the limit @xmath111 . the speed of second sound , like that of ordinary sound , can be computed from derivatives of the state variables . from ref . @xcite , the second sound speed squared in this probe limit should be c_2 ^ 2 = . - |_v=0 . [ speedsecondsound ] from eq . ( [ adictionary ] ) , the values of the charge current @xmath112 and the charge density @xmath113 can be read off from the order @xmath114 pieces of @xmath115 and @xmath42 , respectively . because our system is not rotationally symmetric , the speed of second sound will depend on the direction of propagation . let @xmath116 and @xmath117 be the speeds perpendicular and parallel to the order parameter @xmath51 , respectively . the speed @xmath116 can be computed from the background solution @xmath107 while @xmath60 can be computed from the solution with @xmath118 . in the case @xmath119 , we find that @xmath120 and hence , up to higher order corrections in @xmath76 , c_^2 ^2 ( - 4 ) . [ speedsoundperp ] ( we used the fact that @xmath121 , which can be read off from . ) in the case @xmath122 , at leading order @xmath42 remains the same but now we need a_x^3 = v_- v_(-4 ) u^2 + . [ aypar ] we find that c_^2 ^2 ( -4 ) . [ speedsoundpar ] we confirm these results for @xmath117 and @xmath116 in sections [ sec : transversesound ] and [ sec : longitudinalsound ] through an analysis of the hydrodynamic poles in the current - current correlation functions . for numerical results valid when @xmath76 is not necessarily small , see figure [ c2plot ] . these perturbative solutions in @xmath61 and @xmath60 can also be used to analyze the phase diagram of the system near the critical point @xmath123 . at the critical point , we expect the order parameter to vanish , so @xmath124 . the value of @xmath42 at @xmath125 can be reinterpreted as the value of the chemical potential . these two facts give us a relation between the chemical potential and superfluid velocity along the critical line separating the two phases . for superfluid flow parallel to the order parameter , we expect 4 + v_^2 while for flow perpendicular to the order parameter , we have instead 4 + v_^2 . we compute the contribution to the free energy from the gauge field term in the on - shell action : @xmath126 for a background where @xmath127 and a_t^3 = ( 4 + _ 1 ^2 + _ 2 ^4 ) ( 1-^2 ) , the on - shell action is s_vac = ( 64 + ^2 + ( - + 2 ) ^4 + o(^6 ) ) . here @xmath128 is the spatial volume of the field theory while @xmath129 is the inverse temperature . for the background where @xmath130 has condensed , we find in contrast that s_sf = ( 64 + ^2 + ( - + 2 ) ^4 + o(^6 ) ) . the difference in the values of the two on - shell actions is p = s_vac - s_sf = ( - ^4 + o(^6 ) ) . now @xmath131 can be interpreted also as a difference in the free energies because the free energy ( in the grand canonical ensemble ) is minus the value of the on - shell action . that @xmath132 implies that the free energy of the superfluid phase is smaller and thus the superfluid is stable . moreover , from the fact that the free energy difference scales as @xmath133 , we see that the phase transition is second order . for small @xmath76 , if we restore dimensions , then @xmath48 should be replaced by @xmath135 . thus , @xmath136 . the derivative of @xmath137 with respect to temperature is continuous but non - differentiable . a field theory with a non - abelian global symmetry , such as su(2 ) , has by noether s theorem , a conserved current @xmath138 which transforms under the adjoint representation of this symmetry group . in this paper , we are interested in green s functions for this current , in particular the fourier transformed retarded current - current correlation functions : [ gmunudef ] g^_ab ( p ) = i d^4x e^-i p x ( t ) . if the symmetry is non - anomalous , then we can weakly gauge it by coupling the current to an external gauge field @xmath139 . gauge invariance then implies that the correlation functions obey a series of ward identities . for the one - point function , the covariant derivative of the current vanishes : 0 = ( _ ^c_a + f_ab^c ^b _ ) j^_c . [ d j = 0 ] more usefully for the present discussion , there is also a ward identity for the retarded two - point function . we give here the fourier transformed version : 0=(i p_^c_a + f_ab^c _ ^b ) g^_cd(p ) + f_ad^c j^_c . for our gravitational system beyond the phase transition , the gravitational bulk values of both @xmath42 and @xmath51 are non - zero . the ads / cft dictionary allows us to read @xmath139 and @xmath140 from the near boundary expansion of @xmath67 using the relation ( [ adictionary ] ) . for the system under consideration here , of the components of the external gauge field only @xmath141 is non - zero in the field theory . we have two non - vanishing components of the current , @xmath142 and @xmath143 . below , we compute the green s functions that describe the response of the system to an external gauge field in the third isospin direction : @xmath144 . the relevant ward identities componentwise are @xmath145 our green s functions below obey this set of ward identities . another important observation for the green s functions under consideration is the symmetry under swapping the indices . we observe that g_ab^(p ) = ( -1)^(a , b ) g_ba^(p ) , [ pt ] where @xmath146 is equal to @xmath147 if either @xmath148 or @xmath149 , but not both , and 1 otherwise . this symmetry follows from the discrete symmetries of the system . given that our currents are even under pt , i.e. parity and time reversal , if pt were a symmetry of the state , we would expect the green s functions to be symmetric under an index swap . our state is not symmetric under pt , but it is symmetric under pt times a @xmath150 operation on the @xmath151 lie algebra , @xmath152 and @xmath153 . to compute the current - current correlators in the probe approximation , we perturb the background gauge field by sending @xmath154 consequently , the corresponding field strength @xmath18 changes to @xmath155 , with @xmath156 given by @xmath157 from , one can see that the quadratic action for @xmath158 is @xmath159 which gives the linearized equations of motion @xmath160 the quadratic action is in fact not well defined because the integrand diverges as @xmath161 at small @xmath66 as we will see . we will regulate this divergence using holographic renormalization @xcite . for definiteness , we will only analyze the case where the background gauge field does nt depend on @xmath162 or @xmath163 and where its radial components @xmath164 vanish . we choose a similar gauge for the perturbations by requiring @xmath165 . equations can be solved approximately in the limit of small @xmath66 . an appropriate series expansion in this limit is @xmath166 for some vector - valued functions @xmath167 , @xmath168 , @xmath169 , etc . the values of @xmath170 and @xmath171 are the only ones that can be specified independently ; all the other functions appearing in this expansion , namely @xmath172 and higher order corrections , can be expressed in terms of @xmath170 and @xmath171 . plugging into and looking at the term with the lowest power of @xmath66 in the equation with @xmath173 , one finds a relation between @xmath174 and @xmath175 : @xmath176 \bigg|_{u = 0 } \ .\ ] ] upon integration by parts in , the unregularized on - shell quadratic action can be written as @xmath177 the divergence that arises as one takes @xmath178 comes from the @xmath179 term whose most divergent piece goes like @xmath180 at small @xmath66 . this divergence can be regulated by adding the counterterm @xmath181 \bigg|_{u = 1/\lambda } \ .\ ] ] note that this counterterm depends only on the values of the gauge field on the surface @xmath182 and on its derivatives along this surface , as required by holographic renormalization . as a side note , a simpler formula for @xmath174 can be found if only @xmath42 and @xmath183 approach non - zero values at the boundary of ads . in this case , only @xmath184 is non - zero and is given by @xmath185 assuming that @xmath186 then @xmath187 to compute the fourier transformed two - point function , we first fourier transform the regulated on - shell action s_2^on - shell = ^_a(-p ) ( _ ^a(p ) + c _ ^a(p ) ) , where @xmath91 is an arbitrary constant introduced by the regularization procedure . although such an action is not a generating functional for the retarded green s function , using the procedure outlined by son and starinets @xcite , we can identify the retarded green s function as g^a _ b ( p ) = . the linear response of a system to a perturbation @xmath188 is then a current density of the form j^a_(p ) = g^a _ b(p ) _ ^b ( p ) . [ twopointformula ] for most physical questions , the ambiguity in the choice of @xmath91 should be irrelevant . more precisely , one can see from that schematically @xmath189 , so @xmath190 is ambiguous up to an additive term analytic in @xmath0 . its fourier transform @xmath191 is ambiguous up to an additive term of the from @xmath192 , where @xmath193 , @xmath194 , and @xmath195 are constants that depend on the particular green s function we are computing . since in position space equation reads j^_a(x ) = d^4x g^_ab(x - x ) _ ^b(x ) , it follows that the ambiguity in the choice of @xmath91 does not affect the result of @xmath196 if @xmath197 . in particular , the late - time , large - distance response of the system to localized sources is not affected by this ambiguity . there are many subtleties in these calculations . to calculate the fourier transformed retarded current - current correlation functions , we need to study fluctuations of the su(2 ) gauge fields @xmath198 in our black hole background . in the superfluid phase , the expectation value of the order parameter @xmath130 breaks rotational symmetry and makes our task richer and more complicated than in the rotationally symmetric case where only @xmath199 . in the rotationally symmetric case , it would be enough to consider a fluctuation with a time and space dependence of the form @xmath200 . given the breaking of rotational symmetry , we should in principle consider a more general dependence where we allow for motion both parallel and transverse to the order parameter : @xmath201 . because of the complexity of the full result , we shall not present a full accounting of all the green s functions here . instead we will content ourselves by studying various informative limits where either @xmath202 or @xmath203 . we make a few other additional simplifying restrictions . following in the footsteps of refs.@xcite where the third isospin direction was interpreted as the u(1 ) of electricity and magnetism , we will consider green s functions where at least one of the su(2 ) isospin indices is equal to three . in other words , we are interested in the linear response of the system to external electric and magnetic fields . the last simplifying restriction is to limit our study to the hydrodynamic regime , where the order parameter , the frequency , and the wave - vector are small compared to the temperature . in our dimensionless notation , @xmath204 . it is only in this limit that we have analytic results although it is straightforward to calculate the green s functions numerically beyond this regime . we work out the green s functions in five cases . the first and simplest case , for which we give the most detailed description of the calculation , is for a fluctuation transverse to the order parameter and a wave vector transverse to both the order parameter and the polarization of the fluctuation . we call this fluctuation the pure transverse mode . we next consider fluctuations that correspond to a second sound mode in two different limits , one where the sound is propagating parallel to the order parameter and one where the sound is propagating transverse . these two sets of fluctuations give us independent confirmation of the speeds of second sound computed in section [ sec : superfluidflow ] from thermodynamics . finally we consider fluctuations that correspond to a diffusive mode , again in two different limits , one where the diffusion is parallel to the order parameter , one in which the diffusion is transverse . ( in appendix [ sec : generalcase ] , we attempt to give a picture of the general case and how the different limits of these green s functions fit together . ) in section [ sec : conductivities ] , we discuss conductivities and the london equations . in what follows , to avoid cumbersome indices , we define new variables for the background values of the gauge field : a_x^1 w a_t^3 . the pure transverse mode is described by fluctuations of the field @xmath205 with only @xmath206 spatial dependence . we decompose the fluctuations into fourier modes : a_y^3(u , t , z ) = a_y(u ) e^-i t + i k z . these modes transverse to the order parameter @xmath51 decouple from the other fluctuations of the gauge field and are governed by the differential equation : _ y a_y = a_y , [ ptshear ] where @xmath207 was defined in eq . ( [ cald ] ) . near the horizon @xmath208 , we find that @xmath209 satisfies either ingoing or outgoing plane wave type boundary conditions . consistent with the presence of an event horizon , it is natural to choose ingoing boundary conditions ( the minus sign in the exponent ) . this choice leads to retarded , as opposed to advanced , green s functions in the dual field theory @xcite . at the boundary @xmath125 of ads , we would like the freedom to set @xmath210 to some arbitrary value of our choosing , corresponding to perturbing the dual field theory by a small external field strength . these two boundary conditions along with the differential equation uniquely specify the functional form of @xmath211 . while an analytic solution to eq . ( [ ptshear ] ) does not appear to be available , one can easily solve this equation in the limit of small @xmath70 , @xmath71 , and @xmath76 . we can write the solution for @xmath211 , valid to order @xmath212 , @xmath213 , @xmath214 , and @xmath215 , in the form [ ayshear ] a_y = a_y0 ( ) ^-i / 4 ( 1 + ^2 a_y + ^2 a_y + k^2 a_y k + ^2 a_y + ) . we find @xmath216 the expression for @xmath217 is too cumbersome to give here . near the boundary , this solution ( [ ayshear ] ) has the expansion a_y = a_y0 + a_y0 ( - - - + ( ^2-k^2 ) ( - ( ) ) ) ^2 + from this near boundary expansion and eq . ( [ twopointformula ] ) , we can calculate the two - point function for the current in the hydrodynamic limit : g^yy_33(,k ) = ( - - - + ( ^2-k^2 ) c ) + . [ puretransverseg ] note that the counter - term ambiguity , proportional to an arbitrary constant @xmath91 , is of the form predicted in eq . ( [ tildealpha3 ] ) . in general , second sound modes are expected to produce poles in the density - density correlation function . we thus need to consider fluctuations in the conjugate field @xmath42 . if we consider sound modes moving transverse to the order parameter , we can take the fluctuations to have a @xmath218 dependence but no @xmath219 dependence . the self - consistent set of fluctuations to consider that couple to @xmath220 are @xmath221 where @xmath222 . the four fluctuations satisfy four second order ordinary differential equations and one first order constraint : @xmath223 where @xmath224 , @xmath225 , and @xmath207 were defined in eq . ( [ cald ] ) . we checked that the derivative of the constraint equation ( [ constraintperp ] ) with respect to @xmath66 is a linear combination of all five differential equations . thus if a solution of the first four differential equations satisfies the constraint for some @xmath226 , it will satisfy the constraint equation at all @xmath226 . there are seven integration constants associated with this linear system . if we look at the horizon of the black hole at @xmath227 , we find seven different kinds of behavior . there exist six solutions that have plane wave behavior for @xmath228 , @xmath229 , and @xmath230 near the horizon of the form ( 1-u)^i / 4 . there is also a pure gauge solution , a_t^3 = - i , a_x^3 = i k , a_x^2 = -w . as in the pure transverse case , we choose pure ingoing boundary conditions corresponding to @xmath231 behavior . at the boundary @xmath125 of our asymptotically ads space , we would like to be able to perturb the system with arbitrary boundary values of @xmath232 and @xmath230 but set the `` unphysical '' components of the gauge field @xmath228 and @xmath229 to zero . these are four constraints and we have only three ingoing solutions . thus we will also need to make use of the pure gauge solution to enforce our @xmath125 boundary conditions . we solved the system perturbatively in @xmath70 , @xmath71 , and @xmath76 . we present the results here in the limit where @xmath233 . the near boundary expansion ( @xmath125 ) of the solution takes the form @xmath234 note , the expression @xmath235 is not homogeneous in our scaling limit . we have included the leading corrections proportional to @xmath236 and @xmath237 . there are terms in the expansion proportional to @xmath238 but they are subleading in @xmath70 , @xmath71 , and @xmath76 . the pole in this limit takes the form @xmath239 let us study this cubic polynomial in @xmath70 in two different limits . first , if @xmath240 , we find three poles with the asymptotic form @xmath241 the first two poles are propagating modes that we identify with second sound . indeed , the speed of second sound agrees with the earlier result ( [ speedsoundperp ] ) from section [ sec : superfluidflow ] . the position of the third pole in this limit is determined mostly by the size of the order parameter @xmath76 and so we associate it with the zero mode that causes the phase transition from the superfluid phase back to the normal phase . in the opposite limit , @xmath242 , where the order parameter is small , the behavior should be close to that of the normal fluid . in this limit , we find @xmath243 the first two poles are associated with the zero modes that cause the phase transition from the normal phase to the superfluid phase and were discussed in ref . @xcite while the third pole is associated to the diffusive mode of our conserved charge density . indeed , the location of this diffusive pole is determined by the dynamics of the normal phase and was calculated , without the order @xmath244 correction , long ago in ref . as we vary @xmath76 and @xmath71 the number of poles can not change . the two zero mode poles evolve into the sound poles of the previous limit while the diffusive pole becomes the zero mode pole of the previous limit . from these small @xmath66 expansions , we can read off the eight green s functions @xmath245 , @xmath246 , @xmath247 , @xmath248 , @xmath249 , @xmath250 , @xmath251 , and @xmath252 . from the discrete symmetries ( [ pt ] ) , we can also read off four more green s functions with the indices swapped . note the prefactor @xmath253 in the small @xmath66 expansion . this structure is necessary to satisfy the ward identities ( [ wardone ] ) . as a further check , we consider a particular static limit of the density - density correlation function . from eqs . ( [ twopointformula ] ) and ( [ at3perpsound ] ) , we can read off the green s function , g^tt_33 = - ( 1120 k^4 + 3k^2(117 ^2 - 1120 i ) + ( ^2 - 24 i ) ( 843 ^2 - 72,800 i ) ) . we are interested in the long wave - length limit of this green s function : _ k 0 g^tt_33 ( 0 , k ) = . [ chi ] this long wave - length limit is equal to a thermodynamic susceptibility , _ k 0 g^tt_33 ( 0 , k ) = = . given this relation , we see that eq . ( [ atsusc ] ) agrees with eq . ( [ chi ] ) . longitudinal sound modes correspond to the case where the fluctuations in @xmath42 depend only on @xmath219 . a self - consistent set of perturbations in this case is given by @xmath254 where @xmath255 . these fields satisfy the following six second order equations and three constraints : @xmath256 with @xmath224 and @xmath225 as defined in . again , the three constraint equations are consistent with the second order equations in the sense that if they hold at some @xmath66 , they hold at all @xmath66 . the system has nine integration constants . the nine possible behaviors at the horizon are of two types : six plane wave solutions for @xmath228 , @xmath229 , and @xmath257 that behave as ( 1-u)^i / 4 close to @xmath258 , and three pure gauge solutions given by @xmath259 where @xmath260 are arbitrary constants . as in the previous sections , we require no outgoing modes at the horizon , which amounts to specifying three of the nine integration constants . the other six integration constants are specified in terms of the values of the fields at @xmath261 . in order to examine fluctuations in @xmath232 and @xmath257 , we set their boundary values to @xmath236 and @xmath262 , respectively , and the boundary values of the other four fields to zero . solving the system perturbatively in @xmath70 , @xmath71 , and @xmath76 under the scaling assumption @xmath233 , we find that the boundary behavior of the fluctuations is @xmath263 the pole here is again a cubic polynomial in @xmath70 : @xmath264 we consider the roots of the polynomial first in the limit @xmath240 : @xmath265 the first pair of poles correspond to second sound propagating in the direction parallel to the order parameter with a speed consistent with our earlier result ( [ speedsoundpar ] ) . the third pole is related to the zero mode that causes a phase transition from the superfluid to the normal phase . next we consider the limit @xmath242 : @xmath266 the two sound poles have evolved into the zero mode poles , while the zero mode pole has evolved into a diffusive pole . from the small @xmath66 expansion , we can read off a large number of green s functions which we shall not bother to list . similar to the transverse sound case considered above , the prefactor @xmath267 in the expansion means that the ward identities ( [ wardone ] ) will be satisfied . however , there is more structure here . note that @xmath268 and @xmath269 . in our hydrodynamic limit at leading order in @xmath70 and @xmath71 , these two equations are the ward identities ( [ wardtwo ] ) and ( [ wardthree ] ) , respectively . before moving on , we note that _ k 0 g^tt_33 ( 0 , k ) = , which agrees with eq . ( [ chi ] ) , but @xmath71 here is parallel rather than transverse to the order parameter . in addition to the sound mode found above , in the limit @xmath240 , we expect to find a diffusive mode in the current - current correlator . we begin with the slightly simpler case of a mode polarized transverse to the order parameter but propagating parallel to it , and follow in the next section with a mode polarized longitudinal to the order parameter but propagating transversely . thus first we look for fluctuations in @xmath270 and any other modes that couple to it . a self - consistent set of fluctuations to consider is a_y^a(u , t , x ) = a_y^a(u ) e^-i t + i k x , where @xmath271 . this set of fluctuating modes gives rise to the three differential equations at linear order : @xmath272 as before , we solve this set of equations perturbatively in the limit @xmath233 . the small @xmath66 expansion of the solutions , from which we may read off the green s functions , takes the form : @xmath273 the poles at leading order in this perturbative expansion come from a quadratic polynomial in @xmath70 : = 65 ^2 + - . as before , we consider the roots of this polynomial in two limits . first we consider @xmath240 , in which case we find @xmath274 the first pole is associated with the zero mode that causes the phase transition from the superfluid phase to the normal phase while the second pole comes from a diffusive mode of the charge density . next , we consider the limit @xmath242 where we recover the zero modes of the normal phase , = k^2 + ^2 + . at leading order , the location of the pole is the same as that of eq . ( [ zmodetsound ] ) . however , the subleading order @xmath244 corrections are different . we continue the discussion by looking at modes polarized longitudinal to the order parameter but propagating transversely . we consider fluctuations @xmath275 and all others coupled to it : @xmath276 where @xmath277 . this set of fluctuations obeys the five second order equations and two first order constraints : @xmath278 at the horizon , there are two pure gauge solutions , three ingoing solutions , and three outgoing solutions . we discard the outgoing solutions and use the remaining degrees of freedom to choose the boundary values of the five fluctuations . in particular , we set the boundary values of all the fluctuations to zero save for @xmath257 , which we set to @xmath262 . the near boundary expansion of the solution takes the form : @xmath279 the pole , similar to the case considered previously , is a quadratic polynomial in @xmath70 : = 130 ^2 + ( 6 i k^2 + ) - k^2 ^2 - k^4 . in the limit @xmath240 we find a zero mode and a diffusive mode : @xmath280 in the opposite limit , we find two zero modes : = k^2 + ^2 + . the structure of the small @xmath66 expansion of the gauge fields is again related to the ward identities . we see that @xmath281 and @xmath282 , which are restatements of the ward identities ( [ wardtwo ] ) and ( [ wardthree ] ) , respectively . we have thus far considered the current - current correlation functions in five special cases . while the results are simpler , our presentation has the disadvantage of obscuring the relationship between the various cases . we make some remarks about and provide some results for the general case in appendix [ sec : generalcase ] . in this section , we begin by studying the response of the system to a homogeneous , time dependent electric field , @xmath283 , and end with a discussion of the london equations . a homogeneous electric field should produce a current in the system via ohm s law . to investigate the conductivity in this long wavelength limit , we set @xmath284 for the two - point functions computed above . the case of an electric field orthogonal to the order parameter is simple ; a current and nothing more is produced . from the pure transverse mode and eq . ( [ puretransverseg ] ) , we have g_33^yy ( ) = ( - + i ( - ) + c ^2 ) + . reassuringly , this result agrees with the @xmath285 limit of the green s functions associated to transverse sound propagation and the transverse diffusive mode . for an electric field parallel to the order parameter , the physics is richer . we find a current in the @xmath219 direction but also oscillating ( or precessing ) charge densities associated with the one and two isospin directions : @xmath286 this near boundary expansion agrees with the @xmath285 limit of the expansions for longitudinal sound and diffusion considered above . the associated green s functions are @xmath287 identifying the electric field @xmath288 and recalling ohm s law , the conductivities are related via eq . ( [ twopointformula ] ) to the retarded green s functions , [ sigmafromg ] _ xx ( ) = _ yy ( ) = . the terms proportional to @xmath244 in @xmath289 and @xmath249 thus produce a pole in the imaginary part of the respective conductivities . as discussed in refs . @xcite , by the kramers - kronig relations ( or by properly regularizing the pole ) there must be a delta function in the real part of the conductivity , indicating the material loses all resistance to dc currents and suggesting the phase transition is to a superconducting state . while in refs . @xcite , the pole was seen only numerically , here we can calculate the strength of the pole analytically close the phase transition . its residue is given by @xmath290 in figure [ numericalresidues ] we show a comparison between numerical computations of the residues of the poles at @xmath291 in @xmath292 and @xmath293 , along with the analytic approximation close to @xmath294 . in the drude model for an ideal metal , the conductivity takes the form @xmath295 where @xmath296 is the charge density and @xmath297 is the mass of the charge carrier . in the superconductivity literature ( see for example @xcite ) , the pole in the imaginary part of the conductivity is thus often related to a superfluid density . because our system is not rotationally symmetric , the density to mass ratio defined in this way will depend on the orientation of the superfluid velocity with respect to the order parameter . the proper way to interpret this situation is probably that a suitably defined effective mass of the superfluid depends on the direction of propagation . an important observation is that in our system , the @xmath16 and @xmath17 limits of the green s functions commute . the residue of the pole in the conductivity is related to the long wavelength limit of the current - current correlation function in the following way : i res_= 0 _ jj = _ 0 _ k 0 g^jj_33 ( , k ) . the limit in the opposite order is related to a thermodynamic susceptibility : _ k_y 0 g^xx_33 ( 0 , k_y ) = _ k_x 0 g^yy_33 ( 0 , k_x ) = , where @xmath60 and @xmath61 are superfluid velocities . that is transverse to the polarization of the current - current correlation function . a perturbation of the form @xmath298 or @xmath299 is gauge equivalent to zero and does not produce a response from the system . the green s function in this limit vanishes . ] it follows from eqs . and that = = - = = - . when combined with eq . , these results confirm eq . . to see how these limits commute in greater detail , the reader is referred to appendix [ sec : generalcase ] and eqs . that give the general green s functions in the inviscid limit @xmath300 . as emphasized in this context in ref @xcite , that the limits commute implies the system really does become a superconductor below @xmath8 . given that the limits commute , the system obeys a london type equation for small @xmath71 and @xmath70 : j_x^3 - _ x^3 j_y^3 - _ y^3 . if we now imagine the u(1 ) subgroup generated by @xmath301 is weakly gauged , these london equations imply not only infinite dc conductivity but also a meissner effect with london penetration depths that scale as @xmath302 . and @xmath303 as functions of temperature ( solid lines ) , as well as analytical approximations at small @xmath76 ( dotted lines ) given by equations . , width=480 ] one of the nicest features of our results is their analytic nature . we were able to confirm a number of previous numeric observations @xcite of this superfluid phase transition . in particular , we saw explicitly that the phase transition was second order ; the difference in free energy between the phases scaled as @xmath304 below the phase transition . we saw the order parameter grew as @xmath305 below @xmath8 and thus has a mean field critical exponent . we calculated the speed of second sound near the phase transition and observed that it vanished linearly with the reduced temperature @xmath306 . we looked at the pole at @xmath291 in the imaginary part of the conductivity and saw the same scaling , @xmath307 , that had been observed numerically in a related model @xcite and confirmed that the london penetration depth scales as @xmath308 . this laundry list of scalings ( or critical exponents ) is the same observed in the mean - field landau - ginzburg model of a superconductor . close to @xmath8 , eqs . and indicate that @xmath309 , so one may wonder whether such a formula is valid away from @xmath8 as well . numerical evaluations show that this is not the case : see figure [ c2plot ] . at small temperatures , @xmath310 approaches @xmath311 . our numerical evaluations are not sufficiently reliable at small temperatures to see whether @xmath312 has the same limit . as @xmath313 . this limit follows from eq . ( [ speedsecondsound ] ) and two observations : 1 ) at @xmath314 the lorentz symmetry breaking due to the temperature disappears and the pressure can depend on @xmath48 and @xmath315 only as @xmath316 . 2 ) by dimensional analysis , when @xmath317 , @xmath318 . we would like to thank amos yarom for discussion on this point . ] and @xmath310 as functions of the reduced temperature @xmath319 ( solid lines ) as well as analytical approximations given by eqs . and close to @xmath8 ( dotted lines ) . , width=480 ] there are some other results in this paper that are worth emphasizing . for superfluid velocities that are not too large , we were able to determine analytically the critical line in the temperature - superfluid velocity plane separating the normal phase from the superfluid phase . we calculated a large number of current - current correlation functions in the hydrodynamic limit and verified that they satisfied the non - abelian ward identities . we also investigated how the hydrodynamic poles in these correlation functions move around as a function of @xmath71 and @xmath76 . optimistically , we hope that someday this system will be more than a toy model . the introduction described possible similarities of this system to helium-@xmath1 and @xmath0-wave superconductors . here we add a speculation about a possible connection with qcd . the su(2 ) global symmetry of our model could be thought of as the residual approximate isospin symmetry of qcd at low energies and our chemical potential an isospin chemical potential . the phase structure of qcd at non - zero isospin chemical potential has been discussed by ref . alas , there is no persistent current in the stable phases they discuss . in the future , there are some questions that we would like to address . while we focused on current - current correlation functions in the third isospin direction in this paper , it would be good to study the full set of green s functions more carefully . because of the characteristic magnetic properties of superconductors , it would be interesting to investigate the dependence of the correlation functions on an external magnetic field . another interesting direction to pursue is the connection between this work and the membrane paradigm @xcite where the horizon of a black hole , rather than the boundary of an asymptotically anti - de sitter space , is thought of as a fluid . related to this direction is the observation of ref . @xcite that the fraction of the total charge density outside the black hole horizon scales as @xmath320 close to @xmath8 , suggesting that this quantity might be related to the superfluid density . we would like to know if this analogy can be made more precise . we would like to thank steve gubser , duncan haldane , david huse , nikola kamburov , igor klebanov , and amos yarom for discussion . the work of cph was supported in part by the us nsf under grant no . ssp was supported in part by the nsf under award number phy-0652782 . one can also find other analytical solutions to when the chemical potential is @xmath321 , where @xmath71 is an integer , corresponding to higher zero modes of @xmath51 . in fact , numerical exploration shows that these are the only values of @xmath48 for which zero modes occur . the corresponding backgrounds are probably unstable , because they have higher free energy . when @xmath322 , the corresponding zero mode can be written as @xmath323 where @xmath76 is an arbitrary constant and @xmath324 is a polynomial of degree @xmath325 in @xmath114 . it is not hard to find a recursion formula for the coefficients of @xmath326 . writing @xmath327 and using equation , it follows that @xmath328 a_{n-2}^{(k ) } = 0 \ .\ ] ] the boundary conditions needed to solve this recursion relation can be taken to be @xmath329 and @xmath330 , the latter representing the normalization of the series . one can check that for each integer value of @xmath71 the series terminates at order @xmath331 , the last non - zero coefficient being @xmath332 . in table [ pktable ] we provide analytic expressions for the first few @xmath324 . .formulas for @xmath324 with @xmath333 . [ cols="^,<",options="header " , ] [ pktable ] consider a general fluctuation of the form @xmath334 for @xmath335 and @xmath336 , in the limit where @xmath337 . solving the system of differential equations in the hydrodynamic limit leads to a set of green s functions with a pole that is a fifth order polynomial in @xmath70 . three of the roots of this polynomial correspond to the roots of the third order polynomial we found for the sound modes . two of the roots of this polynomial correspond to the roots of the second order polynomial we found for the diffusive modes . in other words , the six modes we found in the longitudinal and transverse sound cases are actually different limits ( @xmath338 and @xmath339 , respectively ) of three of the roots of this fifth order polynomial in @xmath70 . similarly , the four modes we found in the longitudinal and transverse diffusion cases are different limits of two of the roots of this fifth order polynomial . in the limit @xmath340 , we find that the roots are @xmath341 in the limit @xmath342 , we find in contrast that @xmath343 the case @xmath340 admits an inviscid limit where the two - point functions of the currents in the third isospin direction become @xmath344 from eqs . one can easily see that the limits @xmath16 and @xmath17 commute , implying that there is a meissner effect ; see section [ sec : conductivities ] . s. s. gubser , `` breaking an abelian gauge symmetry near a black hole horizon , '' arxiv:0801.2977 [ hep - th ] . s. a. hartnoll , c. p. herzog and g. t. horowitz , `` building a holographic superconductor , '' phys . * 101 * , 031601 ( 2008 ) [ arxiv:0803.3295 [ hep - th ] ] . j. m. maldacena , `` the large n limit of superconformal field theories and supergravity , '' adv . * 2 * , 231 ( 1998 ) [ int . j. theor . phys . * 38 * , 1113 ( 1999 ) ] [ arxiv : hep - th/9711200 ] . s. s. gubser , i. r. klebanov and a. m. polyakov , `` gauge theory correlators from non - critical string theory , '' phys . b * 428 * , 105 ( 1998 ) [ arxiv : hep - th/9802109 ] . e. witten , `` anti - de sitter space and holography , '' adv . * 2 * , 253 ( 1998 ) [ arxiv : hep - th/9802150 ] . j. bardeen , l. n. cooper and j. r. schrieffer , `` theory of superconductivity , '' phys . rev . * 108 * , 1175 ( 1957 ) . s. s. gubser and s. s. pufu , `` the gravity dual of a p - wave superconductor , '' jhep * 0811 * , 033 ( 2008 ) [ arxiv:0805.2960 [ hep - th ] ] . m. m. roberts and s. a. hartnoll , `` pseudogap and time reversal breaking in a holographic superconductor , '' jhep * 0808 * , 035 ( 2008 ) [ arxiv:0805.3898 [ hep - th ] ] . c. p. herzog , p. k. kovtun and d. t. son , `` holographic model of superfluidity , '' arxiv:0809.4870 [ hep - th ] . f. denef and s. a. hartnoll , `` landscape of superconducting membranes , '' arxiv:0901.1160 [ hep - th ] . p. basu , j. he , a. mukherjee and h. h. shieh , `` superconductivity from d3/d7 : holographic pion superfluid , '' arxiv:0810.3970 [ hep - th ] . m. ammon , j. erdmenger , m. kaminski and p. kerner , `` superconductivity from gauge / gravity duality with flavor , '' arxiv:0810.2316 [ hep - th ] . s. s. gubser and f. d. rocha , `` the gravity dual to a quantum critical point with spontaneous symmetry breaking , '' arxiv:0807.1737 [ hep - th ] . k. maeda and t. okamura , `` characteristic length of an ads / cft superconductor , '' phys . d * 78 * , 106006 ( 2008 ) [ arxiv:0809.3079 [ hep - th ] ] . a. buchel , j. jia and v. a. miransky , `` dynamical stabilization of runaway potentials at finite density , '' phys . b * 647 * , 305 ( 2007 ) [ arxiv : hep - th/0609031 ] . d. vollhardt and p. wlfle , _ the phases of superfluid helium-3 _ , taylor and francis , london , 1990 . r. bartnik and j. mckinnon , `` particle - like solutions of the einstein yang - mills equations , '' phys . rev . lett . * 61 * , 141 ( 1988 ) . p. bizon , `` colored black holes , '' phys . rev . lett . * 64 * , 2844 ( 1990 ) . m. s. volkov and d. v. galtsov , `` gravitating non - abelian solitons and black holes with yang - mills fields , '' phys . rept . * 319 * , 1 ( 1999 ) [ arxiv : hep - th/9810070 ] . e. winstanley , `` classical yang - mills black hole hair in anti - de sitter space , '' lect . notes phys . * 769 * , 49 ( 2009 ) [ arxiv:0801.0527 [ gr - qc ] ] . k. skenderis , `` lecture notes on holographic renormalization , '' class . * 19 * , 5849 ( 2002 ) [ arxiv : hep - th/0209067 ] . d. t. son and a. o. starinets , `` minkowski - space correlators in ads / cft correspondence : recipe and applications , '' jhep * 0209 * , 042 ( 2002 ) [ arxiv : hep - th/0205051 ] . c. p. herzog and d. t. son , `` schwinger - keldysh propagators from ads / cft correspondence , '' jhep * 0303 * , 046 ( 2003 ) [ arxiv : hep - th/0212072 ] . k. skenderis and b. c. van rees , `` real - time gauge / gravity duality : prescription , renormalization and examples , '' arxiv:0812.2909 [ hep - th ] . s. a. hartnoll , c. p. herzog and g. t. horowitz , `` holographic superconductors , '' jhep * 0812 * , 015 ( 2008 ) [ arxiv:0810.1563 [ hep - th ] ] . m. tinkham , _ introduction to superconductivity _ , 2nd edition , dover , new york , 1996 . p. basu , a. mukherjee and h. h. shieh , `` supercurrent : vector hair for an ads black hole , '' arxiv:0809.4494 [ hep - th ] . g. t. horowitz and m. m. roberts , `` holographic superconductors with various condensates , '' phys . d * 78 * , 126008 ( 2008 ) [ arxiv:0810.1077 [ hep - th ] ] .

the paper is devoted to the numerical solution of elastoplastic constitutive initial value problems . an improved form of the implicit return - mapping scheme for nonsmooth yield surfaces is proposed that systematically builds on a subdifferential formulation of the flow rule . the main advantage of this approach is that the treatment of singular points , such as apices or edges at which the flow direction is multivalued involves only a uniquely defined set of non - linear equations , similarly to smooth yield surfaces . this paper ( part i ) is focused on isotropic models containing : yield surfaces with one or two apices ( singular points ) laying on the hydrostatic axis ; plastic pseudo - potentials that are independent of the lode angle ; nonlinear isotropic hardening ( optionally ) . it is shown that for some models the improved integration scheme also enables to a priori decide about a type of the return and investigate existence , uniqueness and semismoothness of discretized constitutive operators in implicit form . further , the semismooth newton method is introduced to solve incremental boundary - value problems . the paper also contains numerical examples related to slope stability with available matlab implementation . keywords : elastoplasticity , nonsmooth yield surface , multivalued flow direction , implicit return - mapping scheme , semismooth newton method , limit analysis

1503.03605

the paper is devoted to the numerical solution of small - strain quasi - static elastoplastic problems . such a problem consists of the constitutive initial value problem ( civp ) and the balance equation representing the principle of virtual work . a broadly exploited and universal numerical / computational concept includes the following steps : * time - discretization of civp leading to an incremental constitutive problem ; * derivation of the constitutive and consistent tangent operators ; * substitution of the constitutive ( stress - strain ) operator into the balance equation leading to the incremental boundary value problem in terms of displacements ; * finite element discretization and derivation of a system of nonlinear equations ; * solving the system using a nonsmooth variant of the newton method . civp satisfies thermodynamical laws and usually involves internal variables such as plastic strains or hardening parameters . several integration schemes for numerical solution of civp were suggested . for their overview , we refer , e.g. , @xcite and references introduced therein . if the implicit or trapezoidal euler method is used then the incremental constitutive problem is solved by the elastic predictor / plastic corrector method . the plastic correction leads to the return - mapping scheme . we distinguish , e.g. , implicit , trapezoidal or midpoint return - mappings depending on a chosen time - discretization ( * ? ? ? * chapter 7 ) . in this paper , we assume that the plastic flow direction is generated by the plastic potential , @xmath3 . if @xmath3 is smooth then the corresponding plastic flow direction is uniquely determined by the derivative of @xmath3 and consequently , the plastic flow rule reads as follows , e.g. ( * ? ? ? * chapter 8) : @xmath4 here , @xmath5 , @xmath6 , @xmath7 , and @xmath8 denotes the plastic strain rate , the plastic multiplier rate , the stress tensor and the hardening thermodynamical forces , respectively . the corresponding return - mapping scheme is relatively straightforward and leads to solving a system of nonlinear equations . a difficulty arises when @xmath3 is _ nonsmooth_. mostly , it happens if the yield surface contains singular points , such as apices or edges . then the function @xmath3 is rather pseudo - potential than potential and its derivative need not exist everywhere . in such a case , the rule ( [ eqn_flow_rule ] ) is usually completed by some additive formulas depending on particular cases of @xmath3 and @xmath7 in an ad - hoc manner . for example , the implementation of the mohr - coulomb model reported in ( * ? ? ? * chapter 6 , 8) employs one , two , or six plastic multipliers @xmath9 , depending on the location of @xmath7 on the yield surface . since the stress tensor @xmath7 is unknown in civp one must blindly guess its right location . moreover , for each tested location , one must usually solve an auxilliary system of nonlinear equations whose solvability is not guaranteed in general . _ these facts are evident drawbacks of the current return - mapping schemes . _ in associative plasticity , it is well - known that the plastic flow rule ( [ eqn_flow_rule ] ) together with a hardening law and loading / unloading conditions can be equivalently replaced by the principle of maximum plastic dissipation within the constitutive model . this alternative formulation of civp does not require special treatment for nonsmooth @xmath3 and enables to solve civp by techniques based on mathematical programming @xcite . in particular , if the implicit or trapezoidal euler method is used then the incremental constitutive problem can be interpreted by a certain kind of the closest - point projection @xcite . for some nonassociative models , civp can be re - formulated using a theory of bipotentials that leads to new numerical schemes @xcite . these alternative definitions of the flow rule enable a variational re - formulation of the initial boundary value elastoplastic problem . consequently , solvability of this problem can be investigated ( see , e.g. , @xcite ) . therefore , the corresponding numerical techniques are usually also correct from the mathematical point of view . on the other hand , such a numerical treatment is not so universal and its implementation is more involved / too complex in comparison with standard procedures of computational inelasticity . the approach pursued in this paper builds on the subdifferental formulation of the plastic flow rule , e.g. ( * ? ? ? * section 6.3.9 ) , @xmath10 for nonsmooth @xmath3 . here , @xmath11 denotes the subdifferential of @xmath3 at @xmath12 with respect to the stress variable . if @xmath3 is convex at least in vicinity of the yield surface then this definition is justified , e.g. , by ( * ? ? ? * corollary 23.7.1 ) and is valid even when @xmath3 is not smooth at @xmath7 . on the first sight , it seems that ( [ inclusion_flow_rule ] ) is not convenient for numerical treatment due to the presence of the multivalued flow direction . the main goal of this paper is to show that the _ opposite is true _ , by demonstrating that the implicit return - mapping scheme based on ( [ inclusion_flow_rule ] ) leads to solving a just one system of nonlinear equations regardless whether the unknown stress tensor lies on the smooth portion of the yield surface or not at least for a wide class of models with nonsmooth plastic pseudo - potentials . using this technique , we eliminate the blind guessing and thus considerably simplify the solution scheme . moreover , the new technique enables to investigate some useful properties of the constitutive operator , like uniqueness or semismoothness , that are not obvious for the current technique . first of all , we illustrate the new technique on a simple 2d projective problem that mimics the structure of an incremental elastoplastic constitutive problem . consider the convex set @xmath13 and define the projection @xmath14 of a point @xmath15 as follows : @xmath16 the scheme of the projection is depicted in figure [ fig_projection ] . clearly , the function @xmath17 is convex in @xmath18 , nondifferentiable at @xmath19 and @xmath20 if @xmath21 then @xmath22 . conversely , if @xmath23 it follows from the karush - kuhn - tucker conditions and ( [ motivation0 ] ) that the projective problem can be written as follows : _ find @xmath24 and the lagrange multiplier @xmath25 : @xmath26 where _ @xmath27 [ -1,1 ] , & w_2^*=0 , \end{array } \right.\ ] ] to find a solution to ( [ motivationb2 ] ) , it is crucial to rewrite the inclusion ( [ motivationb2])@xmath28 as an equation . observe that @xmath29 where @xmath30 denotes a positive part of a function . this crucial transformation will be derived in detail in section [ sec_dp ] on an analogous elastoplastic example . thus ( [ motivationb2 ] ) leads to the following system of equations : @xmath31 since ( [ motivationb3])@xmath28 implies @xmath32 , the system of three nonlinear equations reduces to a single one @xmath33 consequently , @xmath9 can be found in the closed form as @xmath34 from which one can easily compute @xmath35 by ( [ motivationb3])@xmath36 and ( [ motivationb3])@xmath28 . the presented idea is systematically extended on some elastoplastic models . this paper , part i , is focused on isotropic models containing : @xmath0 yield surfaces with one or two apices ( singular points ) laying on the hydrostatic axis ; @xmath1 plastic pseudo - potentials that are independent of the lode angle ; @xmath2 nonlinear isotropic hardening ( optionally ) . such models are usually formulated by the haigh - westergaard coordinates . further , the implicit euler discretization of civp is considered and thus two types of return on the yield surface within the plastic correction are distinguished : @xmath37 return to the smooth portion of the yield surface ; @xmath38 return to the apex ( apices ) . the paper is organized as follows . section [ sec_preliminaries ] contains some preliminaries related to invariants of the stress tensor and semismooth functions . section [ sec_dp ] is devoted to the drucker - prager model including the nonlinear isotropic hardening . although the plastic corrector can not be found in closed form , the new technique enables to a priori decide about the return type and prove existence , uniqueness and semismoothness of the implicit constitutive operator . the consistent tangent operator is also introduced . in section [ sec_jg ] , we derive similar results for the perfect plastic part of the jirsek - grassl model @xcite . in section [ sec_general ] , the new technique is extended on an abstract model written by the haigh - westergaard coordinates . in particular , within the plastic correction , we formulate a unique system of nonlinear equations which is common for the both type of the return . it can lead to a more correct and/or simpler solution scheme in comparison with the current technique . section [ sec_realization ] is devoted to numerical realization of the incremental boundary value elastoplastic problem using the semismooth newton method . in section [ sec_experiments ] , illustrative numerical examples related to a slope stability benchmark are considered . here , limit load is analyzed by an incremental method depending on a mesh type and mesh density for the drucker - prager and jirsek - grassl models . within this paper , second order tensors , matrices and vectors are denoted by bold letters . as usual , small letters are used for vectors and capitals for matrices ( see section [ sec_realization ] ) . further , the fourth order tensors are denoted by capital blackboard letters , e.g. , @xmath39 or @xmath40 . the symbol @xmath41 means the tensor product @xcite . we also use the following notation : @xmath42 and @xmath43 for the space of symmetric , second order tensors . consider a stress tensor @xmath44 and its splitting into the volumetric and deviatoric parts : @xmath45 here , @xmath46 , @xmath47 , @xmath48 and @xmath49 denote the identity second order tensor , the fourth order deviatoric projection tensor , the hydrostatic pressure , and the deviatoric stress , respectively . _ the haigh - westergaard coordinates _ are created by the invariants @xmath48 , @xmath50 and @xmath51 , where @xmath52 @xmath53 clearly , @xmath54 and @xmath55 $ ] . since @xmath51 is not well - defined when @xmath56 , the lode angle is included in elastoplastic models indirectly . usually , it is considered another invariant in the form @xmath57 where @xmath58 is assumed to be a smooth function such that @xmath59 is at least continuous . however , as we will see below , it is more convenient to assume _ strong semismoothness _ of @xmath59 . as a particular case , it will be considered the invariant @xmath60 where @xmath61 the function @xmath62 was proposed in @xcite and contains the excentricity parameter @xmath63 $ ] . it holds : @xmath0 @xmath64 is a bounded and smooth function for any @xmath44 , @xmath65 ; @xmath1 @xmath66 when @xmath56 ; @xmath2 @xmath67 , @xmath68 when @xmath69 . we will also use the following derivatives : @xmath70 @xmath71,\quad \frac{\partial r_e}{\partial{\mbox{\boldmath$\sigma$}}}=-r_e'(\cos\theta)\sin\theta\frac{\partial \theta}{\partial{\mbox{\boldmath$\sigma$}}}. \label{inv_der2}\ ] ] notice that the derivatives of @xmath50 , @xmath72 , @xmath51 and @xmath62 do not exist when @xmath56 . further , @xmath51 is not differentiable when @xmath7 satisfies either @xmath73 or @xmath74 . on the other hand , @xmath62 has derivatives for such stresses @xcite . for purposes of this paper , it is crucial to derive the subdifferential of @xmath50 at @xmath7 when @xmath75 : @xmath76 if @xmath65 then @xmath77 by ( [ inv_der1])@xmath78 . it is readily seen that @xmath79 regardless @xmath80 or not . semismoothness was originally introduced by mifflin @xcite for functionals . qi and j. sun @xcite extended the definition of semismoothness to vector - valued functions to investigate the superlinear convergence of the newton method . we introduce a definition of _ strongly semismooth _ functions @xcite . to this end , consider finite dimensional spaces @xmath81 and @xmath82 with the norms @xmath83 and @xmath84 , respectively . in the context of this paper , the abstract spaces @xmath85 represent either subspaces of @xmath86 or the space @xmath87 . let @xmath88 be locally lipschitz function in a neighborhood of some @xmath89 and @xmath90 denote the generalized jacobian in the clarke sense @xcite . we say that @xmath91 is strongly semismooth at @xmath92 if 1 . @xmath91 is directionally differentiable at @xmath92 , 2 . for any @xmath93 , @xmath94 , and any @xmath95 , @xmath96 [ def_semi ] notice that the estimate ( [ semi_1 ] ) is called the quadratic approximate property in @xcite or the strong @xmath97-semismoothness in @xcite . in literature , there exists several equivalent definitions of strongly semismooth functions , see @xcite . for example the condition ( [ semi_1 ] ) can be replaced with @xmath98 where @xmath99 is the subset of @xmath81 , @xmath91 is frchet differentiable and @xmath100 denotes the directional derivative of @xmath91 at a point @xmath92 and a direction @xmath101 . we say that @xmath88 is strongly semismooth on an open set @xmath102 if @xmath91 is strongly semismooth at every point of @xmath103 . since it is difficult to straightforwardly prove ( [ semi_1 ] ) or ( [ semi_3 ] ) , we summarize several auxilliary results . firstly , piecewise smooth ( @xmath104 ) functions with lipschitz continuous derivatives of selected functions belong among strongly semismooth functions @xcite . especially , we mention the max - function , @xmath105 , @xmath106 . further , scalar product , sum , compositions of strongly semismooth functions are strongly semismooth . finally , we will use the following version of the implicit function theorem @xcite . let @xmath107 be a locally lipschitz function in a neighborhood of @xmath108 , which solve @xmath109 . let @xmath110 denote the generalized derivatives of @xmath111 at @xmath112 with respect to the variables @xmath92 . if @xmath113 is of maximal rank , i.e. the following implication holds , @xmath114 then there exists an open neighborhood @xmath115 of @xmath116 and a function @xmath117 such that @xmath91 is locally lipschitz continuous in @xmath115 , @xmath118 and for every @xmath119 in @xmath115 , @xmath120 . moreover , if @xmath111 is strongly semismooth at @xmath108 , then @xmath91 strongly semismooth at @xmath116 . [ implicit ] the semismoothness of constitutive operators in elastoplasticity has been studied e.g. in @xcite . namely in @xcite , one can find an abstract framework how to investigate it for operators in an implicit form . however , the framework can not be straightforwardly used for models investigated in this paper . therefore , we introduce the following auxilliary result . let @xmath121 , @xmath122 , @xmath123 , @xmath124 , @xmath125 , @xmath126 and @xmath127 be the functions introduced in section [ sec_invariants ] . further , ler @xmath128 be strongly semismooth functions and assume that @xmath129 vanishes for any @xmath44 , @xmath80 . define , @xmath130 then the functions @xmath50 , @xmath131 , @xmath132 and @xmath133 are strongly semismooth in @xmath87 . [ prop_semi ] since the functions @xmath125 and @xmath134 are bounded and have lipschitz continuous derivatives in @xmath135 , it is easy to see that the functions @xmath50 , @xmath131 , @xmath132 and @xmath133 are locally lipschitz continuous in @xmath87 and strongly semismooth for any @xmath44 , @xmath136 . therefore , it remains to show strong semismoothness at @xmath44 , @xmath80 . to this end , we show that ( [ semi_3 ] ) holds for @xmath50 , @xmath131 , @xmath132 and @xmath133 at such @xmath7 . let @xmath137 be such that @xmath138 . then @xmath139 hence @xmath140 , @xmath141 and @xmath142{\mbox{\boldmath$r$}}_e(\cos\theta({\mbox{\boldmath$\tau$}})){=}o(\|{\mbox{\boldmath$\tau$}}\|^2),\\ { \mbox{\boldmath$s$}}'({\mbox{\boldmath$\sigma$}}+{\mbox{\boldmath$\tau$}};{\mbox{\boldmath$\tau$}})-{\mbox{\boldmath$s$}}'({\mbox{\boldmath$\sigma$}};{\mbox{\boldmath$\tau$}})&=&[\hat p'({\mbox{\boldmath$\sigma$}}+{\mbox{\boldmath$\tau$}};{\mbox{\boldmath$\tau$}})-\hat p'({\mbox{\boldmath$\sigma$}};{\mbox{\boldmath$\tau$}})]{\mbox{\boldmath$i$}}+[\hat\varrho'({\mbox{\boldmath$\sigma$}}+{\mbox{\boldmath$\tau$}};{\mbox{\boldmath$\tau$}})-\hat\varrho'({\mbox{\boldmath$\sigma$}};{\mbox{\boldmath$\tau$}})]{\mbox{\boldmath$n$}}({\mbox{\boldmath$\tau$}}){=}o(\|{\mbox{\boldmath$\tau$}}\|^2),\end{aligned}\ ] ] since the functions @xmath143 , @xmath144 satisfy ( [ semi_3 ] ) by the assumption . the function @xmath133 introduced in proposition [ prop_semi ] has the same scheme as a mapping between trial and unknown stress tensors for models introduced in section [ sec_dp][sec_general ] . here , the trial stress is represented by @xmath7 and the unknown stress is in the form @xmath145 . therefore , it is sufficient to prove only semismoothness of the scalar functions @xmath143 , @xmath144 representing invariants of the unknown stress tensor . the semismoothness of @xmath146 has been derived to prove theorem [ th_semi_jg ] . we consider the elastoplastic problem containing the drucker - prager criterion , a nonassociative plastic flow rule and a nonlinear isotropic hardening . within a thermodynamical framework with internal variables , we introduce the corresponding constitutive initial value problem , see @xcite : 1 . _ additive decomposition of the infinitesimal strain tensor @xmath147 on elastic and plastic parts _ : @xmath148 2 . _ linear isotropic elastic law between the stress and the elastic strain _ : @xmath149 where @xmath150 denotes the bulk , and shear moduli , respectively . 3 . _ non - linear isotropic hardening _ : @xmath151 here @xmath152 denotes an isotropic ( scalar ) hardening variable , @xmath153 is the corresponding thermodynamical force and @xmath154 is a nondecreasing , strongly semismooth function satisfying @xmath155 . drucker - prager yield function _ : @xmath156 here , the parameters @xmath157 are usually calculated from the friction angle using a sufficient approximation of the mohr - coulomb yield surface and @xmath158 denotes the initial cohesion . _ plastic pseudo - potential_. @xmath159 here @xmath160 denotes a parameter depending on the dilatancy angle . _ nonassociative plastic flow rule _ @xmath161 where @xmath162 is a multiplier and @xmath163 denotes the subdifferential of the convex function @xmath3 at @xmath7 . using ( [ inv_der1 ] ) , ( [ rho_subgrad ] ) and ( [ potential_function ] ) , the flow rule ( [ flow_rule ] ) can be written as @xmath164 consequently by ( [ split ] ) , ( [ elastic_law ] ) and ( [ rho_subgrad_prop ] ) , @xmath165 7 . _ associative hardening law : _ . _ loading / unloding criterion : _ @xmath167 then the elastoplastic constitutive initial value problem reads as follows : _ given the history of the strain tensor @xmath168 , @xmath169 $ ] , and the initial values @xmath170 find @xmath171 such that @xmath172 hold for each instant @xmath173$]_. we discretize civp using the implicit euler method . to this end we assume a partition @xmath174 of the pseudo - time interval and fix a step @xmath175 . for the sake of brevity , we omit the index @xmath175 and write @xmath176 , @xmath177 , @xmath178 and @xmath179 . further , we define the following trial variables : @xmath180 , @xmath181 and @xmath182 . then the discrete elastoplastic constitutive problem for the @xmath175-step reads as follows : _ given @xmath147 , @xmath183 and @xmath184 . find @xmath7 , @xmath185 and @xmath186 satisfying : _ @xmath187 notice that the remaining input parameter for the next step , @xmath188 , can be computed using the formula @xmath189 after finding a solution to problem ( [ k - step_problem ] ) . we standardly use the elastic predictor / plastic corrector method for solving ( [ k - step_problem ] ) . * elastic predictor * applies when @xmath190 then the triplet @xmath191 is the solution to ( [ k - step_problem ] ) . * plastic corrector * applies when ( [ trial_admissibility ] ) does not hold . then @xmath192 and ( [ k - step_problem ] ) reduces into @xmath193 since the functions @xmath17 and @xmath3 depend on @xmath7 only through the variables @xmath50 and @xmath48 , it is natural to reduce a number of uknowns in problem ( [ k - step_problem_plast ] ) . to this end , we split ( [ k - step_problem_plast])@xmath36 into the deviatoric and volumetric parts : @xmath194 where @xmath195 , @xmath196 denotes the deviatoric stress , and the hydrostatic stress related to @xmath183 , respectively . using ( [ inv_der1])@xmath78 , the equality ( [ flow_dev ] ) yields @xmath197 \triangle\lambda g\sqrt{2}\hat{{\mbox{\boldmath$n$}}}&\mbox{if } & \varrho= 0 . \end{array}\right . \label{flow_s_k}\ ] ] denote @xmath198 and recall that @xmath199 for @xmath56 by ( [ rho_subgrad ] ) . then from ( [ flow_s_k ] ) we obtain @xmath200 following the arguments developed in section 1.1 , we now rewrite ( [ flow_rho ] ) as follows : @xmath201 notice that ( [ flow_rho ] ) and ( [ flow_rho_improve ] ) are equivalent . further from ( [ flow_s_k])@xmath36 , we standardly have : @xmath202 the following theorem summarizes and completes the derived results . let @xmath203 . if @xmath204 is a solution to problem ( [ k - step_problem_plast ] ) and @xmath205 , @xmath206 , @xmath207 , then @xmath208 is a solution to the following system : @xmath209 conversely , if @xmath208 is a solution to ( [ k - step_problem_plast_red ] ) then @xmath204 is the solution to ( [ k - step_problem_plast ] ) where @xmath210 [ lem_auxilliary_problem ] notice that the knowledge of the subdifferential of @xmath50 enables us to formulate the plastic corrector problem as a unique system of nonlinear equations in comparison to the current technique introduced in @xcite . moreover , one can eliminate the unknowns @xmath211 similarly as for the current return - mapping scheme of this model . inserting of ( [ k - step_problem_plast_red])@xmath212 into ( [ k - step_problem_plast_red])@xmath213 leads to the nonlinear equation @xmath214 where @xmath215 using ( [ yield_function ] ) . we have the following solvability result . let @xmath203 . then there exists a unique solution , @xmath192 , of the equation @xmath214 . furthermore , problems ( [ k - step_problem_plast_red ] ) , ( [ k - step_problem_plast ] ) and ( [ k - step_problem ] ) have also unique solutions . in addition , if @xmath216 then @xmath217 and @xmath218 . conversely , if @xmath219 @xmath220 then @xmath221 and @xmath56 . [ th_solvability_dp ] from ( [ q_dp ] ) and the assumptions on @xmath222 , it is readily seen that @xmath223 is a continuous and decreasing function . further , @xmath224 as @xmath225 and @xmath226 . therefore , the equation @xmath214 has just one solution in @xmath227 . if @xmath216 then @xmath217 . otherwise , @xmath221 . the rest of the proof follows from theorem [ lem_auxilliary_problem ] and the elastic prediction . the second part of theorem [ th_solvability_dp ] is very useful from the computational point of view : one can a priori decide whether return to the smooth portion of the yield surface happens or not . this is the main difference in comparison with the current return - mapping scheme . the improved return - mapping scheme reads as follows . * return to the smooth portion * 1 . _ necessary and sufficient condition _ : @xmath228 and @xmath216 . _ find _ @xmath217 : @xmath229 3 . _ * return to the apex * 1 . _ necessary and sufficient condition _ : @xmath231 . find _ @xmath221 : @xmath232 3 . @xmath233 nonlinear equations ( [ lambda_problem1 ] ) and ( [ lambda_problem2 ] ) can be solved by the newton method . then it is natural to use the initial choice @xmath234 , @xmath235 for ( [ lambda_problem1 ] ) , and ( [ lambda_problem2 ] ) , respectively . in case of perfect plasticity , @xmath236 , or linear hardening , @xmath237 , @xmath238 , equations ( [ lambda_problem1 ] ) and ( [ lambda_problem2 ] ) are linear and thus @xmath186 can be found in the closed form . solving the problem ( [ k - step_problem ] ) , we obtain a nonlinear and implicit operator between the stress tensor , @xmath239 , and the strain tensor , @xmath240 . the stress - strain operator , @xmath241 , also depends on @xmath242 and @xmath243 through the trial variables . to emphasize this fact we write @xmath244 . from the results introduced in section [ subsec_const_sol ] , we have @xmath245 , where @xmath246 where @xmath186 is the solution to ( [ lambda_problem1 ] ) , ( [ lambda_problem2 ] ) in ( [ s_def])@xmath28 , and ( [ s_def])@xmath78 , respectively , i.e. @xmath214 . the function @xmath241 is strongly semismooth in @xmath247 with respect to @xmath147 . [ th_semi_dp ] we use the framework introduced in section [ sub_semi ] . consider the function @xmath248 satisfying @xmath249 if @xmath250 , otherwise @xmath214 . applying theorem [ implicit ] on the implicit function @xmath223 , one can easily find that the function @xmath186 is strongly semismooth . consequently , the functions @xmath251 and @xmath252 are strongly semismooth . since @xmath253 , we obtain strong semismoothness of the functions @xmath133 and @xmath241 using proposition [ prop_semi ] . notice that @xmath241 is not smooth if @xmath254 or @xmath255 or if @xmath222 has not derivative at @xmath256 . we introduce the derivative @xmath257 under the assumption that any of these conditions does not hold . set @xmath258 . using ( [ inv_der1 ] ) , ( [ inv_der2 ] ) , ( [ elastic_law ] ) and the chain rule , we obtain the following auxilliary derivatives : @xmath259 we distinguish three possible cases : 1 . let @xmath260 ( _ elastic response _ ) . then clearly , @xmath261 2 . let @xmath228 and @xmath262 ( _ return to the smooth surface _ ) . then the derivative of ( [ sigma_smooth ] ) reads @xmath263 applying the implicit function theorem on ( [ lambda_problem1 ] ) , we obtain @xmath264 hence , @xmath265 3 . let @xmath266 ( _ return to the apex _ ) . then the derivative of ( [ sigma_apex ] ) yields @xmath267 applying the implicit function theorem on ( [ lambda_problem2 ] ) , we obtain @xmath268 hence , @xmath269 the derivatives ( [ deriv_elast])([deriv_apex ] ) define the consistent tangent operator , @xmath270 . it is readily seen that the tangent operator is symmetric if @xmath271 , i.e. for the associative plasticity . for purposes of section [ sec_realization ] , it is useful to extend the definition of @xmath270 for nondifferential points . for example , one can write @xmath272 ( \ref{deriv_smooth } ) & \mbox{if}&q_{tr}(0)>0,\;q_{tr}\left({\varrho^{tr}}/{g\sqrt{2}}\right)<0,\\[2pt ] ( \ref{deriv_apex } ) & \mbox{if } & q_{tr}\left({\varrho^{tr}}/{g\sqrt{2}}\right)\geq0 , \end{array } \right.\ ] ] where @xmath273 in ( [ deriv_smooth ] ) , ( [ deriv_apex ] ) is the derivative from left of @xmath222 at @xmath256 . notice that @xmath274 the jirsek - grassl model was introduced in @xcite . it is a plastic - damage model proposed for complex modelling of concrete failure . the model has been further developed . for example , unteregger and hofstetter @xcite have improved a hardening law and used the model in rock mechanics . for the sake of simplicity , we only consider a perfect plastic part of this model to illustrate the suggested idea and improve the implicit return - mapping scheme . the whole plastic part of the jirsek - grassl model can be included to an abstract model studied in the next section . the perfect plastic model contains the yield function proposed in @xcite : @xmath275 where @xmath276 is the friction parameter and @xmath277 is the uniaxial compressive strength . the invariants @xmath48 , @xmath50 and @xmath278 were introduced in section [ sec_preliminaries ] . notice that the couple @xmath279 defines the apex of the yields surface generated by the function @xmath280 . the yield surface is not smooth only at this apex . scheme of the yield surface can be found in @xcite . further , the following plastic pseudo - potential is considered @xcite : @xmath281 where @xmath282 the subdifferential of @xmath3 consists of the following directions : @xmath283 where @xmath284 is defined by ( [ rho_subgrad ] ) and @xmath285 the @xmath175-step of the incremental constitutive problem received by the implicit euler method reads as follows . _ given @xmath177 and @xmath286 . find @xmath239 and @xmath186 satisfying : @xmath287,\quad \hat{{\mbox{\boldmath$n$}}}\in\partial\varrho({\mbox{\boldmath$\sigma$}}),\\[5pt ] \triangle\lambda\geq0,\quad \hat f(p({\mbox{\boldmath$\sigma$}}),\varrho({\mbox{\boldmath$\sigma$}}),\varrho_e({\mbox{\boldmath$\sigma$}}))\leq0,\quad \triangle\lambda\hat f(p({\mbox{\boldmath$\sigma$}}),\varrho({\mbox{\boldmath$\sigma$}}),\varrho_e({\mbox{\boldmath$\sigma$}}))=0 . \end{array } \right\ } \label{k - step_problem_jg}\ ] ] _ we solve this problem again by the elastic predictor / plastic corrector method . within the plastic correction , we define the trial variables @xmath195 , @xmath288 , @xmath289 , @xmath290 , @xmath291 , @xmath292 and @xmath293 associated with @xmath183 and obtain the following result . let @xmath294 . if @xmath295 is a solution to problem ( [ k - step_problem_jg ] ) and @xmath205 , @xmath206 , @xmath207 , then @xmath296 is a solution to the following system : @xmath297 \varrho=\left[\varrho^{tr}-\triangle\lambda 2g\left(\frac{3\varrho}{\bar f_c^2}+\frac{m_0}{\sqrt{6}\bar f_c}\right)\right]^+,\\[5pt ] \hat f(p,\varrho,\varrho r_e^{tr})=0 . \end{array } \right\ } \label{k - step_problem_plast_jg}\ ] ] conversely , if @xmath296 is a solution to ( [ k - step_problem_plast_jg ] ) then @xmath295 is the solution to ( [ k - step_problem_jg ] ) where @xmath298 p{\mbox{\boldmath$i$ } } & \mbox{if } & \varrho^{tr}\leq\triangle\lambda 2g\left(\frac{3\varrho}{\bar f_c^2}+\frac{m_0}{\sqrt{6}\bar f_c}\right).\\ \end{array } \right.\ ] ] [ lem_auxilliary_problem_jg ] to prove theorem [ lem_auxilliary_problem_jg ] we use the same technique as in section [ subsec_const_sol ] . it is based on the splitting the stress tensor on the deviatoric and volumetric parts , and on using linear dependence between @xmath49 and @xmath195 to reduce a number of unknowns . in particular , we have @xmath299 for @xmath218 . consequently , we obtain ( [ k - step_problem_plast_jg])@xmath28 , @xmath300 and also @xmath301 for @xmath218 using ( [ theta ] ) . finally , notice that @xmath302 as @xmath303 . indeed , @xmath304 as @xmath303 and the function @xmath305 is bounded . analogously to the drucker - prager model , one can analyze existence and uniqueness of a solution to problem ( [ k - step_problem_plast_jg ] ) , and a priori decide whether the return to the smooth portion of the yield surface happens or not . to this end , we define implicit functions @xmath306 and @xmath307 such that @xmath308^+=0,\ ] ] respectively , for any @xmath309 . the following lemma is a consequence of the implicit function theorem . the functions @xmath310 and @xmath311 are well - defined in @xmath227 . further , @xmath310 is smooth and decreasing in @xmath227 , @xmath311 is decreasing in the interval @xmath312 and its closed form reads as follows : @xmath313 [ lem_jg ] now , consider the function @xmath314 , @xmath315 let @xmath294 . then there exists a unique solution , @xmath192 , of the equation @xmath214 . furthermore , problems ( [ k - step_problem_plast_jg ] ) and ( [ k - step_problem_jg ] ) have also unique solutions . in addition , if @xmath316 then @xmath317 and @xmath218 . conversely , if @xmath318 then @xmath319 and @xmath56 . [ th_solvability_jg ] since @xmath320 and @xmath321 in @xmath312 , the functions @xmath311 , @xmath322 are decreasing in this interval . for @xmath323 , these functions vanish . therefore , from ( [ q_jg ] ) and lemma [ lem_jg ] , it is follows that @xmath223 is a continuous and decreasing function in @xmath227 . furthermore , @xmath224 as @xmath225 and @xmath324 . hence , the equation @xmath214 has a unique solution in @xmath227 . if @xmath316 then @xmath317 otherwise , @xmath319 . the rest of the proof follows from theorem [ lem_auxilliary_problem_jg ] and the elastic prediction . although the function @xmath223 is implicit the decision criterion introduced in theorem [ th_solvability_jg ] can be found in closed form . @xmath325 since @xmath326 , @xmath327 hence , @xmath328 . using the definitions of @xmath310 and @xmath329 , we have @xmath330 by theorem [ lem_auxilliary_problem_jg ] and theorem [ th_solvability_jg ] , the return - mapping scheme reads as follows . * return to the smooth portion * 1 . _ necessary and sufficient condition _ : @xmath228 and @xmath316 . 2 . _ find _ @xmath331 , @xmath218 and @xmath317 : @xmath332 \varrho+\triangle\lambda 2g\left(\frac{3\varrho}{\bar f_c^2}+\frac{m_0}{\sqrt{6}\bar f_c}\right)-\varrho^{tr}=0,\\[5pt ] \frac{3}{2}\left(\frac{\varrho}{\bar f_c}\right)^2+m_0\left(\frac{\varrho r_e^{tr}}{\sqrt{6}\bar f_c}+\frac{p}{\bar f_c}\right)-1=0 . \end{array } \right\ } \label{lambda_problem1_jg}\ ] ] 3 . * return to the apex * 1 . _ necessary and sufficient condition _ : @xmath318 . set _ @xmath334 the system ( [ lambda_problem1_jg ] ) of nonlinear equations can be solved by the newton method with the initial choice @xmath335 , @xmath336 , @xmath234 . it was shown that the system has a unique solution subject to @xmath228 and @xmath316 . without these conditions , one can not guarantee existence and uniqueness of the solution to ( [ lambda_problem1_jg ] ) . solving the problem ( [ k - step_problem_jg ] ) , we obtain a nonlinear and implicit operator between the stress tensor , @xmath239 , and the strain tensor , @xmath240 . the stress - strain operator , @xmath241 , also depends on @xmath242 through the trial stress . to emphasize this fact we write @xmath337 . we have @xmath338 where the function @xmath223 is defined by ( [ q_jg ] ) and @xmath48 , @xmath50 are components of the solution to ( [ lambda_problem1_jg ] ) . the function @xmath241 is strongly semismooth in @xmath247 with respect to @xmath147 . [ th_semi_jg ] consider the function @xmath339 satisfying @xmath249 if @xmath250 , otherwise @xmath214 . to apply theorem [ implicit ] on the implicit function @xmath314 , it is necessary to show that @xmath340 is strongly semismooth w.r.t . the variables @xmath341 . this follows from ( [ hat_varrho ] ) and proposition [ prop_semi ] . the rest of the proof coincides with the proof of theorem [ th_semi_dp ] . if @xmath254 or @xmath342 then @xmath241 is not smooth . we derive the derivative @xmath257 under the assumption that any of these conditions does not hold . if @xmath260 ( elastic response ) then @xmath343 . if @xmath344 ( return to the apex ) then @xmath345 ( vanishes ) . let @xmath228 and @xmath316 , i.e. , return to the smooth portion happens . then the derivative @xmath257 can be found as follows . 1 . find the solution @xmath296 to ( [ lambda_problem1_jg ] ) . 2 . use ( [ inv_der1 ] ) , ( [ inv_der2 ] ) , ( [ elastic_law ] ) and the chain rule and compute : @xmath346 @xmath347,\quad \frac{\partial r_e^{tr}}{\partial{\mbox{\boldmath$\varepsilon$}}}=-r_e'(\cos\theta^{tr})\sin\theta^{tr}\frac{\partial \theta^{tr}}{\partial{\mbox{\boldmath$\varepsilon$}}}.\ ] ] 3 . compute : @xmath348 \frac{\partial \varrho}{\partial{\mbox{\boldmath$\varepsilon$}}}\\[5pt ] \frac{\partial \triangle\lambda}{\partial{\mbox{\boldmath$\varepsilon$ } } } \end{array } \right)=\left ( \begin{array}{c c c } 1+\triangle\lambda k\frac{m_g''(p)}{\bar f_c } & 0 & k\frac{m_g'(p)}{\bar f_c}\\ 0 & 1+\triangle\lambda \frac{6g}{\bar f_c^2 } & 2g\left(\frac{3\varrho}{\bar f_c^2}+\frac{m_0}{\sqrt{6}\bar f_c}\right)\\ \frac{m_0}{\bar f_c } & \frac{3\varrho}{\bar f_c^2}+\frac{m_0r_e^{tr}}{\sqrt{6}\bar f_c } & 0 \end{array } \right)^{-1}\left(\begin{array}{c } \frac{\partial p^{tr}}{\partial{\mbox{\boldmath$\varepsilon$}}}\\[5pt ] \frac{\partial \varrho^{tr}}{\partial{\mbox{\boldmath$\varepsilon$}}}\\[5pt ] -\frac{m_0\varrho}{\sqrt{6}\bar f_c}\frac{\partial r_e^{tr}}{\partial{\mbox{\boldmath$\varepsilon$ } } } \end{array}\right ) . \label{linearized_system}\ ] ] notice that the matrix in ( [ linearized_system ] ) arises from linearization of ( [ lambda_problem1_jg ] ) around the solution @xmath296 . the matrix is invertible since its determinant is negative . 4 . compute @xmath349 for numerical purposes , we use the following generalized consistent tangent operator : @xmath350 ( \ref{deriv_smooth_jg } ) & \mbox{if}&q_{tr}(0)>0,\;q_{tr}\left(\sqrt{6}\bar f_c\varrho^{tr}/2gm_0\right)<0,\\[2pt ] \mathbb o & \mbox{if } & \qquad \qquad \quad q_{tr}\left(\sqrt{6}\bar f_c\varrho^{tr}/2gm_0\right)\geq0 . \end{array } \right.\ ] ] the aim of this section is an extension of theorem [ lem_auxilliary_problem ] and [ lem_auxilliary_problem_jg ] on a specific class of elastoplastic models that are usually formulated in the haigh - westergaard coordinates . we consider an abstract model containing the isotropic hardening and the plastic flow pseudo - potential . _ given the history of the strain tensor @xmath168 , @xmath169 $ ] , and the initial values @xmath351 find the generalized stress @xmath352 and the generalized strain @xmath353 such that @xmath354 hold for each instant @xmath173$]_. further , we have the following assumptions on ingredients of the model : 1 . @xmath355 where @xmath280 is increasing with respect to @xmath50 and @xmath356 , convex and continuously differentiable at least in vicinity of the yield surface . 2 . @xmath357 where @xmath358 is an increasing function with respect to @xmath50 , convex and twice continuously differentiable at least in vicinity of the yield surface . 3 . @xmath222 is a nondecreasing , continuous and strongly semismooth function satisfying @xmath155 . 4 . @xmath359 is a positive function . invariants @xmath48 , @xmath50 , @xmath131 and @xmath356 are the same as in section [ sec_preliminaries ] . notice that the assumptions on @xmath280 and @xmath358 guarantee convexity of @xmath17 and @xmath3 using properties of @xmath62 introduced in @xcite . let @xmath360 then one can write the plastic flow rule as follows : @xmath361,\quad \hat{{\mbox{\boldmath$n$}}}\in\partial\varrho({\mbox{\boldmath$\sigma$ } } ) . \label{flow_rule2_abstract}\ ] ] the @xmath175-th step of the incremental constitutive problem received by the implicit euler method reads as follows . _ given @xmath177 , @xmath286 and @xmath180 . find @xmath239 , @xmath362 and @xmath186 satisfying : @xmath363,\quad \hat{{\mbox{\boldmath$n$}}}\in\partial\varrho({\mbox{\boldmath$\sigma$}}),\\[3pt ] \bar{\varepsilon}^{p}=\bar{\varepsilon}^{p , tr}+\triangle\lambda \ell({\mbox{\boldmath$\sigma$}},\kappa),\\[3pt ] \triangle\lambda\geq0,\quad f({\mbox{\boldmath$\sigma$}},h(\bar{\varepsilon}^{p}))\leq0,\quad \triangle\lambda f({\mbox{\boldmath$\sigma$}},h(\bar{\varepsilon}^{p}))=0 . \end{array } \right\ } \label{k - step_problem_abstract}\ ] ] _ if we use the elastic predictor / plastic corrector method then we derive the following straightforward extension of theorem [ lem_auxilliary_problem ] and theorem [ lem_auxilliary_problem_jg ] within the plastic correction . let @xmath203 . if @xmath364 is a solution to problem ( [ k - step_problem_abstract ] ) then @xmath365 , @xmath192 , is a solution to the following system : @xmath366 \varrho=\left[\varrho^{tr}-\triangle\lambda2g\hat g_\varrho(p,\varrho)\right]^+,\\[3pt ] \bar{\varepsilon}^{p}=\bar{\varepsilon}^{p , tr}+\triangle\lambda \hat\ell\left(p,\varrho,\varrho\tilde r(\cos \theta^{tr})\right),\\[3pt ] \hat f\left(p,\varrho,\varrho r_e(\cos \theta^{tr}),h(\bar{\varepsilon}^{p})\right)=0 . \end{array } \right\ } \label{k - step_problem_abstract_red}\ ] ] conversely , if @xmath367 is a solution to ( [ k - step_problem_abstract_red ] ) then @xmath368 solves ( [ k - step_problem_abstract ] ) , where @xmath369 p{\mbox{\boldmath$i$ } } & \mbox{if } & \varrho=0 . \end{array } \right . \label{sigma_k_abstract}\ ] ] [ lem_auxilliary_problem_abstract ] notice that it is generally impossible to a priori decide about the type of the return as in the models introduced above . to be in accordance with the current approach introduced e.g. in @xcite one can split ( [ k - step_problem_abstract_red ] ) into the following two systems : @xmath370 @xmath371 and guess which of these systems provides an admissible solution . beside the blind guessing , the current approach has another drawback : it can happen that ( [ k - step_problem_abstract_red ] ) has a unique solution and mutually one of the systems ( [ f_equality_apex ] ) , ( [ f_equality_smooth ] ) does not have any solution or have more than one solution . therefore , we recommend to solve ( [ k - step_problem_abstract_red ] ) directly by a nonsmooth version of the newton method with the standard initial choice @xmath335 , @xmath336 , @xmath372 and @xmath234 . consider an elasto - plastic body occupying a bounded domain @xmath373 with the lipschitz continuous boundary @xmath374 . it is assumed that @xmath375 , where @xmath376 and @xmath377 are open and disjoint sets . on @xmath376 , the homogeneous dirichlet boundary condition is prescribed . surface tractions of density @xmath378 are applied on @xmath377 and the body is subject to a volume force @xmath379 . notice that the above defined stress , strain and hardening variables depend on the spatial variable @xmath380 , i.e. @xmath381 , etc . let @xmath382 ^ 3\ |\;{\mbox{\boldmath$v$ } } = { \mbox{\boldmath$0$}}\ \mbox{on } \gamma_d \right\}$ ] denote the space of kinematically admissible displacements . under the infinitesimal small strain assumption , we have @xmath383 substitution of the stress - strain operator @xmath241 into the principle of the virtual work leads to the following problem at the @xmath175-th step : @xmath384 where @xmath385 and @xmath386 are the prescribed volume , and surface forces at @xmath387 , respectively . after finding a solution @xmath388 , the remaining unknown fields @xmath389 important for the next step can be computed at the level of integration points . problem @xmath390 can be standardly written as the operator equation in the dual space @xmath391 to @xmath392 : @xmath393 , where @xmath394 since we plan to use the semismooth newton method , we also introduce the operator @xmath395 as follows : @xmath396 to discretize the problem in space we use the finite element method . then the space @xmath392 is approximated by a finite dimensional one , @xmath397 . if linear simplicial elements are not used then it is also necessary to consider a suitable numerical quadrature on each element . let @xmath398 , @xmath399 , @xmath400 denote the approximation of operators @xmath401 , @xmath402 , @xmath403 , respectively , and @xmath404 , @xmath405 , @xmath406 be their algebraic counterparts . then the discretization of problem @xmath390 leads to the system of nonlinear equations , @xmath407 , and the semismooth newton method reads as follows : 1.2 initialization : @xmath408 find @xmath409 : @xmath410 compute @xmath411 set @xmath412 . if @xmath241 is strongly semismooth in @xmath87 then @xmath413 is strongly semismooth in @xmath86 . notice that the strong semismoothness is an essential assumption for local quadratic convergence of this algorithm . in numerical examples introduced below , we observe local quadratic convergence when the tolerance is sufficiently small . in particular , we set @xmath414 . the improved return - mapping schemes in combination with the semismooth newton method have been partially implemented in codes sifel @xcite and matsol @xcite . here , for the sake of simplicity , we consider the slope stability benchmark ( * ? ? ? * page 351 ) for the presented models , the drucker - prager ( dp ) and the jirasek - grassl ( jg ) ones . the benchmark is formulated as a plane strain problem . we focus on : a ) incremental limit analysis and b ) dependence of loading paths on element types and mesh density . for purposes of such an experiment , special matlab codes have been prepared to be transparent . these experimental codes are available in @xcite together with selected graphical outputs . the geometry of the body is depicted in figure [ fig.mesh_p1 ] or [ fig.mesh_q2 ] . the slope height is 10 m and its inclination is @xmath415 . on the bottom , we assume that the body is fixed and , on the left and right sides , zero normal displacements are prescribed . the body is subjected to self - weight . we set the specific weight @xmath416kn / m@xmath417 with @xmath418 being the mass density and @xmath3 the gravitational acceleration . such a volume force is multiplied by a scalar factor , @xmath419 . the loading process starts from @xmath420 . the gravity load factor , @xmath419 , is then increased gradually until collapse occurs . the initial increment of the factor is set to 0.1 . to illustrate loading responses we compute settlement at the corner point @xmath8 on the top of the slope depending on @xmath419 . as in ( * ? ? ? * page 351 ) , we set @xmath421kpa , @xmath422 , @xmath423 and @xmath424kpa , where @xmath425 denotes the cohesion for the perfect plastic model . hence , @xmath426kpa and @xmath427kpa . in comparison to @xcite , we use the presented models instead of the mohr - coulomb one . the remaining parameters for these models will be introduced below . we analyze the problem for linear triangular ( @xmath428 ) elements and eight - pointed quadrilateral ( @xmath429 ) elements . in the latter case , @xmath430-point gauss quadrature is used . for each element type , a hierarchy of four meshes with different densities is considered . the @xmath428-meshes contain 3210 , 12298 , 48126 , and 190121 nodal points , respectively . the @xmath429-meshes consist of 627 , 2405 , 9417 , and 37265 nodal points , respectively . the coarsest meshes for @xmath428 and @xmath429 elements are depicted in figure [ fig.mesh_p1 ] and [ fig.mesh_q2 ] . let us complete that the mesh in figure [ fig.mesh_p1 ] is uniform in vicinity of the slope and consists of right isoscales triangles with the same diagonal orientation . further , it is worth mentioning that the @xmath428-meshes are chosen much more finer in vicinity of the slope than their @xmath429-counterparts within the same level . the drucker - prager parameters @xmath431 , @xmath432 and @xmath433 are computed from the friction angle , @xmath434 , and the dilatancy angle , @xmath435 , as follows @xcite : @xmath436 at first , we introduce results obtained for the model with associative perfect plasticity . in such a case , @xmath437 , @xmath438 and @xmath439kpa . the received loading curves for the investigated meshes and elements are depicted in figure [ fig.load_path_h_p1 ] and [ fig.load_path_h_q2 ] . although @xmath428-meshes are much finer , we observe more significant dependence of the curves on the mesh density for @xmath428-elements than for @xmath429-elements . also computed limit load factors are greater and tend more slowly to a certain value as @xmath440 for @xmath428-meshes than for @xmath429-meshes . the expected limit value is 4.045 as follows from considerations introduced in @xcite . using the finest @xmath428 and @xmath429 meshes , we receive the values 4.230 , and 4.056 , respectively . in general , higher order elements are recommended when a locking effect is expected . in this example , it can be caused due to the presence of the limit load and/or @xmath441 . on the other hand , the strong dependence on mesh density is influenced by other factors like mesh structure or choice of a model . for example , this dependence is not so significant for the jirasek - grassl model ( see the next subsection ) . further , in @xcite , there is theoretically justified and illustrated that the dependence of the limit load on the mesh density is minimal for bounded yield surfaces and that an approximation of unbounded yield surfaces by bounded ones ( the truncation ) leads to a lower bound of the limit load . for illustration , we add figure [ fig.multiplier_perf_plas ] and [ fig.displacement_perf_plas ] with plastic multipliers and total displacements at collapse , respectively . the figures are in accordance with literature . to compare the current return - mapping scheme with improved one , we have also considered the nonassociative model with nonlinear hardening where @xmath442 here , @xmath443 represents the initial slope of the hardening function and the material response is perfect plastic for sufficiently large values of the hardening variable . this nonassociative model yields a slightly lower values of the limit load factors and also the other results are very similar to the associative model . the related graphical outputs are available in ( * ? ? ? * ss - dp - nh ) . further , in vicinity of the limit load , we have observed lower rounding errors for the improved return - mapping scheme and thus lower number of newton steps is necessary to receive the prescribe tolerance than for the current scheme . however , the computational time for both schemes are practically the same since return to the apex happens only on a few elements lying in vicinity of the yield surface . to be the simplified jirasek - grassl ( jg ) model applicable for the investigated soil material we fit its parameters using the associative perfect plastic drucker - prager ( dp ) model as follows : @xmath69 , @xmath444 , @xmath445 , @xmath446 , @xmath447 , @xmath448 and @xmath449 . recall that @xmath69 implies @xmath67 . further the value of @xmath277 corresponds to the uniaxial compressive strength computed from the drucker - prager model . to eliminate the influence of the exponential term in the function @xmath329 , the value of @xmath450 is chosen sufficiently large . then the model is insensitive on @xmath451 and one can vanish it . finally , we require the same flow direction for both the models under the uniaxial compressive strength . since the yield function in the jg model is normalized in comparison to the dp model it is convenient to introduce the following relation between the plastic multipliers : @xmath452 , where @xmath453 is a scale factor . then the values of @xmath276 and @xmath454 are determined from the following equations : @xmath455 to be @xmath276 positive , @xmath453 must be greater than @xmath456 . to be in accordance with results of the dp model , we set @xmath457 . , the limit load factor is underestimated and for greater values of @xmath453 the limit load factor is overestimated . ] comparison of yield surfaces ( in the meridean plane ) and flow directions for the dp and jg models is illustrated in figure [ fig_model_comparison ] . here , the fixed value @xmath458 is used for vectors representing the flow directions . loading curves for the investigated @xmath428 , @xmath429 meshes and the jg model are depicted in figure [ fig.load_path_h_jg ] and [ fig.load_path_h_q2_jg ] . we observe much faster convergence of the @xmath428-loading curves than for the dp model . moreover , the results for @xmath428 and @xmath429 elements are comparable . the computed values of the limit load factor on the finest @xmath428 and @xmath429 meshes are 4.124 , and 4.107 , respectively . the main idea of this paper is that the subdifferential formulation of the plastic flow rule is also useful for computational purposes and numerical analysis . namely , it has been shown that such an approach improves the implicit return - mapping scheme for non - smooth plastic pseudo - potentials as follows . * the unique system of nonlinear equations is solved regardless on a type of the return . * it can be a priori determined the type of the return from a given trial state for some models ( without knowledge of the solution ) . * the scheme can be more correct than the current one , and its form enables to study properties of constitutive operators like existence , uniqueness and semismoothness . in this paper ( part i ) , the new technique has been systematically built on a specific class of models containing singularities only along the hydrostatic axis . beside an abstract model , two particular models have been studied : the drucker - prager and the simplified jirasek - grassl model . however , the presented idea seems to be more universal . for example , it has been successfully used for the mohr - coulomb model in `` part ii '' @xcite . the authors would like to thank to pavel marlek for generating the quadrilateral meshes with midpoints . this work has been supported by the project 13 - 18652s ( ga cr ) and the european regional development fund in the it4innovations centre of excellence project ( cz.1.05/1.1.00/02.0070 ) . armero f , prez - foguet a. on the formulation of closest - point projection algorithms in elastoplasticity - part i : the variational structure . _ international journal for numerical methods in engineering _ 2002 ; * 53 * : 297 - 329 . de angelis f , taylor rl . an efficient return mapping algorithm for elastoplasticity with exact closed form solution of the local constitutive problem , _ engineering computations _ 2015 ; * 32 * : 2259 - 2291 . de saxc g. the biponential method , a new variational and numerical treatment of the dissipative laws of materials . in : _ 10th . int . conf . on mathematical and computer modelling and scientific computing _ , pp . 1 - 6 , 1995 . hjiaj m , fortin j , de saxc g. a complete stress update algorithm for the non - associated drucker - prager model including treatment of the apex . _ international journal of engineering science _ 2003 * 41 * : 1109 - 1143 . prez - foguet a , armero f. on the formulation of closest - point projection algorithms in elastoplasticity - part ii : globally convergent schemes . _ international journal for numerical methods in engineering _ 2002 ; * 53 * : 331 - 374 . willam kj , warnke ep . constitutive model for the triaxial behavior of concrete . i. : concrete structure subjected to triaxial stresses . 19 of iabse report , international association of bridge and structural engineers _ , zurich , 1974 , pp . 1 - 30 . zouain n. some variational formulations of non - associated hardening plasticity . _ mechanics of solids in brasil _ 2009 , h.s . da costa mattos and m. alves eds . , _ brasilian society of mechanical sciences and engineering _ , pp . 503 - 512 . isbn 978 - 85 - 85769 - 43 - 7 .

the amoeba of an affine algebraic variety is the image of under the map . we give a characterisation of the amoeba based on the triangle inequality , which we call ` testing for _ lopsidedness _ ' . we show that if a point is outside the amoeba of , there is an element of the defining ideal which witnesses this fact by being lopsided . this condition is necessary and sufficient for amoebas of arbitrary codimension , as well as for compactifications of amoebas inside any toric variety . our approach naturally leads to methods for approximating hypersurface amoebas and their spines by systems of linear inequalities . finally , we remark that our main result can be seen a precise analogue of a nullstellensatz statement for tropical varieties .

math0603201

let @xmath0 be an algebraic variety , defined by an ideal @xmath3}$ ] . the * amoeba * of @xmath1 , is defined to be the image of @xmath1 under the map @xmath4 defined at the point @xmath5 by @xmath6 we denote the amoeba of @xmath1 by either of @xmath7 or @xmath8 . if @xmath9 is a hypersurface , the zero locus of a single function @xmath10 , we will also use the notation @xmath11 . we refer the reader to mikhalkin s survey article @xcite for a broad discussion of amoebas and their applications . consider @xmath12}$ ] and a point @xmath13 . write @xmath10 as a sum of monomials @xmath14 . define @xmath15 to be the list of positive real numbers @xmath16 note that since the @xmath17 are monomials , this is well defined even though @xmath18 is not injective . [ def : lopsided ] we say that a list of positive numbers is * lopsided * , if one of the numbers is greater than the sum of all the others . equivalently , a list of numbers @xmath19 is not lopsided , if it is possible to choose complex phases @xmath20 ( @xmath21 ) , so that @xmath22 . this follows from the triangle inequality . we also define @xmath23 one can easily see that if @xmath24 then @xmath15 can not be lopsided ; in other words @xmath25 . indeed if @xmath26 then @xmath27 so it is giving a way to assign complex phases to the list @xmath28 such that the sum is @xmath29 . thus one should think of @xmath30 as a crude approximation to the amoeba @xmath11 . suppose @xmath31 , and let @xmath32 . for any complex phases @xmath33 , there exist @xmath34 such that @xmath35 and @xmath36 . thus @xmath24 if and only if @xmath37 is non - lopsided ; i.e. @xmath38 . in the above example we have enough freedom to choose the phases of the monomials @xmath39 for @xmath40 that @xmath41 . however , this only works because @xmath10 has very few non - zero terms . in general @xmath30 can be quite different from @xmath11 ( see figure [ fig : la ] ) . nevertheless , we will show that for a suitable multiple of @xmath10 , we can use this lopsidedness test to get very good approximations for @xmath11 . let @xmath42 be a positive integer . we consider the polynomials @xmath43 these @xmath44 are cyclic resultants @xmath45 and as such can be practically computed . our main result for amoebas of hypersurfaces is roughly the following . the precise version is stated and proved in section [ subsec : hypersurface ] . theorem [ thm : hypersurface ] ( rough version ) as @xmath46 , the family @xmath47 converges uniformly to @xmath11 . there exists an integer @xmath48 such that to compute @xmath11 to within @xmath49 , it suffices to compute @xmath47 for any @xmath50 . moreover , @xmath48 depends only on @xmath49 and the newton polytope ( or degree ) of @xmath10 , and can be computed explicitly from these data . this will lead us to the following characterisation of the amoeba of a general subvariety of @xmath51 . theorem [ thm : generalamoeba ] let @xmath3}$ ] be an ideal . a point @xmath13 is in the amoeba @xmath8 if and only if @xmath52 is not lopsided for every @xmath53 . phrased another way , if a point @xmath54 is outside the amoeba @xmath8 , a polynomial @xmath55 may witness this fact by being lopsided at @xmath54 . theorem [ thm : generalamoeba ] then states that there is always a witness . we will actually show something slightly stronger , in both theorems [ thm : hypersurface ] and [ thm : generalamoeba ] . we will show that there is a witness @xmath10 such that @xmath15 is `` superlopsided '' , according to the following definition : let @xmath56 . we say that a list of positive numbers @xmath57 is if there exists some @xmath58 such that @xmath59 for all @xmath60 . if @xmath61 , we will simply say the list is superlopsided . as before , we also define @xmath62 if a list of positive numbers is superlopsided , it is certainly lopsided ; hence @xmath63 ( see figure [ fig : la ] ) . david speyer observed that each component of the complement of @xmath64 is given by a system of _ linear _ inequalities , making it easier than @xmath30 to compute explicitly . hence theorem [ thm : hypersurface ] actually prescribes a method for approximating @xmath11 to within @xmath49 by systems of linear inequalities . similar ideas lead to a method for approximating the _ spine _ of a hypersurface amoeba . we discuss these constructions in section [ sec : approx ] . the motivation for these results comes from tropical algebraic geometry , and from this viewpoint , lopsidedness ( rather than superlopsidedness ) is the more natural condition to consider . in tropical algebraic geometry , we work with the semiring @xmath65 . this a semiring whose underlying set is @xmath66 , but whose operations are given by * @xmath67 * @xmath68 . the operations @xmath69 and @xmath70 are known as tropical addition and tropical multiplication . one can easily check that they satisfy the usual commutative , associative , and distributive laws ; however there are no additive inverses . a polynomial @xmath71 $ ] is therefore a piecewise linear function on @xmath72 : if @xmath73 then , translated into the usual operations on @xmath66 , @xmath74 the tropical variety associated to @xmath75 is then defined to be the singular locus of this piecewise linear function . a tropical variety associated to a single polynomial @xmath75 in this way is called a * tropical hypersurface*. thus there is a simple nullstellensatz for tropical hypersurfaces . a point @xmath76 is outside the tropical variety of @xmath75 , if there is a single monomial term of @xmath75 which is strictly larger than each of the others when evaluated at the point @xmath76 . in terms of the tropical operations this term is strictly greater than the tropical sum of the other terms ( c.f . definition [ def : lopsided ] ) . more generally the principal results in this paper can be seen of as an analytic analogue of a theorem of speyer and sturmfels @xcite for tropical varieties of arbitrary codimension , also known as non - archimedian amoebas . we discuss this connection in section [ sec : tropical ] . i am deeply grateful to david speyer , for providing a number of useful insights , and in particular for drawing my attention to the existence of lemma [ lem : averaging ] . i would also like to thank allen knutson providing feedback on early drafts , matthias beck for answering my questions on ehrhart polynomials , and bernd sturmfels for being particularly supportive of this project and for suggesting the title . this research was partially supported by nserc . the idea of the one variable case is simple enough . suppose that @xmath78 , and for sake of argument , assume that the absolute values of the @xmath79 are all distinct , say @xmath80 . then @xmath81 for @xmath42 large , the terms @xmath82 are small in comparison with the other terms and so this is approximately @xmath83 suppose @xmath84 . consider @xmath85 : as @xmath46 , every term tends to @xmath29 except for the constant term which is @xmath86 . thus for @xmath42 large , there is a single term in @xmath87 and likewise in @xmath44 which is much bigger in absolute value than all the others . we now formalise this heuristic argument in a way that will be useful in proving theorem [ thm : hypersurface ] . at the crux of the heuristic argument are the following three key facts about @xmath44 : 1 . it has no roots inside a certain annulus ( in the heuristic argument the annulus is @xmath88 } ) . the only non - zero terms which appear are of the form @xmath89 . the number of terms is not too large . ( this approach fails if instead of @xmath90 , we try to use @xmath91 with @xmath92 . ) to get a result which we can apply to the multivariable case , we need to be able to make a uniform statement about polynomials with these properties . this is precisely captured by the next two lemmas . by applying lemma [ lem : onevariable ] directly to the family of functions @xmath90 one immediately obtains a complete proof of theorem [ thm : hypersurface ] in the @xmath77 case . [ lem : goodestimate ] let @xmath93 be an open annulus in @xmath94 . we allow the possibilities of @xmath95 , and @xmath96 . let @xmath97 be a real number with @xmath98 , and let @xmath99 denote the closed interval @xmath100 $ ] . suppose @xmath101 is a polynomial such that 1 . @xmath101 has no roots in @xmath102 , and 2 . the only terms which appear in @xmath101 are of the form @xmath103 , i.e. @xmath104 , where @xmath105 , for some positive integer @xmath42 . then there is some @xmath106 such that @xmath107 for all @xmath108 . a similar statement can be made for laurent polynomials : if @xmath101 is a laurent polynomial satisfying properties 1 and 2 , and @xmath109 , the conclusion remains valid . here @xmath110 and @xmath111 refer respectively to the largest and smallest exponents which appear in @xmath101 . the notation in our proof assumes that @xmath101 is actually a polynomial . we can write @xmath112 where @xmath113 . we adopt the convention that @xmath114 , and @xmath115 . since @xmath116 has no roots in @xmath102 we have @xmath117 for some @xmath106 ( @xmath118 ) . we show that this @xmath106 has the desired property . consider @xmath119 . write @xmath120 where @xmath121 is a monomial of degree @xmath122 . we now estimate the size of @xmath123 at some point @xmath124 . if @xmath125 , clearly holds , so we only need to consider @xmath126 . first , we will show that if @xmath126 , we have @xmath127 certainly for all @xmath126 we have @xmath128 we use the following facts : * if @xmath129 , we have @xmath130 ( since @xmath131 ) . * if @xmath132 and @xmath133 , we have @xmath134 . * if @xmath135 , then @xmath136 . * if @xmath137 , then @xmath138 . if @xmath139 we write as @xmath140 this last equation comes from the fact that @xmath141 is the number of @xmath142 with @xmath143 . similarly , if @xmath144 we write @xmath145 which establishes . thus we obtain @xmath146 in particular for each @xmath126 we have latexmath:[\[\label{eqn : badestimate } however , we can do slightly better than this , by noting that the smallest power of @xmath97 that appears on the right side of the inequality is @xmath148 . thus , whereas tells us that @xmath149 , in fact we have : @xmath150 ( although is a better estimate than , the latter will prove to be more useful to us . ) for @xmath151 we have the following estimate : @xmath152 combining the estimates and we obtain @xmath153 for @xmath126 , as required . [ rmk : domterm ] it should be noted that we have actually determined which term is the special term @xmath154 . this is done in . if @xmath155 is a polynomial , then @xmath156 is the number of roots ( counted with multiplicity ) of @xmath155 inside the disc @xmath157 . if @xmath155 is a laurent polynomial , @xmath158 will be the number of roots inside @xmath159 . [ lem : onevariable ] as before , let @xmath93 . let @xmath97 be a real number with @xmath98 , and @xmath160 $ ] . fix positive integers @xmath161 . suppose @xmath116 is a family of polynomials such that 1 . @xmath116 has no roots in @xmath102 , 2 . the only terms which appear in @xmath116 are of the form @xmath89 . 3 . @xmath162 . more generally , if @xmath155 are laurent polynomials , we can ask that @xmath163 . assume @xmath42 is large enough that @xmath164 . then @xmath165 is @xmath166-superlopsided for all @xmath167 . from lemma [ lem : goodestimate ] , if we write @xmath168 , there is some @xmath106 such that for @xmath126 , @xmath169 here we are using the weaker estimate in to bound the numerator , and for the denominator . thus , @xmath170 the last step uses calculation [ cal:1 ] , which can be found in appendix [ app : calcdetails ] . this establishes that @xmath171 is @xmath166-superlopsided . this proof of lemma [ lem : onevariable ] uses a number of inequalities to give an answer to the question ` how large does @xmath42 have to be ? ' many of these inequalities , at first glance , appear not to be very tight . the answer we obtain is certainly not the best possible , and in most situations we should not expect to need quite such a large @xmath42 . however , without assuming more about @xmath155 it is surprisingly close to the best answer , particularly as @xmath172 . the best general answer for the question will be of the same form , i.e. @xmath173 one can see this by performing the requisite analysis on the polynomials @xmath174 . as a general heuristic , the more closely the roots of @xmath155 are packed , the larger @xmath42 will have to be ; thus the family of polynomials @xmath175 , where every root has as high a multiplicity as possible , is where we should expect our worst case behaviour to occur . suppose we want @xmath42 large enough to guarantee that @xmath176 is @xmath166-superlopsided for @xmath177 . write @xmath178 . we know @xmath86 will be the dominant term as @xmath42 gets large ( since @xmath179 ) , thus we need @xmath180 , or equivalently @xmath181 if we only want to guarantee that @xmath165 is lopsided for @xmath167 , we need @xmath42 large enough so that @xmath182 or equivalently @xmath183 this will hold if we have @xmath184 so @xmath42 only needs to be about half as big to guarantee that @xmath165 is lopsided as it does to guarantee that @xmath165 is superlopsided . again , we can show see that this is fairly close to the best answer by considering @xmath185 . in this section we prove our main theorem characterising the amoeba of a hypersurface . if @xmath186}$ ] , we consider the laurent polynomials @xmath187 theorem [ thm : hypersurface ] states that for @xmath188 , and a point @xmath13 in the complement of the amoeba @xmath11 whose distance from @xmath11 is at least @xmath49 , we can choose @xmath42 large enough so that @xmath189 is superlopsided . moreover , the theorem gives a upper bound on how large @xmath42 needs to be , based only on @xmath49 and the _ newton polytope _ of @xmath10 . the idea behind the proof of theorem [ thm : hypersurface ] is to look at the family of @xmath190 and interpret this as a function of a single variable @xmath191 . at the point @xmath192 , we define @xmath193 we will apply lemma [ lem : onevariable ] to these , and a find a single dominant term in this polynomial of one variable . then by an averaging argument , we show that this implies that @xmath44 has a single dominant term . first , however , we need a few simple observations . [ prop : sameamoeba ] @xmath194 @xmath190 is a product of terms @xmath195 , where @xmath196 . since @xmath197 , @xmath198 , and so @xmath199 . we will also need to know some information about the number and degree of the terms which appear in @xmath44 . first note the following important fact . [ prop : powersofn ] the only monomials which appear in @xmath44 are of the form @xmath200 . in particular the only terms appearing in @xmath201 are of the form @xmath89 . let @xmath202 denote the cyclic group of roots of @xmath203 . @xmath44 is manifestly invariant under the group action of @xmath204 acting on @xmath205}$ ] by @xmath206 . thus each monomial of @xmath44 must be invariant under this action . the only monomials with this property are of the form @xmath200 . the statement about @xmath201 follows immediately . recall that if @xmath207}$ ] its * newton polytope * , denoted @xmath208 , is the subset of @xmath72 defined as the convex hull of the exponent vectors of the monomials which appear in @xmath75 . for any polytope @xmath209 , let @xmath210 be any upper bound on @xmath211 . in general , it is not easy to find a tight upper bound for this number . if one can compute the ehrhart polynomial of @xmath209 explicitly , than an easy upper bound is the sum of the positive coefficients . otherwise , it is possible to bound the coefficients of the ehrhart polynomial in terms of the volume of @xmath209 ( see @xcite ) . using these estimates , for each @xmath212 one can compute constants @xmath102 and @xmath213 such that @xmath214 . clearly we have @xmath215 . this gives us an upper bound on the number of terms that @xmath44 can have : [ prop : newtonpolytope ] let @xmath216 . then @xmath44 has at most @xmath217 terms . by proposition [ prop : powersofn ] , the number of terms in @xmath44 is at most the number of integral points in @xmath218 . this is less than or equal to @xmath217 . finally we need to know something about @xmath219 . let @xmath220 where @xmath221 denotes the @xmath222 coordinate function on @xmath72 . [ prop : projnewtonpolytope ] @xmath223 @xmath224 armed with these facts and lemmas [ lem : goodestimate ] and [ lem : onevariable ] we are now in a position to precisely state and prove our main result for amoebas of hypersurfaces . [ thm : hypersurface ] let @xmath188 . suppose @xmath225 is a point in the amoeba complement whose distance from @xmath11 is at least @xmath49 . let @xmath216 , and @xmath226 . 1 . if @xmath42 is large enough so that @xmath227 then @xmath189 is lopsided . 2 . if @xmath42 is large enough so that @xmath228 then @xmath189 is superlopsided . ( in fact it is @xmath229-superlopsided ) . the key to reducing to the one variable case is the following basic result from complex analysis . [ lem : averaging ] let @xmath230 be a laurent polynomial , and write @xmath231 , where @xmath232 . suppose for all @xmath233 we have @xmath234 . then for each @xmath235 , @xmath236 . we integrate the equations @xmath237 over the set @xmath238 . @xmath239 let @xmath240 , and let @xmath241 . take @xmath233 . the function @xmath242 has no roots in the annulus @xmath243 . if it did , say at @xmath244 , @xmath44 would have a corresponding root at @xmath245 . so @xmath246 , hence by assumption , @xmath247 . but , since @xmath248 , @xmath249 . we ll first prove statement 2 . assume that holds . by propositions [ prop : powersofn ] and [ prop : projnewtonpolytope ] , we are in a position to apply lemma [ lem : onevariable ] to the laurent polynomial @xmath201 , taking @xmath102 , and @xmath97 as above , and @xmath250 , @xmath251 , @xmath252 and @xmath253 . with these choices , @xmath42 is large enough so that @xmath254 this is true for any @xmath233 , and since @xmath255 is connected , we see that for any fixed @xmath58 , it must be the same monomial term which dominates in each @xmath242 , independent of the choice of @xmath256 . let this be the @xmath257-term , and let @xmath258 . write @xmath259 where @xmath260 is the monomial @xmath261 . let @xmath262 . note this does not depend on the particular choice of @xmath256 . let @xmath263 we wish to show that @xmath264 . since the number of terms in @xmath44 is at most @xmath265 , this will imply that @xmath44 is superlopsided . we consider the terms of @xmath201 , and for simplicity of notation , we will temporarily assume that @xmath266 . the @xmath267-term in @xmath268 is @xmath269 for @xmath270 , tells us that @xmath271 since this is true for all @xmath256 , by lemma [ lem : averaging ] , @xmath272 provided @xmath270 . but , this argument is valid for any @xmath58 , so in fact we have @xmath273 for all @xmath274 . in particular , since @xmath275 for some @xmath274 , we deduce @xmath276 as required . we now prove statement 1 . although the approach is essentially the same , it is slightly more difficult , and hence requires some additional lemmas ( calculations [ cal : lopbounds ] and [ cal : expderiv ] , which can be found in appendix [ app : calcdetails ] ) . the reason for this is that we can not get these bounds by appealing directly to lemma [ lem : onevariable ] . instead , we use lemma [ lem : goodestimate ] , which gives better estimates for the coefficients of @xmath242 . write @xmath277 , where @xmath278 . then for each @xmath58 , @xmath279 . so by lemma [ lem : goodestimate ] there is some @xmath280 such that @xmath281 and in fact it is the same @xmath280 for all choices of @xmath256 . as before , let @xmath262 , and let @xmath282 . we have for any @xmath256 that @xmath283 ; thus @xmath284 this is true for all @xmath233 , thus by lemma [ lem : averaging ] we actually have @xmath285 the inequality is valid for all @xmath58 . in particular we should take @xmath58 to be the value which maximises @xmath286 , as this will be the one which minimises the right hand side . as @xmath287 is a sum of terms @xmath288 , can use to get an upper bound on @xmath289 . to do this , note that the right hand side of depends on @xmath290 and @xmath235 only insomuch as it depends on the maximum value of @xmath286 . thus we need to count , for each @xmath291 , how many values of @xmath235 in this sum satisfy @xmath292 an upper bound for this answer , is the number of integral points in @xmath293 with @xmath294 . there are @xmath295 of these , and this number is in turn less than @xmath296 . using @xmath296 as our upper bound for the number of @xmath235 with @xmath297 , gives us @xmath298 and by calculation [ cal : expderiv ] ( see appendix [ app : calcdetails ] ) this becomes @xmath299 assume now that holds . by calculation [ cal : lopbounds ] ( see appendix [ app : calcdetails ] ) , @xmath42 is large enough so that the right hand side is less than @xmath300 . thus we have @xmath301 , or equivalently , @xmath302 , as required . as an immediate consequence of theorem [ thm : hypersurface ] we get : [ cor : uniform ] both the families @xmath303 and @xmath304 converge uniformly to @xmath11 . we have @xmath305 , and for @xmath42 sufficiently large , the distance from any point in @xmath306 to @xmath11 will be less than @xmath307 , which tends to @xmath29 as @xmath46 . although theorem [ thm : hypersurface ] only showed that @xmath189 is _ superlopsided _ for @xmath42 sufficiently large . we can easily deduce something stronger . [ cor : verylopsided ] let @xmath308 . if @xmath42 sufficiently large , then @xmath189 is @xmath309-superlopsided . equivalently , if we write @xmath310 , then for @xmath42 sufficiently large , some @xmath311 will dominate , and in fact @xmath312 to see this , one simply has to look at . as @xmath46 , the right hand side approaches @xmath29 , and thus so does @xmath313 . alternatively , one can note that any @xmath314 can be used in place of @xmath315 in theorem [ thm : hypersurface ] . thus we see that it suffices to take @xmath42 so that @xmath316 just as was the case in lemma [ lem : onevariable ] , the bounds on @xmath42 given in theorem [ thm : hypersurface ] are not quite optimal : there are a number of places in which the inequalities can obviously be made tighter . however , as @xmath317 , the bounds are at least asymptotically correct . to see this for the superlopsided case , we can consider the example @xmath318 the amoeba @xmath11 is the union of all coordinate hyperplanes in @xmath72 . we can easily compute @xmath319 if our point is @xmath320 , with each @xmath321 , the dominant term of @xmath44 , will be @xmath86 . the number of non - zero monomial terms in @xmath44 is greater than @xmath322 ; thus to have @xmath323 superlopsided , we certainly need @xmath324 for all @xmath58 . this implies that @xmath325 in contrast , says that in this example we should take @xmath42 so that @xmath326 our bound for lopsidedness appears to be slightly less satisfactory . in the above example , to guarantee lopsidedness , one needs @xmath42 large enough so that @xmath327 this will hold when @xmath328 noting that @xmath329 , and approximating @xmath330 , this condition becomes @xmath331 if we take @xmath332 then this simplifies to @xmath333 in contrast , theorem [ thm : hypersurface ] tells us that it is sufficient to take @xmath42 so that @xmath334 again , this shows that the bounds in theorem [ thm : hypersurface ] are asymptotically correct , at least for any fixed @xmath212 . we suspect , however , that the correct general answer does not have this last term , or any term which is linear in @xmath212 . instead of the family @xmath44 , one may wish to consider a more general family of cyclic resultants . let @xmath335 be positive integers , and consider @xmath336 unfortunately it is not true that the family @xmath337 converges uniformly to @xmath11 , as @xmath338 . trouble occurs if some of the @xmath339 are significantly larger than others . for example , consider the amoeba of @xmath340 at a point @xmath341 , with @xmath342 . then @xmath343 if @xmath344 , then the first two terms above will have the same order of magnitude . thus @xmath345 will not be superlopsided , even if @xmath346 ( and hence @xmath347 ) are large . it will not be even be lopsided . however , if we restrict ourselves to the situation in which each @xmath339 is bounded by some polynomial in each of the other @xmath348 , then an analogous statement to theorem [ thm : hypersurface ] will be true . for example , we could let @xmath339 be any polynomial function of a single parameter @xmath42 . we will not compute explicit bounds for approximating the amoeba to within @xmath49 , in this more general situation ; however , the answer will depend upon these polynomials . it will certainly still be true that @xmath337 converges uniformly to @xmath11 , as this argument really only depends upon the fact that degrees of @xmath349 are growing only polynomially , while the terms are becoming suitably sparse . theorem [ thm : hypersurface ] tells us that for @xmath42 sufficiently large , one term of @xmath44 dominates , but does not specify which one . the answer will depend upon in which component of the amoeba complement our point @xmath54 lies . since @xmath350 varies continuously with @xmath54 , it depends only on the component of the amoeba complement . now the number of components is relatively small compared to the number of terms of @xmath44 . there is a natural injective map @xmath351 ( c.f . this is called the * index * of the component a complete definition is given below . so only a few of the terms of @xmath44 can possibly be dominant terms . fortunately it is relatively simple to determine which these are . the newton polytope of @xmath44 is @xmath352 , and candidates for dominant term are in fact what one would expect them to be : namely , they are the images of the integral points of @xmath353 under this scaling . let @xmath354 , and and let @xmath355 be the corresponding point in @xmath353 . if @xmath189 is lopsided , then the term of @xmath190 which dominates has exponent vector @xmath356 , ( i.e. it is the @xmath357-term ) . in order to make complete sense of the statement we need to know a definition of the index @xmath290 . there are a number of equivalent definitions , but the simplest for our purposes is the following . let @xmath233 . for each @xmath358 , consider the polynomial , @xmath359 if @xmath10 is a polynomial , then @xmath280 is the number of roots ( with multiplicity ) of @xmath360 inside the open disc @xmath361 . since @xmath360 never has a root on the circle @xmath362 for any @xmath256 , this number is independent of @xmath256 . if @xmath10 is a laurent polynomial , then @xmath363 it suffices to prove this if @xmath10 is a polynomial . if @xmath10 is a laurent polynomial , we multiply by some monomial to make it a polynomial , and note that this shifts both @xmath290 and the dominant term of @xmath44 appropriately . now @xmath364 is a product of @xmath365 terms , each of which is of the form @xmath366 , for some @xmath367 , @xmath368 . thus the number of roots of @xmath201 inside the disc @xmath369 is @xmath370 . as we noted in remark [ rmk : domterm ] , the number of roots in this disc is actually the exponent of the dominant term . since the dominant term of @xmath44 must project to the dominant term of each @xmath242 , the result follows . fix @xmath42 , and let @xmath372 where @xmath373 and @xmath315 are defined as in theorem [ thm : hypersurface ] . suppose we wish to approximate the component of @xmath371 corresponding to @xmath374 . from theorem [ thm : hypersurface ] , we know that @xmath375 and approximates @xmath11 to within @xmath49 . thus each component of @xmath371 is approximated by a component of @xmath376 . write @xmath377 where @xmath378 is the candidate for the dominant term in this component , and @xmath379 . are the other monomials . the corresponding component of @xmath376 is the set @xmath380 where @xmath381 is the number terms in @xmath44 . equivalently , this is the set of @xmath382 such that @xmath383 for all @xmath384 . this is a system of linear inequalities in the variable @xmath76 , so the solutions to these equations are a convex polyhedron which approximates the component of the amoeba to within @xmath49 . if there is no component of the @xmath371 corresponding to @xmath290 then this system of equations will have no solutions . conversely , if this system of inequalities has no solutions , then this component of the amoeba ( if it exists ) is not large enough to contain a ball of radius @xmath49 . thus we can realise any component of the @xmath371 as an increasing union of convex polyhedra . this gives an independent proof of the basic fact @xcite that the the components of the @xmath371 are convex . we must admit , however , that there are simpler proofs of this fact . note that in theorem [ thm : hypersurface ] , we actually show that @xmath44 is @xmath385-superlopsided . thus we can in fact take @xmath386 in and the set of solutions to this system of inequalities will still approximate the component of @xmath371 to within @xmath49 . one of the primary tools for studying amoebas has been the ronkin function @xmath387 , defined in @xcite . for @xmath12}$ ] , @xmath387 is defined to be the pushforward of @xmath388 under the map @xmath18 : @xmath389 ronkin shows in @xcite that @xmath387 is a convex function , and it is affine - linear precisely on the components of @xmath371 . when restricted to a single component of @xmath390 of @xmath371 , @xmath391 . passare and rullgrd use this function to define the spine of the amoeba @xcite , as follows . for each component @xmath392 of @xmath371 , extend the locally affine - linear function of @xmath393 to an affine - linear function @xmath394 on all of @xmath72 . let @xmath395 this is a convex piecewise linear function on @xmath72 , superscribing @xmath387 . the * spine * of the amoeba @xmath11 is defined to be the set of points where @xmath396 is not differentiable , and is denoted @xmath397 . the spine of the amoeba @xmath397 is a strong deformation retract of @xmath11 @xcite . also note that @xmath397 is actually a tropical hypersurface , as defined in the introduction : i.e. it is the singular locus of the maximum of a finite set of linear functions , where the gradient of each linear function is a lattice vector . now observe that @xmath398 can be thought of as a riemann sum for @xmath387 . in particular , we might expect @xmath399 to converge pointwise to @xmath400 . this will certainly be true provided that @xmath401 , is bounded on @xmath402 , which is the case when @xmath403 . suppose @xmath404 is in the component of @xmath371 of index @xmath374 . assume that @xmath76 has distance at least @xmath405 from the amoeba , where @xmath406 is fixed . for any @xmath188 we can find @xmath42 sufficiently large so that @xmath407 where each @xmath408 is relatively small , i.e. @xmath409 ( see corollary [ cor : verylopsided ] ) . thus we have @xmath410 thus we see that as @xmath411 , the values of @xmath399 , @xmath412 and @xmath413 all converge upon @xmath414 . we can use this fact obtain good approximations for the spine of the amoeba . for each @xmath42 we consider the function @xmath415 given by @xmath416 where the maximum is taken over all components of @xmath371 . this is a piecewise linear function . we define the approximate spine of the amoeba @xmath417 to be the set of points where @xmath418 is not smooth . equivalently @xmath417 is the set of points where the maximum in equation is attained by two distinct values of @xmath290 . 1 . @xmath419 . 2 . @xmath420 . statement 1 is true because on the component of @xmath421 of index @xmath290 , @xmath422 for any other @xmath423 . thus the maximum value in equation can not be attained by two distinct @xmath290 if @xmath424 . statement 2 follows from the fact that @xmath425 is a constant function and is less than @xmath49 for @xmath42 large . let @xmath426 and @xmath427 be components of @xmath371 of index @xmath428 and @xmath429 respectively . consider the hyperplane @xmath430 where @xmath431 and @xmath432 coincide , and the hyperplane @xmath433 where @xmath434 and @xmath435 coincide . the two hyperplanes @xmath436 and @xmath433 are parallel , and their distance apart is most @xmath437 , where @xmath99 is some constant depending only on @xmath428 and @xmath429 . as there are only finitely many @xmath438 , these distances can be made uniformly small . for practical reasons , we may wish use an alternate definition of @xmath439 , in which one takes the maximum in equation over only those components which appear in @xmath376 . if we do , statement 1 may be false for small @xmath42 , and we must settle for saying that @xmath440 . one might hope to be able to simplify this construction by taking the maximum in equation over all @xmath441 , rather than just those which actually correspond to components . it appears , however , that this does not give the same answer . with this alternate definition of @xmath439 , the ` approximate spine ' will have false chambers for all @xmath42 : i.e. the complement of this approximate spine will have components which do not correspond to components of @xmath371 . we might still hope that these false chambers would shrink to zero volume as @xmath42 gets large . unfortunately , experimental evidence suggests that the limit of these false chambers , as @xmath46 , can sometimes contain a ball of positive radius , and so this method does not produce a good approximation of the spine . the higher codimension statement ( theorem [ thm : generalamoeba ] ) follows fairly quickly from the hypersurface statement . let @xmath442 be a variety , which is the zero locus of an ideal @xmath443}$ ] . [ prop : reduction ] for every @xmath13 , there exists @xmath444 , such that @xmath445 for any laurent polynomial @xmath446},\ ] ] let @xmath447 denote its complex conjugate @xmath448 we define @xmath449 to be @xmath450 clearly @xmath449 is a laurent polynomial and is in @xmath451 . moreover , if we restrict @xmath452 to @xmath255 , then @xmath453 , so @xmath454 thus @xmath455 if and only if @xmath456 for all @xmath58 . this result is also true for ideals in @xmath457 $ ] : one can find a suitable monomial @xmath458 such that @xmath459 is a polynomial for all @xmath58 , and a similar argument holds if @xmath460 as an immediate consequence of proposition [ prop : reduction ] we have the following . [ cor : intersection ] for any ideal @xmath3}$ ] , @xmath461 it is also now a simple task to prove our second main result . [ thm : generalamoeba ] let @xmath3}$ ] be an ideal . a point @xmath13 is in the amoeba @xmath8 if and only if @xmath52 is not ( super)lopsided for every @xmath53 . if @xmath462 , then @xmath15 can not be lopsided for any @xmath55 , since @xmath24 for every @xmath55 . on the other hand , suppose @xmath463 . then by proposition [ prop : reduction ] if we take @xmath464 , then @xmath465 . by theorem [ thm : hypersurface ] , if @xmath42 is sufficiently large then @xmath466 is ( super)lopsided , and @xmath467 . [ rmk : summary ] in summary , we have three coincident sets , any of which can be used to define the amoeba @xmath7 of a variety @xmath468 . 1 . @xmath469 . 2 . @xmath470 . @xmath471 is not lopsided , for all @xmath472 . in section [ sec : tropical ] we shall see that this is precisely analogous to a theorem for tropical algebraic varieties . if a point @xmath54 is in @xmath473 , the proof of theorem [ thm : generalamoeba ] also tells us where to look for a witness to this fact : namely , we should look at @xmath474 for all @xmath42 . for some sufficiently large @xmath42 , this list will be lopsided . one unfortunate misfeature of this proof is that it requires us to use a different @xmath75 for every point @xmath475 . thus this statement is purely local . it does not give any clues as to how to produce a global uniform approximation to @xmath8 . however , in general we can not expect there to be any finite set of elements @xmath476 such that if @xmath463 , then some @xmath477 is lopsided for @xmath42 sufficiently large . if it were so , this would imply that @xmath8 is always an intersection of finitely many hypersurface amoebas , and this is certainly not true for dimensional reasons if @xmath478 . the best one could for is that we could find some family of sets @xmath479 such that @xmath480 approximates @xmath8 uniformly to within @xmath49 for @xmath42 sufficiently large . we present it as an open problem to find such a family explicitly . if we endow @xmath51 with the symplectic form @xmath481 , then the action of @xmath482 is hamiltonian , and its moment map is @xmath18 . suppose @xmath483 is another torus with a homomorphism @xmath484 . then the induced action of @xmath483 on @xmath51 is also hamiltonian , and we denote its moment map by @xmath485 . in this section we describe how to compute the image of @xmath1 under the map @xmath485 . the image @xmath486 is always a linear transformation of @xmath7 . if @xmath487 denotes the adjoint map to @xmath488 , then @xmath489 ( see e.g. ( * ? ? ? 3.2.8 ) ) . more concretely if we identify @xmath483 with @xmath490 , we can write @xmath491 where @xmath492 are the integer entries of a matrix @xmath102 the matrix representation of @xmath493and @xmath494 we could also take a matrix @xmath102 with integer entries as our starting point , and construct @xmath483 , @xmath488 , and the map @xmath485 so that holds . let @xmath3}$ ] denote the ideal of v. let @xmath495 denote the variety of the ideal @xmath496},\ ] ] where @xmath497 . now consider the projection of @xmath498 onto the @xmath499-coordinates @xmath500 . the image of @xmath498 under this projection is a variety @xmath501 . standard techniques of elimination theory allows us to compute its ideal @xmath502 ( see e.g. @xcite ) . [ prop : lintransform ] @xmath503 . a point in @xmath498 is a simply pair @xmath504 where @xmath505 and @xmath506 . thus we have @xmath507 projecting onto the @xmath499-coordinates , we obtain @xmath508 it is interesting to note that this construction is closely related to the cyclic resultants used in the proof of theorem [ thm : hypersurface ] . suppose @xmath509 is a hypersurface , and @xmath510 is the map @xmath511 . in this case , @xmath512 is a multiple of the identity matrix , and the variety @xmath501 is the zero locus of the function @xmath44 . intuitively , we should of think that the linear transformation is zooming in on the amoeba @xmath513 ; as we zoom in , theorem [ thm : hypersurface ] tells us that we see more and more detail in the approximations @xmath514 and @xmath515 . the most natural generalisation of amoebas in the compact setting , is to subvarieties of projective toric varieties . each projective toric variety is a compactification of @xmath51 , with an @xmath516 action which extends the @xmath516 action on @xmath51 . it also carries a natural symplectic form @xmath517 , for which the @xmath516 action is hamiltonian . we may therefore use the moment map for this hamiltonian action to replace the map @xmath18 . our goal in this section is to give a concrete description of this more general setting , and observe that our results still hold . this follows fairly easily from the non - compact case . our construction of toric varieties and their moment maps roughly follows a combination of @xcite and @xcite . let @xmath518 be a lattice polytope , i.e. the vertices of @xmath209 have integral coordinates . to every such @xmath209 we can associate the following data : 1 . a set of lattice points @xmath519 . 2 . a semigroup ring @xmath520 if @xmath521 , this is defined to be the quotient ring @xmath522/j,\ ] ] where each @xmath523 has degree @xmath86 , and @xmath524 is generated by all ( homogeneous ) relations of the form @xmath525 note that @xmath520 $ ] carries an action of the complex torus @xmath526 , given by @xmath527 3 . a toric variety @xmath528)$ ] . 4 . a projective embedding @xmath529)$ ] , induced by the map on rings @xmath530 \to { \mathbb{c}}[a]$ ] given by @xmath531 . 5 . a symplectic form @xmath532 , where @xmath533 is the fubini - study symplectic form on @xmath534 . 6 . a moment map @xmath535 for the @xmath516 action on @xmath536 . we can in fact write down the moment map @xmath535 explicitly . @xmath537 to evaluate the right hand side , we must choose a lifting of @xmath538 to @xmath539)$ ] . however , since this expression is homogeneous of degree @xmath29 in the @xmath540 , it is in fact well defined . it is well known that @xmath541 , and that if @xmath542 is any other projective toric variety with @xmath543 , then @xmath544 as toric varieties . let @xmath545 $ ] be a homogeneous ideal , and @xmath546/i)$ ] its variety inside @xmath547 . the * compactified amoeba * of @xmath1 is @xmath548 . we denote the compactified amoeba of @xmath1 by either @xmath549 or @xmath550 ( or by @xmath551 if @xmath552 is principal ) . let @xmath553 $ ] be a homogeneous polynomial of degree @xmath499 . we can again decompose @xmath10 as a sum of monomials , i.e. write @xmath554 , where each @xmath17 is a @xmath555-weight vector in @xmath520 $ ] . each of these @xmath17 is a well defined function on @xmath556 . let @xmath557 . we define @xmath558 is any preimage of @xmath559 in the composite map @xmath560 . of course , @xmath561 will depend on the choice of lifting under @xmath562 but only up to rescaling . thus the notions of @xmath561 being lopsided or superlopsided are still well defined . we define @xmath563 and @xmath564 to the set of points @xmath557 such that @xmath561 is non - lopsided , and non - superlopsided respectively . let @xmath565 denote the intersection of @xmath1 with the open dense subset of @xmath547 on which @xmath555 acts freely . we can identify this open dense subset of @xmath51 , and therefore consider @xmath566 . as both @xmath18 and @xmath567 are both submersions with fibres @xmath516 , it follows that @xmath566 is diffeomorphic to @xmath568 , where @xmath569 denotes the interior of @xmath209 . let @xmath570 denote this diffeomorphism . moreover any face @xmath571 of @xmath209 , corresponds to a toric subvariety @xmath572 . and @xmath573 @xcite . thus for every point @xmath557 , we can determine whether @xmath559 is in the compactified amoeba @xmath549 as follows . first we determine the face @xmath574 for which @xmath575 . then @xmath576 identifies @xmath577 with @xmath578 in such a way that @xmath579 is identified with @xmath580 . we then have @xmath581 if and only if @xmath582 . [ lem : uniformtransform ] the map @xmath583 is uniformly continuous . the projective embedding @xmath584 induces a map @xmath585 , which is define by monomials . this induces a linear map from @xmath586 . we also have a map from the moment polytope @xmath587 of @xmath588 to @xmath209 which is the projection induced by the inclusion of tori @xmath589 . the composite @xmath590 is @xmath583 . since the first and last maps are uniformly continuous , it suffices to show that @xmath591 is uniformly continuous . this is fairly straightforward . write @xmath592 . for @xmath593 we may write @xmath594 , and @xmath595 . if @xmath596 then @xmath597 i.e. @xmath598 . so @xmath599 . and the result follows . it should therefore come as no surprise that theorems [ thm : hypersurface ] and [ thm : generalamoeba ] have analogues in the compact setting . [ cor : compacthypersurface ] the families @xmath600 and @xmath601 converge uniformly to @xmath551 . moreover , to approximate the @xmath551 , to within @xmath49 , one can to choose @xmath602 , where @xmath603 can be determined solely from @xmath209 and @xmath604 . one can easily verify that @xmath605 is ( super)lopsided if and only if @xmath606 is ( super)lopsided , where @xmath575 . thus , since the lopsidedness of @xmath607 can be used to test membership in @xmath608 , @xmath609 can be used to determine membership in @xmath551 . the fact that the convergence is uniform follows from lemma [ lem : uniformtransform ] . moreover , as the data @xmath209 and @xmath604 , are enough to determine an outer bound on the newton polytope of @xmath10 , the rate of convergence can be determined from these data . we also immediately have : [ cor : compactamoeba ] @xmath610 . in particular @xmath611 it should be noted that corollary [ cor : compactamoeba ] holds for all toric varieties with a moment map , not just the compact ones . however the statement of uniform convergence in corollary [ cor : compacthypersurface ] does not hold in general for non - compact toric varieties . for example , if one considers the toric variety @xmath612 , with the standard moment map @xmath613 , the convergence of the family @xmath600 will almost never be uniform . one can even see this in the simple example @xmath614 . the failure is that lemma [ lem : uniformtransform ] does not hold : the map @xmath615 is not uniformly continuous , and so the uniform convergence does not carry over . it is most unfortunate that proposition [ prop : lintransform ] does not easily carry over to the compact case . the use of elimination theory appears only to be well suited to the study of @xmath51 with its particular standard symplectic form . in this section we show that theorem [ thm : generalamoeba ] is the analytic counterpart to a theorem for tropical varieties . we have already seen examples of tropical hypersurfaces . tropical varieties in general , can be thought of as a generalisation of amoebas , where one replaces the norm @xmath616 by a valuation in some non - archimedian field . for this reason , tropical varieties are also known as non - archimedian amoebas . let @xmath99 be an algebraically closed field , with valuation @xmath617 . for our purposes , a valuation on @xmath99 is a map @xmath618 , which satisfies the following conditions : * @xmath619 * @xmath620 . this differs from the usual definition of a valuation in two purely cosmetic ways . first , a valuation is traditionally given as a map to @xmath621 ; we have simply translated into the operations of @xmath622 . second , this is @xmath623 times the usual notion of a valuation . our reasons for making these cosmetic changes will become abundantly clear by the end of this section . to every @xmath624}$ ] we can associate a tropical polynomial as follows . if @xmath625 , write @xmath626 and call it the * tropicalisation * of @xmath10 . we denote the tropical hypersurface associated to @xmath627 by @xmath628 . if @xmath629 , we can assign a weight to every monomial @xmath630}$ ] : define the weight of @xmath631 at @xmath54 to be @xmath632 if @xmath633 where @xmath17 are monomials , let @xmath634 recall that in @xmath622 , a list of numbers @xmath635 is ( tropically ) lopsided if the maximum of element of this list does not occur twice ( in which case the maximum element is greater than the tropical sum of all the other elements ) . thus @xmath636 is lopsided if and only if @xmath637 . let @xmath3}$ ] be an ideal , and @xmath638 be its affine variety . let @xmath639 be the map @xmath640 the following theorem , as stated , most closely resembles the formulation in @xcite , though variants of it have also appeared in @xcite . [ thm : tropical ] the following subsets of @xmath641 coincide : 1 . the closure of the set @xmath642 . the intersection of all tropical hypersurfaces @xmath643 3 . the set of points @xmath644 , such that @xmath636 is not lopsided for all @xmath55 . this set is called the * tropical variety * of the ideal @xmath451 . in fact a stronger result than theorem [ thm : tropical ] ( as stated here ) is shown in @xcite . let @xmath106 denote the residue field of @xmath99 . if @xmath645 $ ] , then one can construct an initial ideal of i , in @xmath646 $ ] , corresponding to any weight @xmath647 . one can equivalently describe the tropical variety of @xmath451 as the set of points @xmath647 such that the associated initial ideal contains no monomial . thus it is possible to determine membership in a tropical variety using grbner basis techniques . one can easily see that theorem [ thm : tropical ] is precisely analogous to the summary given in remark [ rmk : summary ] . the proofs of these results , however , are extremely different . an obvious question , therefore , is whether analogous statements can be made in other contexts . the following is a general context in which one might hope for such a theorem to be true . suppose that @xmath99 is an algebraically closed field , and let @xmath648 be a totally ordered semiring . suppose @xmath649 satisfies the following conditions : 1 . @xmath650 for all @xmath651 . 2 . for all @xmath652 we have @xmath653 in particular condition 2 implies that @xmath654 for all @xmath655 . thus @xmath656 is an @xmath657-valued norm . let @xmath658}$ ] , and write @xmath659 as a sum of monomials . for any point @xmath660 , let @xmath256 be such that @xmath661 . we define @xmath662 as @xmath656 is multiplicative , this is independent of the choice of @xmath256 . since @xmath657 is totally ordered , we can define a list to of elements of @xmath657 to be _ lopsided _ if and only if one number is greater than the sum of all the others . let @xmath663 be a variety defined by an ideal @xmath664}$ ] . we can consider the following sets : * the closure of @xmath665 * @xmath666 is not lopsided @xmath667 the question is whether these two sets are equal for a particular @xmath668 . in this paper , we primarily discussed the example in which @xmath669 , @xmath670 , and @xmath671 , and showed that they are equal . we have also just seen that this is true if @xmath99 is a non - archimedian field , with @xmath656 as its valuation and @xmath672 . many ( though not quite all ) of the elements of the proof of theorem [ thm : hypersurface ] are valid in a more general context . suppose that in addition to being a totally ordered semiring , @xmath657 is a @xmath673-module ( i.e. we can make sense of such things as @xmath674 for @xmath675 ) ; for example @xmath622 is a @xmath673-module with the trivial @xmath673-action . define a binary operation @xmath676 on @xmath657 by @xmath677 whenever this set is nonempty . ( we need not overly concern ourselves with the fact that a precise minimum may not exist : one can always get around this by treating this set as a dedekind cut . ) then the triangle inequality @xmath678 is valid ( assuming @xmath679 ) . to see this note that @xmath680 implies @xmath681 ; thus @xmath682 clearly @xmath683 which implies that @xmath684 a closer examination of the proofs of lemma [ lem : goodestimate ] and calculation [ cal : expderiv ] now reveal that they are also valid ( almost word for word ) for a general @xmath685 . we can also prove lemma [ lem : averaging ] in general , by replacing the integral over the torus @xmath686 by a discrete average over a finite subgroup of the torus @xmath687 if @xmath48 is suitably large this discrete average will have the same effect as the integral ( i.e. picking out a single term from the polynomial ) . in fact we can follow the proof statement 1 of theorem [ thm : hypersurface ] , up to and including the inequality . all that remains to show that the right hand side of becomes sufficiently small as @xmath42 gets large . unfortunately , in general , this is not always true for all @xmath688 ( e.g. if @xmath689 $ ] with @xmath690 if the leading coefficient of @xmath691 is positive , this fails for @xmath692 ) . it is true , however , for @xmath693 , so our proof of theorem [ thm : hypersurface ] actually works in the tropical setting as well . this suggests that it may be possible to prove a more general statement , with perhaps some additional restrictions on @xmath657 . on the other hand , our proof of corollary [ cor : intersection ] , which is the key to theorem [ thm : generalamoeba ] , is very highly dependent on the fact that our underlying field is the complex numbers . a very different proof is required in the tropical setting , and it would be interesting to try and unite these two . in this appendix we give the mundane , technical calculations which are used in the proofs of lemma [ lem : onevariable ] and theorem [ thm : hypersurface ] .

we show that , in a system with defects , two - particle states may experience destructive quantum interference , or antiresonance . it prevents an excitation localized on a defect from decaying even where the decay is allowed by energy conservation . the system studied is a qubit chain or an equivalent spin chain with an anisotropic ( ) exchange coupling in a magnetic field . the chain has a defect with an excess on - site energy . it corresponds to a qubit with the level spacing different from other qubits . we show that , because of the interaction between excitations , a single defect may lead to multiple localized states . the energy spectra and localization lengths are found for two - excitation states . the localization of excitations facilitates the operation of a quantum computer . analytical results for strongly anisotropic coupling are confirmed by numerical studies .

quant-ph0303130

one of the most important potential applications of quantum computers ( qc s ) is studies of quantum many - body effects . it is particularly interesting to find new many - body effects in condensed - matter systems that could be easily simulated on a qc . in the present paper we discuss one such effect : antiresonance , or destructive quantum interference between two - particle excitations in a system with defects . we also study interaction - induced two - particle localization on a defect and discuss implications of the results for quantum computing . the basic elements of a qc , qubits , are two - state systems . they are naturally modeled by spin-1/2 particles . in many suggested realizations of qc s , the qubit - qubit interaction is `` on '' all the time @xcite . in terms of spins , it corresponds to exchange interaction . the dynamics of such qc s and spin systems in solids have many important similar aspects that can be studied together . in most proposed qc s the energy difference between the qubit states is large compared to the qubit - qubit interaction . this corresponds to a system of spins in a strong external magnetic field . however , in contrast to ideal spin systems , level spacings of different qubits can be different . a major advantageous feature of qc s is that the qubit energies can be often individually controlled @xcite . this corresponds to controllable disorder of a spin system , and it allows one to use qc s for studying a fundamentally important problem of how the spin - spin interaction affects spin dynamics in the presence of disorder . several models of qc s where the interqubit interaction is permanently `` on '' are currently studied . in these models the effective spin - spin interaction is usually strongly anisotropic . it varies from the essentially ising coupling @xmath1 in nuclear magnetic resonance and some other systems @xcite ( @xmath2 enumerate qubits , and @xmath3 is the direction of the magnetic field ) to the @xmath4-type ( i.e. , @xmath5 ) or the @xmath0-type ( i.e. , @xmath6 ) coupling in some josephson - junction based systems @xcite . the ising coupling describes the system in the case where the transition frequencies of different qubits are strongly different . then @xmath1 is the only part of the interaction that slowly oscillates in time , in the heisenberg representation , and therefore is not averaged out . if the qubit frequencies are close to each other , the terms @xmath7 become smooth functions of time as well . they lead to resonant excitation hopping between qubits . in a multiqubit system with close frequencies , both ising and @xmath4 interactions are present in the general case , but their strengths may be different @xcite . in this sense the @xmath0 coupling is most general , at least for qubits with high transition frequencies . the interqubit interaction often rapidly falls off with the distance and can be approximated by nearest neighbor coupling . many important results on anisotropic spin systems with such coupling have been obtained using the bethe ansatz . initially the emphasis was placed on systems without defects @xcite or with defects on the edge of a spin chain @xcite . more recently these studies have been extended to systems with defects that are described by integrable hamiltonians @xcite . however , the problem of a spin chain with several coupled excitations and with defects of a general type has not been solved . in this paper we investigate interacting excitations in an anisotropic spin system with defects . we show that the excitation localized on a defect does not decay even where the decay is allowed by energy conservation . we also find that , in addition to a single - particle excitation , a defect leads to the onset of two types of localized two - particle excitations . the analysis is done for a system with the @xmath0 coupling . the coupling anisotropy is assumed to be strong , as in the case of a qc based on electrons on helium , for example @xcite . the ground state of the system corresponds to all spins pointing in the same direction ( downwards , for concreteness ) . a single - particle excitation corresponds to one qubit being excited , or one spin being flipped . if the qubit energies are tuned in resonance with each other , a qc behaves as an ideal spin system with no disorder . a single - particle excitation is then magnon - type , it freely propagates through the system . in the opposite case where the qubit energies are tuned far away from each other ( as for diagonal disorder in tight - binding models ) , all single - particle excitations are localized . if the excitation density is high , the interaction between them may affect their localization , leading to quantum chaos , cf . understanding the interplay between interaction and disorder is a prerequisite for building a qc . we will consider the case where the excitation density is low , yet the interaction is important . in particular , excitations may form bound pairs ( but the pair density is small ) . one of the important questions is whether the interaction leads to delocalization of excitations . more specifically , consider an excitation , which is localized on a defect in the absence of other excitations . we now create an extended magnon - type excitation ( a propagating wave ) , that can be scattered off the localized one . the problem is whether this will cause the excitation to move away from the defect . we show below that , due to unexpected destructive quantum interference , the scattering does not lead to delocalization . we consider a one - dimensional array of qubits which models a spin-1/2 chain . for nearest - neighbor coupling , the hamiltonian is @xmath8 here , @xmath9 are the pauli matrices and @xmath10 . the parameter @xmath11 characterizes the strength of the exchange coupling , and @xmath12 determines the coupling anisotropy . we assume that @xmath13 ; for a qc based on electrons on helium , @xmath14 lies between @xmath15 and @xmath16 , for typical parameter values @xcite . we will consider effects due to a single defect . respectively , all on - site spin - flip energies @xmath17 are assumed to be the same except for the site @xmath18 where the defect is located , that is , @xmath19 in order to formulate the problem of interaction - induced decay of localized excitations , we preview in fig . [ fig : energies ] a part of the results on the energy spectrum of the system . in the absence of the defect , the energies of single - spin excitations ( magnons ) lie within the band @xmath20 , where @xmath21 ( the energy is counted off from the ground - state energy ) . the defect has a spin - flip energy that differs by @xmath22 ( a qubit with a transition frequency different from that of other qubits ) . it leads to a localized single - spin excitation with no threshold in @xmath22 , for an infinite chain . the energy of the localized state is shown by a dashed line on the left panel of fig . [ fig : energies ] . centered at @xmath23 . the dashed line indicates the energy of the excitation localized on the defect . right panel : the two - excitation energy spectrum . the band @xmath24 is formed by uncoupled magnons . the band centered at @xmath25 is formed by localized - delocalized pairs ( ldp s ) in which one excitation is localized on the defect and the other is in an extended state . the narrow band centered at @xmath26 is formed by propagating bound pairs ( bp s ) of excitations . the dashed lines show the energies of the states where both excitations are bound to the defect.,width=307 ] we now discuss excitations that correspond to two flipped spins . a defect - free @xmath0 system has a two - magnon band of independently propagating noninteracting magnons . however , the anisotropy of the exchange coupling leads also to the onset of bound pairs ( bp s ) of excitations . the bp band is much narrower than the two - magnon band and is separated from it by a comparatively large energy difference @xmath27 , see the right panel of fig . [ fig : energies ] . in the presence of a defect , there are two - excitation states with one excitation localized on the defect and the other being in an extended state . we call them localized - delocalized pairs ( ldp s ) . an interplay between disorder and interaction may lead to new types of states where both excitations are localized near the defect . their energies are shown in the right panel of fig . [ fig : energies ] by dashed lines . a localized one - spin excitation can not decay by emitting a magnon , by energy conservation . but it might experience an induced decay when a magnon is inelastically scattered off the excited defect into an extended many - spin state . magnon - induced decay is allowed by energy conservation when the total energy of the localized one - spin excitation and the magnon coincides with the energy of another two - particle state . in the @xmath0 model the total number of excitations ( flipped spins ) is conserved , and therefore decay is only possible into extended states of two bound magnons . in other words , it may only happen when the ldp band overlaps with the bp band in fig . [ fig : energies ] . decay into bp states may occur directly or via the two - excitation state located next to the defect . the amplitudes of the corresponding transitions turn out to be nearly equal and opposite in sign . as a result of this quantum interference , even though the band of bound magnons is narrow and has high density of states , the ldp to bp scattering does not happen , i.e. , the excitation on the defect is not delocalized . the bp to ldp scattering does not happen either , i.e. , a localized excitation is not created as a result of bp decay . in sec . ii below and in the appendix we briefly analyze localization of one excitation in a finite chain with a defect , for different boundary conditions . in sec . iii we discuss the two - excitation states localized near a defect . in sec . iv we consider the resonant situation where the energy band of extended bound pair states is within the band of energies of the flipped defect spin plus a magnon , i.e. , where the bp band overlaps with the ldp band in fig . [ fig : energies ] . we find that the localized excitation remains on the defect site in this case . analytical results for a chain with strong anisotropy @xmath14 are compared with numerical calculations . section v contains concluding remarks . in order to set the scene for the analysis of the two - excitation case , in this section and in the appendix we briefly discuss the well - known case of one excitation ( flipped spin ) in an @xmath0 spin chain with a defect @xcite and the role of boundary conditions . the hamiltonian of the chain with the defect on site @xmath28 has the form ( [ hamiltonian ] ) , ( [ chain_energies ] ) . we assume that the excitation energy @xmath29 largely exceeds both the coupling constant @xmath30 and the energy excess on the defect site @xmath31 . in this case the ground - state of the system corresponds to all spins being parallel , with @xmath32 irrespective of the signs of @xmath33 , and @xmath12 . without a defect , one - spin excitations are magnons . they freely propagate throughout the chain . the term in the hamiltonian ( [ hamiltonian ] ) responsible for one - excitation hopping is @xmath34 , with @xmath35^{\dagger}= \sigma_{n+1}^{+ } \sigma_{n}^{-}.\nonumber\end{aligned}\ ] ] the operators @xmath36 and @xmath37 cause excitation shifts @xmath38 and @xmath39 , respectively . a defect leads to magnon scattering and to the onset of localized states . both propagation of excitations and their localization are interesting for quantum computing . coherent excitation transitions allow one to have a qc geometry where remote `` working '' qubits are connected by chains of `` auxiliary '' qubits , which form `` transmission lines '' @xcite . localization , on the other hand , allows one to perform single - qubit operations on targeted qubit . a qc makes it possible to model spin chains with different boundary conditions . the simplest models are an open spin chain with free boundaries , which is mimicked by a finite - length array of qubits ( for electrons on helium , it can be implemented using an array of equally spaced electrodes , cf . ref . ) or a periodic chain , which can be mimicked by a ring of qubits . an open @xmath40-spin chain is described by the hamiltonian ( [ hamiltonian ] ) , where the first sum runs over @xmath41 and the second sum runs over @xmath42 ( the edge spins have neighbors only inside the chain ) . in what follows , we count energy off the ground - state energy @xmath43 , i.e. , we replace in eq . ( [ hamiltonian ] ) @xmath44 . the eigenfunctions of @xmath45 in the case of one excitation can be written as @xcite @xmath46 where @xmath47 corresponds to the spin on site @xmath48 being up and all other spins being down . the schrdinger equation for @xmath49 has the form @xmath50=e_1a(n ) , \label{eq_a}\end{aligned}\ ] ] where @xmath51 is the energy of a flipped spin in an ideal infinite chain in the absence of excitation hopping and @xmath52 is the one - excitation energy eigenvalue . for an open @xmath40-spin chain we set @xmath53 in eq . ( [ eq_a ] ) . the hamiltonian of a closed @xmath40-spin chain has the form ( [ hamiltonian ] ) , where both sums over @xmath48 go from 1 to @xmath40 and the site @xmath54 coincides with the site 1 . here , the defect location @xmath28 can be chosen arbitrarily . the wave function can be sought in the form ( [ one_exc ] ) . the schrdinger equation then has the form ( [ eq_a ] ) , except that there are no terms proportional to @xmath55 from the end points of the chain . it has to be solved with the boundary condition @xmath56 . for an open chain , the solution of the schrdinger equation ( [ eq_a ] ) can be sought in the form of plane waves propagating between the chain boundaries and the defect , @xmath57 the subscripts @xmath58 and @xmath59 refer to the coefficients for the waves to the left ( @xmath60 ) and to the right ( @xmath61 ) from the defect . the interrelations between these coefficients and the coefficient @xmath62 follow from the boundary conditions and from matching the solutions at @xmath28 . they are given by eqs . ( [ bound_cond ] ) and ( [ boundary_defect ] ) . for a closed chain , on the other hand , the solution can be sought in the form @xmath63 the energy @xmath52 as a function of @xmath64 can be obtained by substituting eqs . ( [ left_right ] ) and ( [ right ] ) into eq . ( [ eq_a ] ) . both for the open and closed chains it has the form @xmath65 the eigenfunctions ( [ left_right ] ) , ( [ right ] ) with real @xmath64 correspond to sinusoidal waves ( extended states ) . from ( [ energy_one ] ) , their energies lie within the band @xmath66 . in contrast , localized states have complex @xmath64 , and their energies lie outside this band . the corresponding solutions for both types of chains are discussed in the appendix . in a sufficiently long chain , there is always one localized one - excitation state on a defect . its energy is given by eq . ( [ e_d ] ) . in the case of an open chain with the coupling anisotropy parameter @xmath67 , there are also localized states on the chain boundaries . their energy is given by eq . ( [ e_s ] ) . a spin chain with a defect displays rich behavior in the presence of two excitations . it is determined by the interplay between disorder and inter - excitation coupling . solutions of the two - excitation problem have been obtained in the case of a disorder potential of several special types , where the system is integrable @xcite . here we will study the presumably nonintegrable but physically interesting problem where the on - site energy of the defect differs from that of the host sites . the system is described by hamiltonian ( [ hamiltonian ] ) . in order to concentrate on the effects of disorder rather than boundaries , we will consider a closed chain of length @xmath40 . we will also assume that the anisotropy is strong , @xmath68 . the wave function of a chain with two excitations is given by a linear superposition @xmath69 where @xmath70 is the state where spins on the sites @xmath48 and @xmath71 are pointing upward , whereas all other spins are pointing downward . in a periodic chain , the sites with numbers that differ by @xmath40 are identical , therefore we have @xmath72 . from eq . ( [ hamiltonian ] ) , the schrdinger equation for the coefficients @xmath73 is @xmath74(1-\delta_{m , n+1 } ) + a(n , m+1)\bigr ] = e_2a(n , m).\end{aligned}\ ] ] here , @xmath75 is the energy of a two - excitation state [ @xmath76 is the on - site one - excitation energy , cf . ( [ eq_a ] ) ] . as before , we assume that the defect is located on site @xmath28 . in the absence of a defect the system is integrable . the solution of eq . ( [ a(n , m ) ] ) can be found using the bethe ansatz @xcite , @xmath77 the energy of the state with given @xmath78 is obtained by substituting eq . ( [ two_bethe_ideal ] ) into eq . ( [ a(n , m ) ] ) written for @xmath79 . this gives @xmath80 by requiring that the ansatz ( [ two_bethe_ideal ] ) apply also for @xmath81 , we obtain an interrelation between the coefficients @xmath82 and @xmath83 , @xmath84 with account taken of normalization , eqs . ( [ two_bethe_ideal ] ) and ( [ cc ] ) fully determine the wave function . the states with real @xmath85 form a two - magnon band with width @xmath86 , as seen from the dispersion relation ( [ e2 ] ) . the magnons are not bound to each other and propagate independently . for @xmath87 , eq . ( [ cc ] ) also has a solution @xmath88 , which gives a complex phase @xmath89 with @xmath90 . this solution corresponds to the wave function @xmath91 $ ] ; we have @xmath92 . from eqs . ( [ two_bethe_ideal])([cc ] ) we obtain @xmath93 eq . ( [ bound ] ) describes a bound pair of excitations . such a pair can freely propagate along the chain . the wave function is maximal when the excitations are on neighboring sites . the size of the bp , i.e. the typical distance between the excitations , is determined by the reciprocal decrement @xmath94 , and ultimately by the anisotropy parameter @xmath12 . for large @xmath14 , the excitations in a bp are nearly completely bound to nearest sites . then the coefficients @xmath95 , to the lowest order in @xmath96 . the distance between the centers of the bp band and the two - magnon band @xmath97 is given approximately by the bp binding energy @xmath27 . the width of the bp band @xmath98 is parametrically smaller than the width of the two - magnon band @xmath86 , see fig . [ fig : energies ] . for nearest - neighbor coupling and for large @xmath14 , transport of bound pairs can be visualized as occurring via an intermediate step . first , one of the excitations in the pair makes a virtual transition to the neighboring empty site , and as a result the parallel spins in the pair are separated by one site . the corresponding state differs in energy by @xmath27 from the bound - pair state . at the next step the second spin can move next to the first , and then the whole pair moves by one site . from perturbation theory , the bandwidth should be @xmath99 , which agrees with eq . ( [ bound ] ) . the above arguments can be made quantitative by introducing an effective hamiltonian @xmath100 of bp s . it is obtained from the hamiltonian @xmath101 ( [ transfer_one ] ) in the second order of perturbation theory in the one - excitation hopping constant @xmath11 , @xmath102 \\ & + & \frac{j } { 64 \delta } [ t^{(l)}_{n } t^{(l)}_{n-1 } + t^{(r)}_{n } t^{(r)}_{n+1 } ] . \nonumber\end{aligned}\ ] ] the operators @xmath103 of excitation hopping to the right or left are given by eq . ( [ transfer_one ] ) . the first pair of terms in @xmath104 [ eq . ( [ transfer_bound ] ) ] describes virtual transitions in which a bp dissociates and then recombines on the same site . this leads to a shift of the on - site energy level of the bp by @xmath105 . the second pair of terms describes the motion of a bp as a whole to the left or to the right . the action of the hamiltonian @xmath106 on the wave function @xmath107 is given by @xmath108 . \nonumber\end{aligned}\ ] ] the schrdinger equation for a bp is given by the sum of the diagonal part ( the first term ) of eq . ( [ a(n , m ) ] ) with @xmath81 and the right - hand side of eq . ( [ transferbp ] ) . in this approximation a bp eigenfunction is @xmath109 , and the dispersion law is of the form ( [ bound ] ) . we now consider excitations in the presence of a defect . in this section we assume that the defect excess energy @xmath22 is such that @xmath110 cf . ( [ one_localized ] ) ] . the second condition in eq . ( [ nonresonant_cond ] ) can be understood by noticing that , in a chain with a defect , there is a two - excitation state where one excitation is localized on the defect whereas the other is in an extended magnon - type state . these excitations are not bound together . the energy of such an unbound localized - delocalized pair ( ldp ) should differ from the energy of unbound pair of magnons by @xmath111 , and from the energy of a bound pair of magnons ( [ bound ] ) by @xmath112 . in this section we consider the case where both these energy differences largely exceed the magnon bandwidth @xmath11 ( the case where @xmath113 will be discussed in the following section ) . in the neglect of excitation hopping , the energy of an excitation pair where one excitation is far from the defect ( @xmath114 ) and the other is localized on the defect is @xmath115 . [ this can be seen from eq . ( [ a(n , m ) ] ) for the coefficients @xmath73 in which off - diagonal terms are disregarded . ] at the same time , if one excitation is on the defect and the other is on the neighboring site @xmath116 , this energy becomes @xmath117 . the energy difference @xmath27 largely exceeds the characteristic bandwidth @xmath11 . therefore , if the excitation was initially far from the defect , it will be reflected before it reaches the site @xmath116 . the above arguments suggest to seek the solution for the wave function of an ldp in a periodic chain in the form @xmath118 with the boundary condition @xmath119 . this boundary condition and the form of the solution are similar to what was used in the problem of one excitation in an open chain . from eq . ( [ a(n , m ) ] ) , the energy of an ldp is @xmath120 the wave number @xmath64 takes on @xmath121 values @xmath122 , with @xmath123 . as expected , the bandwidth of the ldp is @xmath124 , as in the case of one - excitation band in an ideal chain . we have compared eq . ( [ energy_ldp ] ) with numerical results obtained by direct solution of the eigenvalue problem ( [ a(n , m ) ] ) . for @xmath125 we obtained excellent agreement once we took into account that the energy levels ( [ energy_ldp ] ) are additionally shifted by @xmath126 . this shift can be readily obtained from eq . ( [ a(n , m ) ] ) as the second - order correction ( in @xmath11 ) to the energy @xmath127 of the excitation localized on the defect . it follows also from eq . ( [ e_d ] ) for @xmath128 . we note that the result is trivially generalized to the case of a finite but small density of magnons . the wave function @xmath129 of @xmath130 uncoupled magnons and an excitation on the site @xmath28 is given by a sum of the appropriately weighted permutations of @xmath131 over the site numbers @xmath132 ( these numbers can be arranged so that @xmath133 ) , with real @xmath134 . the weighting factors for large @xmath14 are found from the boundary conditions and the condition that @xmath135 whenever any two numbers @xmath136 differ by 1 . when the ratio of the number of excitations @xmath130 to the chain length @xmath40 is small , the energy is just a sum of the energies of uncoupled magnons and the localized excitation , i.e. , it is @xmath137 . for small @xmath138 , scattering of a magnon by the excitation on a defect occurs as if there were no other magnons , i.e. , the probability of a three - particle collision is negligibly small . for large @xmath139 , a bound pair of neighboring excitations should be strongly localized when one of the excitations is on the defect site . indeed , if we disregard intersite excitation hopping , the energy of a bp sitting on the defect is @xmath140 . it differs significantly from the energy of freely propagating bp s ( [ bound ] ) , causing localization . the major effect of the excitation hopping is that the pair can make resonant transitions between the sites @xmath141 and @xmath142 . such transitions lead to splitting of the energy level of the pair into a doublet . to second order in @xmath11 the energies of the resulting symmetric and antisymmetric states are @xmath143 the energy splitting between the states is small if @xmath27 and @xmath22 have same sign . if on the other hand , @xmath144 , the theory has to be modified . here , the bound pairs with one excitation localized on the defect are resonantly mixed with extended unbound two - magnon states . we do not consider this case in the present paper . we note that , in terms of quantum computing , the onset of a doublet suggests a simple way of creating entangled states . indeed , by applying a @xmath145 pulse at frequency @xmath146 one can selectively excite the qubit @xmath28 . if then a @xmath145 pulse is applied at the frequency @xmath147 , it will selectively excite a bell state @xmath148/\sqrt{2}$ ] , where @xmath149 describes the state where the qubits on sites @xmath150 and @xmath151 are in the states @xmath152 and @xmath153 , respectively . the excitations on sites @xmath116 can then be separated without breaking their entanglement using two - qubit gate operations . the energy difference @xmath154 between bp states on the defect site ( [ doublet ] ) and extended bp states largely exceeds the bandwidth @xmath98 of the extended states . therefore it is a good approximation to assume that the wave functions of extended bp states are equal to zero on the defect . in other words , such bp s are reflected _ before _ they reach the defect . in this sense , the defect acts as a boundary for them . one may expect that there is a surface - type state associated with this boundary . the emergence of the surface - type state is facilitated by the defect - induced change of the on - site energy of a bp located next to the defect on the sites @xmath155 [ or @xmath156 . this change arises because virtual dissociation of a bp with one excitation hopping onto a defect site gives a different energy denominator compared to the case where a virtual transition is made onto a regular site . it is described by an extra term @xmath157 in the expression ( [ transfer_bound ] ) for the diagonal part of the hamiltonian @xmath104 with @xmath158 [ or @xmath159 , @xmath160 using the transformed hamiltonian ( [ transferbp ] ) , one can analyze bp states in a way similar to the analysis of one - excitation states in an open chain , see the appendix . the schrdinger equation for bp states away from the defect is given by the first term in eq . ( [ a(n , m ) ] ) and by eqs . ( [ transferbp ] ) and ( [ delta_bp ] ) . the bp wave functions can be sought in the form @xmath161 then the bp energy as a function of @xmath64 is given by @xmath162 , cf . ( [ bound ] ) . the values of @xmath64 can be found from the boundary condition that the bp wave function is equal to zero on the defect , i.e. , @xmath163 . from the schrdinger equations for @xmath107 with @xmath158 and @xmath164 , we obtain an equation for @xmath64 of the form @xmath165 ^ 2e^{i\theta ( n-3)}.\end{aligned}\ ] ] equation ( [ theta_bp ] ) has @xmath166 solutions for @xmath167 . the solutions @xmath168 are spurious , in the general case . the roots @xmath64 and @xmath169 describe one and the same wave function . therefore there are @xmath170 physically distinct roots @xmath64 , as expected for an @xmath40-spin chain with two excluded bp states located at @xmath171 . depending on the ratio @xmath172 , the roots @xmath64 are either all real or there is one or two pairs of complex roots with opposite signs . real roots correspond to extended states , whereas complex roots correspond to the states that decay away from the defect . the onset of complex roots can be analyzed in the same way as described in the appendix for one excitation . there is much similarity , formally and physically , between the onset of localized bp states next to the defect and the onset of surface states at the edge of an open chain . we rewrite eq . ( [ theta_bp ] ) in the form @xmath173={2\sin\theta \,(\cos \theta - q)\over ( \cos\theta - q)^2 - \sin^2\theta}.\ ] ] for @xmath174 all roots of eq . ( [ tangent_bp ] ) are real . at @xmath175 there occurs a bifurcation where two real roots with opposite signs merge ( at @xmath176 , for @xmath177 , or at @xmath178 , for @xmath179 ) . they become complex for @xmath180 . for @xmath181 two other real roots coalesce at @xmath176 or @xmath145 and another pair of complex roots emerges . as the length of the chain increases , the difference between the pairs of complex roots decreases . in the limit @xmath182 the roots merge pairwise . the imaginary part of one of the roots is@xmath183 the second root has opposite sign . one can show from the schrdinger equation for the bp s that the solution ( [ bp_s ] ) describes a `` surface - type '' bp state localized next to the defect on sites @xmath155 . the solution with @xmath184 describes the surface state on @xmath185 . the amplitudes of these states exponentially decay away from the defect . for @xmath186 or @xmath187 we have re @xmath188 . for @xmath189 or @xmath190 , we have re @xmath191 , and decay of the localized state is accompanied by oscillations . the complex roots in a finite chain can be pictured as describing those same states on the opposite sites of the defect . but now the states are `` tunnel '' split because of the overlap of their tails inside the chain , which leads to the onset of two slightly different localization lengths . the energy of the localized state in a long chain is @xmath192 it lies outside the band @xmath193 ( [ bound ] ) of the extended bp states . the distance to the band edge strongly depends on the interrelation between the defect excess energy @xmath22 and the bp binding energy @xmath27 . it is of the order of the bp bandwidth @xmath98 , except for the range where the difference between @xmath27 and @xmath22 becomes small . in this range the energy of the surface - type state sharply increases in the absolute value . this is illustrated in fig . [ fig : bp_s ] . the localization length @xmath194 is large when the state energy is close to the band edge and shrinks down with decreasing @xmath195 , i.e. , with increasing @xmath196 . , vs the scaled defect excess energy @xmath197 . the localized state exists in an infinite chain for @xmath198 . the vertical and horizontal dot - dashed lines show the asymptotes for @xmath199 and @xmath200 , respectively . the results refer to the range where the bp and ldp bands are far from each other , compared to the ldp bandwidth @xmath30 . these bands are sketched in the inset ( the bp band is above the ldp band for @xmath201 ) . the dashed line in the inset shows where the energy level of the localized bp state is located with respect to the bands.,width=268 ] the analysis of the preceding section does not apply if the pair binding energy @xmath27 is close to the defect excess energy @xmath22 . when @xmath202 is of the order of the ldp bandwidth @xmath30 , the bp and ldp states are in resonance , their bands overlap or nearly overlap with each other . one might expect that there would occur mixing of states of these two bands . in other words , a delocalized magnon in the ldp band might be scattered off the excitation on the defect , and as a result they both would move away as a bound pair . however , as we show , such mixing does not happen . in order to simplify notations we will assume in what follows that @xmath203 . in the resonant region we should reconsider the analysis of the next - to - the - defect bound pair localized on sites @xmath155 or @xmath185 . as @xmath202 decreases , the energy of this pair ( [ bp_s_energy ] ) moves away from the bp band , see fig . [ fig : bp_s ] . at the same time , the distance between the pair energy and the ldp band @xmath204 becomes _ smaller _ with decreasing @xmath202 as long as latexmath:[$|j\delta - g| > pairs are hybridized with ldp s . the hybridization occurs in first order in the nearest - neighbor coupling constant @xmath11 . to describe the hybridization , we will seek the solution of the schrdinger equation ( [ a(n , m ) ] ) in the form of a linear superposition of an ldp state ( [ ldp_wave ] ) and a pair on the next to the defect sites , @xmath206 ( we remind that @xmath207 ) . the energy of a state with given @xmath64 can be found from ( [ a(n , m ) ] ) as before by considering @xmath71 far away from @xmath28 . it is given by eq . ( [ energy_ldp ] ) . the interrelation between the coefficients @xmath208 , and @xmath209 , as well as the values of @xmath64 should be obtained from the boundary conditions . these conditions follow from the fact that the energy of a pair on sites @xmath171 and the energies of the pairs described by eq.([hybridization ] ) differ by @xmath210 , cf . ( [ doublet ] ) . therefore the pairs on sites @xmath171 are decoupled from the states ( [ hybridization ] ) , and in the analysis of the ldp s we can set @xmath211 . decoupled also are unbound two - excitation states with no excitation on the defect , i.e. , two - magnon states . therefore @xmath212 if simultaneously @xmath213 and @xmath214 . from eq . ( [ a(n , m ) ] ) written for @xmath215 and @xmath216 [ with account taken of the relation @xmath217 we obtain @xmath218 the interrelation between @xmath82 and @xmath83 and the equation for @xmath64 follow from eqs . ( [ hybridization ] ) , ( [ hybr_interrelation ] ) and ( [ a(n , m ) ] ) written for @xmath219 and @xmath220 . they have the form @xmath221 and @xmath222 ^ 2.\nonumber\end{aligned}\ ] ] equation ( [ hybr_theta ] ) is an @xmath223th order equation for @xmath167 . its analysis is completely analogous to that of eq . ( [ theta_bp ] ) . the roots @xmath224 are spurious , and the roots @xmath64 of opposite signs describe one and the same wave function . therefore eq . ( [ hybr_theta ] ) has @xmath225 physically distinct roots @xmath64 . real @xmath64 s correspond to extended states . complex roots appear for @xmath226 . these roots , @xmath227 , describe localized states . in the limit of a long chain , @xmath182 , the imaginary part of one of them is @xmath228 whereas the other root has just opposite sign . the wave function of the localized state is maximal either on sites @xmath155 or @xmath185 and exponentially decays into the chain . for @xmath229 this decay is accompanied by oscillations , re @xmath230 . the energy of the localized state is @xmath231 the localized state ( [ hybridization ] ) , ( [ hybr_decrem ] ) is a `` surface - type '' state induced by the defect . it is the resonant - region analog of the localized next - to - the - defect state discussed in section iii.b3 . the wave function of the latter state ( [ bpfull ] ) , ( [ bp_s ] ) was a linear combination of the wave functions of bound pairs . in contrast , the state given by eqs . ( [ hybridization ] ) , ( [ hybr_decrem ] ) is a combination of the wave functions of the bound pair located next to the defect and a localized - delocalized pair . the evolution of the surface - type state is controlled by the difference between the excess energies of binding two excitations in a pair or localizing one of them on the defect @xmath232 . as @xmath232 varies , the state changes in the following way . it first splits off the bp band when @xmath232 becomes less than @xmath22 , see fig . [ fig : bp_s ] . its energy moves away from the band of extended bp states with decreasing @xmath232 and the localization length decreases [ cf . ( [ bp_s ] ) ] . well before @xmath232 becomes of order @xmath11 , the state becomes strongly localized on the sites @xmath155 or @xmath233 . in the region @xmath234 the localized state becomes stronger hybridized with ldp states than with extended bp states . this hybridization occurs in first order in @xmath11 , via a transition @xmath235 [ or @xmath236 . in this region the localization length increases with decreasing @xmath232 , cf . ( [ hybr_decrem ] ) . ultimately , for @xmath237 the localized surface - type state disappears , as seen in fig . [ fig : ldp ] . the evolution of the energy of the localized state with @xmath238 is shown in fig . [ fig : ldp ] . , vs the energy mismatch @xmath239 . in the region between the dot - dashed lines the bp state is delocalized . the results refer to the case where the bp and ldp bands are close to each other or overlap , as sketched in the inset ( the bp band is above the ldp band for @xmath240 ) . the dashed line in the inset shows where the energy level of the localized bp state is located with respect to the bands.,width=268 ] the crossover from hybridization of the localized surface - type bp state with extended bp states to that with ldp states occurs in the region @xmath241 . it is described using a different approach in ref . . we note that the expressions for the energy of the localized states ( [ bp_s_energy ] ) and ( [ hybr_s_energy ] ) go over into each other for @xmath242 . we are now in a position to consider resonant coupling between extended states of bound pairs and localized - delocalized pairs . to lowest order in @xmath243 , a bp on sites @xmath244 far away from the defect can resonantly hop only to the nearest pair of sites , see eqs . ( [ transfer_bound ] ) , ( [ transferbp ] ) . as noted before , the hopping requires an intermediate virtual transition of the bp into a dissociated state , which differs in energy by @xmath245 . a different situation occurs for the bp on the sites @xmath246 [ or @xmath247 . such bp can hop onto the sites @xmath248 and @xmath249 , as described by ( [ transferbp ] ) . but in addition , for @xmath250 it can make a transition into the ldp state on sites @xmath251 [ or @xmath252 . indeed , such state has the energy @xmath253 , which is close to the bp energy @xmath254 . the transition @xmath255 goes through the intermediate dissociated state @xmath256 , which differs in energy by @xmath257 . it can be taken into account by adding the term @xmath258 to the bp hopping hamiltonian ( [ transferbp ] ) for the sites @xmath246 , @xmath259 extended bp states are connected to ldp states only through a bp on the sites @xmath246 . we are now in a position to analyze this connection . from eqs . ( [ transferbp ] ) and ( [ extra_h_t ] ) , we see that @xmath260a(n_0 + 2,n_0 + 3 ) = ( j/4\delta)\nonumber\\ & & \times[a(n_0,n_0 + 3 ) + a(n_0 + 1,n_0 + 2 ) ] + a,\end{aligned}\ ] ] where @xmath261 is a linear combination of the amplitudes @xmath262 and @xmath263 . the sum @xmath264 in eq . ( [ coupling_resonant ] ) can be expressed in terms of the ldp wave functions ( [ hybridization ] ) . with account taken of the interrelation ( [ hybr_cc ] ) between the coefficients @xmath265 in eq . ( [ hybridization ] ) , we have @xmath266.\end{aligned}\ ] ] equations ( [ coupling_resonant ] ) , ( [ amplitude_resonant ] ) describe the coupling between the bp on the sites @xmath246 and the ldp eigenstates with given @xmath64 . an important conclusion can now be drawn regarding the behavior of bp and ldp states in the resonant region . the center of the bp band lies at @xmath267 ( [ bound ] ) , and the bp band is parametrically narrower than the ldp band @xmath268 ( [ energy_ldp ] ) . when the bp band is inside the ldp band , this means that , for an appropriate wave number @xmath64 of the ldp magnon , @xmath269 , to zeroth order in @xmath243 ( this is the approximation used to obtain the ldp dispersion law ) . it follows from eqs . ( [ coupling_resonant ] ) and ( [ amplitude_resonant ] ) that , for such @xmath270 and @xmath64 there is no coupling between the ldp and extended bp states . the above result means that ldp and extended bp states do not experience resonant scattering into each other , even though it is allowed by the energy - conservation law . such scattering would correspond to the scattering of a magnon off the excitation localized on the defect , with both of them becoming a bound pair that moves away from the defect , or an inverse process . physically , the antiresonant decoupling of bp and ldp states is a result of strong mixing of a bp on the sites @xmath248 and ldp s . because of the mixing , the amplitudes of transitions of extended bp states to the sites @xmath248 and @xmath251 compensate each other , to lowest order in @xmath243 . to illustrate the antiresonant decoupling we show in fig . [ fig : temporal_nonresonant ] two types of time evolution of excitation pairs . in both cases the initial state of the system was chosen as a pair on the sites @xmath248 , i.e. , @xmath271 for @xmath272 . the time - dependent schrdinger equation was then solved with the boundary condition that corresponds to a closed chain . the solid lines refer to the case of nonoverlapping bp and ldp bands , @xmath273 and @xmath274 . in this case there does not emerge a localized surface - type bp state next to the defect . therefore an excitation pair placed initially on the sites @xmath248 resonantly transforms into extended bp states and propagates through the chain . this propagation is seen from the figure as oscillations of the return probability and the probability to find the bp on another arbitrarily chosen pair of neighboring sites @xmath246 . the oscillation rate should be small , of the order of the bandwidth @xmath275 . this estimate agrees with the numerical data . it is seen that the bp state is not transformed into ldp states . the amplitude of ldp states on sites @xmath276 remains extremely small , as illustrated for @xmath277 . the dotted lines in fig . [ fig : temporal_nonresonant ] show a completely different picture which arises when the bp band is inside the ldp band . in this case an excitation pair placed initially on @xmath248 hybridizes with ldp rather than bp states . a transformation of the pair @xmath248 into ldp s with increasing time is clearly seen . the period of oscillations is of the order of the reciprocal bandwidth of the ldp s @xmath278 , it is much shorter than in the previous case . remarkably , as a consequence of the antiresonance , extended states of bound pairs are not excited to any appreciable extent , as seen from the amplitude of the pair on the sites @xmath246 . chain with a defect ; @xmath279 is the occupation of sites @xmath280 . initially an excitation pair is placed next to the defect on sites @xmath248 . the solid and dashed lines refer to the cases where the bp and ldp bands are , respectively , far away from each other ( @xmath281 ) and overlapping ( @xmath282 ) . in the first case a bound pair slowly oscillates between neighboring sites @xmath283 and does not dissociate [ @xmath284 remains very small ] . in the case of overlapping bands , the pair at @xmath248 is hybridized with ldp s , but practically does not mix with bound pairs on other sites [ @xmath262 remains very small ] . the results refer to a ten - site closed chain with @xmath285.,width=307 ] we have analyzed the dynamics of a disordered spin chain with a strongly anisotropic coupling in a magnetic field . a defect in such a chain can lead to several localized states , depending on the number of excitations . this is a consequence of the interaction between excitations and its interplay with the disorder . we have studied chains with one and two excitations . the major results refer to the case of two excitations . here , the physics is determined by the interrelation between the excess on - site energy of the defect @xmath22 and the anisotropic part of the exchange coupling @xmath27 . strong anisotropy leads to binding of excitations into nearest - neighbor pairs that freely propagate in an ideal chain . because of the defect , bp s can localize . a simple type of a localized bp is a pair with one of the excitations located on a defect . a less obvious localized state corresponds to a pair localized next to a defect . it reminds of a surface state split off from the band of extended bp states , with the surface being the defect site . we specified the conditions where the localization occurs and found the characteristics of the localized states . our most unexpected observation is the antiresonant decoupling of extended bp states from localized - delocalized pairs . the ldp s are formed by one excitation on the defect site and another in an extended state . the antiresonance occurs for @xmath286 , when the bp and ldp bands overlap . it results from destructive quantum interference of the amplitudes of transitions of bp s into two types of resonant two - excitation states : one is an ldp , and the other is an excitation pair on the sites next to the defect . as a result of the antiresonant decoupling , extended bp s and ldp s do not scatter into each other , even though the scattering is allowed by energy conservation . this means that an excitation localized on the defect does not delocalize as a result of coupling to other excitations . the occurrence of multiple localized states in the presence of other excitations is important for quantum computing . it shows that , even where the interaction between the qubits is `` on '' all the time , we may still have well - defined states of individual qubits that can be addressed and controlled . one can prepare entangled localized pairs of excitations , as we discussed in sec . iii b 2 , or more complicated entangled excitation complexes . the results of the paper also provide an example of new many - body effects that can be studied using quantum computers with individually controlled qubit transition energies . this research was supported in part by the nsf through grant no . itr-0085922 and by the institute for quantum sciences at michigan state university . in an infinite spin chain in a magnetic field , the anisotropy of the spin - spin interaction does not affect the spectrum and wave functions of one excitation . the matrix element of the term @xmath287 in the hamiltonian ( [ hamiltonian ] ) is just a constant . however , the situation becomes different for a chain of finite length , because the coupling anisotropy can lead to surface states . in the case of two excitations , analogs of surface - type states emerge near defects in an infinite chain , as discussed in sec . iii . here , for completeness and keeping in mind a reader with the background in quantum computing , we briefly outline the results of the standard analysis of a finite - length spin chain with one excitation . the schrdinger equation for an excitation in an open chain has the form ( [ eq_a ] ) , and its solution @xmath288 to the left and to the right from the defect can be written in the form of a superposition of counterpropagating plane waves , eq . ( [ left_right ] ) . the relation between the amplitudes of these waves @xmath289 and @xmath290 follows from the boundary conditions @xmath291 . by substituting eq . ( [ left_right ] ) into eq . ( [ eq_a ] ) with @xmath292 and @xmath293 , we obtain @xmath294/[1-\delta \exp ( -i \theta ) ] .\nonumber\end{aligned}\ ] ] the relations between @xmath289 , @xmath290 , and the amplitude of the wave function on the defect site @xmath62 follow from eqs . ( [ eq_a ] ) and ( [ left_right ] ) for @xmath295 , @xmath296\nonumber\\ & & = c_{r } e^{i\theta n_0}\left[1 - d^{-1 } e^{2i\theta ( n+1-n_0)}\right ] .\end{aligned}\ ] ] with ( [ bound_cond ] ) and ( [ boundary_defect ] ) , all coefficients in the wave function ( [ left_right ] ) are expressed in terms of one number , @xmath62 . it can be obtained from normalization . in a finite chain , the values of @xmath64 are quantized . they can be found from eq . ( [ eq_a ] ) with @xmath18 . with account taken of ( [ boundary_defect ] ) , this equation can be written as @xmath297,\end{aligned}\ ] ] where @xmath298 .\ ] ] the analysis of the roots of eq . ( [ theta ] ) is standard . this is a @xmath299-order equation for @xmath167 , but its solutions for @xmath64 come in pairs @xmath64 and @xmath169 . each pair gives one wave function , as seen from eq . ( [ left_right ] ) . in addition , eq . ( [ theta ] ) has roots @xmath300 ; they are spurious ( unless @xmath301 and @xmath302 ) and appear as a result of algebraic transformations . therefore eq . ( [ theta ] ) has @xmath40 physically distinct roots , as expected for a chain of @xmath40 spins . we note that , for @xmath303 the position of the impurity @xmath28 drops out from eq . ( [ theta ] ) , and then the equation goes over into the result for an ideal chain . along with delocalized states , eq . ( [ theta ] ) describes also localized states with complex @xmath64 . complex roots of eq . ( [ theta ] ) can be found for a long chain , where @xmath304 . they describe surface states , which are localized on the chain boundaries , and a state localized on the defect . the localization length of the states is given by @xmath305 . the surface states arise only for the anisotropy parameter @xmath306 . the corresponding values of @xmath64 are @xmath307 here , the signs @xmath308 and @xmath309 refer to the states localized on the left and right boundaries , respectively , and @xmath310 is the step function . from eq . ( [ energy_one ] ) , the energy of the surface state is @xmath311 it lies outside the energy band of delocalized excitations . we note that , for @xmath312 the surface states decay monotonically with the distance from the boundary ( re @xmath313 ) . for sufficiently large negative @xmath12 , on the other hand , the decay of the wave function is accompanied by oscillations , and @xmath288 changes sign from site to site . a defect in a long chain gives rise to a localized one - spin excitation for an arbitrary excess energy @xmath22 @xcite . the amplitude @xmath288 decays away from the defect as @xmath314 the energy of the localized state is @xmath315 for small @xmath316 , we have i m @xmath317 , i.e. , the reciprocal localization length is simply proportional to the defect excess energy @xmath31 . in the opposite case of large @xmath316 we have i m @xmath318 . in this case the amplitude of the localized state rapidly falls off with the distance from the defect , @xmath319 . if the localization length is comparable to the chain length , the notion of localization is not well defined . however , when discussing numerical results , one can formally call a state localized if its wave function exponentially decays away from the defect and is described by a solution of eq . ( [ theta ] ) with complex @xmath64 . this is equivalent to the statement that the state energy @xmath127 lies outside the band of magnons in the infinite chain . in an open finite chain such localized state may emerge provided the localization length is smaller than the distance from the defect to the boundaries . this means that the defect excess energy @xmath31 should exceed a minimal value that depends on the size of the chain . the comparison of eq . ( [ one_localized ] ) with the numerical solutions of the full equation ( [ theta ] ) for a finite chain is shown in fig . [ fig : loc_length ] . for an infinite chain [ eq . ( [ one_localized ] ) ] as a function of the defect excess energy @xmath22 for @xmath320 ( solid line ) . also shown are the results for an open chain with the number of sites @xmath321 and 12 and @xmath322 ( stars and crosses , respectively ) and a closed chain with 6 and 12 sites ( circles and squares , respectively ) . they are obtained from eq . ( [ theta ] ) with @xmath323 and from eq . ( [ dispersion ] ) . in an open chain , solutions with nonzero i m @xmath324 emerge starting with a certain @xmath325 . in a closed chain with even @xmath40 there is no threshold in @xmath316 for the onset of states with i m @xmath326.,width=307 ] as pointed out in sec . ii , for a closed chain the solution of the schrdinger equation can be also sought in the form of counterpropagating waves with different amplitudes , eq . ( [ right ] ) . clearly , the phases @xmath64 and @xmath169 describe one and the same wave function . the one - excitation energy @xmath52 is given by eq . ( [ energy_one ] ) . the interrelation between the amplitudes of the waves @xmath327 in eq . ( [ right ] ) and the amplitude of the wave function on the defect site @xmath62 can be obtained from eq . ( [ eq_a ] ) with @xmath328 . this equation has two solutions , @xmath329 and @xmath330.\nonumber\end{aligned}\ ] ] in order to fully determine the wave function @xmath288 ( [ right ] ) , eqs . ( [ untouched ] ) , ( [ touched ] ) should be substituted into the schrdinger equation ( [ eq_a ] ) for @xmath18 . equations ( [ eq_a ] ) and ( [ untouched ] ) can be satisfied provided that either @xmath303 , which means that there is no defect , or @xmath331 . the first condition describes excitations in an ideal closed chain and is not interesting for the present paper . the condition @xmath331 corresponds to the wave function @xmath332 , which has a simple physical meaning . it is a standing wave in an ideal chain with a node at the location of a defect . because of the node , the corresponding state `` does not know '' about the defect , and therefore it is exactly the same as in an ideal chain . the equation for @xmath64 that follows from eqs . ( [ eq_a ] ) and ( [ touched ] ) has the form @xmath337 .\ ] ] for @xmath338 this equation has either @xmath339 ( for odd @xmath40 ) or @xmath340 ( for even @xmath40 ) solutions for @xmath341 . therefore the total number of solutions for @xmath64 that follow from eqs . ( [ untouched ] ) and ( [ dispersion ] ) is @xmath40 , as expected . by rewriting eq . ( [ dispersion ] ) as @xmath342 and plotting the left- and right - hand sides as functions of @xmath64 ( cf . , one can see that all physically distinct roots of this equation but one are real and lie in the interval @xmath343 [ except for one case , see below ] . such solutions describe delocalized states with sinusoidal wave functions . the complex root of eq . ( [ dispersion ] ) , @xmath344 , describes a state localized on the defect . for a long chain , i m @xmath345 , the solution has the form ( [ one_localized ] ) , as expected . an interesting situation occurs for a shorter chain . if @xmath40 is even or if @xmath346 , a localized solution with complex @xmath324 emerges for any defect excess energy @xmath22 . thresholdless localization does not happen in an open chain . in a closed chain , it arises because there is no reflection from boundaries . for small positive @xmath347 one obtains the complex solution of eq . ( [ dispersion ] ) in the form @xmath348 . the square - root dependence of i m @xmath324 on @xmath22 is seen from fig . [ fig : loc_length ] . equation ( [ dispersion ] ) has a complex solution also for even @xmath40 and small negative @xmath347 . in this case @xmath349 , i.e. , the decay of the wave function @xmath288 is accompanied by sign flips , @xmath350 . such oscillations can not be reconciled with the periodicity condition for odd @xmath40 . therefore , for odd @xmath40 and negative @xmath347 a decaying solution arises only when @xmath351 exceeds a threshold value . one can show from eq . ( [ dispersion ] ) that this value is @xmath352 , which has also been confirmed numerically . d. vion , a. aassime , a. cottet , p. joyez , h. pothier , c. urbina , d. esteve , and m.h . devoret , science * 296 * , 886 ( 2002 ) ; y. yu , s.y . han , x. chu , s.i . chu , and z. wang , _ ibid _ * 296 * , 889 ( 2002 ) ; y.a . pashkin , t. yamamoto , o. astafiev , y. nakamura , d.v . averin , and j.s . tsai , nature ( london ) * 421 * , 823 ( 2002 ) ; a.j . berkley , h. xu , r.c . ramos , m.a . gubrud , f.w . trauch , p.r . johnson , j.r . anderson , a.j . dragt , c.j . lobb , and f.c . wellstood , science * 300 * , 1548 ( 2003 ) . p.m. platzman and m.i . dykman , science * 284 * , 1967 ( 1999 ) ; m.i . dykman and p.m. platzman , fortschr . phys . * 48 * , 9 ( 2000 ) ; m.i . dykman , p.m. platzman , and p. seddighrad , phys . b * 67 * , 155402 ( 2003 ) . i.l . chuang , l.m.k . vandersypen , x. zhou , d.b . leung , and s. lloyd , nature ( london ) * 393 * , 143 ( 1998 ) ; l.m.k . vandersypen , m. steffen , g. breyta , c.s . yannoni , m.h . sherwood , and i.l . chuang , _ ibid _ * 414 * , 883 ( 2001 ) . cory , r. laflamme , e. knill , l. viola , t.f . havel , n. bouland , g. boutis , e. fortunato , s. lloyd , r. martinez , c. negrevergne , m. pravia , y. sharf , g. teklemariam , y.s . weinstein , and w.h . zurek , fortschr . * 48 * , 875 ( 2000 ) . p. chen , c. piermarocchi , and l.j . sham , phys . * 87 * , 067401 ( 2001 ) ; x. li , y. wu , d.g . steel , d. gammon , t.h . stievater , d.s . katzer , d. park , l.j . sham , and c. piermarocchi , science * 301 * , 809 ( 2003 ) .

we study analytically the time - dependent boundary conditions of superconducting microwave circuit experiments in the high plasma frequency limit , in which the conditions are robin - type and relate the value of the field to the spatial derivative of the field . we give an explicit solution to the field evolution for boundary condition modulations that are small in magnitude but may have arbitrary time dependence , in a formalism that applies both to a semiopen waveguide and to a closed waveguide with two independently adjustable boundaries . the correspondence between the microwave robin boundary conditions and the mechanically - moving dirichlet boundary conditions of the dynamical casimir effect is shown to break down at high field frequencies , approximately one order of magnitude above the frequencies probed in the 2011 experiment of wilson et al . our results bound the parameter regime in which a microwave circuit can be used to model relativistic effects in a mechanically - moving cavity , and they show that beyond this parameter regime moving mirrors produce more particles and generate more entanglement than their non - moving microwave waveguide simulations .

1411.2948

the quantum theory of relativistic fields with moving boundaries was first explored by moore in a remarkably original paper on the quantum formulation of linearly polarized light in a one - dimensional moving cavity @xcite . the primary result of this investigation was the discovery that moving mirrors in vacuum create photons . later , motivated by developments in quantum field theory in curved spacetimes , the specialization to a single moving mirror in minkowski spacetime was carried out by fulling and davies @xcite , who again found that non - uniformly accelerating mirrors generate radiation . these effects , in which particles are produced by moving boundaries , are generally referred to as the dynamical casimir effect ( dce ) or the nonstationary casimir effect @xcite . an experimental verification of the dce in a system with a mechanically moving boundary has remained elusive because of the technological challenges in creating sufficiently large accelerations @xcite . however , an experimental observation of a similar particle creation effect has been reported in a mechanically static semiconductor waveguide where the boundary condition on the field is modulated electrically , by a superconducting quantum - interference device ( squid ) @xcite . a related experimental observation of particle creation in josephson metamaterial has been reported in @xcite . these observations open fascinating prospects for simulating on a mechanically static desktop device quantum phenomena due to motion , including entanglement generation and degradation , in a regime where the moving system would experience significant relativistic effects @xcite . in this paper we address the evolution of a quantum field in waveguides of the type used in the experiment of @xcite in situations where the modulation of the squid(s ) at the end(s ) of the waveguide is small in magnitude but may have arbitrary time - dependence , under the further assumption that the plasma frequency of the squid(s ) is negligibly high compared with the frequencies where the experiment operates . more precisely , recall that the field @xmath0 in the experiment of @xcite satisfies , under the approximations described in @xcite , the @xmath1-dimensional klein - gordon equation @xmath2 with the boundary condition @xmath3 where the cavity is at @xmath4 , the squid is at @xmath5 , the meaning of the positive constants @xmath6 , @xmath7 , @xmath8 and @xmath9 is as described in @xcite , and @xmath10 can be given arbitrary time - dependence by modulating the magnetic field applied to the squid . in the regime where the squid s plasma frequency is large compared to the frequency of @xmath11 , the time derivative term in is negligible , and reduces to @xmath12 where @xmath13}^2 { \bigl[l_0 e_j(t)\bigr]}^{-1 } $ ] . we shall consider the regime in which the boundary condition applies and the time - dependence of @xmath14 is arbitrary in profile but small in magnitude . the regime where the squid s plasma frequency is large but not negligibly so is considered in @xcite . our interest in the boundary condition is twofold . first , considering the condition in its own right , we give an explicit solution to the quantum dynamics to leading perturbative order in the time variation of the boundary condition , using a formalism that allows us to handle both the semiopen waveguide and a closed waveguide , and allowing for the closed waveguide the modulations at the two ends to remain independent of each other . the mathematical observation underlying our formalism is that is a linear relation between the value of the field and the spatial derivative of the field , known as a robin boundary condition , and when this condition is independent of time , it has a well - known role in making the spatial part of the wave operator in self - adjoint and hence ensuring the unitarity of the time evolution @xcite . a time - dependent boundary condition can then be treated perturbatively by combining the spectral methods of @xcite to the techniques developed in @xcite for mechanically moving cavities . for the semiopen waveguide our results agree with those found in @xcite via a different formalism . second , we wish to address the sense in which models a mechanically moving boundary of the dce . for a dirichlet mirror at the time - dependent location @xmath15 , the boundary condition on the field reads @xmath16 if we choose in @xmath17 , reduces to for field frequencies much smaller than @xmath18 , but for higher field frequencies the correspondence no longer holds . we shall see that our perturbative solution of the field evolution with the condition indeed differs from the similar perturbative solution with the condition for frequencies that are not much smaller than @xmath18 , both for a semiopen waveguide and a closed waveguide ; in particular , the large frequency falloff properties of the solution are qualitatively different . simulations of relativistic motion with the mechanically static semiconductor waveguide would hence need to take place in the low frequency regime where the successive approximations from via to hold . both the experiment of @xcite and the proposals of @xcite appear to operate within in this domain by a margin of approximately one order of magnitude . the plan of the paper is as follows : section [ sec : half - open ] considers evolution under a small discontinuous change in the boundary condition for a semiopen waveguide , and section [ sec : closed ] presents the similar analysis for the closed waveguide . evolution under small changes in the boundary condition with arbitrary time - dependence for both types of waveguides is written out in section [ sec : cont - evolution ] . section [ sec : comparison ] compares the evolution to that under dirichlet boundary conditions at one or two mechanically moving boundaries . section [ sec : conclusions ] presents a summary and concluding remarks . appendix [ app : bogoidentities ] collects technical identities in a perturbative expansion of bogoliubov coefficients . appendix [ app : accboundary ] treats the evolution under the dirichlet boundary condition at one mechanically moving boundary in a small acceleration expansion in which velocities and travel distances are unrestricted , adapting to one moving boundary the treatment of mechanically rigid cavities given in @xcite . in appendix [ app : entanglement ] we derive the first order formula for the negativity measure of entanglement in the case when the modes have a continuous spectrum . in this section we discuss a semiopen waveguide under a sudden change in the robin boundary condition . subsection [ subsec : half - open - static ] establishes the notation and reviews the known properties of a time - independent robin boundary condition . the sudden change is implemented in subsection [ subsec : half - open - junction ] , and small sudden changes about the robin boundary condition relevant for the experiment of @xcite and about the dirichlet boundary condition are discussed respectively in subsections [ subsec : half - open - far - from - dir ] and [ subsec : half - open - near - dir ] . we adopt units in which the phase velocity @xmath6 in is set to unity . we may hence think of the field as a real scalar field @xmath19 on a @xmath1-dimensional minkowski spacetime , with the global minkowski coordinates @xmath20 and the metric @xmath21 . we take the boundary to be at @xmath5 and the field to live in the half @xmath22 . we consider the massive klein - gordon field equation @xmath23 where @xmath24 is the mass . the massless special case @xmath25 reduces to . we keep here @xmath26 general , in part because setting @xmath25 does not significantly simplify the analysis , but also in part in view of prospective future comparisons with mechanically moving cavities in situations where transverse dimensions may generate a positive @xmath26 by fourier decomposition @xcite . we introduce at @xmath5 the robin family of boundary conditions @xmath27 where @xmath28 is a constant independent of @xmath29 . the special case @xmath30 gives the dirichlet boundary condition , @xmath31 , the special case @xmath32 gives the neumann boundary condition , @xmath33 , and all other values of @xmath34 mix @xmath35 and @xmath36 . any choice for @xmath34 makes @xmath37 self - adjoint @xcite and would hence yield a consistent quantum theory of a non - relativistic particle on the half - line , although the choice @xmath30 may in that context be considered less fine - tuned than the others @xcite . in the present context of a relativistic quantum field theory , we consider only the values of @xmath34 that make the spectrum of @xmath38 strictly positive . for @xmath25 @xmath39 \cup \{\infty\}$ ] , and for @xmath40 it means @xmath39 \cup \{\infty\ } \cup ( \mu^{-1 } , \infty)$ ] . a negative eigenvalue of @xmath38 would give a tachyonic instability , and the zero eigenvalue that occurs when @xmath40 and @xmath41 would give a zero mode and hence a theory without a fock vacuum . the field is quantised in the usual fashion . a continuum of mode solutions that are positive frequency with respect to @xmath42 are @xmath43 where @xmath44 , @xmath45 , @xmath46 is determined as a function of @xmath47 from @xmath48 and we may fix the phase by choosing @xmath49 for @xmath50 and @xmath51 for @xmath32 . when @xmath40 and @xmath52 , there is in addition a discrete ground state mode , given by @xmath53 writing the klein - gordon inner product in the conventions of @xcite as @xmath54 where the overline denotes complex conjugation , the continuum modes @xmath55 are dirac - orthonormal , @xmath56 and when the discrete ground state @xmath57 exists , it is normalised and orthogonal to the continuum modes . the fock space is built on the vacuum that is annihilated by the annihilation operators associated with the continuum modes @xmath55 and with the discrete mode @xmath57 when the latter exists . suppose that for @xmath58 we use the field modes as introduced above and for @xmath59 we use a similar set of field modes with @xmath34 replaced by @xmath60 . denoting the new continuum modes by @xmath61 and the new discrete mode by @xmath62 , we may match the two sets of modes at @xmath63 in the notation of @xcite as [ eq : half - matching ] @xmath64 \tilde\phi_{g ' } & = \int_0^\infty \bigl ( { { { } _ { \text{o}}\alpha}}_{g ' k } \phi_k + { { { } _ { \text{o}}\beta}}_{g ' k } \overline{\phi_k } \ , \bigr ) \ , dk \ + { { { } _ { \text{o}}\alpha}}_{g ' g } \phi_g + { { { } _ { \text{o}}\beta}}_{g ' g } \overline{\phi_g } \ , \end{aligned}\ ] ] where we have included the lower left subscript @xmath65 in the bogoliubov coefficients @xmath66 and @xmath67 to indicate that the change in the boundary condition is sudden . expressions for the bogoliubov coefficients can be found by taking inner products of with the untilded modes and their complex conjugates @xcite . for the continuum - to - continuum bogoliubov coefficients , we find [ eq : half - c - alphas - and - betas ] @xmath68 { { { } _ { \text{o}}\beta}}_{k ' k } & = \frac{\sin\delta \sin\delta'}{\pi\sqrt{\omega \omega'}\,(\omega+\omega ' ) } \left(\frac{1}{d ' } - \frac{1}{d}\right ) \ , \label{eq : hw - beta - gen}\end{aligned}\ ] ] where @xmath46 and @xmath69 are determined by @xmath70 and @xmath71 in denotes the principal value in the integration over @xmath47 at @xmath72 . in the special cases @xmath30 and @xmath73 the formulas are understood in their well - defined limiting sense . the continuum - to - discrete and discrete - to - continuum bogoliubov coefficients will not be needed below and we omit the formulas . we note that while @xmath74 is a function for all @xmath47 and @xmath75 , @xmath76 has distributional support at @xmath72 because of the dirac delta in the first term and the principal value integral in the second term . we now consider the case where @xmath34 and @xmath60 are negative and close to each other . this is the situation relevant for the the waveguide experiment of @xcite . ( recall that in our conventions the field @xmath19 lives at @xmath77 , while the field @xmath11 in lives at @xmath4 . ) we write @xmath78 where @xmath79 is a positive constant of dimension length and @xmath80 is small . the mode expansion contains no discrete mode for any value of @xmath26 . expanding the bogoliubov coefficients in @xmath80 , we find [ eq : halfline - pert - bogos - far ] @xmath81}^2 } \right\ } \delta(k - k ' ) \notag \\[1ex ] & \hspace{3ex } + \frac{\eta \lambda k k ' } { \pi\sqrt{\omega \omega ' } \ , \sqrt{1 + { ( k\lambda)}^2}\ \sqrt{1 + { ( k'\lambda)}^2 } } \left[1 - \frac{\eta { ( k'\lambda)}^2}{1 + { ( k'\lambda)}^2}\right ] p\left(\frac{1}{\omega-\omega'}\right ) + o\bigl(\eta^3\bigr ) \ , \label{eq : halfline - pert - bogoalpha - far } \\[1ex ] { { { } _ { \text{o}}\beta}}_{k ' k } & = \frac{\eta \lambda k k ' } { \pi\sqrt{\omega \omega'}\,(\omega+\omega ' ) \sqrt{1 + { ( k\lambda)}^2}\ \sqrt{1 + { ( k'\lambda)}^2 } } \left[1 - \frac{\eta { ( k'\lambda)}^2}{1 + { ( k'\lambda)}^2}\right ] + o\bigl(\eta^3\bigr ) \ . \end{aligned}\ ] ] as a consistency check , we have verified that the expansion is consistent with the identities satisfied by the bogoliubov coefficients , collected in appendix [ app : bogoidentities ] , in the sense that the linear order identities hold and the off - diagonal part of the quadratic order identities holds . we are not aware of reasons to suspect inconsistencies in the diagonal part of but we have not undertaken the distributional analysis to examine this . as a second regime of interest , we perturb around the dirichlet boundary condition . this case is not directly relevant for the experiment of @xcite , but we shall see in section [ sec : comparison ] that it exhibits close theoretical similarity with a mechanically moving boundary . we set @xmath30 and assume @xmath60 to be close to @xmath82 . if @xmath60 is negative , the result may be obtained from by writing @xmath83 and letting @xmath84 while @xmath85 remains finite and negative but small : then @xmath86 , and the bogoliubov coefficients are given by [ eq : halfline - pert - bogos - near - dir ] @xmath87 \delta(k - k ' ) - \frac{b k k ' } { \pi\sqrt{\omega \omega ' } } \ , p\left(\frac{1}{\omega-\omega'}\right ) + o\bigl({b}^3\bigr ) \ , \label{eq : halfline - pert - bogoalpha - near - dir } \\[1ex ] { { { } _ { \text{o}}\beta}}_{k ' k } & = - \frac{b k k ' } { \pi\sqrt{\omega \omega'}\,(\omega+\omega ' ) } + o\bigl(b^3\bigr ) \ . \end{aligned}\ ] ] if @xmath60 is positive , the @xmath59 theory has a tachyonic instability ( for all positive @xmath60 if @xmath25 and for @xmath88 if @xmath40 ) because of the negative eigenvalue of @xmath38 ; however , the tachyon is nonperturbative in @xmath60 , and we have verified that just ignoring the tachyon and proceeding directly from leads again to , where now @xmath89 . as a consistency check , we have verified that satisfies the linear order bogoliubov identities and the off - diagonal part of the quadratic order bogoliubov identities , regardless the sign of @xmath85 . in this section we adapt the analysis of section [ sec : half - open ] to a cavity waveguide that is closed at both ends . for simplicity , we treat only the massless field , @xmath25 . we follow the notation of section [ sec : half - open ] , setting @xmath25 . we place the cavity at @xmath90 , where the positive constant @xmath91 is the length of the cavity . we write the static robin boundary conditions at the ends of the cavity as [ eq : boundarycondition ] @xmath92 where @xmath93 and @xmath94 are constants independent of @xmath29 , taking values in @xmath95 . any choice for @xmath93 and @xmath94 makes @xmath37 self - adjoint @xcite . to avoid instabilities and zero modes , we assume initially @xmath93 and @xmath94 to be such that the spectrum of @xmath37 is strictly positive . we shall however see in subsection [ subsec : closed - near - dir ] that a perturbative treatment in @xmath93 and @xmath94 remains consistent even in the presence of a nonperturbative tachyon , on a par with what happened for the semiopen waveguide in subsection [ subsec : half - open - near - dir ] . the field is again quantised in the usual fashion . the klein - gordon inner product is as in but integrated over @xmath96 $ ] . the equation for the eigenvalues can be written down using . we consider the case where @xmath97 and @xmath98 , and there is a sudden but small change in their values . this models a waveguide whose each end terminates at a squid as in the experiment of @xcite . for @xmath99 , we set @xmath100 and @xmath101 , where @xmath102 and @xmath103 are positive dimensionless constants . with this boundary condition @xmath37 is positive definite . the mode functions are @xmath104 where @xmath105 @xmath106 goes over the positive solutions to @xmath107 @xmath108 , and we choose the phase so that @xmath109 . there is exactly one @xmath106 in each interval @xmath110 , @xmath111 . the mode functions are normalised to @xmath112 . for @xmath59 , we set @xmath113 and @xmath114 , where @xmath115 and @xmath116 are dimensionless constants , assumed to be small . we work perturbatively in @xmath115 and @xmath116 , setting both of them to be proportional to a formal expansion parameter @xmath80 which at the end is set to unity . the mode functions are proportional to @xmath117 where @xmath47 and @xmath46 are determined from . we label the mode functions by the positive solutions to @xmath118 such that @xmath119 , we denote them by @xmath120 , and we choose their phase to agree with that of in the zeroth perturbative order . we then find that the eigenfrequencies are given by @xmath121 the bogoliubov coefficients are now defined by matching the modes at @xmath63 as @xmath122 proceeding as in @xcite , we find [ eq : cav - gen - bogos ] @xmath123 { { { } _ { \text{o}}\alpha}}_{pq } & = \frac{\bigl[\eta_1\sqrt{(1+\kappa_2 ^ 2p^2)(1+\kappa_2 ^ 2q^2 ) } -{(-1)}^{\varphi_p+\varphi_q } \eta_2 \sqrt{(1+\kappa_1 ^ 2p^2)(1+\kappa_1 ^ 2q^2)}\ , \bigr ] \sqrt{pq } } { ( p - q)\sqrt{f(p)f(q ) } } \notag \\[1ex ] & \hspace{3ex } + o\bigl(\eta^2\bigr ) \ \ \ \text{for $ p\ne q$}\ , \\[1ex ] { { { } _ { \text{o}}\beta}}_{pq } & = - \frac{\bigl[\eta_1\sqrt{(1+\kappa_2 ^ 2p^2)(1+\kappa_2 ^ 2q^2 ) } - { ( -1)}^{\varphi_p+\varphi_q } \eta_2 \sqrt{(1+\kappa_1 ^ 2p^2)(1+\kappa_1 ^ 2q^2)}\ , \bigr ] \sqrt{pq } } { ( p+q)\sqrt{f(p)f(q ) } } \notag \\[1ex ] & \hspace{3ex } + o\bigl(\eta^2\bigr ) \ , \end{aligned}\ ] ] where the map @xmath124 labels the consecutive solutions to by consecutive integers . the expressions for the order @xmath125 terms in and are lengthy and we suppress them here . as a consistency check , the linear terms in satisfy the linear order bogoliubov identities . we are not aware of reasons to suspect inconsistencies in the quadratic order identities but examining these would require a nontrivial evaluation of the left - hand side in and we have not carried out this evaluation . we consider also a perturbation around the dirichlet boundary condition . if @xmath126 and @xmath127 , the result may be obtained from simply by taking the limit @xmath128 and @xmath129 . the @xmath58 mode functions are given by @xmath130 where @xmath131 . using positive integers to label both the @xmath58 mode functions and the @xmath59 mode functions , we find that the @xmath59 eigenfrequencies are given by @xmath132 where @xmath133 , and the bogoliubov coefficients are given by [ eq : near - dir - bogos ] @xmath134 { { { } _ { \text{o}}\alpha}}_{mn } & = \frac{\bigl(\eta_1 - { ( -1)}^{m+n } \eta_2\bigr ) \sqrt{mn}}{(m - n ) } \notag \\[1ex ] & \hspace{3ex } - \frac{\bigl(\eta_1 - { ( -1)}^{m+n } \eta_2\bigr ) ( \eta_1-\eta_2 ) \ , n \sqrt{mn}}{{(m - n)}^2 } + o\bigl(\eta^3\bigr ) \ \ \ \text{for $ m\ne n$}\ , \\[1ex ] { { { } _ { \text{o}}\beta}}_{mn } & = - \frac{\bigl(\eta_1 - { ( -1)}^{m+n } \eta_2\bigr ) \sqrt{mn}}{(m+n ) } \notag \\[1ex ] & \hspace{3ex } - \frac{\bigl(\eta_1 - { ( -1)}^{m+n } \eta_2\bigr ) ( \eta_1-\eta_2 ) \ , n \sqrt{mn}}{{(m+n)}^2 } + o\bigl(\eta^3\bigr ) \ , \end{aligned}\ ] ] where we have now displayed also the order @xmath125 terms . if @xmath135 and/or @xmath136 , the @xmath59 theory may have tachyonic instabilities or zero modes ; however , both of these are nonperturbative , and we have verified that setting @xmath137 for @xmath58 , @xmath138 and @xmath139 for @xmath59 , ignoring any tachyons or zero modes , and working directly from and , yields and regardless the signs of @xmath115 and @xmath116 . as a consistency check , the expressions in can be verified to satisfy the perturbative bogoliubov identities . the elements of the matrix square on the left - hand side of are given by absolutely convergent sums that can be evaluated by residue techniques . when the boundary condition has arbitrary time dependence but the variations remain so small in magnitude that first - order perturbation theory suffices , the evolution of the field can be obtained by composing the sudden changes of sections [ sec : half - open ] and [ sec : closed ] and passing to the limit @xcite . we discuss first the semiopen waveguide and then the closed waveguide . for the semiopen waveguide of section [ sec : half - open ] , we consider a boundary condition of the form where @xmath34 may change in time but only within the interval @xmath140 . consider first the far - from dirichlet case . we write @xmath141 , where @xmath79 is a positive constant and the function @xmath142 is vanishing outside the interval @xmath140 and satisfies @xmath143 . at @xmath144 , we introduce early time basis modes that are as in but with the replacement @xmath145 . at @xmath146 , we similarly introduce late time basis modes that are as in but with the replacement @xmath147 . let @xmath148 and @xmath149 be the coefficients in the bogoliubov transformation from the early time modes to the late time modes . working perturbatively in @xmath80 , we may proceed as in @xcite , and the outcome can be read off from formulas ( 6 ) and ( 7 ) therein . we find [ eq : half - fardir - tcoeffs ] @xmath150 \beta_{k'k } & = e^{i\omega'(t_f - t_0 ) } \hat b_{k'k } + o\bigl(\eta^2\bigr ) \ , \end{aligned}\ ] ] where [ eq : half - fardir - pcoeffs ] @xmath151 \hat b_{k'k } & = \frac{i\lambda k k ' } { \pi\sqrt{\omega \omega ' } \ , \sqrt{1 + { ( k\lambda)}^2}\ \sqrt{1 + { ( k'\lambda)}^2 } } \int_{t_0}^{t_f } e^{-i(\omega ' + \omega)(t - t_0 ) } \ , \eta(t ) \ , dt \ . \end{aligned}\ ] ] when @xmath25 , and reduce to what was found in @xcite via a different formalism . in the near - dirichlet case , we take @xmath152 , where the function @xmath153 is vanishing outside the interval @xmath140 . proceeding as above , we find [ eq : half - neardir - tcoeffs ] @xmath154 \beta_{k'k } & = e^{i\omega'(t_f - t_0 ) } \hat b_{k'k } + o\bigl(b^2\bigr ) \ , \end{aligned}\ ] ] where [ eq : half - neardir - pcoeffs ] @xmath155 \hat b_{k'k } & = - \frac{i k k ' } { \pi\sqrt{\omega \omega ' } } \int_{t_0}^{t_f } e^{-i(\omega ' + \omega)(t - t_0 ) } \ , b(t ) \ , dt \ . \end{aligned}\ ] ] for the closed waveguide of section [ sec : closed ] , we consider a boundary condition of the form where @xmath93 and @xmath94 may change in time but only within the interval @xmath140 . we may proceed as above . the only new aspect is that the method of @xcite needs to be generalised to accommodate the linear term that appears in the frequencies and . in the far - from dirichlet case , we write @xmath156 and @xmath157 , where @xmath102 and @xmath103 are positive constants and the functions @xmath158 and @xmath159 are vanishing outside the interval @xmath140 and satisfy @xmath160 and @xmath161 . indexing the mode functions in the notation of subsection , writing @xmath162 and proceeding as above , we find that the coefficients in the bogoliubov transformation from the early time modes to the late time modes are given by [ eq : far - dir - cav - bogocontfull ] @xmath163 \beta_{pq } & = e^{i \omega_p(t_f - t_0 ) } \hat b_{pq } + o\bigl(\eta^2\bigr ) \ , \end{aligned}\ ] ] where [ eq : far - dir - cav - bogocontlin ] @xmath164 & \hspace{7ex } \left . - { ( -1)}^{\varphi_p+\varphi_q } \sqrt{(1+\kappa_1 ^ 2p^2)(1+\kappa_1 ^ 2q^2 ) } \int_{t_0}^{t_f } e^{-i(\omega_p - \omega_q)(t - t_0 ) } \ , \eta_2(t ) \ , dt \right ] \ , \\[1ex ] \hat b_{pq } & = - \frac{i\sqrt{pq } } { l\sqrt{f(p)f(q ) } } \left [ \sqrt{(1+\kappa_2 ^ 2p^2)(1+\kappa_2 ^ 2q^2 ) } \int_{t_0}^{t_f } e^{-i(\omega_p + \omega_q)(t - t_0 ) } \ , \eta_1(t ) \ , dt \right . \notag \\[1ex ] & \hspace{7ex } \left . - { ( -1)}^{\varphi_p+\varphi_q } \sqrt{(1+\kappa_1 ^ 2p^2)(1+\kappa_1 ^ 2q^2 ) } \int_{t_0}^{t_f } e^{-i(\omega_p + \omega_q)(t - t_0 ) } \ , \eta_2(t ) \ , dt \right ] \ . \end{aligned}\ ] ] in the near - dirichlet case , we write @xmath165 and @xmath166 . indexing the past and future mode functions by positive integers in the notation of subsection , and writing @xmath167 we find [ eq : near - dir - cav - bogocontfull ] @xmath168 \beta_{mn } & = e^{i \omega_m(t_f - t_0 ) } \hat b_{mn } + o\bigl(\eta^2\bigr ) \ , \end{aligned}\ ] ] where [ eq : near - dir - cav - bogocontlin ] @xmath169 \hat b_{mn } & = - \frac{i\pi \sqrt{mn}}{l } \int_{t_0}^{t_f } e^{-i(\omega_m + \omega_n)(t - t_0 ) } \ , \bigl(\eta_1(t ) - { ( -1)}^{m+n } \eta_2(t ) \bigr ) \ , dt \ . \end{aligned}\ ] ] we are now ready to compare the evolution under the time - dependent robin boundary condition to the evolution under the dirichlet condition at a mechanically moving boundary . for the semiopen waveguide , the evolution of a field with mass @xmath24 under the time - dependent robin boundary condition was given in subsection [ subsec : cont - halfline ] . the evolution of a massless field under a dirichlet condition at a mechanically moving boundary is given in appendix [ app : accboundary ] in terms of the acceleration of the boundary , in a small acceleration approximation that allows the velocity and the travel distance to remain arbitrary and overlaps with the dce literature results @xcite in the common domain of validity . comparing and , we see that the massless field with the mechanically moving boundary can be simulated to the leading order in perturbation theory by the @xmath25 near - dirichlet robin boundary condition provided we choose @xmath153 so that @xmath170 , where @xmath171 is the proper acceleration of the boundary as a function of its proper time @xmath172 , and the modulation starts and ends so gently that both @xmath85 and @xmath173 vanish . this is precisely the relation one would have expected from the low frequency equivalence between the robin boundary condition and the mechanically moving dirichlet boundary condition . the simulation is reliable in the frequency range where the first - order perturbation theory results on both the robin side and on the mechanical side remain reliable ; we shall not attempt to quantify this range more precisely , but for given @xmath171 and @xmath174 the range is unlikely to include arbitrarily high frequencies . comparing further and with , we see that the massless field with the mechanically moving boundary can be simulated by the @xmath25 far - from - dirichlet robin boundary condition provided we choose @xmath142 so that @xmath175 , the frequencies are much smaller than @xmath176 , and the modulation starts and ends so gently that both @xmath80 and @xmath177 vanish . again , this is precisely the outcome one would have expected from and . when the frequencies are not much smaller than @xmath176 , the evolution with the robin boundary condition differs from the evolution with the mechanically moving boundary because the square root factors in differ from unity . investigating these differences further , recall that the beta coefficients are directly related to the total photon production number by : @xmath178 taking @xmath142 to be a sinusoidal function : @xmath179 \ , \ ] ] with positive constant @xmath180 and driving frequency @xmath181 , the integrals in can be performed exactly . the dominant part of the photon flux occurs for frequencies below the driving frequency . we therefore define the photon flux density , @xmath182 , as a function of the reduced frequency @xmath183 , by @xmath184 so that @xmath185 figure [ fig : photonflux ] presents numerical plots of the photon flux density for @xmath186 and @xmath187 . in both cases the spectrum has in the range @xmath188 a distinctive parabolic shape that is qualitatively characteristic of the dce , and for @xmath186 there is good agreement with the zero temperature curve of figure 8 in @xcite . for a quantitative comparison with the dce , figure [ fig : photonflux ] presents also the photon flux density for the mechanically moving dirichlet boundary , obtained from with the matching @xmath189 . when the effective length , @xmath79 , is much smaller than the effective wavelength of the driving frequency ( @xmath186 ) , the moving boundary and robin - boundary systems produce almost identical spectral functions , even though the modulation used for the plot has nonvanishing @xmath190 at the start and end . on the other hand , when the effective length is larger than the effective wavelength of the driving frequency ( @xmath187 ) , the spectral functions from the moving boundary system and the robin - boundary systems differ : the radiation from the moving boundary has a higher intensity , and the spectrum from the robin boundary condition decays more rapidly at high frequencies . both of these behaviours result from the square root factors which suppress the beta coefficients in . photon flux densities @xmath182 for the semiopen waveguide in terms of the reduced frequency @xmath191 for moving boundary ( red dashed ) and non - moving time - varying robin boundary ( blue solid ) . ( left ) in the small effective length limit @xmath186 the two systems display near identical behavior . ( right ) in the large effective length limit @xmath187 the moving and non - moving systems have different photon production profiles . in both plots we use the values ( in units with @xmath192 ) , @xmath193 , @xmath194mm@xmath195 and @xmath196 mm . in the left plot we take @xmath197 mm and in the right plot we take @xmath198 mm . the parameters in the left plot can be matched to those suggested in @xcite by taking the propagation velocity @xmath199mm / s and noting that @xmath200 , @xmath201.,title="fig:",scaledwidth=50.0% ] photon flux densities @xmath182 for the semiopen waveguide in terms of the reduced frequency @xmath191 for moving boundary ( red dashed ) and non - moving time - varying robin boundary ( blue solid ) . ( left ) in the small effective length limit @xmath186 the two systems display near identical behavior . ( right ) in the large effective length limit @xmath187 the moving and non - moving systems have different photon production profiles . in both plots we use the values ( in units with @xmath192 ) , @xmath193 , @xmath194mm@xmath195 and @xmath196 mm . in the left plot we take @xmath197 mm and in the right plot we take @xmath198 mm . the parameters in the left plot can be matched to those suggested in @xcite by taking the propagation velocity @xmath199mm / s and noting that @xmath200 , @xmath201.,title="fig:",scaledwidth=50.0% ] the plots show the negativity function @xmath202 ( see appendix [ app : entanglement ] ) with @xmath203 and @xmath204 in terms of @xmath205 for moving boundary ( red dashed ) and non - moving time - varying robin boundary ( blue solid ) . ( left ) in the small effective length limit @xmath186 the two systems display near identical behavior . ( right ) in the large effective length limit @xmath187 the moving system produces more entanglement than the non - moving system . we use the same parameters as in fig . [ fig : photonflux].,title="fig:",scaledwidth=50.0% ] the plots show the negativity function @xmath202 ( see appendix [ app : entanglement ] ) with @xmath203 and @xmath204 in terms of @xmath205 for moving boundary ( red dashed ) and non - moving time - varying robin boundary ( blue solid ) . ( left ) in the small effective length limit @xmath186 the two systems display near identical behavior . ( right ) in the large effective length limit @xmath187 the moving system produces more entanglement than the non - moving system . we use the same parameters as in fig . [ fig : photonflux].,title="fig:",scaledwidth=50.0% ] one of the signatures of the dce radiation is that emitted photons are entangled . we can quantify the amount of entanglement produced using the negativity @xcite which is defined as minus the sum of the negative eigenvalues of the partially transposed state . we show in appendix [ app : entanglement ] that for perturbative bogoliubov coefficients of the type the leading term in the negativity between sharply peaked field modes @xmath47 and @xmath75 is given by @xmath206 where @xmath207 is the spectral line - width of the frequencies . for the driving function given in equation , @xmath208 is highest for values of @xmath209 near the driving frequency . therefore , we define @xmath210 , and we expand the frequencies near half driving - frequency as : @xmath211 the results for @xmath208 as a function of @xmath205 are shown in fig . [ fig : negativity ] . these plots are qualitatively similar to those of fig . 14 in @xcite for the `` without resonator '' line . we again observe good agreement when @xmath186 between moving mirrors and non - moving microwave waveguides . also , when @xmath187 we find that the moving mirrors generate more entanglement than their associated non - moving microwave waveguide simulations . the entanglement considered above is entanglement between the field modes . this entanglement could be observed directly with a homodyne setup by coupling the emitted radiation into an optics circuit with stationary photodetectors . another possibility would be to harvest the entanglement by inserting localised detectors , such as an atomic qubit . this has been discussed in a range of settings in @xcite and the references therein . for the closed waveguide , the evolution of a massless field under the time - dependent robin boundary condition was given in subsection [ subsec : cont - cavity ] . the evolution of a massless field in a moving , mechanically rigid cavity of proper length @xmath91 with dirichlet boundary conditions is given by @xcite [ eq : acc - cav - cont - bogofulls ] @xmath212 \beta_{mn } & = e^{i\omega_m(\tau_f-\tau_0 ) } \hat b_{mn } + o\bigl(a^2\bigr ) \ , \end{aligned}\ ] ] where [ eq : acc - cav - cont - bogolins ] @xmath213 \hat a_{mn } & = -\frac{i \pi \sqrt{mn } \bigl(1 - { ( -1)}^{m+n}\bigr)}{l^2 { ( \omega_m-\omega_n)}^2 } \ , \int_{\tau_0}^{\tau_f } e^{-i(\omega_m - \omega_n)(\tau-\tau_0 ) } \ , a(\tau ) \ , d\tau \ \ \ \text{for $ m\ne n$ } \ , \\[1ex ] \hat b_{mn } & = \frac{i \pi \sqrt{mn } \bigl(1 - { ( -1)}^{m+n}\bigr)}{l^2 { ( \omega_m+\omega_n)}^2 } \ , \int_{\tau_0}^{\tau_f } e^{-i(\omega_m + \omega_n)(\tau-\tau_0 ) } \ , a(\tau ) \ , d\tau \ , \end{aligned}\ ] ] @xmath214 and @xmath215 are positive integers , @xmath216 and @xmath217 are given by , @xmath172 is the proper time at the centre of the cavity and @xmath171 is the proper acceleration at the centre of the cavity . comparing and , we see that the massless field in the rigidly moving cavity can be simulated to the leading order in perturbation theory by the near - dirichlet robin boundary condition provided we choose @xmath218 and @xmath219 , and the modulation starts and ends so gently that both @xmath80 and @xmath190 vanish . again , this is precisely the relation one would have expected from and , and the simulation is reliable in the frequency range where the first - order perturbation theory results on both sides are reliable . comparing further with , we see that the rigidly moving cavity can be simulated by the far - from - dirichlet robin boundary condition with @xmath218 and @xmath219 when the frequencies are much smaller than @xmath220 and @xmath221 and the modulation starts and ends so gently that both @xmath80 and @xmath190 vanish . we anticipate that the simulation can be extended to a moving cavity that is not rigid , in the small amplitude regime commonly considered in the dce literature @xcite , by equating @xmath222 ( respectively @xmath223 ) to the variation in the position of the left ( right ) boundary . we have however not examined this question systematically . in this paper we have analysed the evolution of a quantum field in @xmath224 dimensions under time - dependent robin boundary conditions that occur in superconducting microwave circuit experiments in the high plasma frequency limit . we solved the evolution explicitly to linear order in the time variation of the robin boundary condition , in a formalism that allowed us to handle both a semiopen waveguide and a closed waveguide , and for the latter allowing the boundary conditions at the two ends to be varied independently . for frequencies that are much smaller than the effective inverse length associated with the squid(s ) at the end(s ) of the waveguide , we verified that a suitable modulation of the squid(s ) allows the waveguide to simulate a dirichlet boundary condition at a mechanically moving boundary of the dce , even in the regime where the mechanical motion is relativistic ; both the experiment reported in @xcite and the experimental proposals of @xcite appear to operate within in this domain by a margin of approximately one order of magnitude . for higher frequencies the waveguide still exhibits particle creation and mode mixing , but these can no longer be quantitatively matched to those of the dce , and in particular the large frequency falloff properties in the evolution are qualitatively different . these features in the large frequency bogoliubov coefficients result in differing particle emission spectrum for the moving and non - moving systems when the effective length is larger than the inverse driving frequency @xmath225 . in this limit , mechanically moving boundary radiation can be characterised as having a larger total flux and a less steep falloff at high frequencies compared to radiation from the static waveguide with time - varying robin boundary conditions . while the analogy between a moving dirichlet boundary and a time - varying robin boundary condition is useful for simulation purposes it is important to keep in mind that the physical systems corresponding to these two situations are different and therefore can lead to different outcomes . on the one hand , our results support proposals to simulate in a mechanically static waveguide quantum phenomena due to motion , including entanglement generation and degradation , even in a regime where the mechanically moving system experiences significant relativistic effects @xcite . on the other hand , our results demonstrate that the interpretation of a waveguide experiment in terms of the simulation of the dce is possible only in certain parameter ranges . within the robin boundary condition approximation that we have analysed , the range of dce interpretation is determined just by the effective inverse length scale set by the squid(s ) at the end(s ) of the waveguide . it remains a subject to future work to establish the range of dce interpretation when all relevant effects beyond the robin boundary condition approximation are considered @xcite , including the effects due to a large but finite squid plasma frequency analysed in @xcite . j.l . thanks ivette fuentes , joel lindkvist and carlos sabn for helpful discussions and correspondence on waveguides and david bruschi for helpful discussions on equations . we thank giuliano benenti , simone felicetti , diego mazzitelli and hector silva for bringing earlier work to our attention . j.d . was supported in the early stages of the work by epsrc ( career acceleration fellowship grant ep / g00496x/2 to ivette fuentes ) . j.l . was supported in part by stfc ( theory consolidated grant st / j000388/1 ) . in this appendix we record identities satisfied by a perturbative expansion of the bogoliubov coefficients . these identities are used in the main text for consistency checks of the perturbative treatment . let @xmath226 and @xmath227 denote the matrices of a bogoliubov transformation in the notation of @xcite , with the matrix elements @xmath228 and @xmath229 . the indices may be discrete or continuous ; in the latter case the matrix is understood as the kernel of an integral operator and may include a distributional part . by construction , the matrices satisfy the bogoliubov identities [ eq : bogo - idents ] @xmath230 where @xmath231 stands for the identity matrix . suppose now that @xmath226 and @xmath227 have the formal power series expansions [ eq : bogo - mat - expansions ] @xmath232 where @xmath233 is a real - valued expansion parameter . substituting in and collecting terms order by order , order @xmath234 is identically satisfied while orders @xmath235 and @xmath236 give [ eq : bogo - pert - complex12 ] @xmath237 \epsilon^2 : & \ \ \begin{cases } 0 = \alpha_1 \alpha_1^\dagger - \beta_1 \beta_1^\dagger + \alpha_2 + \alpha_2^\dagger \ , \\ 0 = \beta_1 \alpha_1^t - \alpha_1 \beta_1^t + \beta_2 - \beta_2^t \ . \end{cases}\end{aligned}\ ] ] when @xmath226 and @xmath227 are real , simplifies to [ eq : bogo - pert - real12 ] @xmath238 \epsilon^2 : & \ \ { ( \alpha_1 \pm \beta_1)}^2 = \alpha_2 + \alpha_2^t \pm \bigl ( \beta_2 - \beta_2^t \bigr ) \ . \label{eq : bogo - pert - real2}\end{aligned}\ ] ] in this appendix we consider a quantised massless scalar field in @xmath1-dimensional minkowski spacetime subject to the dirichlet boundary condition at one accelerated boundary , in the limit where the acceleration is treated perturbatively but may have arbitrary time - dependence , and the velocity and travel distance of the boundary remain arbitrary . we follow the methods that were developed for a mechanically rigid accelerated cavity in @xcite . the results overlap with those in the dce literature @xcite for small amplitude oscillations in the common domain of validity . the corresponding problem for a classical scalar field has been analysed in @xcite . we work with a massless scalar field @xmath19 in @xmath1-dimensional minkowski spacetime , in the notation of the main text : the metric is @xmath239 , and the wave equation is with @xmath25 . for @xmath58 , we take the field to live in the half - space @xmath240 , where @xmath241 is a positive constant of dimension inverse length , and we adopt at @xmath242 the dirichlet boundary condition . we adopt the basis of mode functions @xmath243 \ , \label{eq : r - staticmode}\end{aligned}\ ] ] where @xmath44 . @xmath55 has the positive frequency @xmath47 with respect to @xmath42 , and the normalisation in the klein - gordon inner product is @xmath244 . for @xmath245 , we make the boundary follow the uniformly - accelerated worldline @xmath246 . the proper acceleration of the boundary is equal to @xmath241 . the field lives in the region @xmath247 , and we take the field to satisfy the dirichlet boundary condition at the accelerated boundary . we adopt the basis of mode functions @xmath248^{i k / a } - \bigl[a(x+t)\bigr]^{-i k / a } \right\ } \ , \label{eq : r - rmode}\end{aligned}\ ] ] where @xmath44 . @xmath249 has the positive frequency @xmath250 with respect to the boost killing vector @xmath251 , and it has the positive frequency @xmath47 with respect to the proper time of the boundary . the normalisation in the klein - gordon inner product is @xmath252 . at the junction @xmath63 , we match the two sets of modes by the bogoliubov transformation @xmath253 from the inner products that give the bogoliubov coefficients @xcite , we find [ eq : mr - alphabeta - int ] @xmath254 \ , \label{eq : mr - alpha - int } \\[1ex ] { { { } _ { \text{o}}\beta}}_{k'k } & = \frac{1}{\pi a } \sqrt{\frac{k'}{k } } \ , \operatorname{re}\left [ \int_1^\infty \frac{dy}{y } \ , y^{ik'/a } \ , e^{i(k / a)(y-1 ) } \right ] \ . \label{eq : mr - beta - int}\end{aligned}\ ] ] we wish to find the asymptotic form of @xmath76 and @xmath74 when the acceleration @xmath241 of the boundary is small compared with both @xmath47 and @xmath75 . the small @xmath241 expansion of @xmath74 may be obtained by applying to the method of repeated integration by parts @xcite . the result is @xmath255 the small @xmath241 expansion of @xmath76 is more involved since we expect the coefficients in this expansion to be no longer functions but distributions , the leading term being @xmath256 . we shall not give a rigorous treatment but proceed heuristically as follows . starting from the integral in , we replace @xmath47 in the integrand by @xmath257 where @xmath258 . if @xmath233 is considered fixed , the asymptotic small @xmath241 expansion can be obtained by the method of repeated integration by parts @xcite , with the result @xmath259 + a \operatorname{re}\left [ \frac{k+i\epsilon } { { ( k - k ' + i\epsilon)}^3 } \right ] \vphantom{\left [ \frac{(k+i\epsilon ) \bigl[2(k+i\epsilon)+k'\bigr]}{{(k - k ' + i\epsilon)}^4 } \right ] } \right . \notag \\[1ex ] & \hspace{12ex } + \left . a^2 \operatorname{im}\left [ \frac{(k+i\epsilon ) \bigl[2(k+i\epsilon)+k'\bigr]}{{(k - k ' + i\epsilon)}^5 } \right ] + o\bigl(a^{3}\bigr ) \right\ } \ . \label{eq : mr - alpha-2nd - raw}\end{aligned}\ ] ] each of the three terms shown in has a well - defined distributional limit as @xmath260 , as follows from the identity @xmath261 and its derivatives . assuming that the @xmath260 limit commutes with the small @xmath241 expansion , we obtain @xmath262 where the distributions @xmath263 and @xmath264 are given by @xmath265 \int_0^\infty dk \ , g_2(k',k ) \ , f(k ) & = - \frac{{(k')}^{1/2}}{12 } \ , \partial^4_{k ' } \left ( { ( k')}^{3/2 } f(k')\right ) - \frac{{(k')}^{3/2}}{24 } \ , \partial^4_{k ' } \left ( { ( k')}^{1/2 } f(k')\right ) \ . \end{aligned}\ ] ] both @xmath263 and @xmath264 are hence representable by a function except at the coincidence limit , and we may write @xmath266 g_2(k',k ) & = 0 \hspace*{-15ex } & \text{for $ k\ne k'$ } \ . \end{aligned}\ ] ] as a consistency check , the expansions in and satisfy the linear order bogoliubov identities . the quadratic order bogoliubov identities would require a distributional treatment and we shall not analyse them here . let @xmath172 denote the proper time of the boundary and @xmath171 the proper acceleration of the boundary , such that positive ( negative ) values of @xmath171 mean acceleration towards increasing ( decreasing ) @xmath267 . we assume that @xmath171 is vanishing outside the interval @xmath268 and non - negative within this interval , and we assume that @xmath171 remains much smaller than the frequencies to be considered . we define the early ( respectively late ) time modes by in the inertial frame in which the boundary is at rest in the early ( late ) times . proceeding as in @xcite , or in section [ sec : cont - evolution ] of the present paper , we find that the bogoliubov coefficients from the early time modes to the late time modes are [ eq : half - rel - cont - bogos ] @xmath269 \beta_{k'k } & = e^{ik'(\tau_f-\tau_0 ) } \hat b_{k'k } + o\bigl(a^2\bigr ) \ , \end{aligned}\ ] ] where [ eq : half - rel - cont - bogolins ] @xmath270 \hat b_{k'k } & = \frac{i \sqrt{k k'}}{\pi { ( k+k')}^2 } \ , \int_{\tau_0}^{\tau_f } e^{-i(k ' + k)(\tau-\tau_0 ) } \ , a(\tau ) \ , d\tau \ , \end{aligned}\ ] ] and we omit the examination of the distributional part of @xmath271 at @xmath72 . the above treatment assumes @xmath171 to be non - negative . we shall not examine the validity of when @xmath171 may take negative values . in this appendix we derive the formula for the perturbative approximation to the bipartite mode entanglement of the field , as measured by the entanglement negativity , in the case when the field solutions have continuous spectra . these generalise the negativity formulas found for field solutions with discrete eigenvalues in @xcite . the field is prepared initially in the vacuum state and is subjected to an evolution which can be described by bogoliubov transformations that are assumed to take the general form : where @xmath273 , @xmath47 and @xmath75 are continuous real - valued parameters , @xmath271 and @xmath274 are of linear order in @xmath80 and @xmath275 is a phase taking the general form , @xmath276 , where @xmath277 is a real - valued function of @xmath47 . furthermore , we assume that @xmath278 and @xmath274 satisfy the conditions : the added difficulty of systems with continuous spectra is that the states are dirac @xmath46-normalised which can lead to apparent infinities in formulas for the entanglement if not correctly handled . the infinities are an artefact of working with infinitely precise frequencies which in practice is not possible there is always a spectral line - width , @xmath207 , associated with the measurement of a frequency . experiments which measure bipartite entanglement of the field do so by measuring two distinct frequencies each with a small line - width . for simplicity , we will assume that the modes measured are uniform wavepackets of frequencies having spectral line - width @xmath280 and central frequencies @xmath281 , @xmath282 respectively . we define : @xmath283 0 & \text{otherwise . } \end{cases}\end{aligned}\ ] ] we will also assume that the measured frequencies are sufficiently separated such that there is no overlap of their supports in frequency space : @xmath284 let @xmath285 be the annihilation operator associated with the frequency @xmath47 . after the evolution the annihilation operators are transformed by the bogoliubov transfomations according to : @xmath286 the quadrature operators associated with these frequencies are : as already alluded these frequencies are not measured precisely rather the actual measurement occurs over a small band of frequencies . we can define the quadrature operators associated with a wavepacket centred at the frequency @xmath281 by : since the bogoliubov transformations are linear and the vacuum state is a gaussian state , it follows that the final state of the field will also be a gaussian state . gaussian states are completely characterised by their first and second statistical moments . it is the second moments which are important for determining the amount of entanglement in a gaussian state . the second moments can be arranged into a covariance matrix : we define @xmath291 , then the covariance matrix can be defined by : @xmath292 where curly braces denote anti - commutator and the covariance matrix is normalised such that the covariance matrix of the vacuum state is the identity matrix . expectation values are to be taken with respect to the initial state which in our case is the vacuum state . it is easy to see that the second term in vanishes . to determine the bipartite entanglement , it is necessary to calculate the symplectic eigenvalues , @xmath305 , of the covariance matrix of the partially transposed state , @xmath306 . the partial transpose is implemented by a transformation of the covariance matrix given by @xcite : the zeroth order eigenvalues are @xmath313 and both eigenvalues have double degeneracy . it is therefore necessary to use degenerate perturbation theory to determine the eigenvalues to the linear order . the linear order corrections to the eigenvalues are found to be : @xmath314 where @xmath315 the symplectic eigenvalues are therefore : @xmath316 and the negativity is : @xmath317 where @xmath318 is the smallest of the two symplectic values . let us now assume that the wavepackets are very sharply peaked @xmath319 and @xmath320 , then the integrals in equation can be approximated by : @xmath321 and the negativity ( to first order in @xmath80 ) simplifies to : @xmath322 this is the continuous spectrum formula for the entanglement negativity and holds for non - integer values of the frequencies . it is equivalent to the formula for discrete frequency modes in the case when the modes have opposite parity , i.e. , when @xmath323 is odd @xcite . c. m. wilson , g. johansson , a. pourkabirian , m. simoen , j. r. johansson , t. duty , f. nori and p. delsing , `` observation of the dynamical casimir effect in a superconducting circuit , '' nature * 479 * , 376 ( 2011 ) [ arxiv:1105.4714 [ quant - ph ] ] . n. friis , a. r. lee , k. truong , c. sabn , e. solano , g. johansson and i. fuentes , `` relativistic quantum teleportation with superconducting circuits , '' phys . rev . lett . * 110 * , 113602 ( 2013 ) [ arxiv:1211.5563 [ quant - ph ] ] . j. lindkvist , c. sabn , i. fuentes , a. dragan , i .- m . svensson , p. delsing and g. johansson , `` twin paradox with macroscopic clocks in superconducting circuits , '' phys . rev . a * 90 * , 052113 ( 2014 ) [ arxiv:1401.0129 [ quant - ph ] ] . j. r. johansson , g. johansson , c. m. wilson and f. nori , `` dynamical casimir effect in a superconducting coplanar waveguide , '' phys . rev . lett . * 103 * , 147003 ( 2009 ) [ arxiv:0906.3127 [ cond-mat.supr-con ] ] . s. felicetti , m. sanz , l. lamata , g. romero , g. johansson , p. delsing and e. solano , `` dynamical casimir effect entangles artificial atoms , '' phys . lett . * 113 * , 093602 ( 2014 ) [ arxiv:1402.4451 [ quant - ph ] ] . d. e. bruschi , j. louko , d. faccio and i. fuentes , `` mode - mixing quantum gates and entanglement without particle creation in periodically accelerated cavities , '' new j. phys . * 15 * , 073052 ( 2013 ) [ arxiv:1210.6772 [ quant - ph ] ] . c. farina , h. o. silva , a. l. c. rego and d. t. alves , `` time - dependent robin boundary conditions in the dynamical casimir effect , '' int . j. mod . * 14 * , 306 ( 2012 ) [ arxiv:1201.3846 [ quant - ph ] ] . a. l. c. rego , j. p. d. s. alves , d. t. alves and c. farina , `` relativistic bands in the spectrum of created particles via the dynamical casimir effect , '' phys . rev . a * 88 * , 032515 ( 2013 ) [ arxiv:1309.3159 [ quant - ph ] ] .

we consider the phenomenology of a 750 gev resonance which can be produced at the lhc by only photon fusion and subsequently decay into di - photons . we propose that the spin - zero state is coupled to a heavy lepton that lives in the bulk of a higher - dimensional theory and interacts only with the photons of the standard model . we compute the di - photon rate in these models with two and more compact extra dimensions and demonstrate that they allow for a compelling explanation of the di - photon excess recently observed by the atlas and cms collaborations . the central role in our approach is played by the summation over the kaluza - klein modes of the new leptons , thus providing a significant enhancement of the loops for the production and decay subprocesses . it is expected that the jet activity accompanying these purely electromagnetic ( at the partonic level ) processes is numerically suppressed by factors such as . = 1

1601.07167

in the analysis of the first data obtained in proton - proton collisions with centre of mass energy @xmath3 tev at run 2 of the lhc , the atlas and cms collaborations both reported an excess of events in the invariant mass distribution of two photons near 750 gev @xcite . if confirmed by the bulk of the run 2 lhc data in the near future , this excess indicates the existence of a new boson @xmath0 with mass around 750 gev , that decays into two photons . assuming that this is indeed the case , and not a statistical fluctuation , we may finally have the first hint of physics beyond the standard model . this tantalising possibility led to an unprecedented explosion in the number of exploratory papers dedicated to the 750 gev di - photon resonance . the atlas collaboration analysed 3.2 1/fb of data and reported an excess of 14 events at the di - photon invariant mass of 750 gev , with a best - fit width of approximately 45 gev . the quoted statistical significance of the atlas excess is 3.9 @xmath4 or 2.3 @xmath4 including the look - elsewhere effect . the cms collaboration reports an excess of 10 di - photon events with a narrow - width peak at around 760 gev and a lower statistical significance of 2.6 @xmath4 . the cross - sections for the observed di - photon excess were estimated in ref . @xcite ( see also refs . @xcite ) as @xmath5 which we will take to be 8 fb . the di - photon resonance @xmath0 must be a boson thanks to the landau - yang theorem . in this work we will neglect the possibility of a spin - two particle and focus on a spin - zero state a scalar or a pseudo - scalar . as in most of the di - photon literature we will also assume the mass and the width of the @xmath0 to fit the atlas data , thus in total , ^obs_pp x = 8 fb , m_x= 750 gev , _ tot= 45 gev , though the observational evidence for the latter feature of a relatively large total width @xmath6 , is not strong ; consequently in most of our numerical estimates in the next section we will indicate an appropriate scaling factor between the 45 gev value and a generic @xmath7 . at first sight @xmath0 appears to be very similar to the standard model higgs it is produced in @xmath8 collisions and its decay modes include the @xmath9 channel . hence it is natural to expect that it is dominantly produced from initial protons via the gluon - gluon fusion process ( plus contributions from quark - anti - quark annihilation ) and then decays into two photons . however , because it is six times heavier than the standard model ( sm ) higgs , the new boson @xmath0 provides an interesting complication to model - building it requires the addition of new mediators between @xmath0 and the sm , contributing to _ both _ the production and the decay process . one can not hope to utilise any of the sm fermions either in the loop of the gluon fusion production of @xmath0 , or in the @xmath10 decay loop . since @xmath11 is above all of the sm fermion - anti - fermion thresholds , @xmath0 bosons produced on - shell would rapidly decay into fermion - anti - fermion pairs at tree level , thus wiping out the corresponding @xmath12 and @xmath13 branching ratios which are loop suppressed . hence , both parts of the assumed @xmath14 process , the gluon fusion , and the di - photon decay have to be generated entirely by new heavy mediators with @xmath15 propagating in the loops . futhermore , the mixing of a scalar @xmath0 singlet with the sm higgs must be severely suppressed @xcite , @xmath16 . in summary , the entire parton - level process must be generated by beyond sm ( bsm ) physics with no input from the standard model beyond just providing the external states . in addition , the yukawa couplings of @xmath0 to the mediators that are required to accommodate the observed di - photon rate for a 45 gev resonance are already pushed to large values and become non - perturbative not much above the tev scale @xcite . in this paper we follow an alternative more minimal approach where the @xmath0 ( pseudo)-scalar communicates only to the @xmath17 hypercharge factor of the standard model . in this case one does not require separate bsm - enabled production and decay mechanisms . this idea that the di - photon resonance couples to photons but not gluons was put forward in two pioneering papers @xcite already on december 17 2015 , and considered further in refs . @xcite . the main theoretical challenge facing this ` pure photon ' set - up is that the di - photon channel is subdominant , being suppressed by @xmath18 relative to the gluon fusion rate and this , as we will briefly review in section * [ sec : estimates ] * , puts the model in dangerous territory , with a combination of non - perturbative couplings , very large numbers @xmath19 of mediator flavours and very low scales of new physics . we will propose that these short - comings can be resolved by allowing a lepton mediator that feels large extra dimensions . the resulting tower of kaluza - klein ( kk ) states quite easily enhances the di - photon rate sufficiently to match with observations . in section * [ sec : model ] * we will examine the effect in various dimensions determining how the photon - fusion rates match with the expectations for the observed di - photon resonance . following that , in section * [ sec : p ] * we will derive additional constraints on the uv - cutoff of the theory by requiring perturbativity of the gauge coupling . even though the assumptions about the content of the theory are the most minimal , we find that a significant window remains to fit the di - photon signal in perturbative settings . the effect of radiative corrections on the scalar boson mass and its finite naturalness are discussed in section * [ sec : mx ] * , in a framework where a supersymmetric theory is compactified down to 4d with scherk - schwarz boundary conditions . following this , in section * [ sec : dark ] * , we address the additional invisible decays of @xmath0 to @xmath9 , that can account for the full total width @xmath20 gev , and hence enable the di - photon resonance to play a role as mediator between the sm and dark matter ( dm ) sectors . finally section * [ sec : conc ] * presents our conclusions . for a ( pseudo)-scalar of mass @xmath11 and width @xmath7 one can express the di - photon cross - section using the standard narrow - width approximation expression , _ pp x ( s)= ( _ i = g , q , c_ii _ xii ) _ x [ eq : narw ] where the sum is over all partons and @xmath21 are the dimensionless integrals over the corresponding parton distribution functions . for example for gluons @xmath22 at 13 tev was estimated in @xcite as @xmath23 , for quarks @xmath24 , @xmath25 , and for photons @xmath26 . we now concentrate on the case where the @xmath0 boson is coupled only to photons , thus keeping only @xmath27 in the sum on the right hand side of . in fact , one should also consider corrections to this process sourced by the photon pdf integral @xmath27 , by the ` vbf ' type processes , which are proportional to quark pdfs @xmath28 with each of the initial quarks emitting a photon . a very rough estimate for the relative importance of this effect would be @xmath29 , using the values quoted in @xcite . while a detailed calculation of such effects is beyond the scope of this paper , we will take the above estimate as a hint that the corrections to the leading photo - production process are relatively small and as a result the presence of additional jets ( in the vbf case arising from the initial quark partons ) is suppressed relative to what one would expect for a heavy higgs - like scalar production in the standard model . photon - photon fusion in elastic @xmath8 scattering as searches of new physics was considered even earlier in ref . but in the weakly - coupled sm - like settings , the branching ratio of a scalar resonance to photons is tiny , and hence the rate for such processes is negligible . for example , the sm higgs production via photon fusion in elastic @xmath8 collision is only @xmath30 fb @xcite . the next step is to account for contributions from inelastic as well as elastic @xmath8 collisions . following refs . @xcite we will express the answer in the form _ pp : x ( 13 tev ) 10 pb ( ) ( br_x)^2 . [ eq : xc ] the numerical factor @xmath31 pb on the right hand side includes the contribution from the integral over the parton distribution functions of the photons in the initial state protons . as explained in refs . @xcite , this factor can suffer from a large theoretical uncertainty and arises from accounting for inelastic collisions where one or both initial protons gets destroyed after emitting a photon . the estimates for @xmath32 obtained in the recent literature give @xmath33 , @xmath34 , @xmath35 and the most recent calculation @xcite gives @xmath36 . ( an even earlier estimate made in @xcite , which was based on the ( subdominant ) elastic @xmath8 collisions , gives the rate suppressed by two orders of magnitude relative to . ) the overall factor of the order-10 pb is perhaps on the optimistic side , but for the purposes of our work which aims instead to enhance the @xmath37 fractions , it will not be critical if the overall coefficient in is reduced . the photon fusion cross section computed in the same manner at @xmath38 tev in @xcite is roughly a factor of 2 smaller than the corresponding @xmath39 tev result in , which is consistent with the absence of the di - photon resonance signal in run 1 lhc searches . to match the signal rate quoted in - , a rather large value of @xmath0 to photons branching fraction is required on the right hand side of , _ x:= = ( 0.02 - 0.04 ) , [ eq : bound ] which in the case of the relatively wide 45 gev resonance , as preferred by the current atlas data , and for the 8 fb signal rate amounts to 0.028 i.e. 2.8% . in the usual four - dimensional qft settings , such large branching fractions to photons would require introducing new physics already at the electro - weak scale . to see this , we parametrise the interactions of @xmath0 with photons via the leading - order dimension-5 operators , x f^ f _ , x f^ _ , [ eq : eft ] where @xmath0 is assumed to be a scalar in the first case , and a pseudo - scalar in the second case , and @xmath40 is the new physics scale responsible for generating these eft vertices . greater than both , the @xmath0-resonance mass , @xmath41 gev , and the electroweak scale @xmath42 itself . ] for the partial width of @xmath0 to photons one has , _ = _ em^2 , where @xmath43 is either of the two : @xmath44 , @xmath45 ; and to achieve the required branching ratio in , we need = _ em ( ) ^1/4750 gev 236 gev , [ eq : estl ] with the final expression assuming the 45 gev resonance . thus , we see that to reproduce the experimental signal ( or excess ) attributed to the di - photon resonance with @xmath41 gev and @xmath46 gev in terms of the _ photon fusion _ process on its own , would require the introduction of new physics degrees of freedom already at the electroweak scale , i.e. @xmath47 , which is obviously at odds with the experiment . if the total width of the di - photon resonance turns out to be much less than 45 gev , the value of @xmath43 will be rescaled as indicated in . what is the minimal possible value of the total width ? clearly it is achieved when the branching ratio to photons @xmath48 . it then follows from and , that ( ^min _ tot / 45 gev ) 8 10 ^ -4 . [ eq : minw ] if we restore other theoretical and experimental uncertainties , such as the numerical factor @xmath32 on the right hand side of , and the total experimental cross - section in , which we denote @xmath49 , the estimate in becomes = ( ) ^1/4 ( ) ^1/4 ( ) ^1/4 236 gev . [ eq : estl2 ] the quartic roots ensure that the deviations even by an order of magnitude in any of the three parameters away from the ` central ' values would not lead to any significant deviations from the bound by the electroweak scale in eq . . to 1.4 tev . ] in fact , it is easy to see that the actual mass scale of the new physics degrees of freedom appearing in a hypothetical four - dimensional perturbative extension of the sm appears to be further suppressed compared to the estimate in by a multiplicative factor of @xmath50 , cf eq . below . to verify this , consider augmenting the standard model by a massive dirac fermion @xmath51 charged only under hypercharge @xmath17 and coupled to the spin - zero sm - singlet @xmath0 , @xmath52 here @xmath53 is the @xmath17 covariant derivative and @xmath54 is the hypercharge of @xmath51 , @xmath55 is its mass and @xmath56 is the yukawa coupling to the singlet @xmath0 . the equation describes the scalar @xmath0 while corresponds to the pseudo - scalar di - photon resonance . the dirac fermion , @xmath51 consists of two 2-component weyl spinors , @xmath57 and @xmath58 where @xmath59 transforms in fundamental , and @xmath60 in anti - fundamental representations , in this case only of the hypercharge , being a singlet under the @xmath61 and the @xmath62 gauge factors . the @xmath63 process is described by the triangle diagram with the virtual @xmath51 fermion propagating in the loop . evaluating the diagram for the scalar @xmath0 using the interactions in , one recovers the well - known result @xcite for the @xmath64 formfactor , which we present as the expression for the wilson coefficient of the first operator in : = ( ) [ eq : formfactor ] . here , = , ( ) = ( + ( -1 ) arcsin^2 ( ) ) . [ eq : formfactor2 ] the expression on the right hand side of is presented in a factorised form , such that the formfactor @xmath65 is normalised to one and tends to @xmath66 in the heavy lepton - mass limit @xmath67 . hence , in this limit we have a relation m_l = n_f q^2 y f ( ) , [ eq : mlest ] where we have assumed that there are @xmath19 flavours of the vector - like leptons of mass @xmath55 . using the bound @xmath68 gev in , we are led to an estimate for an effective number of new lepton flavours , n_f 500 ( ) ( ) ( ) , [ eq : mlest2 ] which - interpreted in a four - dimensional perturbative model with the tev - scale vector - like leptons - is unjustifiably high . such a large value for the number of new lepton flavours to match the observed di - photon excess @xcite by photon fusion is clearly unsatisfactory and begs for an alternative explanation . we will propose in the following section that the virtual new lepton states propagating in the loop extend into large extra dimensions . this gives an infinite tower of kaluza - klein states of vector - like leptons populating the loops and , as we will show in the following section , leads to an appropriate enhancement necessary ( and sufficient ) to photo - produce the observed di - photon rate . at the same time , the kaluza - klein lepton tower also contributes to the coupling constants of the theory which , not surprisingly , exhibit power - like ( rather than merely logarithmic ) growth at energies above the kk mass threshold . the theory tends to become strongly coupled at large uv scales . if one requires perturbativity of the theory , this amounts to additional constraints on the uv - cutoff of the theory @xmath69 tev as will be shown in section * [ sec : p ] * based on the analysis of the gauge coupling . even in this simplistic minimal theory , we find that a significant window remains to fit the di - photon signal in perturbative settings . before closing this section , we note that when the di - photon mediator @xmath0 is a pseudo - scalar , the calculation is essentially identical to the scalar case considered above . a completely analogous result to holds for the pseudo - scalar @xmath0 di - photon interaction following , = _ ps ( ) [ eq : ps_formfactor ] , only the functional form of the formfactor @xmath70 is different from the scalar case . but as we are working in the limit of heavy lepton masses ( the kk modes are becoming increasingly heavy ) , the formactor can again be set to one , and the same estimate - applies . as outlined in the introduction , to address these shortcomings we wish to consider a kk tower of vector - like leptons coupled to photons and to @xmath0 . in the simplest scenario , we take the new leptons to extend into the bulk of a @xmath71-dimensional theory @xcite with @xmath72 flat compact extra dimensions . we denote the d - dimensional coordinates , ( x^0,x^1,x^2,x^3,z_1, z_d ) ( x^,z_1, z_d ) , 0z_i 2r_i . the vector - like leptons are the only bulk fields we need to consider , and we assume them to be charged only under the sm @xmath17 hypercharge , and to be colour- and @xmath73-singlets . apart from these new leptons , we can take the entire standard model to be localised in four dimensions . however it is possible ( and probably more natural in string models based on webs of d - branes ) for the hypercharge to also be a bulk field . it will not make any difference to the discussion . the fermions of @xmath74 and @xmath75 dimensions map naturally to dirac fermions in @xmath76 in a vector - like representation of @xmath17 . in the compactified extra dimensions , the boundary conditions are taken to be l_d(z_i+2r_i , x)= e^i q_fi 2 l_d(z_i , x ) , [ eq : bcl ] where @xmath77 are arbitrary phase - shifts . performing the fourier series expansion in each of the compact @xmath78-coordinates , the bulk field yields a tower of kaluza - klein ( kk ) states in the four - dimensional description with masses @xmath79 . for example in the @xmath74 case one has l_d(z_1,x^)= _ n=-^+ e^iz_1(n+q_f)/r l_d^(n)(x^ ) , m_ln^2= m_d^2 + ( n + q_f)^2/r^2 , [ eq : bcl5d ] where @xmath80 is the dirac mass of the leptons ( cf . ) , _ bulk = will for simplicity assume degenerate radii , @xmath81 , and introduce a @xmath82-dimensional vector notation , @xmath83 and @xmath84 . we can then write , l_d(z_i , x^)= _ n e^iz ( n + q ) /r l_d^(n)(x^ ) , [ eq : bcl6d ] with kk masses given by m_ln^2= m_d^2 + ( n + q)^2 m_c^2 , where @xmath85 is the compactification mass scale . the @xmath0 to di - photon interaction is generated by summing over the kk modes of the vector - like lepton propagating in the loop of the triangle diagram . in the @xmath75 theory , the wilson coefficient of the @xmath86 operator is given by ( cf . ) , = _ n e^-(m_ln^2/m_st^2 ) f ( ) [ eq : kkformfactor ] , where the function @xmath87 is the formfactor appearing in . we should remark that this contribution is related to the contribution of the kk modes to the beta function of the gauge coupling itself . therefore one should pay attention to the perturbativity of the latter at some point . indeed we shall see in the following section that the two can be directly related by performing a single beta function calculation . for the moment however we focus on the direct evaluation of the formfactor . note that the kk sum has been regulated in with a factor @xmath88 ; we argue shortly and in appendix a that this regulator is the one that naturally emerges in most string theory calculations . moreover for the results in @xmath74 and @xmath75 the precise form of the regulator will not change the conclusions . note also that the contributions of each individual kk mode to the right hand side of eq . go as @xmath89 and not as @xmath90 as one might have supposed from the characteristic @xmath91 behaviour in . in fact is easy to see that the the triangle diagram vanishes in the @xmath92 limit for any kk occupation number , as a consequence of the fact that only the traces of even numbers of dirac matrices survive so at least one power of @xmath80 is required in the numerator of d^4 p ~ k_1^ k_2^ , where @xmath93 are @xmath71-dimensional scalar products , and @xmath94 are the external momenta of the photons . for a model in @xmath71 dimensions we have to evaluate the @xmath82-fold sum , s = _ ( -m_l,^2/m_st^2 ) , m_l,^2= m_d^2 + ( + ) ^2 /r^2 [ eq : sumgen ] . it is informative to compare with string calculations , and in particular to understand the motivation of the regularisation in eq . , to use schwinger parameterisation to write this expression as @xmath95\ , . \label{eq : sumgen2}\ ] ] note that for @xmath96 this integral is simply the usual zero - mode factor @xmath97 . by inspection , putting a regularization @xmath98 in the kk sum , is precisely equivalent to placing a uv cut - off on this integral of @xmath99 . conversely , it is for this reason that the automatic regularization that occurs in string theory due to for example modular invariance excising the uv divergent regions , typically results in regulating factors @xmath98 appearing in otherwise divergent kaluza - klein sums . indeed such regulating factors appear already in tree - level scattering , because of the `` softening '' effect of string theory ( the fact that d - branes have finite thickness for example ) , which puts a limit on the accessible compton wavelengths @xcite . we discuss this aspect more in appendix a. in @xmath74 the summation in is finite resulting in s = ~ 2 ^ 2 & : 2m_dr1,q_f0 + & : 2m_d r1,q_f=0 . [ eq:5ds ] regardless of the relative sizes of @xmath80 and @xmath100 , there is no significant enhancement over simply taking @xmath101 as the indicative value of @xmath102 . in @xmath75 ( and higher ) dimensions we expect to find logarithmic ( and higher @xmath103 ) dependence on the uv theory , and we therefore reinstate the @xmath104 cut - off in . performing poisson resummation we have @xmath105 where @xmath106 denotes the dual integer lattice sum with the same dimensionality as @xmath107 . the contributions for non - zero @xmath106 are negligible , and for the @xmath108 contributions we find @xmath109.\ ] ] expanding for small @xmath110 , we find @xmath111 & : \,\,d=2\\ \frac{2\pi^{\frac{d}{2}}}{d-2}\frac{1}{m_{d}}(m_{d}/m_{\rm c})^{2}\left(m_{\rm st}/m_{\rm c}\right)^{d-2 } & : \,\,d>2 \end{cases } \ , \qquad { \rm where}\quad m_{\rm c } \,=\ , 1/r\,.\ ] ] note that for these approximations to be valid , including the original poisson resummation , we need only assume that @xmath112 , but do not need to make any assumption about the relative values of @xmath80 and @xmath100 . moreover it is important to appreciate that the kk tower is preventing large @xmath80 killing the sum since , it is the uv end of the tower that dominates the contributions , and this depends solely on the separations @xmath100 regardless of the dirac mass . we thus find = & : d=2 + ( m_st / m_c)^d-2 & : d>2 . [ eq : kk4 ] we will now trade the the 4d dirac mass @xmath80 for the mass of the lightest kk mode @xmath113 via , m_l , = = m_d . we can now determine the lepton mass necessary to reproduce the di - photon signal in terms of @xmath68 gev in our earlier estimate ; m_l , = q^2 y ( 1 + ( ) ^2 ) ^1/2 & : d=2 + ( m_st / m_c)^d-2 & : d>2 . [ eq : fin1 ] [ cols="^,^ " , ] -.4 cm it can be checked that this yields precisely the previous result for the @xmath114 coupling , as a term in addition to a power - law contribution to @xmath115 . clearly then the kk enhancement is associated with an enhanced gauge beta function . for perturbativity we should also then ensure @xmath116 in other words the best one can do is to take @xmath117 to represent a landau pole at the scale @xmath118 ; numerically this translates as @xmath119 , which implies @xmath120 in general the naive assumption that kk modes contribute to the gauge coupling requires a string scale one or perhaps two orders of magnitude higher than the compactification scale . we can now re - write the equation for the model with @xmath121 extra dimensions in the form , d=2 : m_l , = 4 x^2 ( z -x -_e/2 ) , [ eq : fin2 ] where we have defined the two mass ratios , x= m_d / m_c z= m_st / m_c in figure [ fig : onep ] we plot the values of @xmath122 resulting from ( as required to fit the di - photon production rate ) now taking into account the perturbativity constraint . we see that even in this class of the relatively low - cutoff perturbative models , one can obtain @xmath122 in the range from above the @xmath123 gev to up to 1200 gev and for string scales @xmath118 up to @xmath124 tev . now in the model with @xmath125 extra dimensions we can the second equation in ( see also ) in the form , @xmath126 \ , < \ , 450 - 500 \ , { \rm gev}\ , , \label{eq : fin4mc } \end{aligned}\ ] ] where the bound on the right hand side of comes from the perturbativity bound @xmath127 in the second equation in . this amounts to the uv - cutoff scale bound @xmath128 tev . finally , for the the zero mode of the kk lepton in the eight - dimensional theory one can choose any value in the range between @xmath129 and @xmath130 where @xmath131 . for example , for @xmath132 , the lowest kk lepton modes are in the range of 1.5 tev ( times @xmath133 ) . our approach so far has been based on the idea that the high multiplicity of kk states of the vector - like leptons can significantly enhance their 1-loop contributions to the coupling of the spin - zero @xmath0 field to two photons . we must be careful , however , to ensure that the same tower of kk leptons does not at the same time introduce unwanted large effects elsewhere . the most obvious and potentially most dangerous such effect would be an exploding contribution to the 750 gev mass of the @xmath0-resonance itself . indeed , the vector - lepton contributions to the self - energy of @xmath0 have two ( rather than three , which was the case before ) propagators in the loop and produce relevant ( i.e. mass dimension-2 ) operators . uv - cutoff - dependent contributions @xmath135 would be very unwelcome . such contributions can however be removed by a ` lepton - partners ' mechanism , as we will now outline . a simple implementation of such a mechanism begins with a supersymmetric wess - zumino model in @xmath75 , with superpotential , = y x l+ m_dl , [ eq : w ] where @xmath59 and @xmath60 are two chiral superfields , which can be thought of as forming an @xmath136 hypermultiplet . as before , the fermionic components of @xmath59 and @xmath60 form a single dirac multiplet , but also of course provide scalar lepton partners . imposing boundary conditions as in eq . with @xmath137 ( for bosons and fermions respectively ) breaks supersymmetry spontaneously by the scherk - schwarz mechanism @xcite , for early string theory realisation see @xcite . typically @xmath138 and @xmath139 are taken to have anti - periodic or periodic entries . at 1-loop level the contribution to the @xmath0-boson mass from the kk towers of the lepton and lepton partner states is , m_x^2 = y^2 _ ( - . ) [ eq : susykk ] the final expression is known to be entirely finite thanks to the spontaneous nature of the breaking @xcite . indeed one can easily calculate the mass by again poisson resumming the schwinger integrals ( as in appendix b ) , to find m_x^2 = 0 & : _ b = _ f + y^2 m_c^2 ( 1+d/2 ) & : _ b = 0 _ f=1/2 . in the first case boundary conditions do not break supersymmetry and there are no radiative corrections to the mass of the di - photon resonance at the order @xmath140 , i.e. when we neglect the back reaction from the standard model . in the second case , where supersymmetry in the bulk theory is broken by @xmath141 , the finite naturalness for the di - photon resonance requires that 750 gev y ( ( 1+d/2 ) ) ^1/2 m_c 0.05y m_c & : d=2 + 0.08y m_c & : d=4 , where @xmath142 is a multidimensional sum derived in appendix b : its first few values are @xmath143 , @xmath144 , @xmath145 , @xmath146 . this is only a mild constraint on the size of the compactification mass , or equivalently the lightest kk mode of the lepton . it is also interesting to estimate the back reaction on @xmath11 of the susy - breaking scale implicit the standard model . the leading - order contribution is the two - loop effect , similar to sfermion masses in gauge mediation . in our case @xmath0 directly interacts only with the lepton multiplets @xmath147 which then interact with the @xmath17 photon - photino loop . thus we have a rough estimate m_x^2 ~ m^2 _ , where @xmath148 is the photino mass in the standard model . the finite naturalness bound for the 750 gev scalar would then imply an upper bound on the photino mass of m _ ( 4/ ) m_x ( 1/ ) 110 tev , which should not cause any problems with the current limits . let us also comment on the soft masses of the scalar leptons @xmath149 . these are susy - breaking contributions first generated at two loops in @xmath150 by back reaction from the standard model . analogously to ordinary slepton mass - squared soft terms in gauge - mediated supersymmetry breaking , we can parameterise these contributions as , m_l_sc^2 ~ m^2_soft sm , where @xmath151 are the slepton masses in the standard model . even at 100 tev a hypothetical sm slepton would result only in a @xmath152 gev contribution compared to the mass of @xmath149 , which does not impose any significant phenomenological constraints . having generated the required @xmath153% branching ratio of @xmath0 to @xmath9 necessary to explain the di - photon excess observed by atlas and cms in terms of solely photon - photon fusion and decay , both mediated by a tower of kk states of vector - like leptons , we now address the remaining 97% of the 45 gev total width of @xmath0 . an obvious and interesting possibility is that the di - photon spin - zero state @xmath0 decays into dark matter particles . more generally @xmath0 can decay invisibly into any particles of the dark sector , which includes but is not necessarily limited to cosmologically stable dark matter . over the last two years so - called simplified models describing scalar and pseudo - scalar mediators to dark sectors and their searches at the lhc @xcite and future hadron colliders @xcite have attracted a fair amount of interest . the role of the 750 gev resonance as the possible mediator to dm was studied in the recent work , including refs . @xcite . the simplified models describing elementary interactions of spin - zero mediators with the the dark sector particles , which for simplicity we take to be dirac fermions @xmath154 , @xmath155 , contain the interactions @xmath156 the partial decay widths of @xmath0 into these fermions ( for a single flavour ) are given by [ eq : gs ] _ = m_x(1-)^ , _ = m_x(1-)^ , for the scalar and the pseudo - scalar cases respectively . it is straightforward to achieve 97% of the 45 gev @xmath7 from these decays alone . for example for @xmath19 light dm fermions one would need n_f^1/2 g_dm 1.21 , which is easy to accommodate in a weakly coupled theory . it is also possible , if so desired , to arrange for dm particles to propagate in large extra dimensions assuming that the lowest kk modes are kinematically accessible to the 750 gev particle . three comments are in order . first , following the same logic as in refs . @xcite , we will not insist that the relic density of the dark sector fermions computed using the lagrangian in eq . should match the observed cosmological abundance of dm . such calculations have already been carried out elsewhere , e.g. in refs . @xcite , but more importantly , it is also possible that the dark sector particles @xmath154 are only the distant parents of the cosmologically stable dm and can undergo further decays . our second comment is that for searches of these dark sectors at the lhc which are mediated by the @xmath0 particle produced via photon fusion , the standard strategy of jets plus missing energy searches used in @xcite will not be directly applicable . the reason is that at the partonic level the process @xmath157 is entirely electromagnetic and we expect that additional qcd jet activity will be suppressed numerically ( as already mentioned in section * 2 * ) . finally , one should ensure that the new heavy leptons do not themselves introduce stable charged matter that violates cosmological and astrophysical bounds . this requires either sufficiently low reheat temperatures , small freeze - out densities or perhaps decay through higher dimensional operators . see for example refs.@xcite . we proposed a scenario where the spin - zero 750 gev state is coupled to a new heavy lepton which lives in the bulk of a higher - dimensional theory and interacts only with the photons of the standard model . we found that they allow for a minimal and compelling explanation of the di - photon resonance via photo - production and decay . the central role in this effect is played by the summation over the kaluza - klein modes of these leptons appearing in the loops of the @xmath1 production and decay subprocesses . the set - up requires only a minimal extension of the standard model in the sense that the same mechanism is used for the production and the decay of the @xmath0 resonance . with the new lepton being coupled only to the @xmath17 gauge sector in the standard model one can explain the absence of other resonances at 8 and 13 tev in the vicinity of the 750 gev di - photon invariant mass . the decay channel into @xmath158 is suppressed by @xmath159 , while other potential candidates , such as two jets , @xmath160 and 2 leptons are either absent or suppressed by powers of @xmath150 . this feature is particularly important for suppressing standard model jets which would otherwise accompany the di - photon production mediated by @xmath0 . since at the parton level the entire @xmath161 process is electromagnetic , any additional qcd jet activity would be suppressed by extra powers of @xmath150 or a combination of @xmath162 ( for the vbf process ) relative to the @xmath27 factor in our leading order photon fusion process . thus the relative absence of additional jets which would accompany the di - photon resonance if it was produced in gluon fusion relative to the case of pure photo - production considered here , is a distinguishing feature of our model . similarly , while mono - jet searches disfavour a large invisible width @xmath163 gev , this conclusion @xcite does not apply directly to our model , as the mono - jet would have to originate e.g. from a quark in one of the initial protons @xmath164 in addition to the photon which participates in the photon fusion . this is again suppressed by the fine structure constant times the corresponding ratio of pdfs . at the same time , the approach presented here can be easily applied to the case where the di - photon ( pseudo)-scalar @xmath0 is produced in the gluon fusion channel . in this case one would simply substitute the vector - like lepton by a single species of a new quark @xmath54 , @xmath165 in the bulk of an extra - dimensional theory . the di - photon rate in the gluon fusion process would of course be greater relative to the photon fusion , and will be even easier to fit in our kk model . the extension of the standard model by a new spin - zero singlet state @xmath0 also sits very well with other items on the bsm wish list : @xmath0 can play the role of the inflaton as the singlet degree of freedom which is non - minimally coupled to gravity see e.g. @xcite . it can also help to stabilise the sm higgs potential @xcite , assist the first order phase transition and provide additional sources of cp violation for baryogenesis . finally as was already noted in section * [ sec : dark ] * , a ( pseudo)-scalar @xmath0 is also an obvious candidate for being a mediator to the dark matter sector @xcite . we are grateful to valery khoze , alberto mariotti , misha ryskin and michael spannowsky for useful discussions and comments . this work is supported by the stfc through the ippp grant , and for vvk in part by the royal society wolfson research merit award . the expressions that one derives in the effective field theories are dominated by two elements : the kk tower and the magnitude and nature of the uv cut - off . in view of the latter , one should check that these approximations adequately model what happens when a uv completion is added to the theory . we have already argued that the physical manifestation of the uv finiteness of string theory is typically equivalent to a simple cut - off on schwinger integrals , and that this in turn is equivalent to inserting factors @xmath166 in otherwise divergent kk sums . however we would like briefly to confirm that this expectation holds for the wilson coefficient we are interested in . a simple equivalent calculation to the field theory one can be performed in the heterotic string , where the uv cut - off is simply understood as a consequence of modular symmetry . in general one can find the wilson coefficient by obtaining the threshold corrections to @xmath150 from the two - point function of the photon in the standard way , but inserting the resonance as continuous wilson line backgrounds @xmath167 ( that break some non - abelian gauge symmetry ) around compact dimensions @xmath168 . the standard starting point is therefore @xmath169.\nonumber \end{aligned}\ ] ] where @xmath54 refers to the charges , and @xmath170 refers to spin structure on the two cycles of the torus where @xmath171 is the @xmath172-scale ; of course this is not required for the current problem . ] . the integration over the real part of the modular parameter , @xmath173 , projects onto physical states , while the integral over the imaginary part , @xmath174 is precisely equivalent to the schwinger integral in field theory , with the fundamental domain @xmath175 providing a natural cut - off . expanding in the canonically normalized @xmath0 fields corresponding to the wilson - lines , @xmath176 where @xmath177 is the radius of the dimension associated with wilson line @xmath0 , and neglecting the exponentially suppressed string modes as well , we find @xmath178 where the formfactor of interest arises from the @xmath179 cross - term with @xmath180 . a non - negligible result requires @xmath177 to be small ( in other words we did nt need to resum that dimension ) , leading to a formfactor of order @xmath181 in agreement with the field theory . note that the compactification was somewhat asymmetric , but the crucial factor was the appearance of the bulk volume in string units @xmath182 , which had to be orthogonal to the wilson line . a similar situation holds in any situation where there is a large bulk volume dependence in the hypercharge gauge thresholds . for example in type i or type ii models with d - branes , the dependence of threshold corrections on wilson lines and/or d - brane displacements has been examined in the literature in various contexts , for example @xcite . the typical dependence goes as @xmath183 where @xmath184 is now the brane displacement , @xmath185 is the volume of neumann directions , and @xmath186 the volume of dirichlet ones . as for the previous example , we can expand for small displacements and recover a wilson coefficient proportional to the volume . we should note in this context that there has been some interest in the 750 gev di - photon as a pseudo - scalar axion that participates in generalized green - schwarz anomaly cancellation ( i.e. it appears as a closed string r - r state in the wess - zumino d - brane action ) @xcite . that possibility can be understood also entirely within an effective field theory in which heavy chiral modes are integrated out due to a large yukawa coupling @xcite . however in the present case where we consider the state to be a scalar coupling to kk modes , it would correspond to a _ dilatonic _ closed string mode , more along the lines of the scalar proposed in ref . @xcite . finally it is interesting in this context naively to estimate parameters under the assumption that the vector - like leptons and @xmath187 all occupy the same relatively large volumed d - brane . then one has @xmath188 with @xmath189 and @xmath190 denoting volumes of the respective branes in string units @xcite , leading to an estimate of m_d ~ ( ) ^2 ~ y()^2 , assuming @xmath191 . given the possibility of non - zero scherk - schwarz contributions to supersymmetry breaking , it is interesting to ask what contributions to the @xmath0 scalar mass can arise from the kaluza - klein lepton tower in . like the 5d coupling computed earlier in @xcite the result is expected to be finite , but in this case for arbitrary dimensions due to the spontaneous nature of supersymmetry breaking in scherk - schwarz mechanism . as we shall see , the soft - masses are dominated by the kk modes and consequently the field theory calculation contains all the necessary physics provided the compactification radii are relatively large compared to the fundamental scale . ( in other words any string calculation would yield the same result . ) to see this explicitly , and to obtain precise results for arbitrary dimensions , let us briefly derive the soft - mass of the scalar in field theory using the earlier schwinger approach . the object of interest is m_x^2 = y^2 _ . using schwinger parametrization each integral is of the form _ _ 0^(-t(+m_d^2))- , and poisson resumming as before we find @xmath192e^{-m_{d}^{2}t-\mathbf{\bfell}^{2}\pi^{2}r^{2}/t}.\ ] ] the terms with zero @xmath106 vanish ( thanks to the supersymmetry being spontaneously broken ) and the integral is then rendered finite ( hence no uv regularisation is required ) . let us assume that @xmath193 for every compact radius . then the result for general @xmath82 is @xmath194}{16\pi^{4+d/2}}\sum_{\mathbf{\bfell}}\frac{(1-(-1)^{2\mathbf{q}_{f}.\bfell})}{(\bfell.\bfell)^{1+d/2}},\ ] ] which converges . in 5 dimensions , for @xmath195 , we find the result of @xcite @xmath196 for @xmath197 dimensions , assuming that @xmath198 for all compact dimensions we have @xmath194}{8\pi^{4+d/2}}\xi(1+d/2),\ ] ] where are @xmath143 , @xmath144 , @xmath145 , @xmath146 . ] @xmath199 . r. franceschini _ et al . _ , `` what is the gamma gamma resonance at 750 gev ? , '' arxiv:1512.04933 [ hep - ph ] . j. ellis , s. a. r. ellis , j. quevillon , v. sanz and t. you , `` on the interpretation of a possible @xmath201 gev particle decaying into @xmath202 , '' arxiv:1512.05327 [ hep - ph ] . r. s. gupta , s. jaeger , y. kats , g. perez and e. stamou , `` interpreting a 750 gev diphoton resonance , '' arxiv:1512.05332 [ hep - ph ] . a. falkowski , o. slone and t. volansky , `` phenomenology of a 750 gev singlet , '' arxiv:1512.05777 [ hep - ph ] . w. altmannshofer , j. galloway , s. gori , a. l. kagan , a. martin and j. zupan , `` on the 750 gev di - photon excess , '' arxiv:1512.07616 [ hep - ph ] . j. h. davis , m. fairbairn , j. heal and p. tunney , `` the significance of the 750 gev fluctuation in the atlas run 2 diphoton data , '' arxiv:1601.03153 [ hep - ph ] . l. berthier , j. m. cline , w. shepherd and m. trott , `` effective interpretations of a diphoton excess , '' arxiv:1512.06799 [ hep - ph ] . j. gu and z. liu , `` running after diphoton , '' arxiv:1512.07624 [ hep - ph ] . m. son and a. urbano , `` a new scalar resonance at 750 gev : towards a proof of concept in favor of strongly interacting theories , '' arxiv:1512.08307 [ hep - ph ] . f. goertz , j. f. kamenik , a. katz and m. nardecchia , `` indirect constraints on the scalar di - photon resonance at the lhc , '' arxiv:1512.08500 [ hep - ph ] . s. fichet , g. von gersdorff and c. royon , `` scattering light by light at 750 gev at the lhc , '' arxiv:1512.05751 [ hep - ph ] . c. csaki , j. hubisz and j. terning , `` the minimal model of a diphoton resonance : production without gluon couplings , '' arxiv:1512.05776 [ hep - ph ] . l. a. anchordoqui , i. antoniadis , h. goldberg , x. huang , d. lust and t. r. taylor , `` 750 gev diphotons from closed string states , '' arxiv:1512.08502 [ hep - ph ] . c. csaki , j. hubisz , s. lombardo and j. terning , `` gluon vs. photon production of a 750 gev diphoton resonance , '' arxiv:1601.00638 [ hep - ph ] . s. fichet , g. von gersdorff and c. royon , `` measuring the diphoton coupling of a 750 gev resonance , '' arxiv:1601.01712 [ hep - ph ] . j. jaeckel , m. jankowiak and m. spannowsky , `` lhc probes the hidden sector , '' phys . dark univ . * 2 * , 111 ( 2013 ) doi:10.1016/j.dark.2013.06.001 [ arxiv:1212.3620 [ hep - ph ] ] . v. a. khoze , a. d. martin and m. g. ryskin , `` prospects for new physics observations in diffractive processes at the lhc and tevatron , '' eur . j. c * 23 * ( 2002 ) 311 doi:10.1007/s100520100884 [ hep - ph/0111078 ] . l. a. harland - lang , v. a. khoze and m. g. ryskin , `` the production of a diphoton resonance via photon - photon fusion , '' jhep * 1603 * ( 2016 ) 182 doi:10.1007/jhep03(2016)182 [ arxiv:1601.07187 [ hep - ph ] ] . i. antoniadis , `` a possible new dimension at a few tev , '' phys . b * 246 * ( 1990 ) 377 . doi:10.1016/0370 - 2693(90)90617-f n. arkani - hamed , s. dimopoulos and g. r. dvali , `` the hierarchy problem and new dimensions at a millimeter , '' phys . b * 429 * ( 1998 ) 263 doi:10.1016/s0370 - 2693(98)00466 - 3 [ hep - ph/9803315 ] . i. antoniadis , n. arkani - hamed , s. dimopoulos and g. r. dvali , `` new dimensions at a millimeter to a fermi and superstrings at a tev , '' phys . b * 436 * ( 1998 ) 257 doi:10.1016/s0370 - 2693(98)00860 - 0 [ hep - ph/9804398 ] . i. antoniadis , k. benakli and a. laugier , `` contact interactions in d - brane models , '' jhep * 0105 * ( 2001 ) 044 doi:10.1088/1126 - 6708/2001/05/044 [ hep - th/0011281 ] . d. m. ghilencea , h. p. nilles and s. stieberger , `` divergences in kaluza - klein models and their string regularization , '' new j. phys . * 4 * ( 2002 ) 15 doi:10.1088/1367 - 2630/4/1/315 [ hep - th/0108183 ] . s. a. abel and a. w. owen , `` interactions in intersecting brane models , '' nucl . phys . b * 663 * , 197 ( 2003 ) doi:10.1016/s0550 - 3213(03)00370 - 5 [ hep - th/0303124 ] . k. r. dienes , e. dudas and t. gherghetta , `` gut precursors and nontrivial fixed points in higher dimensional gauge theories , '' phys . * 91 * , 061601 ( 2003 ) doi:10.1103/physrevlett.91.061601 [ hep - th/0210294 ] . h. partouche , `` large volume supersymmetry breaking without decompactification problem , '' arxiv:1601.04564 [ hep - th ] . j. scherk and j. h. schwarz , `` how to get masses from extra dimensions , '' nucl . b * 153 * ( 1979 ) 61 . doi:10.1016/0550 - 3213(79)90592 - 3 ; `` spontaneous breaking of supersymmetry through dimensional reduction , '' phys . lett . b * 82 * ( 1979 ) 60 . doi:10.1016/0370 - 2693(79)90425 - 8 m. r. buckley , d. feld and d. goncalves , `` scalar simplified models for dark matter , '' phys . d * 91 * ( 2015 ) 015017 doi:10.1103/physrevd.91.015017 [ arxiv:1410.6497 [ hep - ph ] ] . p. harris , v. v. khoze , m. spannowsky and c. williams , `` constraining dark sectors at colliders : beyond the effective theory approach , '' phys . d * 91 * ( 2015 ) 055009 doi:10.1103/physrevd.91.055009 [ arxiv:1411.0535 [ hep - ph ] ] . p. harris , v. v. khoze , m. spannowsky and c. williams , `` closing up on dark sectors at colliders : from 14 to 100 tev , '' arxiv:1509.02904 [ hep - ph ] . y. mambrini , g. arcadi and a. djouadi , `` the lhc diphoton resonance and dark matter , '' arxiv:1512.04913 [ hep - ph ] . m. backovic , a. mariotti and d. redigolo , `` di - photon excess illuminates dark matter , '' arxiv:1512.04917 [ hep - ph ] . c. han , h. m. lee , m. park and v. sanz , `` the diphoton resonance as a gravity mediator of dark matter , '' arxiv:1512.06376 [ hep - ph ] x. j. bi , q. f. xiang , p. f. yin and z. h. yu , `` the 750 gev diphoton excess at the lhc and dark matter constraints , '' arxiv:1512.06787 [ hep - ph ] . m. bauer and m. neubert , `` flavor anomalies , the diphoton excess and a dark matter candidate , '' arxiv:1512.06828 [ hep - ph ] . d. barducci , a. goudelis , s. kulkarni and d. sengupta , `` one jet to rule them all : monojet constraints and invisible decays of a 750 gev diphoton resonance , '' arxiv:1512.06842 [ hep - ph ] . j. c. park and s. c. park , `` indirect signature of dark matter with the diphoton resonance at 750 gev , '' arxiv:1512.08117 [ hep - ph ] . x. j. huang , w. h. zhang and y. f. zhou , `` a 750 gev dark matter messenger at the galactic center , '' arxiv:1512.08992 [ hep - ph ] . f. deramo , j. de vries and p. panci , `` a 750 gev portal : lhc phenomenology and dark matter candidates , '' arxiv:1601.01571 [ hep - ph ] . k. ishiwata and m. b. wise , `` phenomenology of heavy vectorlike leptons , '' phys . d * 88 * , no . 5 , 055009 ( 2013 ) doi:10.1103/physrevd.88.055009 [ arxiv:1307.1112 [ hep - ph ] ] . w. altmannshofer , m. bauer and m. carena , `` exotic leptons : higgs , flavor and collider phenomenology , '' jhep * 1401 * , 060 ( 2014 ) doi:10.1007/jhep01(2014)060 [ arxiv:1308.1987 [ hep - ph ] ] . a. falkowski , d. m. straub and a. vicente , `` vector - like leptons : higgs decays and collider phenomenology , '' jhep * 1405 * , 092 ( 2014 ) doi:10.1007/jhep05(2014)092 [ arxiv:1312.5329 [ hep - ph ] ] . i. antoniadis , s. dimopoulos , a. pomarol and m. quiros , `` soft masses in theories with supersymmetry breaking by tev compactification , '' nucl . b * 544 * ( 1999 ) 503 doi:10.1016/s0550 - 3213(99)00037 - 1 [ hep - ph/9810410 ] . d. s. salopek , j. r. bond and j. m. bardeen , `` designing density fluctuation spectra in inflation , '' phys . d * 40 * ( 1989 ) 1753 . m. p. hertzberg , `` on inflation with non - minimal coupling , '' jhep * 1011 * ( 2010 ) 023 arxiv:1002.2995 [ hep - ph ] . v. v. khoze , `` inflation and dark matter in the higgs portal of classically scale invariant standard model , '' jhep * 1311 * ( 2013 ) 215 doi:10.1007/jhep11(2013)215 [ arxiv:1308.6338 [ hep - ph ] ] . j. elias - miro , j. r. espinosa , g. f. giudice , h. m. lee and a. strumia , `` stabilization of the electroweak vacuum by a scalar threshold effect , '' jhep * 1206 * ( 2012 ) 031 doi:10.1007/jhep06(2012)031 [ arxiv:1203.0237 [ hep - ph ] ] . o. lebedev , `` on stability of the electroweak vacuum and the higgs portal , '' eur . j. c * 72 * ( 2012 ) 2058 doi:10.1140/epjc / s10052 - 012 - 2058 - 2 [ arxiv:1203.0156 [ hep - ph ] ] . v. v. khoze , c. mccabe and g. ro , `` higgs vacuum stability from the dark matter portal , '' jhep * 1408 * ( 2014 ) 026 doi:10.1007/jhep08(2014)026 [ arxiv:1403.4953 [ hep - ph ] ] . a. salvio and a. mazumdar , `` higgs stability and the 750 gev diphoton excess , '' arxiv:1512.08184 [ hep - ph ] . d. lust and s. stieberger , `` gauge threshold corrections in intersecting brane world models , '' fortsch . * 55 * ( 2007 ) 427 doi:10.1002/prop.200310335 [ hep - th/0302221 ] . s. a. abel and b. w. schofield , `` brane anti - brane kinetic mixing , millicharged particles and susy breaking , '' nucl . b * 685 * ( 2004 ) 150 doi:10.1016/j.nuclphysb.2004.02.037 [ hep - th/0311051 ] . m. berg , m. haack and b. kors , `` loop corrections to volume moduli and inflation in string theory , '' phys . d * 71 * ( 2005 ) 026005 doi:10.1103/physrevd.71.026005 [ hep - th/0404087 ] . r. blumenhagen , s. moster and t. weigand , `` heterotic gut and standard model vacua from simply connected calabi - yau manifolds , '' nucl . b * 751 * ( 2006 ) 186 doi:10.1016/j.nuclphysb.2006.06.005 [ hep - th/0603015 ] . s. a. abel , m. d. goodsell , j. jaeckel , v. v. khoze and a. ringwald , `` kinetic mixing of the photon with hidden u(1)s in string phenomenology , '' jhep * 0807 * ( 2008 ) 124 doi:10.1088/1126 - 6708/2008/07/124 [ arxiv:0803.1449 [ hep - ph ] ] . m. cvetic , j. halverson and p. langacker , `` string consistency , heavy exotics , and the @xmath203 gev diphoton excess at the lhc , '' arxiv:1512.07622 [ hep - ph ] . p. anastasopoulos , m. bianchi , e. dudas and e. kiritsis , `` anomalies , anomalous u(1 ) s and generalized chern - simons terms , '' jhep * 0611 * ( 2006 ) 057 doi:10.1088/1126 - 6708/2006/11/057 [ hep - th/0605225 ] . l. e. ibanez , c. munoz and s. rigolin , `` aspect of type i string phenomenology , '' nucl . b * 553 * ( 1999 ) 43 doi:10.1016/s0550 - 3213(99)00264 - 3 [ hep - ph/9812397 ] .

plasmons are commonly interpreted with classical electrodynamics , while nonclassical effects may influence the dynamics of plasmon resonances as the plasmon confinement is approaching the few nanometer scale . however , an unambiguous approach to quantify the degree of nonclassical dynamics remains . we propose a _ nonclassical - impact parameter _ ( nci ) to characterize the degree of nonclassical effects from an energy perspective , i.e. which fraction of the total electromagnetic energy is attributed to classical electrodynamic terms and which fraction is correspondingly to be assigned to nonclassical degrees of freedom ? we show that the nci relates directly to two fundamental parameters of plasmon resonances : the loss function and the quality factor . guided by the nci , we discuss the nonclassical effects of plasmon waveguiding modes of metallic slab waveguides , and highlight the general features of the nonclassical effects at different microscopic levels by contrasting the numerical results from the semi - classical hydrodynamic drude model ( hdm ) and the microscopic random - phase approximation ( rpa ) . the formal relation between the hdm and the rpa is also established for metals by exploring the limit of an infinite work function .

1511.08390

plasmon resonances , collective oscillations of a free - electron gas against a positive ion background , are subject to significant attentions.@xcite plasmon resonances , including both localized - plasmon resonances and plasmon waveguiding modes , can confine electric fields beyond the optical diffraction limit.@xcite this leads to numerous applications such as cancer therapy,@xcite nanophotonics circuits,@xcite nanolasers,@xcite and quantum information processing.@xcite the majority of these developments rely on our insight from classical electrodynamics where the drude model for the intraband plasmonic response of metals stands as a cornerstone.@xcite this celebrated approach explains noble - metal plasmon phenomena extremely well for plasmonic structures with characteristic dimensions well above @xmath0.@xcite however , the maturing nanofabrication allows the realization and exploration of yet smaller feature sizes.@xcite when approaching the few nanometer or even sub nanometer scale , an increasing importance of nonclassical degrees of freedom ( i.e. beyond the drude model ) can be anticipated and with increasing weights.@xcite examples of such nonclassical degrees of freedom include the kinetics associated with the finite compressibility of the quantum electron gas@xcite and the inhomogeneous microscopic equilibrium distribution of the quantum electron gas in the vicinity of a surface.@xcite in maxwell s equations , such aspects modify the local drude permittivity to a generalized nonlocal form.@xcite with respect to different nonlocal models , such as the semi - classical hydrodynamic drude model ( hdm ) and the microscopical random - phase approximation ( rpa ) , the commonly employed drude model is also referred to as the local response approximation ( lra ) . in recent years , efforts have been directed to theoretically@xcite and experimentally@xcite investigating nonclassical effects of plasmon resonances in metals and with interesting extensions to two dimensional ( 2d ) plasmonic materials such as graphene.@xcite in addition to the fundamental interest in quantum plasmonic phenomena , we note that quantum electronic control may open new avenues in applications of nanoscale light confinement@xcite and plasmon - induced hot electrons.@xcite for resonant phenomena , the nonclassical effects usually manifest in broadening and shifting of plasmon resonances.@xcite in turn , nonclassical effects also smear singularity phenomena predicted by the drude model , such as perfect imaging,@xcite nanofocusing@xcite and the field enhancement at the center of a touching dimer@xcite that would otherwise be singular.@xcite nonclassical dynamics may also be associated with phenomena without any classical counterparts , such as the mutipole surface - plasmon resonances at the interface of simple metals,@xcite the charge transfer plasmon resonances for dimers with sub - nanometric gaps,@xcite and the quantized bulk - plasmon resonances above the plasma frequency.@xcite the theoretical understandings of nonclassical effects of plasmon resonances usually employ the semi - classical hdm,@xcite the rpa,@xcite or the time - dependent density - functional theory ( dft).@xcite elevating us beyond these different models , it is interesting to ask if we can introduce a measure to quantify the degree of nonclassical contributions to the plasmon dynamics ? one immediate candidate at hand is to associate such a parameter with the resonance shift of plasmon modes when compared to the lra . obviously , this would work for well developed resonances existing within the lra , while it would not help to appreciate quantum plasmon phenomena not holding classical counterparts . to address a broader variety of nonclassical phenomena , we here propose a _ nonclassical - impact parameter _ ( nci ) to characterize the nonclassical effects from a total energy ( @xmath1 ) perspective : @xmath2 , where @xmath3 for classical ( c ) dynamics , while @xmath4 for entirely nonclassical ( nc ) dynamics . as a key result we show that @xmath5}{q_{\rm pr}}\label{eq : ncirpa}\end{aligned}\ ] ] where @xmath6 is the quality factor of the plasmon resonance while @xmath7 $ ] is the loss function ; both established quantities in the fields of plasmonics and electron - energy loss spectroscopy of plasmons.@xcite the remaining part of the paper is organized as follows : in section ii , with the hdm as the starting point , we partition different forms of the energies of plasmon resonances including the contributions from the nonclassical effects . based on the energy considerations in section ii , we define the nci to characterize the nonclassical effects beyond the hdm assumption in section iii . in section iv , we discuss the relation between the nci and the electron - pressure wave , and demonstrate the fundamental upper limit of the nci for plasmon waveguiding modes in the hdm . in section v , with the nci , we investigate in detail the nonclassical effects by numerical analysis of plasmon waveguiding modes of metallic slab waveguides . in section vi , the formal relation between the hdm and the rpa is discussed in the limit of an infinite work function . in section vii , a summary and conclusions are given . some details of our derivations are left for appendices a , b and c. we start by introducing the hdm to describe the free - electron gas . in short , the hdm seeks to account for the finite compressibility of the quantum electron gas by incorporating a pressure term in the equation - of - motion of electrons subject to electromagnetic fields . commonly , thomas fermi theory is used to describe the nonclassical ( quantum ) freedom of the kinetic energy , which arises from the statistical distribution of the free - electron gas e.g. the fermi dirac statistics in equilibrium . the internal kinetic energy per particle is proportional to @xmath8 , where @xmath9 is the fermi energy , while @xmath10 is the characteristic thermal energy . for metals such as sodium , aluminum , silver and gold , with the fermi velocity @xmath11 around @xmath12 and the effective electron mass close to the electron rest mass , we typically have @xmath13 at room temperatures,@xcite and consequently the nonclassical kinetic energy is governed by @xmath9 . in the present paper , we focus our attention on the above mentioned metals . in the hdm , the classical drude model is accompanied by an additional diffusion - like gradient term@xcite @xmath14 here , @xmath15 and @xmath16 denote the electric current density and the electric charge density , respectively . furthermore , @xmath17 is the damping rate and @xmath18 is the classical drude conductivity with @xmath19 being the plasma frequency . the parameter @xmath20 , associated with the nonclassical kinetic energy , has the expression @xmath21,@xcite and gives the strength of the nonlocal relationship between @xmath22 and @xmath23 . the hdm is usually combined with the assumptions of a uniform equilibrium - electron density and an infinite work function . the two assumptions lead to ( 1 ) all material parameters in eq . ( [ eq : currenthdm ] ) , i.e. , @xmath24 , @xmath17 , and @xmath20 , are set to the bulk values of the corresponding metal ; ( 2 ) an additional boundary condition is imposed , i.e. the vanishing of the normal component of @xmath15 at the metal surface.@xcite before proceeding , we note that the hdm constitutes a lowest - order correction of the lra to include nonlocal effects due to the finite compressibility of the quantum electron gas . apart from recent generalizations,@xcite the hdm still neglects several important microscopic effects including the inhomogeneity of the equilibrium - electron density and its related correction to the nonclassical kinetic energy , and the electron spill - out . nevertheless , the main motivation here of employing the hdm is to illustrate how the nonclassical degrees of freedom are modifying the dynamics of plasmons . this insight also leads us to our introduction of the @xmath25 . at a later stage , the assumptions associated with the hdm will be relaxed . now , consider an arbitrary free - electron gas embedded in a dielectric background with a relative permittivity @xmath26 ( possibly varying in space ) . in the electrostatic limit , the electromagnetic fields associated with plasmon resonances respect@xcite @xmath27=0.\nonumber\end{aligned}\ ] ] utilizing eq . ( [ eq : currenthdm ] ) , the above identity yields @xmath28 with @xmath29 here , @xmath30 is the spatial integration performed over the volume of the free - electron gas , with @xmath31 being the equilibrium - electron density , while @xmath32 is the effective electron mass . we note that the same expressions can also be derived from the poynting theorem.@xcite the total energy @xmath33 and the classical energy @xmath34 are @xmath35 the physical meanings of @xmath36 are indicated by their expressions . in particular , @xmath37 is the time average of the classical kinetic energy of the free - electron gas , and @xmath38 relates with the energy of the electric field . @xmath39 is the nonclassical part of the energy . to appreciate the nonclassical nature of @xmath39 , we consider a free - electron gas with a density @xmath40 , where @xmath31 and @xmath41 are two constants with @xmath41 being a small perturbation . assuming that electrons are obeying fermi dirac statics , the system has the nonclassical energy density @xmath42 . accordingly , the time average of the nonclassical kinetic energy density @xmath43 contributed from the dynamic perturbation @xmath41 is @xmath44 in the limit where @xmath45 with @xmath46 , one finds @xmath47 by recalling that @xmath48 and @xmath49 . this directly reveals the nonclassical origin of @xmath39 . if on the other hand @xmath50 with @xmath51 , then @xmath52 . here , the prefactor @xmath53 is deviating from @xmath54 owing to the fact that the fermi dirac statistics no longer applies as @xmath50.@xcite the total energy of the plasmon resonance is given by eq . ( [ eq : ublance ] ) . the drude model invokes the lra and is entirely classical , i.e. @xmath55 . plasmon resonances are characterized by the harmonic energy transfer between the classical kinetic energy of free electrons and the electric field energy of the whole system , in analogy with optical resonances in dielectrics with the energy transfer between the electric field energy and the magnetic energy.@xcite including the nonclassical degrees of freedom , @xmath39 has a final share of the total energy , and this modifies the classical plasmon dynamics accordingly . based on the above energy perspective , we here introduce the _ nonclassical - impact parameter _ ( nci ) that we already highlighted in the introduction : @xmath56 in the classical drude model , @xmath57 as a direct result of @xmath55 . including the nonclassical effects in the hdm , @xmath25 deviates from @xmath58 , and more specifically becomes a positive value . in the limiting case @xmath59 or equivalently @xmath60 , we have @xmath61 . in a hydrodynamic description this limit corresponds to the regime where plasmons are fully longitudinal oscillations , i.e. electron - pressure waves as discussed in the next section . the @xmath62 is bounded between @xmath58 and @xmath54 . the larger value of the nci , the more noticeable nonclassical effects . as an example , we consider a metallic slab waveguide imbedded in a free - space background medium . for simplicity we consider a loss - less ( @xmath63 ) free - electron metal with material parameters @xmath64ev and @xmath65 m/s ( corresponding to sodium ) . [ fig : slab](a ) depicts the dispersions of the symmetric surface plasmon waveguiding modes ( pwms ) for waveguide widths of both @xmath66 nm and @xmath67 nm , contrasting both the lra classical drude model ( dash - dotted lines ) and the hdm ( solid lines ) . the wavenumber @xmath68 of the waveguide mode is normalized by the fermi wavenumber @xmath69 and likewise , the frequency @xmath70 is conveniently normalized by the plasma frequency @xmath19 . in the drude model , @xmath68 shows a linear relation with @xmath71 , indicating that @xmath68 can be increased to any value without limitations by decreasing @xmath72 . in the hdm , such linear relation is broken due to nonlocal effects which cut off the large@xmath68 divergence ( see ref . for details ) . accordingly , there is a ceiling to @xmath68 ( a maximal permitted value of @xmath68 ) , as we will discuss in the next section . [ fig : slab](b ) plots the nci predicted by the hdm as a function of the frequency @xmath70 . it is seen that the nci follows a similar trend as the wavenumber @xmath68 does in fig . [ fig : slab](a ) . in particular , the nci dramatically increases as @xmath70 exceeds the classical surface - plasmon frequency @xmath73 . below this frequency the plasmon field is predominantly of a transverse nature in good accordance with the lra , while it turns gradually more longitudinal beyond this frequency . [ fig : slab](c ) depicts the nci as a function of @xmath74 . the nci increases as the @xmath68 increases . this directly illustrates how its nonclassical nature increases as @xmath68 becomes larger . the observation can be directly understood from eq . ( [ eq : currenthdm ] ) where the nonclassical term , i.e. , the gradient term , is roughly proportional to @xmath68 . as another example , we consider the nci for the generic problem of a free - electron metallic sphere imbedded in free space . using a mie formulation of the hdm,@xcite we find to lowest order in the particle radius @xmath75 that the nci for localized plasmon resonances directly manifests the nonlocal blueshifting with @xmath76 here , @xmath77 is the plasmon resonance frequency in the lra limit ( @xmath78 for the dipole resonance with @xmath79 ) , @xmath80 is the hdm frequency blueshift with respect to @xmath81 , and @xmath82 is the angular momentum number of the plasmon resonance . this is a quite intuitive , but also remarkable result which in turn shows how the loss function , the quality factor , and the nonlocal blueshift constitute mutually linked quantities , i.e. @xmath83/q_{\rm pr}$ ] [ see eq . ( [ eq : ncirpa ] ) or the following subsection for definitions of @xmath84 and @xmath6 ] . nm and 5 nm ) imbedded in a free - space background . ( a ) dispersion relations within the hdm and the lra classical drude model are plotted as the solid lines and the dash - dotted lines , respectively . ( b ) the nci illustrated as a function of the plasmon wavenumber . the upper right inset illustrates the mode profile of the symmetric surface pwm . the metal has the material parameters @xmath64ev and @xmath65 m/s . , scaledwidth=42.0% ] above , we used the semi - classical hdm as our starting point for our introduction of the nci . however , our energy considerations ( involving @xmath86 and @xmath87 ) go beyond this particular model . thus , the @xmath25 can also be derived for microscopic quantum models , where more subtle nonclassical effects can be included . here , we turn to the random - phase approximation as the most common way of calculating the electromagnetic response with a starting point in the single - electron eigenstates of the equilibrium system . in the rpa , the exchange - correlation contributions to the plasmon dynamics are omitted . we note that for our metals of interest with only modest electronic correlations , the induced error is of only quantitative rather than qualitative character . to discuss the nci in the rpa , we introduce the spectral representation of the dielectric function operator @xmath88 with@xcite @xmath89 where @xmath90 and @xmath91 represent the states of the external incident and resulting total electric potentials , respectively . here , @xmath92 , where @xmath93 is the coulomb interaction operator defined by @xmath94 , and @xmath95 is defined by @xmath96 , where @xmath97 is the non - interacting response function constructed from the single - electron orbitals of the underlying equilibrium system . plasmon resonances are associated with the poles of @xmath98 . in particular , they are the eigenstates of @xmath88 @xmath99 with @xmath100 being zero . however , due to the loss , e.g. the electron - momentum relaxation associated with electron - phonon scattering and the electron - hole pair excitations , the pole will be slightly away from the real axis in the complex frequency plane . since this complicates the evaluation of the complex - valued resonance frequency we here make the simplification to focus on the real part of the frequency ( where @xmath101 has a complex value ) . we define plasmon resonances in a loose mathematical sense as the eigenstates with the loss function @xmath102}$ ] exhibiting a local spectral peak.@xcite those plasmon resonance ( pr ) eigenstates and eigenvalues are specifically denoted as @xmath103 and @xmath84 , respectively . the loss function @xmath104 $ ] is a common concept in electron microscopy of plasmons@xcite where it quantifies the ability of external free carriers ( such as a focused electron beam in an electron microscope where the swift electrons act as the sources of @xmath90 ) to couple to plasmons and in this way dissipate energy . within the above framework , we find that the nci can be expressed in a rather elegant form , eq . ( [ eq : ncirpa ] ) that is @xmath5}{q_{\rm pr}},\nonumber\end{aligned}\ ] ] comprising the loss function @xmath105 $ ] and the quality factor @xmath6 , which characterizes the plasmon resonance lifetime . for the details of the derivation we refer to appendix b. again , we emphasize that in this way our new measure of quantum effects is related directly to long - established key quantities for plasmon resonances ; quantities that can in principle be evaluated by various methods and techniques ranging all the way from microscopic theories and _ ab - initio _ approaches over semi - classical models to even experiments measuring the far - field optical spectra and electron - energy loss spectra of plasmon resonances . note that compared to eq . ( [ eq : nci ] ) the form of eq . ( [ eq : ncirpa ] ) no longer involves @xmath87 , for which we do not find a simple expression in the rpa , while quite conveniently the quality factor and the loss function can be evaluated ( see appendix a for details ) . since we have arrived at eq . ( [ eq : ncirpa ] ) with more general arguments , the expression is also valid for the classical drude model and the semi - classical hdm , since both models represent different levels of approximations of the rpa . in particular , the classical drude model is the local - response approximation of the rpa , while the hdm is a @xmath106-approximation ( @xmath107 being the wavenumber of the electric potential after the fourier transformation ) of the rpa . thus , eq . ( [ eq : ncirpa ] ) offers a generic recipe to quantify the nonclassical effects . as an immediate consequence of eq . ( [ eq : ncirpa ] ) , we have ( in agreement with ref . ) @xmath108\;\text { ( within the lra ) } \label{eq : qpdrude}\end{aligned}\ ] ] since the drude model is classical and @xmath57 within the lra . as an example , consider the bulk plasmon resonance for a homogenous free - electron gas . in the classical drude model , the dielectric function operator @xmath88 is simply the drude permittivity @xmath109 . at the bulk plasmon resonance @xmath110 , it is evident that @xmath105=\omega_{\rm p}/\gamma$ ] , which also gives the value of the quality factor @xmath6 according to eq . ( [ eq : qpdrude ] ) . thus , our considerations are for the lra in perfect line with the elegant treatment of plasmon resonances in ref . . further , to confirm the equivalence between eqs . ( [ eq : nci ] ) and ( [ eq : ncirpa ] ) , we use the hdm to calculate the nci of the symmetric surface pwm for a metallic slab waveguide imbedded in a free - space background . the material parameters of the metal are the same as in fig . [ fig : slab ] and for the width of the waveguide we consider 5 nm . the results are plotted in fig . [ fig : benchmark ] , where we observe that the values of the nci from either eq . ( [ eq : nci ] ) or eq . ( [ eq : ncirpa ] ) agree perfectly with each other . ) [ solid line ] and eq . ( [ eq : nci ] ) [ data points ] illustrated by hdm calculation of the nci for a symmetric surface pwm of a metallic 5 nm slab imbedded in a free - space background . the material parameters of the metal are the same as in fig . [ fig : slab ] . a phenomenological damping rate @xmath111 is added when using eq . ( [ eq : ncirpa ] ) , while the damping is neglected when evaluating eq . ( [ eq : nci ] ) . , scaledwidth=42.0% ] in this section , we return to the hdm and discuss the nci in the context of longitudinal plasmon excitations . in the limiting case with @xmath112 , we have @xmath113 as indicated by eq . ( [ eq : nci ] ) . in this case , plasmon resonances exhibit a harmonic energy transfer mainly between @xmath37 and @xmath39 , but with negligible @xmath86 . consequently , the electric field term in eq . ( [ eq : currenthdm ] ) can be simply neglected . then , by applying the gradient operator to eq . ( [ eq : currenthdm ] ) we get @xmath114 the solutions to eq . ( [ eq : preeq ] ) are charge - density waves that we shall here refer to as _ electron - pressure waves _ since the nonclassical gradient term in eq . ( [ eq : currenthdm ] ) is commonly referred to as the pressure term of the free - electron gas.@xcite thus , as @xmath113 , the linear - response dynamics of the free - electron gas has the property of an electron - pressure wave . as mentioned above , the hdm is the @xmath106-approximation of the rpa of a homogenous electron gas . consequently , it is important to see whether the value of @xmath115 predicted by the hdm is close to that of the rpa . in this context , we observe that the longitudinal permittivity in the hdm is given by @xmath116 clearly , @xmath115 is nothing but the resonance pole of @xmath117 if @xmath63 . within the rpa this inspires us to extract @xmath115 in a similar way by solving for the resonance of @xmath117 . as an example , we considering a metal with properties as in fig . [ fig : slab ] . the left panel of fig . [ fig : rh ] depicts @xmath117 as a function of @xmath107 at @xmath118 for both the hdm and the rpa . the shaded regions represent the intraband electron - hole ( e - h ) pair continuum , which contributes to the imaginary part of @xmath117 in the rpa . in the hdm , a phenomenological damping @xmath111 is added . clearly , for both the hdm and the rpa , @xmath117 exhibits the typical resonance features : the abrupt sign change of @xmath119 $ ] accompanied by a peak in @xmath120 $ ] . the latter feature is now employed to identify the resonance position ( marked by squares ) , i.e. , the value of @xmath115 . we see that @xmath115 in the rpa is close to that in the hdm with @xmath121 . the right panel of fig . [ fig : rh ] depicts @xmath117 as a function of @xmath107 at @xmath122 . comparing with the case of @xmath118 , we emphasize two observations : ( 1 ) @xmath115 in the rpa deviates more from that in the hdm with @xmath123 ; ( 2 ) the resonance of the electron - pressure wave in the rpa is more damped due to the enhanced e - h pair excitations . in summary , the @xmath115 predicted by the hdm is qualitatively accurate and compares well with the rpa , especially in the low - frequency regime . ( left panels ) and @xmath122 ( right panels ) . the material parameters are as in fig . [ fig : slab ] , while using @xmath111 in the hdm . the squares mark the positions of @xmath115 while shaded regions represent the rpa intraband electron - hole continuum.,scaledwidth=48.0% ] in fig . [ fig : benchmark ] , we observe that the nci increases as the pwm wavenumber @xmath68 grows . thus , it is reasonable to conjecture that the ceiling to @xmath68 should correspond to the limiting case @xmath124 , i.e. , dynamics characteristic to an electron - pressure wave . this immediately suggests the following relation : @xmath125 for a slab metallic waveguide in a homogenous dielectric background , eq . ( [ eq : kspineq ] ) has been demonstrated explicitly by analyzing the dispersion equation.@xcite it has also been shown that such a ceiling is responsible for the limiting value of the photonic density - of - states of layered hyperbolic metamaterials.@xcite here , based on the arguments from the energy perspective , eq . ( [ eq : kspineq ] ) is provided without any detailed specifications of either the underlying waveguide geometry or the imbedding medium s dielectric properties . this suggests that eq . ( [ eq : kspineq ] ) should be universally valid within the hdm , which is rigorously proven in appendix a. to allow @xmath68 to approach @xmath115 , fig . [ fig : slab ] tells us that one should either make the waveguide s transverse dimensions small or alternatively aim for the high - frequency regime . however , even if the waveguide would hypothetically be as thin as one atomic layer , @xmath68 would still be far away from @xmath115 , as will be demonstrated in the next section . additionally , in the high - frequency regime , the enhanced loss due to the e - h pair excitations can reduce the plasmon lifetime greatly ( as observed in fig . [ fig : rh ] ) , which would make experimental observations difficult . in this sense , the fundamental ceiling defined in eq . ( [ eq : kspineq ] ) is perhaps not posing a real practical limitation in the presence of realistic damping , but it rather serves as a theoretical concept manifesting the more ultimate limitation associated with the spatial dispersion of electron - pressure waves and their nonclassical kinetic energy . the hdm invokes the following approximations and assumptions : ( a ) the @xmath106-approximation of the rpa ; ( b ) a uniform equilibrium - electron density ; ( c ) an infinite work function . to investigate the nonclassical effects beyond these simplifications , we here employ the microscopic rpa for numerically analyzing the pwms of metallic slab waveguides . the rpa automatically lifts the @xmath106-approximation inherent to the hdm . procedures to further relax the remaining approximations and assumptions will depend on the particular choice of the single - electron potential @xmath126 . here , two different models for @xmath126 are considered . one approach is the infinite work function ( ifw ) potential well , in which @xmath127 inside the metal , i.e. , the region occupied by the jellium ion background , while @xmath128 outside the metal , as illustrated in the left inset of fig . [ fig : mode ] . in this case , the assumption of the uniform electron density is relaxed . this allows for friedel oscillations in the equilibrium - electron density , while quantum spill - out is suppressed due to the iwf ( for an early application of the infinite - barrier idea to plasmons , see ref . ) . the other choice involves a single - electron potential @xmath126 treated self - consistently within dft a more accurate description,@xcite as illustrated in the right inset of fig . [ fig : mode ] . in the dft , the local - density approximation of the exchange - correlation energy is employed . in this case , the assumptions of the uniform electron density and the infinite work function are both elevated . in the following , our comparison of the two rpa descriptions ( the ifw rpa and the dft rpa ) and the hdm will exhibit nonclassical effects of plasmon resonances at different microscopic levels . figure [ fig : mode ] depicts the dispersions of the symmetric pwms for a metallic slab of width @xmath129 , contrasting the results from the hdm ( dash - dotted lines ) , the ifw rpa ( left ) and the dft rpa ( right ) . in the dft , we employ the jellium approximation for the ion lattices , and self - consistently obtain a work function of 3ev ( see the inset of the right part of fig . [ fig : mode ] ) . the material parameters of the metal are as in fig . [ fig : slab ] , while a phenomenological damping rate @xmath130 is included . for the ifw rpa and the dft rpa , we illustrate the largest values of the loss function @xmath131}$ ] ( see sec . [ sec : nipb ] ) divided by the frequency , whose local peaks represent the pwms . the demonstrated pwms include both the surface pwms and the bulk pwms above @xmath19 . .,scaledwidth=48.0% ] we focus on the surface pwms ( the mode of the low - frequency branch ) . the results of the hdm and the ifw rpa agree with each other . this indicates the quantitative agreement between the two models when it comes to characterizing the properties of the surface screening of the free - electron gas . this will be further discussed in sec.[sec : rhdmrpa ] . for the dft rpa , the predicted surface pwms exhibit : ( 1 ) redshifting with respect to the hdm and ifw rpa , ( 2 ) more damping as @xmath68 increases . these two observations are attributed to the electron spill - out permitted by a finite work function . in particular , the increased damping relates with the enhanced e - h pair excitations near the jellium boundary ( surface scattering ) due to the electron spill - out , as recently discussed in refs . . contrasting the hdm , the ifw rpa , and the dft rpa results . ( lower panel ) @xmath132 the damping rate of the surface pwms contributed from the e - h pair excitations contrasting the ifw rpa and the dft rpa . rpa results are illustrated as the stripes with the vertical width representing the numerical uncertainty of determining the quality factor of the pwm . , scaledwidth=42.0% ] to quantify the nonclassical effects , we next compute the nci . [ fig : qi1 ] depicts the nci for the surface pwms considered in fig . [ fig : mode ] , contrasting the hdm , the ifw rpa , and the dft rpa . to compute the nci in the rpa , the quality factor of the pwms is needed . while this is indeed feasible , we note that the numerical determination of the quality factor comes with a small numerical uncertainty , especially for the dft rpa ( see appendix b for details ) . to visualize the numerical accuracy , the nci for the dft rpa is accompanied by a shaded region representing the numerical uncertainty . rpa gives an nci close to that of the hdm , which is consistent with the agreement of their mode dispersions in fig . [ fig : mode ] . this suggests that the nonclassical energy of the pwms in the ifw rpa is mainly in the form of the hydrodynamic - like nonclassical kinetic energy . rpa reveals a higher value of the nci , which underlines the account for extra nonclassical degrees of freedom . it is reasonable to deduce that such extra degrees of freedom are associated with the finite work function and the associated electron spill - out , which constitute the main difference between the dft rpa and the other two models . as an evidence for this , we also plot the damping rate @xmath132 due to e - h pair excitations in fig . [ fig : qi1 ] . it is seen that @xmath132 is higher within the dft rpa than in the ifw rpa . thus , for the dft rpa , the proportion of energy carried by the e - h pairs is larger , which of course in turn serves to increase the nci . in section [ sec : fceil ] , we discussed the fundamental ceiling @xmath115 to the wavenumber of pwms as desribed within the hdm . to approach this fundamental ceiling , it is preferable to have metallic waveguides with reduced transverse dimensions . considering a metallic slab , the thinnest imaginable slab is conceptually that of a two dimensional ( 2d ) atomic monolayer.@xcite in this subsection , we will investigate the pwms of such a monolayer and discuss the practical feasibility of approaching @xmath115 . is also shown ( solid lines ) . the atomic monolayer is chosen to represent the ( 100 ) surface of sodium , as illustrated in the inset in the left part panel.,scaledwidth=48.0% ] for the free - electron gas supported by a 2d atomic monolayer , the hydrodynamic equation , eq . ( [ eq : currenthdm ] ) , requires a slight modifications owing to the dimensional reduction . in particular , @xmath133 and @xmath134 in eq . ( [ eq : currenthdm ] ) are the induced electric current density and electric charge density on the 2d plane . additionally , the drude conductivity is expressed as @xmath135 where @xmath136 is the equilibrium - electron density of the 2d monolayer . furthermore , the nonlocal parameter @xmath20 is @xmath137 for @xmath50 , and @xmath138 for @xmath45 , where @xmath139 is the fermi velocity of the 2d free - electron gas.@xcite besides the hdm , we also employ the ifw rpa and the dft rpa to characterize the pwms . in the ifw rpa , the width of the infinite work function potential well @xmath140 is chosen by @xmath141 . for the dft rpa , the ion pseudo - potential proposed by ashcroft@xcite is used to describe the ion lattices beyond the jellium approximation . finally , for numerical convenience we neglect the dependence of the pseudo - potential on the 2d lattice by spatial averaging within the plan while preserving the out - of plane modulation of the potential ( see right inset in fig . [ fig : monlayer ] ) . the atomic monolayer of the ( 100 ) surface of sodium is taken as a thought example . such a monolayer would have a square lattice with lattice constant @xmath142 and accordingly @xmath143 . [ fig : monlayer ] depicts the dispersions of the symmetric pwms , contrasting the hdm ( dash - dotted lines ) , the ifw rpa ( left ) and the dft rpa ( right ) , and the ceiling wavenumber @xmath115 ( solid lines ) . with our eyes guided by the mode dispersion of the hdm , we observe that the results predicted by the hdm and the ifw rpa have a good mutual agreement , while the mode dispersion of the dft rpa exhibits a redshifting and an enhanced damping as @xmath68 increases . additionally , we see that the fundamental ceiling wavenumber @xmath115 is much larger than @xmath68 . this consolidates our previous statement that @xmath115 mainly constitutes an ultimate limitation for plasmon focusing beyond the diffraction limit of light.@xcite figure [ fig : qi2 ] depicts the nci for the pwms in fig . [ fig : monlayer ] . similar to our observations in fig . [ fig : qi1 ] , the values of the nci are close to each other for the hdm and the ifw rpa , while the dft rpa predicts a higher value due to the inclusion of additional nonclassical degrees of freedom . in fig . [ fig : qi2 ] we also plot @xmath132 , contrasting the ifw rpa and the dft rpa . we see that @xmath132 for the ifw rpa is nearly zero and this also holds approximately in the dft rpa for @xmath68 smaller than @xmath144 ( marked by the vertical dash - dotted line ) . the reason is that the plasmon dynamics of such a monolayer resembles the intraband dynamics of an ideal 2d electron gas as long as the excitation energy is smaller than the minimum energy @xmath145 required for e - h interband excitations . for the ifw rpa the energy should exceed @xmath146ev , while for dft rpa the energy is much reduced to @xmath147ev . accordingly , when the energy is smaller than @xmath145 , the e - h pair excitations can only originate from intraband excitations , which is impossible for @xmath148.@xcite for the dft rpa , the e - h interband excitations are possible when the excitation energy exceeds 1.8ev , and can be further enhanced as the excitation energy approaches and eventually exceeds the work function of 2.32ev . this explains our observations discussed earlier in this paper : the @xmath132 remains low unless @xmath149 , in which case the excitation energy favors noticeable e - h interband excitations . finally , it is interesting to note that the numerical uncertainties of the nci and @xmath132 exhibit a correlation with the magnitude of @xmath132 , which is discussed in more detail in appendix a. , contrasting the hdm , the ifw rpa , and the dft rpa . the damping rate @xmath132 ( lower panel ) of the surface pwms contributed from the e - h pair excitations , contrasting the ifw rpa and the dft rpa . for the dft rpa results , the shaded region illustrates the numerical uncertainty of determining the quality factor of the pwm . , scaledwidth=42.0% ] as another example of a 2d material , we briefly discuss doped graphene nanostructures.@xcite on the one hand , classical electrodynamics will be unable to resolve atomic - scale details . on the other hand , the underlying graphene lattice can in fact be cut into flakes in two distinct different ways which host slightly different electronic properties : flakes where edge - atoms are configured in either an armchair or zigzag fashion . the latter is giving rise to localized electronic edge states not supported by the former . these zigzag edge - states host additional optical transitions not found for armchair structures.@xcite this is indeed also seen for the plasmon spectra of ribbons@xcite and triangularly shaped flakes,@xcite where zigzag structures exhibit additional quantum dynamics not exhibited by the armchair counterparts.@xcite in fact , the different magnitude in loss functions reported in ref . are in agreement with our anticipation that the zigzag structures exhibit a larger nci than their armchair counterparts . the numerical results of the pwms for a metallic slab show an excellent agreement between the hdm and the ifw rpa , see figs . [ fig : mode ] and [ fig : monlayer ] . is this just by pure accidence or is it a manifestation of two more closely connected approaches ? here , we argue that the latter is indeed the case . we note that a formal relation between the hdm and the rpa is already established in section ii b of ref . , while we in appendix [ app : rhdmrpa ] revisit the problem to offer a more explicit connection between the hdm and the ifw rpa . it should be emphasized that the ( semi-)infinite electron gas obviously does not support any discretization of electron orbitals , i.e. , quantum - size effects associated with the particle - in - a - box picture.@xcite for an arbitrary confined electron gas occupying a finite region of space , quantum size effects would be captured in the rpa of the bounded electron gas , but not in the corresponding hdm . this can lead to an additional discrepancy between the hdm and rpa for the bounded electron gas . however , only minor differences are to be anticipated if quantum confinement is limited , i.e. for dimensions of the metal volume significantly exceeding the fermi wavelength . for noble metals , the fermi wavelength is sub - nanometric ( @xmath150 nm ) and in few - nanometer sized metal particles , thermal smearing would typically exceed the level spacing of the particle - in - a - box picture . for weakly doped semiconductors and low - dimensional materials , the fermi wavelength can be much longer and quantization effects can indeed appear ( quantized conductance is a famous consequence of this@xcite ) . additionally , in the extreme case like a 2d monolayer where strong out - of - plane quantum confinement may serve to only have electrons occupying a single 2d sub - band , the hdm for 2d in - plane bounded electron gas again serves the @xmath106-approximation of the rpa for the corresponding bounded electron problem . from this perspective , the observed numerical agreement between the hdm and the ifw rpa in figs . [ fig : mode ] and [ fig : monlayer ] is to be expected . for doped graphene nanostructures ( without edge - state contributions associated with zigzag edge termination ) we have also reported similar good agreement between hdm and rpa.@xcite in conclusion , we have proposed a _ nonclassical - impact parameter _ ( nci ) to quantify the nonclassical effects of plasmon resonances from an energy perspective . importantly , we have provided a general expression for the nci which links up to quantities commonly employed within classical electrodynamics in the discussion of plasmon resonances : the loss function and the quality factor . we have discussed the relation between the limiting value of the nci and the electron - pressure wave in the hydrodynamic drude model ( hdm ) and explored the ultimate ceiling to the confinement of the plasmon waveguiding modes . guided by the nci , nonclassical effects have been explored numerically for plasmon waveguiding modes , contrasting the hdm and the microscopic random - phase approximation ( rpa ) . finally , we have detailed the formal relation between the hdm and the rpa for a metal slab represented as an infinite work function potential well . we acknowledge thomas christensen for catalyzing our conceptual development of the nonclassical - impact parameter and thank martijn wubs for insightful comments on our manuscript . the center for nanostructured graphene is sponsored by the danish national research foundation , project dnrf58 . the work was also supported by the danish council for independent research natural sciences , project 1323 - 00087 . is financially supported by the lundbeck foundation , grant no . for plasmon resonances , the total energy @xmath87 and the power dissipation @xmath151 are related by the quality factor @xmath6 through the relation @xmath152 thus , we already now see how the quality factor enters eq . ( [ eq : nci ] ) . next , as mentioned in section [ sec : micro ] , plasmon resonances relate with the eigenstate of the dielectric function operator @xmath153 , which in its spectral representation can be expressed as@xcite @xmath154 here @xmath155 represents the eigenstate of @xmath153 . the @xmath156 and @xmath157 satisfy the potential density relation @xmath158 among @xmath156 , the electric potential state of the plasmon resonance is denoted as @xmath159 . the eigenvalue is @xmath160 with @xmath161 $ ] exhibiting a peak at @xmath162 . the related electron charge density state @xmath160 at @xmath162 is defined by @xmath163 next , consider an arbitrary external electric potential state @xmath164 at the resonant frequency @xmath162 incident on the plasmonic structure . the induced potential state of plasmon resonances , denoted as @xmath165 , has the expression @xmath166 the dissipation energy contributed from @xmath165 is @xmath167,\end{aligned}\ ] ] and the electric field energy associated with @xmath165 is @xmath168 with eq . ( [ eq : utn ] ) , expressions for @xmath151 and @xmath87 can be derived and eq . ( [ eq : nci ] ) is eventually rewritten as @xmath169 . \label{eq : ncia1}\end{aligned}\ ] ] to excite plasmon resonances , the most efficient way is to choose @xmath170 . in this way we finally arrive at eq . ( [ eq : ncirpa ] ) . we note that the nci is an intrinsic property of plasmon resonances , and should be independent of the external potential . however , the nci in eq . ( [ eq : ncia1 ] ) shows a dependence on the external potential , and eq . ( [ eq : ncirpa ] ) is derived with a special choice of @xmath171 . this is because @xmath172 is only a close approximation to plasmon resonances with losses . evidently , if the loss is zero , we have @xmath173 , and the dependence on the external potential of the nci is removed . the final expression of the nci is again eq . ( [ eq : ncirpa ] ) . to compute the nci , @xmath6 and @xmath174 $ ] are needed . $ ] can be directly computed by solving the eigenvalues of the operator @xmath153 . for @xmath6 , it can be expressed as @xmath175 with @xmath176 $ } .$ ] to determine @xmath177 , we use a lorentzian function to fit the spectrum of @xmath178 $ ] , and the width of the fitted lorentzian function is @xmath177 . from the numerical observations , it is found that such a scheme to determine @xmath177 is appropriate for the ifw rpa with the fitting error below @xmath179 . however , for the dft rpa , the fitting error for parameters in figs . [ fig : qi1 ] and [ fig : qi2 ] is larger with the highest value approaching approximately @xmath180 . the fitting error gives the numerical uncertainty of @xmath177 , which is visualized by the shaded regions in figs . [ fig : qi1 ] and [ fig : qi2 ] . further , we deduce that the larger fitting error in the dft rpa relates with the stronger e - h excitations , which distort the resonance spectrum from the lorentzian shape . the deduction is indirectly evidenced by the observations in fig . [ fig : qi2 ] that the numerical uncertainty increases with @xmath132 . to prove eq . ( [ eq : kspineq ] ) , we first rewrite eq . ( [ eq : currenthdm ] ) in terms of the polarization field @xmath181 defined by @xmath182 and the electric field @xmath22 : @xmath183 here , we have without loss of generality suppressed the damping @xmath17 . next , by applying the operation @xmath184 to both sides of eq . ( [ eq : hdmp ] ) we arrive at @xmath185 here , the cross - sectional plane of the waveguide is chosen to be the @xmath186-@xmath187 plane and subscripts are used to indicate vector components parallel to the cross - sectional plane . to derive eq . ( [ eq : ksphdm ] ) , we employed the additional boundary condition that the normal component of @xmath181 is vanishing at the boundary . next , we focus on the integrand ( in the following denoted by @xmath188 ) in the numerator on the right - hand side of eq . ( [ eq : ksphdm ] ) : @xmath189 where @xmath190 for @xmath191 , we may write the electrical field as @xmath192 , where @xmath193 represents the electric potential , and for the polarization we have @xmath194 from eq . ( [ eq : hdmp ] ) . in this way we have @xmath195 using the identity @xmath196 , the integral @xmath197 is found to be @xmath198 here , @xmath199 is the integration along the boundary of the metal side , and @xmath200 is the surface normal pointing outward the metal . employing the boundary conditions that @xmath201 and @xmath202 , where the subscripts `` m '' and `` d '' indicate the quantities at the boundary belonging to the metal and the imbedding dielectric medium , respectively , the integration @xmath203 can be rewritten as @xmath204 in summary , we have @xmath205 for the dielectric background with @xmath206 , it is clear that @xmath207 . next , we turn to @xmath208 . the potential @xmath193 in the metal can be decomposed into the transverse component @xmath209 and the longitudinal component @xmath210 , respectively . these potentials satisfy @xmath211 and @xmath212 , where @xmath213 . in terms of @xmath209 and @xmath210 , @xmath181 is expressed as @xmath214.\nonumber\end{aligned}\ ] ] the above expression directly leads to @xmath215 which implies that @xmath216 . with eq . ( [ eq : ksphdm ] ) and the inequalities @xmath207 and @xmath216 , we now finally arrive at eq . ( [ eq : kspineq ] ) . first , considering an extended electron gas with a uniform equilibrium - electron density . in this case , it is known that the hdm is the @xmath106-approximation of the rpa . we represent this symbolically as @xmath217 with the subscript reminding us of the underlying equilibrium assumption of an infinite homogenous electron gas . next , consider the bounded electron gas , where the spatial extension is now finite . in this case , the hdm assumes a uniform equilibrium - electron density while the act of an infinite work function is incorporated in the equations through an additional boundary condition that accounts for the vanishing electron flux through the surface of the metal . for clarity , we symbolically denote this model as `` hdm@xmath218 '' in order to clearly distinguish it from the case of the infinite electron gas . to establish the relation between the hdm@xmath218 and the rpa , we turn to the hdm@xmath218 in an integral form @xmath219\nonumber\end{aligned}\ ] ] resembling the form within rpa . for an arbitrary electron gas , we do not find a general way to derive the expression of @xmath220 . to make progresses , we put the generality aside , and focus on the specific case of a semi - infinite electron gas with a planar boundary . the new light shed by this specific example will then guide us to a more general conclusion . the semi - infinite electron gas is located at @xmath221 . with eq . ( [ eq : currenthdm ] ) and the additional boundary condition , @xmath220 is then derived : @xmath222 here , @xmath223 is the response function of the hdm for the infinite homogenous electron gas , and @xmath224 is the parity operator on the @xmath186 coordinate defined by @xmath225 . the first response function has the expression @xmath226 where @xmath115 is the wavenumber of the electron - pressure wave defined in eq . ( [ eq : preeq ] ) . ( [ eq : rpahdm0 ] ) and ( [ eq : xhdmb ] ) together suggest that the hdm@xmath218 can be considered as the @xmath106-approximation of a specific rpa denoted as rpa@xmath218 , with the response function @xmath227 being @xmath228 here , @xmath229 is the response function of the rpa for the infinite homogenous electron gas . with the expression of @xmath229 , the @xmath227 can be written as @xmath230 width @xmath231 being the response function for the ifw rpa , whose physical meaning is presented in sec.[sec : micro ] . in the second term , @xmath232 is the response function for the bounded electron gas satisfying the boundary condition that the normal component of the derivative of the electron wave function is vanishing at the surface . this boundary condition is mathematically well defined , provided that the electron mass approaches zero at the boundary , i.e. , a _ zero - mass boundary _ ( zmb ) . the relation between the hdm@xmath218 and the rpa@xmath218 for the semi - infinite electron gas can now be symbolically summarized as @xmath233 below , we provide an intuitive shortcut to better appreciate eq . ( [ eq : rpahdm ] ) . first , in the ifw rpa and the zmb rpa , the electron wave functions @xmath234 are determined by the boundary conditions @xmath235 and @xmath236 , respectively . thus , we have @xmath237 and @xmath238 , where the subscripts indicate the parallel components of the vectors in the @xmath187-@xmath239 plane . the wave functions ensure that the normal component of the electron current vanishes at the boundary , which is consistent with the additional boundary condition in the hdm@xmath218 . additionally , a uniform equilibrium - electron density is assumed within the hdm@xmath218 . this feature is respected by the average of the equilibrium - electron density of the ifw rpa and the zmb rpa , i.e. , @xmath240 . thus , important features of the hdm@xmath218 are all covered by the average of the ifw rpa and the zmb rpa , which highlights the validity of eq . ( [ eq : rpahdm ] ) . 84ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop _ ( , , ) link:\doibase 10.1038/nphoton.2009.282 [ * * , ( ) ] link:\doibase 10.1038/nmat2630 [ * * , ( ) ] link:\doibase 10.1039/c5fd90020d [ * * , ( ) ] link:\doibase 10.1016/j.physrep.2015.07.002 [ * * , ( ) ] link:\doibase 10.1016/j.canlet.2004.02.004 [ * * , ( ) ] link:\doibase 10.1038/nphoton.2014.2 [ * * , ( ) ] link:\doibase 10.1039/c5nr01282a [ * * , ( ) ] link:\doibase 10.1038/nature08318 [ * * , ( ) ] link:\doibase 10.1038/nature08364 [ * * , ( ) ] link:\doibase 10.1038/nphys708 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/27/18/183204 [ * * , ( ) ] link:\doibase 10.1021/nl302315 g [ * * , ( ) ] link:\doibase 10.1038/nature11653 [ * * , ( ) ] link:\doibase 10.1021/nl304078v [ * * , ( ) ] link:\doibase 10.1038/ncomms5125 [ * * , ( ) ] link:\doibase 10.1038/nphoton.2015.141 [ * * , ( ) ] link:\doibase 10.1038/nphys2615 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1103/physrevb.84.121412 [ * * , ( ) ] link:\doibase 10.1103/physrevb.1.4555 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.115.193901 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.115.137403 [ * * , ( ) ] link:\doibase 10.1038/ncomms4809 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1021/nl300269c [ * * , ( ) ] link:\doibase 10.1038/ncomms1806 [ * * , ( ) ] link:\doibase 10.1021/jp401887y [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.263901 [ * * , ( ) ] link:\doibase 10.1021/acsphotonics.5b00246 [ * * , ( ) ] link:\doibase 10.1557/jmr.2015.232 [ * * , ( ) ] link:\doibase 10.1038/nature10904 [ * * , ( ) ] link:\doibase 10.1126/science.1224823 [ * * , ( ) ] link:\doibase 10.1515/nanoph-2012 - 0032 [ * * , ( ) ] link:\doibase 10.1021/nl4018463 [ * * , ( ) ] link:\doibase 10.1126/science.1248797 [ * * , ( ) ] link:\doibase 10.1038/ncomms9788 [ * * , ( ) ] link:\doibase 10.1021/acsnano.5b05568 [ ( ) , 10.1021/acsnano.5b05568 ] link:\doibase 10.1021/nn204780e [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.241414 [ * * , ( ) ] link:\doibase 10.1038/srep09535 [ * * , ( ) ] link:\doibase 10.1038/nnano.2015.165 [ * * , ( ) ] link:\doibase 10.1038/nnano.2014.311 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.2200 [ * * , ( ) ] link:\doibase 10.1016/0079 - 6816(82)90001 - 6 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.50.1316 [ * * , ( ) ] link:\doibase 10.1016/0039 - 6028(91)90142-f [ * * , ( ) ] link:\doibase 10.1021/nl047957a [ * * , ( ) ] link:\doibase 10.1021/nl301478n [ * * , ( ) ] link:\doibase 10.1039/c4fd00193a [ * * , ( ) ] link:\doibase 10.1364/oe.20.004176 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.106802 [ * * , ( ) ] link:\doibase 10.1364/oe.14.009988 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.197601 [ * * , ( ) ] link:\doibase 10.1515/nanoph-2013 - 0014 [ * * , ( ) ] link:\doibase 10.1364/oe.21.001430 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.115401 [ * * , ( ) ] link:\doibase 10.1143/jpsj.80.044606 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.245129 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.235433 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.82.209 [ * * , ( ) ] link:\doibase 10.1103/physrevb.51.7497 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.205429 [ * * , ( ) ] link:\doibase 10.1038/ncomms8132 [ * * , ( ) ] link:\doibase 10.1103/physrevb.91.115416 [ * * , ( ) ] link:\doibase 10.1364/oe.23.029738 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.97.206806 [ * * , ( ) ] link:\doibase 10.1088/0305 - 4608/11/1/021 [ * * , ( ) ] link:\doibase 10.1063/1.3328126 [ * * , ( ) ] link:\doibase 10.1038/nnano.2014.310 [ * * , ( ) ] link:\doibase 10.1021/nn406153k [ * * , ( ) ] link:\doibase 10.1016/0030 - 4018(93)90389-m [ * * , ( ) ] \doibase http://dx.doi.org/10.1103/physrev.140.a1133 [ * * , ( ) ] link:\doibase 10.1038/ncomms4548 [ * * , ( ) ] link:\doibase 10.1016/0003 - 4916(73)90161 - 9 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1016/0031 - 9163(66)90251 - 4 [ * * , ( ) ] link:\doibase 10.1364/oe.21.015026 [ * * , ( ) ] link:\doibase 10.1038/srep02526 [ * * , ( ) ] link:\doibase 10.1021/ph400147y [ * * , ( ) ] link:\doibase 10.1103/physrevb.19.1689 [ * * , ( ) ] link:\doibase 10.1038/483417a [ * * , ( ) ] link:\doibase 10.1103/physrevlett.60.848 [ * * , ( ) ] link:\doibase 10.1038/nphys2009 [ * * , ( ) ]

a quantization of-dimensional gravity with negative cosmological constant is presented and quantum aspects of the-dimensional black holes are studied thereby . the quantization consists of two procedures . one is related with quantization of the asymptotic virasoro symmetry . a notion of the virasoro deformation of-geometry is introduced . for a given black hole , the deformation of the exterior of the outer horizon is identified with a product of appropriate coadjoint orbits of the virasoro groups . its quantization provides unitary irreducible representations of the virasoro algebra , in which state of the black hole becomes primary . to make the quantization complete , , the global degrees of freedom , are taken into account . by an identification of these topological operators with zero modes of the liouville field , the aforementioned unitary representations reveal , as far as , as the hilbert space of this two - dimensional conformal field theory . this conformal field theory , living on the cylinder at infinity of the black hole and having continuous spectrums , can recognize the outer horizon only as a _ one - dimensional object _ in and realize it as insertions of the corresponding vertex operator . therefore it can not be a conformal field theory on the horizon . two possible descriptions of the horizon conformal field theory are proposed . -5 mm -2 mm addtoresetequationsection ou - het 317 + hep - th/9903259 + march 1999 * three - dimensional black holes * + * and * + * liouville field theory * t. nakatsu , h. umetsu and n. yokoi +

hep-th9903259

it has been proposed @xcite that supergravity ( and string theory ) on @xmath6 times a compact manifold is equivalent to a conformal field theory living on the boundary of @xmath6 . supergravity ( and string theory ) on asymptotically @xmath6 space - time is also expected @xcite to be realized by a boundary field theory on a suitable vacuum . investigations of the bulk quantum gravity , in particular , to clarify @xmath7 black holes as the quantum gravitational states , will lead a deeper insight on our understanding of the @xmath8 correspondence . in three dimensions , a description of gravity in terms of gauge theory @xcite is known , by which quantum gravity in three dimensions seems to be tractable and thus consideration about such quantum gravitational states will become possible . in this article we present a quantization of @xmath0-dimensional gravity with negative cosmological constant @xmath9 and thereby investigate quantum aspects of the @xmath0-dimensional black holes from the perspective of the @xmath10 correspondence . the quantization consists of the following two procedures . the first is related with quantization of the asymptotic virasoro symmetry found by brown and henneaux @xcite . to describe it we introduce in section 2 a notion of the virasoro deformation of three - geometry . it is originally defined as a deformation of a suitable non - compact simply - connected region in @xmath5 and is given by the coadjoint action of the virasoro groups @xmath11 . induced metric from @xmath5 provides a metric of the deformed region . given a black hole , we can take , as such a non - compact simply - connected region , a covering space of the exterior of the outer horizon . its virasoro deformation commutes with the projection . therefore we can obtain a family of the deformed quotients . this defines the virasoro deformation of the exterior of the outer horizon of the black hole . the induced metric of the deformed region becomes that of its quotient , which can be understood as a deformed metric of the black hole obtained in @xcite . the asymptotic virasoro symmetry turns out to be the infinitesimal form of this deformation . the family of the deformed quotients will be identified with a product of the coadjoint orbits of the virasoro groups @xmath2 labeled by mass and angular momentum of the black hole together with the cosmological constant . in section 3 we provide quantization of the asymptotic virasoro symmetry or the virasoro deformation of the exterior of the outer horizon . it is prescribed by quantization of the corresponding coadjoint orbits . quantization of each orbit turns out to give a unitary irreducible representation of the virasoro algebra with central charge @xmath12 . state of the black hole becomes the primary state of the representation . all the excited states of the representation correspond to the local excitations by the virasoro deformation , that is , as noted in @xcite , gravitons . @xmath13 is also examined . quantization of its virasoro deformation turns out to provide a unitary irreducible representation , in which state of @xmath13 is the @xmath14-invariant primary state . although the quantization of the virasoro deformations leads the unitary irreducible representations of the virasoro algebra , it is not sufficient as a quantization of three - dimensional gravity . if one takes the perspective of @xmath14 chern - simons gravity @xcite , the virasoro deformation corresponds to local degrees of freedom of the theory and , in order to make the quantization complete , we must take into account of the global degrees of freedom , _ i.e. _ , holonomies . we start section 4 by discussing their quantization . having quantized holonomies , it is very reasonable to expect from the perspective of the @xmath15 correspondence that , by an identification of these topological operators with the zero modes of a suitable two - dimensional quantum field , the unitary irreducible representations obtained by the quantization of the virasoro deformations could be reproduced as the hilbert space of the two - dimensional conformal field theory . this expectation turns out to be true at least when @xmath4 . the holonomy variables will be identified with the zero modes of a real scalar field @xmath16 , which is interpreted as the liouville field . the state of @xmath13 is identified with the @xmath17-invariant vacuum of this conformal field theory . the states of the black holes are the primary states obtained from the vacuum by an operation of the corresponding vertex operators . excitations of the liouville field are equivalent with the generators of the virasoro deformation . this field theory admits to have continuous spectrums as expected . having obtained the quantization of three - dimensional gravity by the liouville field theory , the following question may arise : how can one understand the three - dimensional black holes in two dimensions where the liouville field lives ? the liouville field theory is living on a two - sphere which is obtained by a compactification of the boundary cylinder at infinity of the black holes . to answer this question it becomes necessary to understand how the virasoro deformation recognizes the outer horizon of the black hole . it will be shown at the end of section 4 that it recognizes the outer horizon not as a two - dimensional object but as a one - dimensional object in @xmath5 . therefore the boundary liouville field @xmath16 can see the outer horizon only as a one - dimensional object . compactification of the boundary cylinder to a sphere simultaneously makes the solid cylinder to a three - ball , by which this one - dimensional outer horizon intersects with the boundary sphere at two points . the boundary liouville field theory recognizes its intersections as insertions of the corresponding vertex operator . a microscopic description of the three - dimensional black holes has been proposed by carlip @xcite and strominger @xcite using a conformal field theory on the horizon . since the virasoro deformation of the exterior of the outer horizon can not recognize the horizon as a two - dimensional object , the boundary liouville field theory can not be equivalent to the horizon conformal field theory . in section 5 we propose two possible descriptions of the horizon conformal field theory . one is a string theory in the background of a macroscopic string living in @xmath5 . worldvolume of the macroscopic string should be identified with the two - dimensional outer horizon , by which we can obtain the virasoro algebra on the horizon as its fluctuation by microscopic string . the other is a conformal field theory which could be obtained by the virasoro deformation of the region between the inner and outer horizons of the black hole . the second possibility seems to imply an interpretation of the conjecture made by maldacena @xcite , in terms of a gauge transformation . the btz black holes . ] @xcite are three - dimensional black holes specified by their mass and angular momenta . the btz black hole with mass @xmath18 and angular momentum @xmath19 will be denoted by @xmath20 . one can avoid a naked singularity by imposing the bound , @xmath21 , on the allowed value of its angular momentum . in terms of the schwarzschild coordinates @xmath22 , where the ranges are taken @xmath23 , @xmath24 and @xmath25 , the black hole metric @xmath26 has the form @xmath27 @xmath28 and @xmath29 are functions of the radial coordinate @xmath30 and given by @xmath31 the outer and inner horizons are located respectively at @xmath32 and @xmath33 . information about the mass and angular momentum is encoded in @xmath34 by @xmath35 where @xmath36 is newton constant . in the case of @xmath37 , the black hole is called extremal . it holds @xmath38 and then the outer and inner horizons coincide with each other . when @xmath39 , it is called non - extremal . the exterior of the outer horizon @xmath32 of the btz black hole will play an important role throughout this paper . we denote the exterior of the outer horizon of @xmath20 by @xmath40 . that is , @xmath41 . it is known @xcite that the black hole can be identified with an appropriate quotient of @xmath5 . we will need a further detailed description of this relation between @xmath40 and @xmath5 . in this subsection we provide it for the case of the non - extremal black holes , that is , the case of @xmath42 . discussion presented here becomes a prototype for the other cases , which will be treated briefly in the next section . @xmath5 is the three - dimensional non - compact group manifold admitted to have invariant metrics which scalar curvatures are negative constant . these metrics are the killing metric and its rescalings . among them , let @xmath43 be the metric which scalar curvature equals to @xmath44 . it is given by @xmath45 this metric is an exact solution of the vacuum einstein equation with the negative cosmological constant @xmath9 . any element @xmath46 of @xmath5 can be written as @xmath47 , where @xmath48 and @xmath49 . , a basis of the lie algebra over @xmath50 is taken by the three matrices @xmath51 @xmath52 these enjoy @xmath53= \epsilon^{abc}j_c$ ] , where @xmath54 is a totally anti - symmetric tensor normalized by @xmath55 and the @xmath56-indices are lowered ( or raised ) by @xmath57 ( or @xmath58 , the inverse matrix of @xmath59 ) . ] with this cartan decomposition one can see that @xmath5 is topologically a solid torus . see fig.1 . let @xmath60 be a non - compact simply - connected region of @xmath5 consisting of the group elements @xmath61 where @xmath62 and @xmath63 are real parameters . their ranges are taken @xmath64 and @xmath65 . @xmath66 is illustrated in fig.2 . notice that the line @xmath67 ( @xmath68 ) in fig.2 is excluded by the definition . the invariant metric ( [ ds sl2 ] ) defines an induced metric on @xmath60 . we call it @xmath69 . it is merely a restriction of ( [ ds sl2 ] ) on @xmath60 . regarding the parameters @xmath70 and @xmath63 as coordinates of @xmath60 , it can be written in the form @xmath71 we can provide a description of the above region together with the induced metric in terms of @xmath14 chern - simons gravity . to present the description , it is convenient to introduce a mapping , @xmath72 , by @xmath73 this defines a pairing of @xmath74 which is invariant under the left diagonal @xmath5 action , @xmath75 . it is also convenient to regard the group elements ( [ z0 non - ex ] ) as an image of a function @xmath76 which takes its values in @xmath5 as presented in ( [ z0 non - ex ] ) . explicitly @xmath76 is given by @xmath77 with this perspective , we can consider the following decomposition of @xmath76 into a product of @xmath74-valued functions @xmath78 : @xmath79 this decomposition is equivalent to construct a pre - image of @xmath60 in the mapping ( [ pairing of sl2pm ] ) . it is unique up to the left diagonal @xmath5 gauge transformation . this symmetry turns out to be the three - dimensional local lorentz symmetry . given such @xmath74-valued functions , we can associate flat @xmath5 connections @xmath80 as follows : @xmath81 it turns out that @xmath82 is a desired classical solution of the chern - simons gravity . in fact , one can reproduce the metric ( [ ds0 non - ex ] ) by means of the 3-vein @xmath83 which is obtained from @xmath84 through the following correspondence @xcite between the relativity and gauge theory languages : @xmath85 both 3-vein @xmath83 and spin connection @xmath86 are realized as @xmath87valued one - forms , @xmath88 and @xmath89 . to be explicit , by use of this correspondence , the 3-vein @xmath90 has the expression @xmath91 . it can be also expressed as @xmath92 . then the metric @xmath93 @xmath94 acquires the form , @xmath95 . this is exactly the induced metric @xmath69 . the correspondence ( [ a0 by ( e , w ) ] ) also shows that the flatness of the connection @xmath84 is equivalent to the cartan form of the vacuum einstein equation with the negative cosmological constant @xmath96 the non - compact simply - connected region @xmath60 is a covering space of @xmath40 , the exterior of the outer horizon of the non - extremal black hole @xmath20 . the projection @xmath97 can be described by using the relations @xmath98 besides the identification of the schwarzschild angular coordinate , @xmath99 . see fig.3 . this projection map becomes the isometry : @xmath100 . therefore the exterior of the outer horizon can be identified with the quotient of @xmath60 : @xmath101 here the equivalence relation @xmath102 " , which originates in the periodicity of @xmath103 , is given by @xmath104 it turns out convenient to introduce another parametrization of @xmath60 which is slightly different from @xmath105 . let @xmath106 and @xmath107 be the quantities @xmath108 we can write any element of @xmath60 in the form @xmath109 where @xmath110 and @xmath111 are real parameters . to exactly cover @xmath60 with this parametrization , the ranges of @xmath110 and @xmath111 must be taken as @xmath112 and @xmath113 . using this parametrization or coordinates of @xmath60 , let us provide an explicit form of the corresponding classical solution of the chern - simons gravity . the @xmath5-valued function @xmath76 , which is defined in ( [ h non - ex ] ) , can be read in this coordinate as @xmath114 the following decomposition of ( [ h non - ex 2 ] ) may be chosen @xmath115 with this choice , the flat connections @xmath80 , which are defined in ( [ def of a0 ] ) , become @xmath116 then @xmath117 turns out to give the metric @xmath118 after the change of the coordinates , which can be read from ( [ z0 non - ex ] ) and ( [ z0 non - ex 2 ] ) , it coincides with the induced metric @xmath119 . in terms of the coordinates @xmath120 , the equivalence relation ( [ equiv non - ex ] ) used in the identification with the black hole becomes simple : @xmath121 therefore we can also say that the quotient of @xmath60 by the identification ( [ equiv non - ex2 ] ) , together with the induced metric ( [ ds0 - 2 ] ) , is identified with the exterior of the outer horizon of @xmath20 . a generalization of the flat @xmath14 connection ( [ a0 non - ex ] ) is examined @xcite to find a realization of the asymptotic virasoro symmetry @xcite in terms of the chern - simons gravity . it has the following form @xmath122 the suffix @xmath123 and @xmath124 label the flat connections . to be consistent with the equivalence relation ( [ equiv non - ex2 ] ) , we assume that @xmath123 and @xmath124 are respectively real functions of @xmath125 and @xmath126 with @xmath127 periodicity . the asymptotic virasoro algebra may be realized by the following generators of the infinitesimal gauge transformation @xmath128 where , again to be consistent with ( [ equiv non - ex2 ] ) , @xmath129 and @xmath130 are assumed to be real functions with @xmath127 periodicity . the gauge transformation of the connection ( [ a(b , tb ) ] ) equals @xcite to deformations of @xmath123 and @xmath124 @xmath131 these deformations have the forms @xmath132 in the asymptotic region where @xmath111 is very large , the non - zero off - diagonal elements of @xmath133 and @xmath134 become negligible . hence the gauge transformation can be regarded as a asymptotic symmetry . to contact with the asymptotic virasoro symmetry it is convenient to decompose the gauge transformation into the diffeomorphism and the local lorentz transformation . notice that , for a given flat connection @xmath135 , its infinitesimal general coordinates transform ( say , generated by a vector field @xmath136 ) can be interpreted @xcite as its infinitesimal gauge transform . via the sugawara construction the generator of the gauge transformation is given by @xmath137 . decompose the gauge transformation ( [ polyakov transform ] ) to @xmath133 @xmath138 and @xmath134 @xmath139 , where @xmath140 are vector fields and @xmath141 is a generator of the local lorentz transformation ( the diagonal @xmath5 gauge transformation ) . the vector fields turn out to be @xmath142 in the asymptotic region they behave as @xmath136 @xmath143 and @xmath144 @xmath145 . these coincide with the virasoro generators considered in @xcite . given the flat connection ( [ a(b , tb ) ] ) , one can construct a metric according to the prescription of the chern - simons gravity . denote this metric by @xmath146 . it has the form @xmath147 it is also possible to obtain a region of @xmath5 , analogous to @xmath60 , on which the induced metric from @xmath5 is precisely given by ( [ ds(b , tb ) ] ) . since @xmath148 and @xmath149 are flat connections we can always trivialize them by some @xmath74-valued functions @xmath150 and @xmath151 @xmath152 then the desired region of @xmath5 , which we will call @xmath153 , consists of those obtained from @xmath150 and @xmath151 by their local lorentz invariant pairing , @xmath154 . it is easy to see that the induced metric on @xmath153 is nothing but @xmath146 . when both @xmath123 and @xmath124 are constant , @xmath153 coincides with @xmath60 . in particular @xmath155 holds . on the other hand , as far as @xmath123 or @xmath124 is not constant , @xmath153 becomes the region of @xmath5 which is different from @xmath60 . one may suspect that a reparametrization of @xmath60 such as @xmath156 might lead the flat connection ( [ a(b , tb ) ] ) or the metric ( [ ds(b , tb ) ] ) . but it does not so as far as @xmath123 or @xmath124 is not constant . it is an interesting but hard task to describe the region @xmath153 explicitly . nevertheless , the following argument shows that @xmath157 is generically different from @xmath60 in @xmath5 , which is enough for our discussion in this article . we first notice that the completion of @xmath60 at @xmath68 is achived by adding the one - dimensional line @xmath67 ( fig.2 ) in @xmath5 . it is a counterpart of the outer horizon in @xmath5 . , may be a two - dimensinal null surface . ] this can be read from the induced metric @xmath158 of @xmath60 . the counterpart of the horizon is located at @xmath68 , that is , @xmath159 and there the metric degenerates to @xmath160 as regards a generic @xmath153 , the analogue of the horizon will be read from the singularity of the vector fields @xmath136 and @xmath161 . looking at ( [ xi - barxi ] ) , they diverge at @xmath162 . for the diffeomorphism to work , @xmath111 should satisfy @xmath163 . only with this understanding @xmath153 becomes a deformation of @xmath60 . ( the vector fields @xmath136 and @xmath161 can be expected to be integrated step by step from @xmath60 to @xmath153 in @xmath5 . ) the completion of @xmath153 at @xmath159 will be done by adding a two - dimensional space - like surface in @xmath5 . this can be also understood from the behavior of the induced metric @xmath164 the difference under their completions in @xmath5 clearly shows that @xmath153 is the different region from @xmath60 as far as @xmath123 or @xmath124 is not constant . if one identifies @xmath125 and @xmath126 with the coordinates of circles , each deformation @xmath165 and @xmath166 given in ( [ coadjoint action1 ] ) can be regarded as the coadjoint action of the virasoro algebra with the central charge @xmath107 . here @xmath123 and @xmath124 play the role of quadratic differentials on the circles . let us consider the coadjoint orbits of @xmath167 and @xmath168 . we will call them respectively @xmath169 and @xmath170 . their product @xmath171 may be identified with the family of the regions @xmath153 of @xmath5 . strictly speaking , we should consider not @xmath153 itself but its quotient obtained by the identification @xmath172 . this is because the information about @xmath167 and @xmath168 , that is , about mass and angular momentum of the black hole , is encoded not in @xmath60 itself but , through the parametrization ( [ z0 non - ex 2 ] ) , encoded in the way how to take the quotient . the actions of the virasoro algebra are consistent with the equivalence relation used in the quotient . therefore we can discuss about the virasoro deformation of the quotient space . let @xmath173 be the quotient @xmath174 . in particular @xmath175 is identified with @xmath176 , the exterior of the outer horizon of the black hole @xmath20 . the virasoro groups @xmath177 deform the metric @xmath178 of @xmath40 preserving its asymptotic form . the deformed metric @xmath179 is nothing but the metric of @xmath173 which is related with @xmath176 by the coadjoint action . @xmath5 itself plays the role of a classifying space in this argument . namely the coadjoint action of @xmath177 deforms the region @xmath66 without touching the metric of @xmath5 . the metric @xmath180 of each deformed region @xmath153 is the one induced from @xmath181 and determines the metric of @xmath173 by the projection @xmath97 . it is the deformed metric of the black hole . it should be emphasized that the above virasoro deformation can not transform the black hole @xmath182 to the black hole @xmath183 which mass @xmath184 and angular momentum @xmath185 are different from @xmath186 and @xmath19 . this can be understood by the fact that any constant shifts of @xmath167 and @xmath168 are not allowed in ( [ coadjoint action1 ] ) since @xmath187 and @xmath188 are taken to be @xmath127-periodic for the consistency with the equivalence relation used in the projection @xmath97 . what the deformation generates are the oscillating modes of @xmath189 and @xmath190 . the deformed metric @xmath191 is not static since @xmath125 and @xmath126 are originally the light cone coordinates ( [ pi non - ex ] ) . the virasoro algebra is the lie algebra of the virasoro group @xmath192 , which is the central extension of @xmath193 , the group of diffeomorphisms of a circle . it consists of vector fields @xmath194 on the circle together with a central element @xmath107 . a general element has the form @xmath195 with @xmath196 a real number . let us write it by the pair @xmath197 . the commutation relation is given by @xmath198= \left(fg'-f'g,~ \frac{1}{48\pi}\ ! \int_0^{2\pi}\!dw(fg'''-f'''g ) \right ) . \label{virasoro 1}\end{aligned}\ ] ] the coadjoint representation of the virasoro algebra consists of quadratic differentials @xmath199 together with the dual central element @xmath200 . a general coadjoint vector ( an element of the dual of the virasoro algebra ) has the form @xmath201 with @xmath202 a real number . let us denote it by the pair @xmath203 . the pairing between coadjoint and adjoint vectors are given by @xmath204 this is invariant under reparametrizations of the circle . the coadjoint action of the virasoro algebra is @xmath205 where , the action of the central element being trivial , we write @xmath206 as @xmath207 for short . the dual pairing ( [ virasoro dual pairing ] ) is invariant under the action of the virasoro algebra , @xmath208 where @xmath209 is given by ( [ virasoro 1 ] ) . as can be seen in ( [ coadjoint action 2 ] ) the dual central element @xmath200 is invariant under the coadjoint action . the value of @xmath202 determines the central charge . we shall set @xmath210 . according to the definition ( [ b0-tb0-c ] ) we will write this @xmath202 as @xmath107 . and @xmath211 , written as @xmath212 at the end of the previous section , is the coadjoint orbit of @xmath213 , that is , the @xmath214-orbit of @xmath213 . the transformation ( [ coadjoint action 2 ] ) also shows that @xmath211 is a homogeneous space of @xmath193 . therefore we can parametrize the orbit by elements of @xmath193 ( modulo the little group @xmath215 ) : for @xmath216 , letting @xmath217 , the integrated form of ( [ coadjoint action1 ] ) or ( [ coadjoint action 2 ] ) becomes @xmath218(dw)^2 , \label{coadjoint action 3}\end{aligned}\ ] ] where @xmath219 is the schwarzian derivative , @xmath220 @xmath221 @xmath222 . then @xmath223 provides the corresponding element of @xmath211 . according to the kirillov - souriau - kostant theory @xcite , each coadjoint orbit admits to have a canonical symplectic structure defined by the dual pairing . if one regards tangent vectors at @xmath223 as vector fields on the circle , the symplectic form @xmath224 can be described as follows : @xmath225 \rangle , \label{symplectic form}\end{aligned}\ ] ] where @xmath226 denote tangent vectors at @xmath223 . generators of the virasoro algebra are given by hamiltonian functions on @xmath211 . if one takes @xmath227 , the commutation relation ( [ virasoro 1 ] ) becomes @xmath228 = ( m - n)l_{m+n}+\frac{c}{12}m^3\delta_{m+n,0}. \label{virasoro cylinder}\end{aligned}\ ] ] the corresponding hamiltonian functions are given by @xmath229 where @xmath216 is thought as the coordinate of @xmath211 . using the prescribed symplectic structure @xcite they turn out to satisfy the ( classical ) virasoro algebra : @xmath230 we generalize our argument given in the previous section to the cases of the massive - extremal and massless black holes . for each of them there appears an appropriate region @xmath60 of @xmath5 . the virasoro groups @xmath231 can deform each @xmath60 in @xmath5 . this will be described by the coadjoint action of the virasoro group . quotient of the deformed region , obtained in a similar manner as the non - extremal case , provides a deformation of the exterior of the outer horizon . each quotient has a metric induced from @xmath5 , which is nothing but the deformed metric of the black hole considered in @xcite . the family of these quotients will be identified with the product of the coadjoint orbits @xmath171 . the values of @xmath167 and @xmath232 can be read from the values @xmath233 of the black hole by using ( [ b0-tb0-c ] ) . we also examine the case of @xmath13 . in this case it becomes necessary to study @xmath234 , the universal cover of @xmath5 . taking an appropriate region of @xmath235 , it becomes possible to repeat an argument similar to the black holes . in particular the family of the deformed regions will be identified with @xmath236 . + let @xmath60 be a region of @xmath5 consisting of the group elements . ] @xmath237 where @xmath238 and @xmath63 are real parameters . there ranges are taken @xmath239 . it is a non - compact simply - connected region of @xmath5 . regarding these parameters as coordinates of @xmath60 , the induced metric from @xmath5 can be written in the form @xmath240 this non - compact simply - connected region of @xmath5 is a covering space of @xmath241 , the exterior of the outer horizon of the massive extremal black hole @xmath242 . the projection @xmath97 can be described by @xmath243 with the identification @xmath99 . since @xmath97 turns out to be the isometry , we can identify @xmath241 with the quotient of @xmath60 : @xmath244 here the equivalence relation is given by @xmath245 @xmath246 . reparametrization of the group elements ( [ z0 massive - ex ] ) to the form @xmath247 leads the flat @xmath14 connection @xmath248 in this derivation we first decompose the @xmath5-valued function @xmath76 ( [ h massive - ex 2 ] ) , according to the rule ( [ def of hpm ] ) , into the pair of @xmath74-valued functions , @xmath249 and @xmath250 , and then apply the prescription ( [ def of a0 ] ) . the connection ( [ a0 massive - ex ] ) coincides with ( [ a(b , tb ) ] ) at @xmath251 . the equivalence relation used in the quotient ( [ quotient massive - ex ] ) acquires the form ( [ equiv non - ex ] ) in these new coordinates . so our argument completely reduces to that given in the previous section . in particular deformations of the massive - extremal black hole @xmath241 are realized by the quotients @xmath173 which are connected with it by the virasoro group . the family of these @xmath173 can be identified with @xmath252 . + let @xmath60 be a region of @xmath5 consisting of the group elements @xmath253 where @xmath110 and @xmath111 are real parameters . their ranges are taken @xmath254 . it is also a non - compact simply - connected region of @xmath5 . regarding these parameters as coordinates of @xmath60 , the induced metric from @xmath5 can be written as @xmath255 this non - compact simply - connected region is a covering space of @xmath256 , the exterior of the horizon of the massless black hole . the projection @xmath97 is prescribed by @xmath257 with the identification @xmath99 . hence the exterior of the horizon of the massless black hole can be identified with the quotient of @xmath60 : @xmath258 where the equivalence relation is @xmath259 @xmath260 . decomposition of ( [ z0 massless ] ) by @xmath261 and @xmath262 leads the flat @xmath14 connection @xmath263 this is the connection ( [ a(b , tb ) ] ) at @xmath264 . so our argument also reduces to the previous one . deformations of the massless black hole @xmath256 are realized by the quotients @xmath173 which are connected with it by the virasoro group . the family of these @xmath173 can be identified with @xmath265 . @xmath13 is the universal cover of @xmath5 . in terms of the schwarzschild type coordinates @xmath22 , where the ranges are @xmath266 , @xmath267 and @xmath268 , the metric @xmath269 acquires the form @xmath270 notice that @xmath271 is not a true singularity . it is merely a coordinate singularity . nevertheless , to provide the virasoro deformation of this space which is analogous to the black holes , it is required to consider the complement of @xmath271 rather than the whole actually this complement is an analogue of @xmath40 of the black hole . the complement of the line at @xmath271 will be called @xmath272 . let @xmath60 be a region of @xmath5 consisting of the group elements @xmath273 where @xmath110 and @xmath63 are real parameters . notice that ( [ z0 ads ] ) have the form of the cartan decomposition . in order to avoid a doubly parametrization , the ranges of @xmath125 and @xmath126 should be taken within @xmath274 . so we choose @xmath274 for their ranges . as regards @xmath63 it is set to @xmath275 . if one admits @xmath68 in the definition , @xmath60 coincides with @xmath5 itself . due to its exclusion @xmath60 becomes the complement of the circle at @xmath68 . if one regards the time of @xmath13 being @xmath276-periodic , @xmath60 can be identified with @xmath272 . the relation with the schwarzschild type coordinates can be read as @xmath277 introduce a slightly different parametrization of @xmath60 with the form @xmath278 where real parameter @xmath111 satisfies @xmath279 , due to the shift . decomposition of the @xmath5-valued function @xmath76 ( [ h ads 2 ] ) by the pair of @xmath74-valued functions , @xmath280 and @xmath281 leads the flat @xmath14 connection @xmath282 this is the connection ( [ a(b , tb ) ] ) at @xmath283 . @xmath5 has a topologically nontrivial circle . the universal cover @xmath234 can be obtained by making this @xmath284 to a line . this is equivalent to a simple change of the identification of @xmath125 and @xmath126 in the cartan decomposition into an identification , @xmath285 . the universal covering of @xmath60 taken in @xmath234 is the complement of the line at @xmath68 and can be identified with @xmath272 . for a given element @xmath286 @xmath287 , the flat connection ( [ a(b , tb ) ] ) may be thought as a connection on a solid cylinder , where @xmath111 measures its radial direction and @xmath288 are identified with the lightcone coordinates of the cylinder . trivializations of @xmath289 and their local lorentz invariant pairing define a map from the solid cylinder to @xmath5 . the solid cylinder wraps @xmath5 infinitely many times . we may unfold this wrapping by considering @xmath234 in stead of @xmath5 . therefore , letting @xmath60 be the complement of the line in @xmath234 we can construct its deformation by the virasoro group . the family of the deformations can be identified with @xmath290 . for a given coadjoint orbit @xmath211 , the hamiltonian function @xmath291 generates the @xmath284-action and can be considered as an energy function of the orbit . @xmath167 is a fixed point of this circle action and corresponds to the classical vacuum . we will discuss about the stability of this vacuum . write @xmath292 in the form @xmath293 , where @xmath294 are complex numbers satisfying @xmath295 . these @xmath294 with @xmath296 provide local coordinates in the neighborhood of @xmath167 . @xmath297 acts trivially on @xmath167 ( cf.([coadjoint action 2 ] ) ) . the behavior of the energy function in the neighborhood of @xmath167 can be seen by inserting the above @xmath298 into ( [ lm(s ) ] ) and expanding it with respect to @xmath294 . the expansion turns out @xcite to be the form @xmath299 the stability of the classical vacuum will be assured by the condition , @xmath300 being bounded from below at @xmath167 . it is achieved only when @xmath167 satisfies @xmath301 all the coadjoint orbits corresponding to the 3-geometries under consideration are satisfying this stability condition . the orbit @xmath211 which satisfies the condition ( [ stability ] ) can be quantized . this provides a unitary irreducible representation of the virasoro algebra . actually it is quantized @xcite by the khler quantization or geometric quantization method . the coadjoint orbit is topologically the homogeneous space @xmath302 . the little group @xmath215 is @xmath284 for the black holes , and @xmath303 for @xmath13 . can be seen easily by solving @xmath304 ( [ coadjoint action 2 ] ) for each cases . ] for these @xmath215 , the homogeneous space @xmath302 becomes a complex manifold . the complex structure turns out @xcite to be compatible with the symplectic structure ( [ symplectic form ] ) . thereby the orbit @xmath211 which we need , becomes khler . the complex line bundle on the orbit with its @xmath305st chern class @xmath224 provides the unitary irreducible representation on the space of the holomorphic sections . to describe the representations it is convenient to shift the virasoro generator @xmath306 to @xmath307 . with this shift the algebra ( [ virasoro cylinder ] ) becomes the standard form @xmath228 = ( m - n)l_{m+n}+\frac{c}{12}(m^3-m)\delta_{m+n,0}. \label{virasoro plane}\end{aligned}\ ] ] any representation of the virasoro algebra with the central charge @xmath107 can be specified by its highest weight state @xmath308 . the state @xmath309 , which is called primary , satisfies @xmath310 and @xmath311 for @xmath312 . in addition to these conditions the primary state @xmath313 satisfies @xmath314 . so it is a @xmath5 invariant state . unitary representations are the representations in which @xmath315 satisfy the condition @xmath316 . the unitary irreducible representations which are obtained by the quantization of the orbits can be summarized as follows @xcite : is required . ] for the orbit of @xmath211 with @xmath317 , the corresponding unitary irreducible representation is given by the verma module @xmath318 . it is a module obtained by successive actions of @xmath319 ( @xmath320 ) on @xmath321 . it has the form @xmath322 for the orbit @xmath323 , the corresponding unitary irreducible representation is given by an analogue of the verma module @xmath324 . it is obtained by successive actions of @xmath319 ( @xmath325 ) on @xmath326 . we call this module @xmath327 . it has the form @xmath328 gathering these results about quantization of the orbits we can prescribe quantization of the asymptotic virasoro symmetry of the 3-geometries or their virasoro deformations in the following manner : for the btz black hole @xmath20 , the deformations @xmath173 of the exterior of the outer horizon can be identified with the product of the coadjoint orbits @xmath171 . the quantizations of the orbits @xmath169 and @xmath170 provide the unitary irreducible representations @xmath329 and @xmath330 . therefore the quantization of the deformations leads the representation @xmath331 . in particular state of the black hole can be identified with the primary state @xmath332 @xmath333 of the representation . for @xmath13 , the deformations of @xmath272 can be identified with the product of the coadjoint orbits @xmath334 . these orbits provide the unitary irreducible representations @xmath335 and @xmath336 . therefore the quantization of the deformation leads the representation @xmath337 . the state of @xmath13 can be identified with the primary state @xmath338 @xmath339 . it is the state invariant under @xmath340 , which is the isometry of @xmath13 . excitations by @xmath319 correspond to the virasoro deformation of @xmath341 . originally it is the deformation of @xmath60 in @xmath5 which by no means provides any transformation of the black hole to another one having different mass and angular momentum .. ] nevertheless , these degrees of freedom provide the unitary representations of the virasoro algebra . this shows that these degrees of freedom can be included in the physical spectrum as ( massive ) gravitons @xcite . as we have seen , quantizations of the virasoro deformations of the btz black holes and @xmath13 lead the unitary irreducible representations of the virasoro algebra at least when @xmath342 . one may wonder quantization of three - dimensional gravity with negative cosmological constant becomes complete by these quantizations . but it does not so . if one takes the chern - simons gravity viewpoint , these deformations correspond to the local degrees of freedom of the theory ( gravitons ) , that is , oscillating modes of @xmath123 and @xmath124 in the flat connection ( [ a(b , tb ) ] ) . to make the quantization complete , one must take into account of the global degrees of freedom , _ i.e. _ , holonomies . the holonomy variables should be dynamical in the chern - simons gauge theory . in our context this requires an introduction of their conjugates . given a suitable poisson structure on the holonomy variables and their quantizations , it is very reasonable to expect from the perspective of the @xmath10 correspondence that , with an identification of these quantum operators with the zero modes of an appropriate two - dimensional quantum field , the unitary representations obtained in the previous section could be reproduced as the hilbert space of the two - dimensional conformal field theory . this expectation turns out to be true at least when @xmath343 . the holonomy variables can be identified with the zero modes of a real scalar field @xmath16 . this scalar field should be interpreted as the liouville field in the ultimate . we start with the description of the poisson structure of the holonomies . then we turn to construction of the unitary irreducible representations in the framework of the liouville field theory or the coulomb gas formalism . finally we make identifications between the states of the three - dimensional gravity and those of the liouville field theory . for a given closed path @xmath344 , holonomy of a connection @xmath345 along the path is given by @xmath346 \equiv pe^{\int_\gamma a^{(\pm)}},\ ] ] where @xmath347 denotes the path ordering . since we can assume @xmath135 is flat , @xmath348=(h^{(+)}[\gamma ] , h^{(-)}[\gamma])$ ] is invariant up to conjugation under smooth deformations of @xmath344 . in particular @xmath349 are homotopy invariant . the btz black holes originally admit to have a nontrivial holonomy around a closed path connecting @xmath350 and @xmath351 in the constant @xmath111 surface . let us call this path @xmath352 . if there is only one nontrivial holonomy there appears no symplectic structure . in order to make it a dynamical variable we need to introduce another closed " path . let @xmath353 be a path connecting @xmath350 and @xmath354 in the constant @xmath111 surface . if one regards it as a closed path and then considers about a holonomy around this path , the quantities @xmath355 and @xmath356 together with @xmath34 may become dynamical . in this extended phase " space , the btz black holes themselves will become a lagrangian submanifold @xmath357 . this extension is a generalization of the off - shell extension of the euclidean black holes examined by carlip and teitelboim @xcite . let @xmath16 be a real scalar field on @xmath370 . cft we consider the euclidean version . ] the action is given by @xmath371=\frac1{4\pi i}\int_{{\bf p}^1 } \bar{\partial } x \wedge \partial x + \frac{\alpha_0}{2\pi}\int_{{\bf p}^1 } rx , \label{liouville action}\end{aligned}\ ] ] where @xmath372 is the riemann tensor of a fixed khler metric on @xmath370 .. ] @xmath373 is a real number and its value will be specified later . the stress tensor @xmath389 has the form @xmath390 expansion of the stress tensor to the form @xmath391 provides the generators of the virasoro algebra ( [ virasoro plane ] ) with the central charge @xmath392 . in terms of the oscillator modes their expressions become @xmath393 the actions of @xmath315 on the in - fock vacua satisfy the following properties : on the vacuum @xmath382 , using the expressions ( [ ln by an ] ) one can read @xmath394 while on the vacuum @xmath380 with @xmath395 , one obtains @xmath396 these properties besides a comparison between the commutation relations @xmath397=na_{-n}$ ] and @xmath398=nl_{-n}$ ] , show that the in - fock space @xmath399 is equivalent to the verma module @xmath400 with @xmath401 , or @xmath327 if @xmath402 . hermitian conjugation of the mode operators becomes @xmath403 notice that @xmath404 can not be hermitian due to the existence of the background charge @xmath373 . since the action ( [ liouville action ] ) is real , it is expected that the realization ( [ ln by an ] ) of the virasoro algebra becomes unitary . in fact , under this conjugation the realization becomes unitary , that is , @xmath315 satisfy @xmath316 . the unitarity imposes a constraint on the allowed value @xmath405 of the in - fock vacuum . the condition @xmath406 together with the requirement of its eigenvalue being non - negative , restricts @xmath407 in the form ( fig.4 ) @xmath408 or @xmath409 . \label{k for ads}\ ] ] now we introduce the out - fock space @xmath410 . let @xmath411 be the state which satisfies the conditions : @xmath412 . the out - fock vacuum @xmath413 can be introduced by @xmath414 . it satisfies @xmath415 . the out - fock space @xmath410 is the fock space built on @xmath413 . using ( [ conjugate of modes ] ) it has the form , @xmath416 . this space is located at @xmath417 . the actions of @xmath315 on the out - fock vacua satisfy the following properties : on the vacuum @xmath411 , using the expressions ( [ ln by an ] ) one can obtain @xmath418 while on the vacuum @xmath413 with @xmath419 , @xmath420 these properties besides the action of @xmath404 on the out - fock vacuum show that the out - fock space @xmath421 can be identified with @xmath422 , the dual of @xmath399 . , which is dual to @xmath380 , satisfies @xmath423 . ] with this identification the pairing between @xmath424 and @xmath399 becomes non - degenerate . in particular , if one takes the allowed value ( [ k for black holes ] ) or ( [ k for ads ] ) of @xmath405 , the representation becomes unitary . choosing the value of @xmath373 appropriately , it becomes the unitary irreducible representation of the virasoro algebra obtained by the quantization of the coadjoint orbit . given the commutation relations ( [ ccr holonomy ] ) of the quantum holonomy operators @xmath367 and @xmath368 , one is tempted to identify these topological operators with the zero modes of the liouville field @xmath375 . let @xmath426 and @xmath427 be the zero modes of the holomorphic part @xmath428 ( the anti - holomorphic part @xmath429 ) of the liouville field . the identification is precisely given by @xmath430 and becomes consistent with the commutation relations @xmath431=i$ ] and @xmath432=i$ ] . at the end of the previous section , the state of the btz black hole @xmath40 has been identified with the primary state @xmath332 @xmath333 . the values of weights can be expressed in terms of the geometrical data , @xmath433 a comparison of these weights with that of the in - fock vacuum given in ( [ ln on in - vacuum k ] ) makes us adjust the background charge of the liouville field to @xmath434 and , accepting this background charge , it also leads an identification of the black hole state with the following in - fock vacuum : @xmath435 where @xmath436 and @xmath437 are given by @xmath438 these are the values allowed by the unitarity condition ( [ k for black holes ] ) . the region @xmath439 in fig.4 corresponds to the black holes . all the excitations in the fock space @xmath440 can be identified with the excitations owing to the virasoro deformation of @xmath40 . these correspond to gravitons . to work the above machinery completely , the central charge ( [ b0-tb0-c ] ) of the virasoro deformation of the 3-geometries should be matched with that of the two - dimensional conformal field theory . this consistency requires the condition , @xmath4 or equivalently @xmath441 . under this condition the central charge of the liouville field theory can be regarded as @xmath442 . therefore these two coincide in this limit . the deviation at finite @xmath107 is of order @xmath443 . the state of @xmath272 has been identified with the primary state @xmath444 . it is invariant under @xmath340 . thereby we can identify this state with the @xmath17-invariant vacuum of the liouville field theory : @xmath445 the @xmath17-invariant vacuum is the fock vacuum with @xmath446 ( @xmath447 ) in ( [ k for ads ] ) ( the origin in fig.4 ) . from the viewpoint of the unitary representation theory other values of @xmath344 and @xmath448 are also allowed as far as they satisfy @xmath449 . notice that @xmath450 , that is , @xmath451 in ( [ k for black holes ] ) corresponds to the state of the massless black hole @xmath256 . consider a state @xmath452 and @xmath453 have the form given in ( [ k for ads ] ) . the weights of this state are respectively @xmath454 and @xmath455 . using ( [ wt of x(j , m ) ] ) one can read mass and angular momentum as @xmath456 and @xmath457 . therefore the corresponding 3-geometry is conic @xcite . this means that the state @xmath458 in fig.4 corresponds to the conical singularities . rather surprisingly , its virasoro deformation gives rise to the unitary representation . this might imply that conical singularities with the allowed mass and momentum get mild quantum mechanically . our understanding of singularities in classical relativity may be required to change in quantum theory . the values of @xmath167 and @xmath168 which correspond to @xmath459 and @xmath453 are @xmath460 and @xmath461 . the three - dimensional metric becomes @xmath462 in this case @xmath60 will be taken in @xmath463 . if one considers it in @xmath5 , it consists of the elements @xmath464 the three - geometry which we want to describe is @xmath465 . it is obtained from @xmath60 by the projection @xmath97 . a comparison of ( [ conical z0 ] ) with the cartan decomposition of @xmath5 given by @xmath466 shows that @xmath60 can be taken in @xmath463 as the cheese cake " with angle @xmath467 . see fig.5 . the ranges are @xmath468 and @xmath469 . the projection @xmath97 causes a conical singularity at @xmath68 ( fig.6 ) . = 4.5 cm mathematically speaking , @xmath470 is regular but , once we make the completion of @xmath60 at @xmath68 by adding the line @xmath471 and extend @xmath97 to this line , it becomes singular . this singular mapping causes the conical singularity . in particular , when @xmath472 , the appearance of the conical singularity is due to a particle with mass @xmath473 sitting at @xmath68 @xcite . having obtained the correspondence between the states of the three - dimensional quantum gravity and the two - dimensional conformal field theory , one may ask how the three - dimensional black holes can be understood in two dimensions where the conformal field lives . to investigate this question , it is useful to discuss first the case of the conical sigularities . the state @xmath474-invariant vacuum , @xmath475 making the holomorphic and anti - holomorphic pieces together , we will rewrite the above operator into the form , @xmath476 where @xmath477 and @xmath478 ] . if one takes the path - integral formulation of @xmath479 cft , the norm of this state , which should be normalized to unity , can be written in the following manner : @xmath480 where @xmath481 is the action ( [ liouville action ] ) with @xmath482 . the vertex operators inserted at @xmath483 and @xmath484 respectively create the states of the conical singularity from the @xmath17 invariant in- and out- vacua . choosing the background metric so that its curvature concentrates at @xmath484 , we rewrite the expression ( [ norm by path - integral 21 ] ) in the following form : @xmath485 the liouville field @xmath16 treats the conical singularity as an insertion of the corresponding vertex operator . this vertex operator has the following origin in three - dimensions . to explain this , we first remark that the @xmath370 where the liouville field theory lives can be regarded as that obtained by a compactification of the boundary cylinder at infinity this compactification also makes the solid cylinder to a three - dimensional ball . in such a compactification to a three - ball , the conical sigularity located at the center must intersect with the @xmath370 precisely at the two points , @xmath483 and @xmath484 . see fig.7 . now , back to the path - integral expression given in ( [ norm by path - integral 22 ] ) , we can say that these intersections of the conical singularity with the boundary sphere are realized as the vertex operators in the boundary liouville field theory . in particular , the gravitational state of a particle with mass @xmath473 sitting at the center of @xmath13 can be treated by an insertion of the corresponding vertex operator . nextly let us discuss about the case of the black holes . basically we follow the same argument as above . the black hole state ( [ black hole state in 2d ] ) can be obtained by an operation of the corresponding vertex operator on the @xmath486-invariant vacuum , @xmath487 if one takes the path - integral formulation , the norm of the black hole state can be written in the following manner : @xmath488 this path - integral representation also provides some idea about our interpretation of the black holes in the two dimensions . we first remark that the outer horizon of the black hole can not be recognized as a two - dimensional object under the virasoro deformation . see fig.3 for the non - extremal case . it is treated as a one - dimensional object for the massive black hole , and as a point for the massless black hole . the virasoro deformation is originally introduced as a deformation of the non - compact simply - connected region @xmath60 . this region of @xmath5 is the covering space of the exterior of the outer horizon of the black hole . under the virasoro deformation what one can recognize as the outer horizon is the counterpart of the outer horizon on @xmath5 . it is obtained by the completion of @xmath60 at @xmath68 . we can easily see that these completions for the cases of the massive extremal and massless black holes are respectively given by adding one - dimensional and zero - dimensional objects . being zero - dimensional in the massless case may be understood as a degeneration from the massive case . we hope to explain the geometrical origin of the insertion of the vertex operator in ( [ norm by path - integral 2 ] ) . let us examine first the compactification of the boundary cylinder of the black hole @xmath489 to the @xmath490 . @xmath491 circles of the boundary cylinder are mapped to the two points @xmath483 and @xmath484 of the @xmath490 . this compactification of the boundary also provide a compactification of the bulk geometry . now we will ask what it gives rise to . we can assume reasonably that @xmath491 slices of @xmath40 are compactified to @xmath483 and @xmath484 of the @xmath490 . if one accepts this assumption , the outer horizon must be compactified to @xmath483 and @xmath484 of @xmath490 . this is because the @xmath491 slices of @xmath40 include the outer horizon after the completion . see fig.3 . this shows that the intersections of the black hole with the boundary sphere are realized by the vertex operators as described in ( [ norm by path - integral 2 ] ) . see fig.8 . to provide a microscopic description of black holes in quantum gravity or string theory is a very impressive and enlightening issue . for the three - dimensional black holes such a microscopic description has been proposed by carlip @xcite and strominger @xcite . it states that microscopic states of the black holes are the states of a conformal field theory on the horizon . although at present we do not know exactly what conformal fields do live on the horizon , their proposal makes it possible to give statistical mechanical explanations of the thermodynamical properties of the black hole . receiving their proposal , many questions will arise . first of all , can one identify the boundary liouville field theory with a conformal field theory on the horizon ? our answer is no . as explained at the end of the last section , the virasoro deformation of the exterior of the outer horizon can not recognize the horizon as a two - dimensional object . it can recognize the horizon only as a one - dimensional object . this means that the virasoro algebra obtained from the deformation can not be the virasoro algebra on the horizon . on the other hand , this deformation provides the asymptotic virasoro algebra , which generators are gravitons in bulk and are identified with the excitations of the boundary liouville field . this field theory has the continuous spectrums . therefore these two conformal field theories are by no means equivalent to each other . if one accepts this answer , one might be led to another question : what conformal field theory is living on the horizon ? to this question we can not answer definitely . here we would like to propose two possible descriptions of this conformal field theory . the first one is based on a simple hypothesis : the virasoro algebra of the horizon conformal field theory should generate the virasoro deformation of the horizon . to work this hypothesis , we need , first of all , to treat the outer horizon as a _ two - dimensional _ object . if one takes the chern - simons gravity viewpoint , it is doubtful whether one can handle the outer horizon as a two - dimensional object . to be successful , presumably one needs the string theory viewpoint . namely , if the two - dimensional horizon is identified with a macroscopic string in @xmath5 , the horizon conformal field theory would be string theory on this background . it describes quantum fluctuations of the macroscopic string and will lead the microscopic description of the black holes . but it is not clear , at least for us , to what extent this string theory is analyzable . the second possible description of the horizon conformal field theory is based on the hypothesis : the virasoro deformation of the region between the inner and outer horizons should lead the virasoro algebra on the horizon conformal field theory . this hypothesis can be regarded as an analogue which we used in the construction of the boundary liouville field theory . in particular it may allow us to follow the similar step as we took for the study of the exterior of the outer horizon . to be explicit , let us consider the region between the inner and outer horizons of the non - extremal btz black hole @xmath20 . we will call it @xmath492 . @xmath493 @xmath494 . ] a covering space of @xmath492 is given by a non - compact simply - connected region of @xmath5 consisting of the group elements @xmath495 where the ranges of @xmath70 and @xmath63 are taken @xmath496 and @xmath497 . let us call this region @xmath498 . the projection can be described by @xmath499 together with the identification @xmath99 . the induced metric of @xmath498 has the form @xmath500 and is isometric to @xmath501 . therefore @xmath498 is indeed the covering space . @xmath492 is identified with the quotient of @xmath498 , @xmath502 where the equivalence relation is @xmath503 . as is done in the study of @xmath60 , it is convenient to rewrite the group element ( [ z1 non - ex ] ) in the form @xmath504 description of @xmath498 in terms of the chern - simons gravity is given by the flat connection @xmath505 which is determined by the prescription ( [ def of a0 ] ) using a decomposition of @xmath506 into the following @xmath74-valued functions @xmath507 : @xmath508 a slight comparison between @xmath509 and @xmath78 given in ( [ hpm non - ex ] ) shows that the flat connection @xmath510 is gauge - equivalent to @xmath84 ( [ a0 non - ex ] ) , the flat connection which describes @xmath60 , the covering space of the exterior of the horizon : @xmath511 where the gauge transformation @xmath512 is given by @xmath513 and @xmath514 . this gauge equivalence should be understood as a formal one because the allowed ranges of the coordinates are different from each other . ( for @xmath84 ( [ a0 non - ex ] ) the range of @xmath111 is @xmath515 while for @xmath510 , @xmath516 . ) nevertheless , if one takes the existence of such a gauge transformation seriously , one may arrive at an idea that a deformation of @xmath498 could be constructed in such a manner that it is related with the virasoro deformation of @xmath60 by an appropriate gauge transformation . thus obtained deformation will define the virasoro deformation of @xmath498 and thereby that of @xmath492 , the region between the inner and outer horizons of the black hole . the corresponding conformal field theory will be regarded as the horizon conformal field theory . the second description of the horizon conformal field theory may provide another interpretation of the surprising conjecture made by maldacena @xcite , which may be understood in our situation as a correspondence between the boundary and horizon conformal field theories under a suitable renormalization group flow . the second description , if it is correct , implies a possibility of a realization of this correspondence in terms of gauge transformation . we thank s. mano for useful discussions . after this work was completed we received a preprint @xcite in which the content of 3.3 is also discussed . o. coussaert , m. henneaux and p.van . driel , the asymptotic dynamics of three - dimensional einstein gravity with a negative cosmological constant " , class . * 12 * ( 1995 ) 2961 , + m. banados and m. ortiz , the central charge in three - dimensional anti - de sitter space " , hep - th/9806089 .

we present the results of a search for clusters in dwarf irregular galaxies in the virgo and fornax cluster using hst wfpc2 snapshot data . the galaxy sample includes 28 galaxies , 11 of which are confirmed members of the virgo and fornax clusters . in the 11 confirmed members , we detect 237 cluster candidates and determine their v magnitudes , v - i colors and core radii . after statistical subtraction of background galaxies and foreground stars , most of the cluster candidates have v - i colors of -0.2 and 1.4 , v magnitudes lying between 20 and 25th magnitude and core radii between 0 and 6 pc . using h observations , we find that 26% of the blue cluster candidates are most likely hii regions . the rest of the cluster candidates are most likely massive ( m ) young and old clusters . a comparison between the red cluster candidates in our sample and the milky way globular clusters shows that they have similar luminosity distributions , but that the red cluster candidates typically have larger core radii . assuming that the red cluster candidates are in fact globular clusters , we derive specific frequencies ( s ) ranging from - 9 for the galaxies . although the values are uncertain , seven of the galaxies appear to have specific frequencies greater than 2 . these values are more typical of ellipticals and nucleated dwarf ellipticals than they are of spirals or local group dwarf irregulars .

astro-ph0311106

the study of extragalactic globular cluster systems in galaxies of a variety of hubble types has led to the development of a number of scenarios that place globular clusters within the context of galaxy formation and evolution . @xcite and @xcite reviewed the myriad pieces of evidence derived from globular cluster systems and the scenarios designed to explain them . the properties of globular clusters in giant elliptical galaxies , in particular , have led to suggestions that globular cluster systems form 1 ) _ in situ _ , perhaps in the hierarchical collapse of the halo @xcite and possibly in multiple phases @xcite ; 2 ) in galaxy mergers , both through the formation of new clusters in the colliding gas ( e.g. * ? ? ? * ) and through the combination of the pre - existing globular cluster systems ( e.g. * ? ? ? * ) ; 3 ) by accretion of the clusters of smaller galaxies @xcite , as exemplified by the @xmath54 gcs being contributed to the milky way @xcite by the sagittarius dwarf galaxy @xcite . more recently the second two of these ideas have been incorporated into the current cosmological scenario using simulations ( e.g. * ? ? ? * ) and semi - analytic methods ( e.g. * ? ? ? * ) . the generally low globular cluster specific frequencies ( @xmath6 , the number of clusters normalized by the galaxy magnitude ; * ? * ) observed in nearby dwarf and spiral galaxies pose a difficulty for the third of the above possibilities , a point which was raised by @xcite in argument against spiral - spiral mergers explaining the high @xmath6 of cd galaxies . however , it may be misleading to compare the specific frequences of spiral and dwarf galaxies residing in nearby low - density galaxy groups with those of giant elliptical galaxies found in more distant rich clusters of galaxies , since the role of the environment in the assembly of globular cluster systems is not understood . a census of the globular cluster systems of a wide variety of galaxies drawn from the same environment is needed to properly assess the roles that each of the formation mechanisms may play . @xcite embarked on such a census through the study of the globular clusters of dwarf elliptical ( de ) galaxies in the virgo , fornax , and leo clusters . the results of that survey revealed a distinction between the globular cluster populations of nucleated de ( de , n ) and nonnucleated de galaxies ; in particular , de , n galaxies were found to have a factor of 3 higher @xmath6 than nonnucelated des . the high @xmath6 of de , n align them more closely with giant elliptical galaxies than with dwarf irregular ( dirr ) and spiral galaxies , which from measurements in _ local _ galaxies have @xmath7 @xcite ; the nonnucleated de s , on the other hand , _ do _ have @xmath6 that resemble those of dirrs and spiral galaxies , once fading of the young stellar populations is taken into account . the difference in @xmath6 between de and de , n galaxies thus suggests globular cluster formation processes in de , n and giant ellipticals that are different from those of spirals and dirrs . moreover , it argues that not all de galaxies represent the stripped or harassed remnants of dirr galaxies , processes described by e.g. @xcite and @xcite . these conclusions depend , however , to large degree on the validity of comparing globular cluster populations in rich - cluster de galaxies with those of local dirr and spiral galaxies . in this paper , we discuss the results of a _ hubble space telescope _ ( hst ) snapshot survey of dirr galaxies in the virgo and fornax galaxy clusters , the same environments from which many of the de galaxies of @xcite were drawn . we describe the observational sample and data reduction in 2 . in 3 , we describe our technique for detecting candidate globular clusters . we present the properties of our sample of candidate globular clusters , derive specific frequencies , and examine individual galaxies in 4 . some of the more interesting results are discussed in 5 , followed by our conclusions in 6 . the galaxy sample , shown in table 1 , was observed as part of a snapshot program go-7377 ( pi : miller ) using the wfpc2 camera of hst . all hst observations consisted of two 240 second exposures in the f555w filter and one 300 second exposure in the f814w filter . the galaxies , which have typical sizes of @xmath530 arcsec , were in all cases centered and completely contained ( except vcc 1448 ) on the wf3 chip . the f555w observations , with their much more accurate cosmic ray removal , are used for most of the analysis in the paper . the f814w observations are only used to derive the v - i colors . the hst sample originally included 29 galaxies selected from the virgo cluster catalog ( vcc ; * ? ? ? * ) , fornax cluster catalog ( fcc ; * ? ? ? * ) , and dorado group catalog ( dgc ; * ? ? ? * ) in a manner similar to that used to select the de galaxies for @xcite , @xcite , and @xcite . targets classified as i m or de / im in the catalogs and that do not have hst guide stars within 3 arcminutes ( so as to limit the chance of having a highly saturated star in the field ) were selected . since lower luminosity galaxies are more likely to have fewer clusters , more galaxies were selected in lower - luminosity bins . since a minority of galaxies had measured velocities , radial velocity was not used as a selection criteria . in addition , the di / de transition object ngc 4286 @xcite was included as a special target because it fit a project goal of investigating the connections between de and di galaxies . of these 30 galaxies 29 were observed successfully during hst cycle 7 . in this paper we concentrate on the 28 galaxies in the virgo and fornax cluster in this sample . the initial reduction procedure for the hst data was similar to that given in @xcite . the two f555w images were combined and cleaned using idl routines supplied by richard white of stsci . all other processing was done using scripts written in iraf . the cleaned f555w images were then used as templates to identify and remove cosmic rays in the f814w images . an hst image of one of the sample galaxies , vcc 328 , can be seen in figure 1a . h@xmath0 images were taken through a set of four narrow band ( fwhm=20 ) filters on the apache point observatory 3.5 m telescope in april , 2003 . each galaxy was observed through an on filter tuned to the velocity of the galaxy and an off filter , separated by at least 1000 km / sec from the galaxy velocity . the filters had central wavelengths of 6563 , 6585 , 6607 and 6629 . feige 34 was used as a spectrophotmetric standard @xcite . based on the scatter in the photometry of feige 34 , we estimate that absolute fluxes are only accurate to 50% due to the presence of some clouds during the first night of observations . observations on subsequent nights allowed for the bootstrapping of the flux calibration using the galaxy vcc 1374 . bias subtraction and flat fielding were performed using ccdproc . for one filter ( centered at 6585 ) , the flat fielding left 10% variations in the sky level across the field requiring the use of an illumination correction image . this illumination correction image was derived from seven program images with the sources masked using sextractor . after the illumination correction the sky was constant to 2% . following flat - fielding , the sky was subtracted from each image and cosmic rays were removed using the idl routine la_cosmic @xcite . the on and off images were then aligned and scaled using the brightest stars in each field and then were subtracted . the resulting h@xmath0 images ( one example is shown in figure 1b ) have detection limits corresponding to a flux of @xmath8 ergs / sec at a distance of 17 mpc . to determine accurate wcs coordinates for the images , we used the iraf mscred task msczero @xcite . based on the scatter of stars relative to their expected positions within the field , these coordinates are good to within 1@xmath9 . this section will detail the process by which we identify and derive properties of candidate clusters . before explaining the process in detail , we provide a brief overview : using daophot ii @xcite , we derived a point - spread function ( psf ) using hst data of the large magellanic cloud ( lmc ) cluster ngc 1868 . we convolved this psf with king profiles of varying size to create model cluster profiles . then for each of the galaxies in our sample , sources were identified and fit to the model cluster profiles to determine their magnitudes and sizes . to obtain a list of candidate star clusters in each galaxy , we assumed that all point - like sources on the wf3 chip between our completeness limit and an m@xmath10 of -11.5 are potential clusters . we then used the flanking wf2 and wf4 chips on the wfpc2 camera to correct for the presence of background galaxies and foreground stars in our list of cluster candidates . an alternate method for finding cluster candidates in nearby , late - type , external galaxies with hst data is given by @xcite . the methods are similar in that both detect magnitudes and sizes for cluster candidates in galaxies at moderate distance , but differ in the method used to determine those quantities , and the way in which the cluster samples are defined . for simplicity , throughout the paper we assume a distance to virgo and fornax of 17 mpc ( dm of 31.15 ) consistent with the distances given in @xcite and @xcite . differences in the distance modulus of a few tenths of a magnitude should not strongly affect our results . to determine a psf for the wf2 , wf3 and wf4 chips in the f555w filter we used wfpc2 observations of a field of the lmc cluster ngc 1868 that were taken on nov . 12 , 1998 close in time to the majority of our observations ( see table 1 ) . the psfs were characterized in daophot ii using roughly 100 stars distributed across each chip . the derived psfs ( one for each chip ) are a position dependent combination of a moffat function with @xmath11=1.5 and sub - pixel corrections . the psf was fit using a 5 pixel ( 0.5 arcsec ) radius and extends over a 7 pixel radius . we conducted a qualitative test comparing our derived psfs to tiny tim psfs @xcite by subtracting both from point - source fields ( using substar in daophot ii ) and found that our derived psfs left behind much less residual flux . we will discuss results on the stability of the psf in the next section . because we are interested in determining the radii of the star cluster candidates , we created _ model cluster profiles _ by convolving the derived psf with king profiles . seventeen different king profiles were used , all with a tidal radius of 40 pc , a value typical of galactic globular clusters @xcite , and a core - radius of 0 ( the original psf ) , 0.25 , 0.5 , 0.75 , 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 10 , 12 , 15 and 20 pc ( where all physical sizes assume a distance of 17 mpc ) . the model cluster profiles are always referred to by their physical sizes ; note that 1 wfpc2/wf pixel = 0.1 arcsec = 8.2 pc at 17 mpc . the convolution of the psf with the king profiles was done in idl at 5 times the pixel resolution in order to avoid the effects of undersampling . the model cluster profiles were then converted back into daophot s psf format , with a lookup table for the variation of the psf as a function of position in the frame and its sub - pixel position . a pipeline consisting of daophot ii and idl programs was used to identify candidate clusters . sources were found using the daophot ii find routine using a threshold of 5@xmath12 and a fwhm of 1.4 pixels . lower thresholds were found to give too many spurious sources . changing the fwhm was found not to strongly affect the number / quality of detected sources - partially resolved ( broader ) sources were not thrown out by the routine . however , asymmetric sources are rejected by the find routine , perhaps resulting in the elimination of some larger hii regions from our sample . aperture photometry was then done using the phot routine with a 5 pixel / 0.5 arcsec radius , which is the radius at which the wfpc2 standard photometric system is defined @xcite . for the profile photometry , we fit each source in the f555w images to all 17 of the model cluster profiles in turn ( see 3.1 ) . for most sources , the @xmath13 ( the quality - of - fit parameter output by daophot ii ) of the profile fit varies smoothly with the core radius of the model cluster profile . this is shown for a number of sources in vcc 1448 in figure 2 . note that many of the minimum @xmath13 values have values very close to 1 , confirming that the psf and model cluster profiles accurately match the data . the fit with the minimum @xmath13 value was used to derive photometry and to determine the core radii of the cluster candidates . aperture corrections were measured for each of the model cluster profiles and ranged between + 0.026 for the more compact model cluster profiles and + 0.034 for the model cluster profile with r@xmath14=20pc . since the majority of our cluster candidates have r@xmath15pc , we applied an aperture correction of + 0.026 to all data . figure 3 shows the scatter in profile - fit and aperture photometry . the scatter shows an asymmetric distribution with the aperture photometry in general being brighter , probably due to confusion from background and nearby sources . in our final sample ( crosses ) , this scatter is significantly reduced due by our application of a 90% completeness cut ( see 3.4 ) and the exclusion of two high surface brightness regions in vcc 1374 . for both aperture and profile - fit photometry the f555w and f814w fluxes were converted into johnson v and kron - cousins i magnitudes using the procedure given in @xcite . in addition , the charge transfer efficiency was corrected for using the method in @xcite and the geometric distortion was corrected for by multiplying the images by the pixel area map . the v magnitudes of all sources were derived from the profile - fit photometry ( v@xmath16 ) ; however , we do nt derive f814w profile - fit photometry and so aperture magnitudes ( for both v & i ) are used for colors . the median error on the v - i color is 0.15 magnitudes for our full sample of cluster candidates . in the color range between v - i of 0.5 and 1.5 , critical for separating young and old clusters , the median v - i error is 0.12 magnitudes . these errors are derived from the errors in the aperture magnitude as reported by daophot ii . magnitudes for the galaxies as a whole were also derived from the wfpc2 f555w data . the magnitudes were calculated in two ways : ( 1 ) by computing the total flux within a circular aperture with a radius two times the pertrosian radius @xcite , and ( 2 ) by summing up all the light in the wf3 chip after masking out clearly extended background galaxies and very bright foreground stars . we take the mean of these two methods as the galaxy magnitude , v@xmath17 , shown in table 2 . also shown is the difference in the two magnitudes , @xmath18v@xmath17 . this number , which is larger than the formal error in the magnitude calculation , gives a sense of the uncertainty in our magnitudes . for brighter galaxies this difference is typically less than 0.1 magnitudes , while it is over 1 magnitude for the faintest galaxies in our sample . no color or cte corrections were made to the magnitudes , these effects contribute to v@xmath17 below the 0.1 magnitude level . as noted earlier , vcc 1448 is larger than the wf3 chip field of view , and fcc 76 and vcc 888 also may contain some emission off the chip ; for these galaxies , the magnitudes given are lower limits . as noted above , a measure of the core radius of each candidate cluster is obtained using the best fitting model cluster profile . at our assumed distance of 17 mpc , one wf pixel corresponds to 8.2 pc . we expect to be able to determine core radii smaller than this , because the broadening due to the king profiles extends out past the central pixel ; for example , the fraction of light falling outside of the central pixel changes from 63% for the psf to 79% for the model cluster profile with a core radius of 2 pc . nevertheless , it is important to consider the effects of our limited spatial resolution and of the variability in the @xmath19 psf on our measurements of the core radii of cluster candidates . we address the issue of the constancy of the psf here , and address the effects of resolution and random error in the next section . we conducted a test to check whether or not our routine accurately detected point - sources in seven wfpc2 fields taken during the period aug . 2 , 1997 to apr . 3 , 1999 , spanning the dates of our observations . the fields were drawn from hst program pids 6364 , 6804 , 7307 , 7335 , and 7434 . all fields were either galactic globular clusters or located in the large magellanic cloud and thus should be dominated by point sources . we ran them through the same pipeline as we used for our program sources . including all seven fields , @xmath580% of the isolated sources were best fit by model cluster profiles with core radii less than or equal to 1 pc ( 0.012 arcsec ) . as a comparison , only @xmath540% of isolated sources in our program sources were best fit by model cluster profiles with core radii @xmath20 1 pc . this test shows that our method reliably assigns point sources small values for the core radius . it also allows us to examine the variation of the psf over time , which would result in a systematic error in our determinations of the cluster radii . for the seven fields , the peak in the best - fitting core radius distribution varied between 0 and 1 pc . this suggests that the psf is unstable at a level corresponding to @xmath51 pc ( 0.012 arcsec ) in our method . as we will see in the next section , this is roughly the same level as our random error on a single measurement of the core radius . we conducted artificial star and cluster tests using the daophot ii addstar routine . these tests were done by adding grids of objects generated from the original psf ( r@xmath21=0 pc ) and model cluster profiles with r@xmath21 of 1 , 3 , 7 and 12 pc . the grids were added to the wf3 chip galaxy images with a random position offset and typically contained 600 objects each separated by 30 pixels . input magnitudes for the stars were determined randomly and ranged from 18th to 27th magnitude . the fields with the added stars / clusters were then run through the same pipeline outlined in the previous section to identify and determine properties of the objects . because identification and determination of the radius was done using the f555w observations , the tests were only done for those images . for each galaxy , a minimum of 50,000 artificial objects were inserted for each of the five model cluster profiles used . the artificial star and cluster tests were run using the condor distributed computing software . curves of fifty and ninety percent completeness as functions of r@xmath21 and f555w magnitude are shown in figure 4a . the completeness is best for point sources and gets worse with increasing core radius due to the broader profiles and lower surface brightnesses of these objects . this result is generally applicable to all surveys of extragalactic globular clusters the completeness limit for point sources can differ by more than a magnitude from the completeness limit of more extended objects . this makes correcting for this completeness function complicated if there are large numbers of extended sources . we note that the f555w magnitudes differ from the johnson v magnitudes by a color correction term which is typically small ( few @xmath22 magnitudes ) , and therefore can be ignored for the completeness limits . the derived error on the magnitudes is typically @xmath230.01 on the bright end and @xmath50.2 near the 50% completeness limit . our actual errors , especially on the bright end , will be somewhat worse than this due to the effects of pixelization . the artificial cluster tests can also be used to determine the random error in our determination of the core radii of the cluster candidates . figure 4b shows the standard deviation of the difference between the input core radius and the detected core radius ( @xmath12(r@xmath21 ) ) . for point sources , this number is typically less than 1 pc , consistent with the results from our point source fields test . most of the galaxies have similar errors at each core radius , but the brightest galaxies have larger errors caused by the increased background against which the profiles were fitted . the error in core radius also is a function of magnitude - the median value of @xmath12(r@xmath21 ) is shown for each input core radius in figure 5 . the vertical lines indicate the median 90% completeness for each input core radius . given the systematic error of up to 1 pc from the point source fields test , and the random errors shown in figure 4b and figure 5 , we believe our measurements of core radius have an error of @xmath52 pc for objects brighter than the 90% completeness limit . in addition , the decreased random error for small objects means that objects with a best - fitting core radius of @xmath242 pc can reliably be considered resolved . in this section , we describe how we defined a sample of candidate clusters based on their individual properties and then how we statistically corrected that sample in order to determine the number of clusters in each galaxy . we defined the sample of candidate clusters conservatively . using the completeness data above , we interpolated the 90% completeness function for each galaxy to all core radii and used only clusters brighter than this limit . the limit corresponded to a v@xmath16 between 24.15 and 24.55 ( m@xmath25 of -7.0 to -6.6 ) , and was chosen to ensure the accuracy of the measured properties of the cluster candidates . clusters brighter than v@xmath16=19.65 ( m@xmath25=-11.5 ) were also rejected . this limit on bright objects was chosen to include even the brightest globular clusters . it should also include all but the most massive young clusters ( e.g. all the young and old clusters in the lmc would be included ) . the final number of cluster candidates in each galaxy is shown in table 2 . it is important to consider the possible sources of contamination in our sample . 1 . background galaxies - especially compact ellipticals and bulge components of spiral galaxies . 2 . foreground stars - faint stars in our own galaxy . 3 . compact , symmetrical hii regions in the target galaxies . 4 . single , very bright stars in the target galaxies . spurious detections in high surface brightness regions with many overlapping sources . visually examining the data , only one of our galaxies ( vcc 1374 ) contained problematically high surface brightness regions . the sources detected inside two high surface brightness regions in that galaxy were removed from the sample . these two regions can be clearly seen in figure 14 outlined by white boxes . the contamination of the sample by extremely bright stars was minimized by our inclusion of only those objects brighter than m@xmath26 . contamination of our sample by compact hii regions will be examined later using the h@xmath0 data in a subsample of the galaxies . we obtained background and foreground ( bg / fg ) source counts from the flanking ( wf2 and wf4 ) fields . in this section we describe how we found the expected bg / fg level for each galaxy . the expected bg / fg levels are then used in two ways : ( 1 ) to remove the bg / fg sources from the sample of candidate clusters as a whole , described in 4 , and ( 2 ) to determine the number of red candidate clusters in each galaxy for use in computing individual specific frequencies as described in 4.1 . we based the expected bg / fg level on the mean level of sources in the flanking fields in the entire sample . however , because each galaxy has different completeness limits , the bg / fg level varies from galaxy to galaxy . to determine the bg / fg level in a specific galaxy , all the sources in the flanking fields were put through the 90% completeness cut and bright - end [email protected] cut as described earlier in this section . then the bg / fg level and error were determined by taking the mean and standard deviation of the number of remaining sources in each flanking field . the values for each galaxy are shown in table 2 . note that the number of cluster candidates in table 2 is not corrected for the bg / fg level in any way . we also note that because the virgo and fornax fields had nearly identical average bg / fg counts and standard deviations , all fields were used in the bg / fg level determination . figure 6 shows the number of candidate clusters vs. the number of bg / fg sources on each chip . about half of all the galaxies , and 9 of the 11 galaxies with confirmed cluster membership , appear to have significant number of candidate clusters . the remaining galaxies have a number of candidate clusters consistent with the bg / fg contamination level . we note that in the tables , figures , and analysis that follow we excluded the eight candidate clusters with r@xmath21 values of 20pc . these objects were either the bulges of obvious background galaxies , or were in highly confused regions and most likely have inaccurate sizes , magnitudes and colors . for the seven galaxies that are confirmed members of the virgo cluster we have deep h@xmath0 observations . the reduction of these data was described in 2 . the h@xmath0 data were used to find the galaxy - wide h@xmath0 flux assuming a distance of 17 mpc ( table 3 ) . from the h@xmath0 fluxes , we derived star formation rates using the relationship given by @xcite : @xmath28 = \frac{l_{h\alpha}}{1.25 \times 10^{41 } [ ergs / sec ] } 10^{0.4 a_{h\alpha}}\ ] ] where @xmath29 is the h@xmath0 luminosity and @xmath30 is the absorption by dust at h@xmath0 ( 0.78 a@xmath25 ) . the values for the star formation rates given in table 3 were calculated assuming no extinction by dust internal to the galaxies and are thus lower limits . assuming absorption similar to that of hot stars in the lmc , @xmath31 @xcite , neglecting extinction would make the sfrs in table 3 roughly 30% too small . note that the h@xmath0 fluxes and sfrs in table 3 have errors dominated by the absolute calibration of the photometry ( i.e. 50% ) . the h@xmath0 fluxes are in rough agreement with those given in @xcite for vcc 328 , 1374 and 1992 . the largest difference is for vcc 1374 , where our flux is roughly a factor of two ( 2@xmath12 ) smaller than in @xcite . we also detect h@xmath0 in vcc 1822 where they had a just an upper limit of @xmath32 ergs / sec . the values found for star formation rates fall in the same range as the dwarf irregulars in the local group @xcite . as mentioned in 3.4 compact hii regions are one possible source of contamination in our sample . to determine the possible contamination of clusters by hii regions , we first measured the h@xmath0 flux in a 1@xmath9 annulus around the location of each cluster . then , using a balmer decrement of 2.8 @xcite and the wfpc2/f555w filter transmission curve ( @xmath57.5% at h@xmath11 and @xmath51.3% at h@xmath0 ) we calculated the contribution of the h@xmath0 and h@xmath11 line flux to the v - band magnitude , v@xmath33 . the relation between this derived magnitude and the actual contribution to the f555w flux is somewhat uncertain due to three effects : ( 1 ) the absolute calibration of the h@xmath0 is reliable only to about 50% , ( 2 ) any dust will reduce the h@xmath11 flux relative to the h@xmath0 flux and ( 3 ) the poorer resolution of the ground based h@xmath0 data means that any h@xmath0 flux located within 1@xmath9 of the cluster candidate will be included in our value for v@xmath33 . taking into account our errors , we considered a cluster candidate a possible hii region contaminant if it had a v@xmath33 brighter than , or within 0.5 magnitudes of , its v@xmath16 . our detection limit of @xmath8 ergs / sec results in v@xmath33 @xmath5 26 , and is therefore sensitive enough to detect contamination in even our faintest sources . only 32 out of 185 cluster candidates were found to be contaminated by h@xmath0 emission , with 27 of those lying in the two galaxies brightest in h@xmath0 , vcc 1992 and vcc 1374 . figure 7 shows the properties of the h@xmath0-contaminated candidate clusters . of the blue cluster candidates ( v - i @xmath23 0.85 ) , @xmath526% were found to be likely hii regions , while only 5 ( 6% ) of the red candidate clusters were contaminated by h@xmath0 , all located in vcc 1374 . this difference between blue and red cluster candidates is not surprising , since hii regions are expected to appear blue through the f555w and f814w filters . this suggests that our sample of red clusters , which we consider to be candidate globular clusters , are essentially uncontaminated by hii regions . of our galaxy sample , only 11 galaxies , 7 in virgo and 4 in fornax , have known velocities that are less than 3000 ( see table 1 ) . these are the only galaxies for which cluster membership is confirmed and therefore to which the distance is known . we note that most of the other galaxies in our sample ( including three vcc galaxies with high velocities ) have insignficant or barely significant numbers of candidate clusters , as might be expected for more distant galaxies with fainter globular cluster systems , or for fainter galaxies within the virgo and fornax groups . in this section , we consider the bulk properties of the cluster candidates using _ only the 11 likely cluster members _ , fcc 76 , 120 , 247 and 282 and vcc 83 , 888 , 1374 , 1448 , 1822 and 1992 . these 11 galaxies contain 237 cluster candidates and an expected bg / fg contamination ( see table 2 ) of 51 sources . figure 8 shows the properties of the candidate clusters before ( thin ) and after subtraction of the bg / fg ( thick ) . this subtraction was done by selecting the number of expected background sources at random from the bg / fg population and then removing the closest matching cluster candidate ( from the sample as a whole ) in v@xmath16 , v - i and r@xmath21 . error bars on the histograms of the subtracted sources were determined by taking the standard deviation of 200 monte carlo simulations in which the candidate cluster list and the bg / fg list were drawn at random from the initial sample . the most notable difference between the bg / fg sources and the candidate clusters is in color - the bg / fg are mostly red objects , and account for almost all of the candidate clusters with colors of v - i @xmath1 1.6 . the bg / fg also contains many compact ( r@xmath34 pc ) objects , as would be expected from foreground stars . the distribution of cluster candidates peaks at v - i @xmath5 1 and r@xmath35 0 . the leftmost bin in v@xmath16 ( @xmath1 24 mag ) is strongly affected by our completeness - based cuts and therefore the apparent turnover at faint magnitudes is not significant . however , a turnover at that magnitude would be expected for the typical globular cluster system with m@xmath36 - 7.4 . we can divide the sample based on color to attempt to separate old from young clusters . the single stellar population ( ssp ) models of @xcite show that , for any metallicity , clusters with v - i of @xmath1 0.85 are likely to be older than 10@xmath37 years , while clusters bluer than this are likely to be younger . using this cut , we regard the redder / older clusters as globular cluster candidates . even if we are detecting clusters with ages of a few billion years , with our magnitude cutoff ( m@xmath38 ) these clusters will have globular cluster masses . the bluer objects are most likely young open or massive clusters , with a small fraction being hii regions ( see above ) . table 3 gives the numbers of blue and red cluster candidates in each galaxy , while figure 9 shows histograms comparing the magnitudes and sizes of all 114 blue ( thick ) and 72 red ( thin ) cluster candidates left after bg / fg subtraction . the blue objects tend to be dimmer and more compact than the red objects . this result can be explained by either the presence of single bright stars or compact open clusters among the blue cluster candidates . figure 10 shows the radial distribution of the blue vs. red cluster candidates after bg / fg subtraction . the red cluster candidates typically lie farther from the center of the galaxy than the blue cluster candidates . k - s tests show that the distributions are significantly different , with p values ranging between 10@xmath39 and 10@xmath40 depending on the background subtraction . this distribution is what would be expected if the red cluster candidates are an extended population of globular clusters while the blue cluster candidates are younger clusters formed in the body of the galaxy . the color and magnitude information also allows us to roughly approximate the masses of our candidate clusters . figure 11 shows a color - magnitude diagram of background - subtracted cluster candidates with lines representing the color - magnitude evolution of single stellar populations of three masses @xcite ranging in ages between 60 myr and 18 gyr . the blue clusters almost entirely lie inbetween masses of 10@xmath2 and 10@xmath41 m@xmath3 , while the redder clusters all have masses significantly greater than 10@xmath41 m@xmath3 , and thus do represent objects with masses similar to those of globular clusters . the milky way globular cluster system ( mwgcs ) ( * ? ? ? * revised 2003 ) provides a useful point of reference with which to compare our globular cluster candidates . in order to facilitate this comparison , we took the @xcite catalog data and passed it through our completeness - based cuts ( see 3.4 ) assuming a distance modulus of 31.15 . of the 141 clusters with m@xmath25 and r@xmath21 measurements in the catalog , 83 survived the cuts . of the 96 that also had v - i color information , 76 survived the cuts . we then compared the surviving milky way clusters to our bg / fg subtracted sample of candidate clusters . figure 12 shows histograms comparing the v magnitudes , v - i colors and r@xmath21 of the mwgcs to the 72 cluster candidates with v - i @xmath1 0.85 . these plots exhibit some notable similarities and differences between our red cluster candidates and the mwgcs . first , the luminosity function of the red cluster candidates is similar to that of the mwgcs , with a k - s p=0.68 . this indicates that _ the luminosity function of the red candidate clusters is consistent with the universal globular cluster luminosity function _ found in other galaxies @xcite . this similarity also supports the idea that our red candidate clusters are , in fact , globular clusters . second , we consider the color distribution . the mwgcs v - i distribution is shown both before and after subtraction of the foreground reddening , where e(v - i)@xmath42 e(b - v ) @xcite . in both cases , there is a clear color cut off at v - i @xmath5 0.8 , similar to the color cut we used to select our candidate clusters . all the galaxies in our sample have foreground reddenings e(b - v ) @xmath23 0.04 @xcite , with an average of e(b - v ) = 0.022 . however , the internal reddening due to dust is likely to be larger for clusters seen inside / behind the body of the galaxy . corrected only for the mean foreground reddening , the median color of our sample is v - i @xmath43 , corresponding to an [ fe / h ] of -1.2 using the relation from @xcite . this median color is redder than the 0.91 of the dereddened mwgcs , but is less than the average v - i of 1.04 for elliptical galaxies @xcite . the median color is also redder than the globular cluster systems in virgo , fornax and leo cluster dwarf elliptical galaxies @xcite , which are found to have a correlation between host galaxy luminosity and redder colors / higher metallicity . the unknown internal reddening prevents us from determining if their is a similar correlation in our galaxy sample . because of the unknown reddening , our median color is a redward limit and the metallicity estimate is an upper limit . this suggests that the clusters in the dwarf galaxies in our sample are in general metal - poor . finally , the bottom panel of figure 12 shows that the distribution of core radii of the red cluster candidates appears to be broader than that of the mwgcs . however , errors in the determination of the core - radius could make the distribution appear broader than it is . to compensate for this , we took the detected milky way clusters and added to each a random error , the magnitude of which was determined by its core radius . this error was interpolated from the median of the errors shown in figure 4b . a k - s test was performed using the resulting distribution of milky way core radii and that of the virgo / fornax red cluster candidates . the random errors applied to the milky way data resulted in a range of p values with a median value of 0.002 and a standard deviation of 0.009 . this test confirms that the red cluster candidates are significantly broader than the milky way globular clusters . note that the varying completeness limit inhibits our ability to detect broader clusters relative to more compact ones , adding to the significance of the difference between the cluster candidates and the mwgcs . to summarize , comparison of cluster candidates with v - i @xmath1 0.85 to milky way globular clusters suggests that their luminosity functions are similar , while the red cluster candidates have larger sizes . the average color is uncertain due to unknown reddening , but is consistent with the red cluster candidates being relatively metal poor . as discussed above , the number of red ( v - i @xmath1 0.85 ) cluster candidates , when corrected for the expected bg / fg level , provide an estimate of the number of globular clusters in each galaxy in our sample . a common way of expressing the number of clusters in a galaxy relative to its luminosity is the quantity s@xmath44 , the specific frequency @xcite . the specific frequency is defined as : @xmath45 where n@xmath46 is the number of globular clusters in a galaxy that has absolute magnitude m@xmath25 . in order to determine an accurate value for the specific frequency we need to : ( 1 ) remove the bg / fg contamination from our number of red candidate clusters , ( 2 ) correct for the unobserved portion of the luminosity function , and ( 3 ) correct for the incompleteness in our sample . we remove the expected bg / fg level from our red cluster candidate count in a fairly simple way . of all the bg / fg sources , 75% of them have v - i @xmath1 0.85 . therefore we subtracted the expected number of red bg / fg sources ( 75% of the number given in table 2 ) from the number of red candidate clusters ( table 3 ) to obtain n@xmath46 . using this value for n@xmath46 we derived a conservative estimate of the specific frequency , shown as s@xmath47 in table 3 . we consider this estimate conservative because it is uncorrected for both steps ( 2 ) and ( 3 ) above . traditionally , s@xmath44 values are corrected for any unobserved portion of the globular cluster luminosity function . given a milky way globular cluster luminosity function , with m@xmath48=-7.33 and @xmath12=1.23 @xcite , we determined the fraction of sources fainter than the 90% completeness limit for point sources in each galaxy , and then scaled up n@xmath46 by this factor . the resulting luminosity function corrections have values of @xmath51.4 . these corrections and the specific frequencies calculated with them ( `` s@xmath44 w/ corr '' ) are shown in table 3 and plotted in figure 13 . we are unable to correct for the incompleteness in our data because the incompleteness is dependent on the unknown size distribution of the clusters . however , given that most of our sources are fairly compact , and that we define our sample using only clusters brighter than the 90% completeness level , the correction for incompleteness should be small . in both s@xmath44 calculations , m@xmath25 was derived from the magnitude in table 2 assuming a distance modulus of 31.15 ; the error in the bars include the effects of poisson noise , the errors in the bg / fg counts and galaxy magnitudes ( @xmath18v@xmath17 in table 2 ) , and uncertainty introduced by the errors in the v - i colors . our calculated specific frequencies may be in error due to the following effects : ( 1 ) internal reddening in the galaxies can cause some young clusters to have colors that would land them in our red sample . this would result in an overestimate of the specific frequency . given the low star formation rates of most of our galaxies , it is unlikely that there are large numbers of young embedded clusters in our sample . such clusters might also be expected to be associated with h@xmath0 emission . of the seven galaxies in our h@xmath0 sample , only one ( vcc 1374 ) had any red clusters associated with h@xmath0 emission . the effect of the random errors in v - i color on the specific frequency are included in the error estimates given in table 3 and figure 13 . ( 2 ) the lack of a completeness correction . in both estimates of the specific frequency given above , we are unable to correct for the possible presence of faint , extended objects . this could result in an underestimate of the specific frequency . ( 3 ) in order to compare our s@xmath44 values with those of early - type galaxies it is traditional to `` age - fade '' the magnitude of the galaxies to the magnitude it would have if it were an old stellar population @xcite . using a scenario in which a galaxy forms stars at a constant rate for 5 gyr and then fades for 5 gyr , @xcite find a fading of @xmath51.5 magnitudes in v , resulting in factor of @xmath54 increase in specific frequencies . however , without better knowledge of the stellar populations of the galaxies , and the destruction timescales for the globular clusters , it is not possible to determine what the exact increase in specific frequency would be . ( 4 ) intermediate - age ( @xmath49 gyr ) massive clusters may be included in this sample . to get a sense of the number of intermediate age clusters we might be detecting , we used the lmc as a case study . the lmc provides a nearby laboratory in which cluster ages are well known . using the old and intermediate - age lmc clusters given in @xcite and our detection limits , we would detect @xmath510 clusters as red candidate clusters . only half of these are bona fide globular clusters , with the other half being clusters with ages of a few gigayears . so it appears very likely that the presence of intermediate - age clusters contributes to the specific frequency values we measure . however , we note that intermediate - age clusters may also be included in most other extragalactic specific frequency measurements , which typically do nt include a method for estimating the ages and metallicities of the clusters . spectroscopic and ir observations done by @xcite and @xcite show that intermediate - age clusters do exist in some early type galaxies . we discuss the specific frequency values further in 5 . in this section , we consider the distribution of candidate clusters in individual galaxies from our sample . in general , we find that most galaxies have cluster candidates situated both inside and outside their apparent optical boundaries . the candidates located on the galaxies tend to be blue , while those around the periphery are more often red . this can easily be seen in figure 10 which shows a large difference in radial distribution between the red and blue cluster candidates . this distribution is consistent with these galaxies having recent star formation concentrated near their centers and a more widely distributed older globular cluster population . previous classifications of a number of the galaxies is somewhat uncertain . of the sample of 11 galaxies ( table 3 ) , 8 are clearly dwarf irregulars , while two are most likely dwarf ellipticals / spheroidals ( fcc 247 and vcc 1448 ) and one ( fcc 282 ) is a high surface brightness object with some suggestions of spiral structure , but would most likely still be classified as an irregular galaxy . figure 14 shows cmds for the cluster candidates in each galaxy and images with the location of red and blue candidates denoted with circles and squares , respectively . larger images of the galaxies are available from the authors by request . _ fcc 76 _ this fairly high surface brightness irregular galaxy has five blue and two red cluster candidates on its face . however , around the outskirts of the galaxy , five of the seven cluster candidates are red . there are also a fairly large number of clusters found at large radii ( 5 red , 3 blue ) . _ fcc 282 _ as noted above , this galaxy has an unusual morphology and was classified as an im / depec by @xcite . towards the outer portion of the galaxy , tightly wound spiral structure is visible . however there is no bulge component and the high surface brightness central area is mottled without any obvious peak . if fcc 282 is at the distance of the fornax cluster , as suggested by its velocity , it would have an m@xmath10 of about -17 , making it similar in luminosity to the lmc . based on its appearance , this galaxy is almost certainly not an early type galaxy ; we classify it as an irregular . most of the cluster candidates are located around the edge of the inner regions of the galaxy ( figure 14 ) and are predominantly blue . around the outskirts of the galaxy there are 5 red and 2 blue cluster candidates . _ vcc 83 _ this galaxy appears to be sandwiched between a pair of galaxies . given the other two galaxies small size ( @xmath23 15@xmath9 ) , high surface brightness , and spiral appearance , it seems that they are background galaxies in chance aligment . we have attempted to measure the magnitude of vcc 83 without including those galaxies . _ vcc 328 _ this galaxy has a very prominent system of blue clusters . the cluster candidates appear to be located preferentially along the edge of the apparent optical boundary of the galaxy ( particularly towards the lower right side in figure 14 ) , a property shared by the h@xmath0 distribution ( which can be seen in fig . 1 ) . of the 14 clusters located inside the apparent optical boundary of the galaxy , only one is red , while two appear to be hii region contaminants . _ vcc 1374 _ this galaxy contains active star formation and has an apparent diffuse h@xmath0 component . the 64 candidate clusters are the most detected in any galaxy in our sample . this is despite the exclusion of two of the most crowded regions of the galaxy ( seen outlined by white boxes in figure 14 ) in which cluster candidates could not be reliably identified . there is a mixture of blue and red candidates on the galaxy face suggesting either the presence of both young and older populations or large amounts of reddening . of the red candidates , 20% of them are found to be possible hii regions . given the differing resolutions of the ground - based and hst images , this could either be due to chance proximity of older clusters with hii regions , or evidence for highly reddened young clusters . if the red cluster candidates are in fact globular clusters , then the globular cluster system appears flattened , as is found in the lmc @xcite . of nine candidate clusters found around the outskirts of the galaxy , all but one is red , perhaps tracing a halo component to the globular cluster system . _ vcc 1448 _ this galaxy , a.k.a . ic 3475 , is the most well - studied galaxy in our sample . the galaxy is large ( @xmath51.7 ) , has a low surface brightness , is very hi and hii poor , and has smooth isophotes @xcite , all properties which would generally classify it as a de galaxy . however the presence of a central bar and a large number of knots on the surface of the galaxy might suggest the galaxy is a dirr . all the knots that appear in our field of view appear to be candidate clusters . our observations cover the bar of the galaxy , but clearly do not cover the entire halo . therefore , our integrated magnitude for the galaxy predominantly represents the magnitude of the bar ; it is more than a magnitude fainter than the rc3.9 value of v@xmath17=13.11 @xcite . this means that the specific frequency listed in table 3 is a measure of the local specific frequency , because it excludes much of the halo light and any halo clusters that may exist . some bright cluster candidates outside our field of view are clearly visible in fig . 1e of @xcite . the 26 candidate clusters in vcc 1448 all fall between v - i of 0.77 and 1.30 ( with the exception of one likely fg star at v - i=2.35 ) as can be seen in the color - magnitude diagram for vcc 1448 shown in figure 14 . this is consistent with the suggestion of @xcite that the knots were intermediate - age clusters , but is also consistent with the clusters being a metal - poor globular cluster system . @xcite find vcc 1448 to be 0.4 magnitudes bluer than a typical de and suggest that it is a metal - poor galaxy which has recently finished its last episode of star formation . spectroscopic observations or accurate three color photometry ( as in * ? ? ? * ) of these clusters would enable the determination of ages and shed some light on the star formation history of this unusual galaxy . _ vcc 1992 _ is a diffuse irregular galaxy with a number of prominent star formation regions with strong h@xmath0 emission . eight ( more than half ) of the blue cluster candidates are likely hii region contaminants . three of the seven red cluster candidates lie near the center of the diffuse emission , while three others around the outskirts of the galaxy . none are associated with the star forming regions , suggesting that they are , in fact , old or intermediate - age clusters . _ specific frequencies _ based on our results , it seems that dwarf irregular galaxies have a wide range of specific frequencies , including values greater than five . in the local group , the specific frequencies for many of the irregulars are 0 ( e.g. ic 1613 , sextans b ) while the magellanic clouds have [email protected] @xcite and wlm and ngc 6822 both have one detected globular cluster , giving them s@xmath44 values of @xmath52 @xcite . so it appears from the local group that dwarf irregulars typically have s@xmath44 values that range between 0 and 2 . this is similar to the values obtained for spiral galaxies @xcite . although there are many uncertainties in our specific frequency values ( see 4.1 ) , it appears that the specific frequencies found here for dwarf irregular galaxies vcc 83 , 1374 , 1992 and fcc 76 and 120 ( we omit likely de galaxy vcc 1448 from this list ) are higher than their local group counterparts . this result suggests that environment may play an important role in determining the characteristics of a globular cluster system . specifically , the high specific frequencies we observe might be related to the high density environments that these virgo and fornax dwarf irregulars live in . this is a plausible effect given the apparent connection between galaxy interaction and cluster formation @xcite . the high specific frequencies of our sample also affect the conclusions of @xcite on the possible relationship of de and dirr galaxies . @xcite found that the nucleated de galaxies have an average s@xmath44 of [email protected] , while the average for non - nucleated de galaxies is [email protected] . they use this to suggest that the dirr galaxies ( measured locally ) did not have sufficiently high specific frequencies to evolve into nucleated de galaxies . however , with age - fading , they suggested that dirr galaxies could evolve into non - nucleated de galaxies . our results modify this conclusion - given the high s@xmath44 values we observe , it is possible that both types of de galaxies could evolve from dirr galaxies . _ populous clusters _ the best - studied irregular galaxy , the lmc , is known to host a number of luminous , compact , blue clusters , often called populous clusters that are not seen in our own galaxy . from the @xcite catalogue , there are @xmath5 35 objects with m@xmath52 -8.5 and colors bluer than b - v=0.5 , which should be roughly equivalent to our v - i=0.85 cutoff . in our sample , 15 blue candidate clusters are brighter than m@xmath10=-8.5 ( see figure 11 ) , suggesting that a similar , but smaller , population exists among the dwarf irregular galaxies in the virgo & fornax clusters . our sample contains no blue objects as bright as r136 . _ size of red candidate clusters _ figure 12 and our k - s tests ( see 4 ) indicate that the distribution of core radii of our red cluster candidates is larger than the milky way globular clusters . there are also indications that globulars in the lmc , smc and fornax are larger than in the milky way @xcite . using data from @xcite for 12 globular clusters ( with age @xmath1 10 gyr ) in the lmc , we find a mean and median core radius of 2.5 and 2.0 pc respectively . for the milky way globulars in a comparable range of absolute magnitudes ( m@xmath52 -6.5 ) , the mean and median core radius is 1.5 and 0.9 pc @xcite . for elliptical galaxies , globular clusters have been found to have a mean half light radius of 2.4 pc @xcite , similar to , and somewhat smaller than the value of @xmath53 pc found for milky way globulars @xcite . so it seems that irregular ( and perhaps dwarf ellipticals as well ) have larger globular clusters than the milky way and giant elliptical galaxies . @xcite suggests that the reason for the large sizes of clusters in the magellanic clouds may be due to their formation in a lower density environment . however , destruction processes could also create differences in the average size of a population @xcite . it is therefore unclear whether the differences in size distribution are the result of differences in formation environment or evolution of the cluster system . we have studied the cluster systems of dwarf irregular galaxies using hst / wfpc2 observations of 28 galaxies in the virgo and fornax clusters , 11 of which are confirmed members . in these 11 confirmed members we have found 237 cluster candidates and determined their magnitudes , core radii , and v - i colors . based on source counts of neighboring , empty , wfpc2 fields , we expect a contamination of 51 background and foreground sources , which were statistically subtracted from the data . the cluster candidates were divided by color in order to separate red ( with age @xmath53 1 gyr ) and blue ( age @xmath54 1 gyr ) cluster candidates . we consider the red cluster candidates to be globular cluster candidates . the main results presented here are : 1 . the red cluster candidates have the same magnitudes as expected for globular clusters . more specifically , the luminosity function of galactic globular clusters @xcite is well matched by the luminosity function of our red cluster candidates down to our 90% completeness limit at m@xmath55 -7 . 2 . the median color ( v - i = 1.0 ) of the red cluster candidates is bluer than the median colors of globular cluster systems in elliptical galaxies @xcite , even without a correction for internal reddening . this is consistent with the globular cluster systems in these dwarf irregular galaxies being metal - poor . the distribution of core radii for the red candidate clusters is significantly larger than for galactic globular clusters . larger radii are also seen in the globular clusters of the lmc and other local group dwarves @xcite suggesting that this might be a general property of globular clusters in dwarf galaxies . the blue cluster candidates are in general more compact and fainter than the red cluster candidates . based on single - stellar population models , they have masses ranging between 10@xmath2 and 10@xmath41 m@xmath3 . h@xmath0 data on 7 virgo cluster galaxies suggests that 26% of the blue cluster candidates are in fact hii regions . the specific frequencies for the galaxies exhibit a large scatter , but are surprisingly large , with 7 of the 11 galaxies having specific frequency values greater than two . these values are surprising when compared to local group dwarf irregulars and other nearby late - type galaxies , which have specific frequencies less than two . we note , however , that our specific frequency values are uncertain due to the possible presence of reddened young clusters , massive intermediate - age clusters and incompleteness . we plan to follow up this work with spectroscopic observations or infrared imaging of some of these cluster systems . these observations of the cluster candidates will test the accuracy of our photometric results by enabling accurate age and metallicity determination . this work was supported by the national optical astronomy observatory , which is operated by the association of universities for research in astronomy , inc . , under cooperative agreement with the national science foundation . it also has been supported by the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , on behalf of the international gemini partnership of argentina , australia , brazil , canada , chile , the united kingdom , and the united states of america . the authors would like to thank : paul hodge , for his guidance and help observing , peter stetson for help with daophot ii , eduardo bica , for providing us with his tables , nick suntzeff for his helpful comments , ricardo covarrubias , and armin rest . fcc 5 & 03:17:51.8 & -36:45:59 & & 22/10/98 & + fcc 41 & 03:25:35.9 & -32:57:35 & & 16/12/98 & + fcc 76 & 03:29:43.2 & -33:33:25 & 1808 & 12/03/99 & + fcc 120 & 03:33:34.2 & -36:36:21 & 887 & 14/12/98 & + fcc 128 & 03:34:06.9 & -36:27:52 & & 8/01/99 & + fcc 173 & 03:36:43.2 & -34:09:33 & & 31/07/97 & + fcc 224 & 03:39:32.8 & -31:47:32 & & 31/12/98 & + fcc 247 & 03:40:42.4 & -35:39:40 & 1097 & 21/02/99 & + fcc 282 & 03:42:45.5 & -33:55:13 & 1225 & 8/11/98 & + fcc 337 & 03:51:01.2 & -36:41:00 & & 16/12/98 & + vcc 72 & 12:13:02.2 & + 14:55:58 & 6351 & 18/12/98 & + vcc 83 & 12:13:33.5 & + 14:28:51 & 2441 & 18/12/98 & yes + vcc 280 & 12:18:14.6 & + 11:28:52 & & 16/12/98 & + vcc 328 & 12:19:11.1 & + 12:53:05 & 2179 & 4/01/98 & yes + vcc 364 & 12:19:44.0 & + 12:16:54 & & 16/12/98 & + vcc 666 & 12:23:46.1 & + 16:47:25 & & 2/01/99 & + vcc 888 & 12:26:18.3 & + 08:20:57 & 1096 & 13/12/98 & yes + vcc 899 & 12:26:27.4 & + 06:42:32 & 4198 & 11/12/98 & + vcc 1165 & 12:29:10.4 & + 09:16:01 & & 10/12/98 & + vcc 1227 & 12:29:46.8 & + 11:10:02 & & 14/12/98 & + vcc 1374 & 12:31:37.7 & + 14:51:38 & 2559 & 19/12/98 & yes + vcc 1413 & 12:32:07.7 & + 12:26:04 & & 17/12/98 & + vcc 1448 & 12:32:40.7 & + 12:46:16 & 2583 & 13/02/99 & yes + vcc 1822 & 12:40:10.4 & + 06:50:48 & 1012 & 7/12/98 & yes + vcc 1889 & 12:41:46.0 & + 11:15:01 & 4725 & 16/12/98 & + vcc 1992 & 12:44:09.7 & + 12:06:47 & 1010 & 21/03/99 & yes + fcc 5 & 17.5 & 0.09 & 24.38 & 5 & 4.92 @xmath51 2.59 + fcc 41 & 18.2 & 1.20 & 24.50 & 3 & 5.27 @xmath51 2.52 + fcc 76 & 14.6 & 0.04 & 24.38 & 24 & 4.69 @xmath51 2.53 + fcc 120 & 15.9 & 0.23 & 24.50 & 12 & 5.04 @xmath51 2.54 + fcc 128 & 16.3 & 0.02 & 24.50 & 7 & 5.23 @xmath51 2.63 + fcc 173 & 18.7 & 3.79 & 24.45 & 4 & 5.23 @xmath51 2.70 + fcc 224 & 16.4 & 0.17 & 24.49 & 17 & 5.30 @xmath51 2.62 + fcc 247 & 17.6 & 0.10 & 24.45 & 6 & 5.19 @xmath51 2.67 + fcc 282 & 14.2 & 0.01 & 24.36 & 25 & 4.46 @xmath51 2.52 + fcc 337 & 17.0 & 0.41 & 24.53 & 4 & 5.42 @xmath51 2.67 + vcc 72 & 16.2 & 0.07 & 24.30 & 7 & 4.54 @xmath51 2.50 + vcc 83 & 15.5 & 0.03 & 24.31 & 18 & 4.42 @xmath51 2.55 + vcc 280 & 18.4 & 1.58 & 24.29 & 6 & 4.31 @xmath51 2.36 + vcc 328 & 16.0 & 0.12 & 24.37 & 18 & 4.54 @xmath51 2.53 + vcc 364 & 16.9 & 0.05 & 24.31 & 10 & 4.27 @xmath51 2.39 + vcc 666 & 16.6 & 0.06 & 24.36 & 7 & 4.65 @xmath51 2.51 + vcc 888 & 15.3 & 0.06 & 24.22 & 16 & 4.19 @xmath51 2.38 + vcc 899 & 16.8 & 0.17 & 24.21 & 0 & 4.15 @xmath51 2.34 + vcc 1165 & 16.9 & 0.28 & 24.28 & 5 & 4.38 @xmath51 2.56 + vcc 1227 & 18.9 & 0.16 & 24.31 & 5 & 4.58 @xmath51 2.45 + vcc 1374 & 14.6 & 0.01 & 24.20 & 64 & 3.96 @xmath51 2.34 + vcc 1413 & 17.4 & 0.02 & 24.37 & 6 & 4.73 @xmath51 2.43 + vcc 1448 & 15.0 & 0.10 & 24.36 & 25 & 4.58 @xmath51 2.48 + vcc 1822 & 16.1 & 0.01 & 24.19 & 7 & 3.92 @xmath51 2.30 + vcc 1889 & 16.1 & 0.08 & 24.29 & 6 & 4.35 @xmath51 2.54 + vcc 1992 & 15.9 & 0.05 & 24.33 & 22 & 4.27 @xmath51 2.52 + lccccccc fcc76 & 11 & 13 & 2.3 @xmath51 0.9 & 1.36 & 3.1 @xmath51 1.0 & & + fcc120 & 5 & 7 & 2.4 @xmath51 2.0 & 1.28 & 3.0 @xmath51 2.2 & & + fcc247 & 1 & 5 & 3.7 @xmath51 7.1 & 1.31 & 4.9 @xmath51 7.5 & & + fcc282 & 15 & 10 & 1.1 @xmath51 0.6 & 1.36 & 1.5 @xmath51 0.6 & & + vcc83 & 10 & 8 & 2.5 @xmath51 1.7 & 1.40 & 3.5 @xmath51 1.8 & 11.0 & 0.009 + vcc328 & 14 & 4 & 0.5 @xmath51 1.9 & 1.36 & 0.7 @xmath51 1.9 & 7.1 & 0.006 + vcc888 & 13 & 3 & -0.1 @xmath51 0.9 & 1.47 & -0.2 @xmath51 0.9 & 6.3 & 0.005 + vcc1374 & 39 & 25 & 5.3 @xmath51 1.3 & 1.49 & 7.8 @xmath51 1.5 & 54.3 & 0.044 + vcc1448 & 4 & 21 & 6.1 @xmath51 1.7 & 1.36 & 8.3 @xmath51 1.9 & 0.0 & 0.0 + vcc1822 & 3 & 4 & 1.0 @xmath51 1.6 & 1.50 & 1.4 @xmath51 1.7 & 0.2 & 0.0002 + vcc1992 & 15 & 7 & 2.9 @xmath51 2.1 & 1.39 & 4.0 @xmath51 2.3 & 16.0 & 0.013 +

we analyze the magnetic dipole contribution to atom - surface dispersion forces . unlike its electrical counterpart , it involves small transition frequencies that are comparable to thermal energy scales . a significant temperature dependence is found near surfaces with a nonzero dc conductivity , leading to a strong suppression of the dispersion force at . we use thermal response theory for the surface material and discuss both normal metals and superconductors . the asymptotes of the free energy of interaction and of the entropy are calculated analytically over a large range of distances . near a superconductor , the onset of dissipation at the phase transition strongly changes the interaction , including a discontinuous entropy . we discuss the similarities with the casimir interaction between two surfaces and suggest that precision measurements of the atom - surface interaction may shed light upon open questions around the temperature dependence of dispersion forces between lossy media .

0910.3133

ever since the work of lennard - jones @xcite has the interaction between atoms and surfaces been of interest in many fields of physics , chemistry and technology . the seminal work by casimir and polder @xcite demonstrated that the shift in atomic energy levels close to a conductor is a probe for the quantum fluctuations of the electromagnetic field , a key concept of quantum electrodynamics ( qed ) . in this context , a nonzero temperature becomes relevant for several aspects of the atom - surface interaction : thermally excited motion of the surface ( phonons ) and inelastic scattering of atomic beams @xcite , occupation of excited atomic energy levels , and enhancement of field fluctuations due to thermal photons @xcite . the latter aspect is usually associated with distances from the surface larger than the thermal wavelength @xmath1 , approximately @xmath2 at room temperature . the free energy of interaction typically shows a change in power law with distance around this point : generally , it is enhanced with respect to zero temperature and becomes proportional to @xmath3 . this is the classical limit where the interaction is mainly entropic in character @xcite . experimental progress in recent years has achieved the sensitivity required to detect the small energy shifts that occur at distances on the order of @xmath4 , making use of the exquisite control over the motion of atomic beams ( beam deflection @xcite , quantum reflection @xcite ) or clouds of ultracold laser - cooled atoms @xcite . the latter can be handled precisely in miniaturized traps implemented near solid state surfaces known as _ atom chips _ these devices use optical or magnetic fields for trapping and can hold atomic clouds at distances down to a few microns . here , the atom - surface interaction manifests itself typically as a distortion of the trapping potential ( loss of atoms through tunneling to the surface or change in the trap oscillation frequency ) . therefore , the design of such setups requires exact knowledge of atom - surface interactions and conversely , theory predictions can be tested experimentally with high precision . a surprising result of the research on atom chips is the importance of magnetic field fluctuations near the surface arising from thermally excited currents in the material of the chip ( johnson noise related to ohmic dissipation ) . these fluctuations couple to the spin magnetic moment of the trapped atoms and are known to provoke the loss of atoms from the trap by flipping the sign of the potential @xcite . these losses are a main obstacle for technical applications and further down - scaling of atom chips . predictions that superconducting materials reduce the spin - flip induced losses significantly have recently been backed by trap lifetime measurements @xcite . in this paper , we address the magnetic dipole contribution to the atom - surface ( casimir - polder ) interaction including nonzero temperature . one would expect this to be a small correction to the electric dipole coupling @xcite because of the smallness of the transition matrix elements @xcite . yet , the strong magnetic mode density close to a metallic surface @xcite and experimental evidence for magnetic spin flips call for a reconsideration of the magnetic contribution . in addition , the thermal occupation of photonic modes is quite relevant because magnetic transitions occur at much lower frequencies than electric ones , leading to a stronger temperature dependence . finally , it is well - known that dispersion forces between dielectric and magnetic materials can be repulsive @xcite , as has been shown for the magnetic casimir - polder interaction at @xmath5 in ref.@xcite . we were thus led to investigate whether at distances beyond @xmath4 the total atom - surface interaction might be reduced due to the magnetic contribution . in this work , we calculate the magnetic casimir - polder free energy of interaction at different temperatures and consider a few well - known models for the electromagnetic response of the surface . since ohmic losses are crucial for the thermal behavior , it is highly interesting to compare both normal metals and superconductors . the latter are described here in the frame of the two - fluid model and bardeen - cooper - schrieffer ( bcs ) theory @xcite . we demonstrate that the magnetic atom - surface coupling has very peculiar features unknown from its electrical equivalent . we find that for normal conductors at nonzero temperature , the magnetic dipole contribution to the interaction is reduced , while it is enhanced for superconductors and in certain non - equilibrium situations . this resembles the macroscopic casimir interaction between two dissipative plates , where the correct calculation of the force at large distances and nonzero temperatures has been the subject of debate @xcite . this article is organized as follows . in sec . [ atom - surface interaction ] , we give a brief review of the formalism used to calculate atom - surface interactions . section [ polarizabilities and green tensors ] presents the specific forms of the response functions and the experimental setups they describe . we also give expressions for the green s tensor in different asymptotic regimes of the distance between the atom and the surface . the magnetic casimir - polder free energy and entropy of an atom near metallic or superconducting surfaces at zero temperature is calculated in sec . [ magnetic casimir - polder ] . section [ casimir - polder interaction at nonzero temperature ] covers the effects at nonzero temperature and discusses the dissipative reduction in the interaction and questions connected to the entropy . non - thermal ( out - of - equilibrium ) states of the atoms that occur typically in experimental setups are investigated in section [ non - thermal states ] . we conclude summarizing and discussing the main results . further technical details are given in the appendices . quite a lot of research has been done on the interaction between an atom and a surface @xcite . it can been shown from perturbation theory with respect to the multipolar atom - field coupling @xcite that the free energy of a polarizable particle at nonzero temperature @xmath3 has the following general form ( einstein summation convention ) @xmath6 . \label{eq1}\ ] ] here , @xmath7 is the ( magnetic or electric ) polarizability tensor for the atom in a thermal state of temperature @xmath3 , and @xmath8 is the ( corresponding ) green s tensor in the presence of the surface , defined in eq.([eq : def - induced - field ] ) below . in the planar geometry we are interested in , the green s tensor depends only on the atom - surface distance @xmath9 and on frequency . it is well known that eq.([eq1 ] ) has the same form for electric or magnetic dipole couplings @xcite ; our notation is adapted to the magnetic case . a simple and general derivation of eq . can be given following refs.@xcite . the effective interaction potential between a polarizable particle and the ( magnetic or electric ) field @xmath10 @xcite is given by @xmath11 the expectation value @xmath12 is taken in an equilibrium state of the non - coupled system at temperature @xmath3 and implicitly evaluates symmetrically ordered operator products ; @xmath13 is the ( magnetic or electric ) dipole operator and @xmath10 the corresponding field operator , evaluated at the atom position @xmath14 . the factor @xmath15 arises from a coupling constant integration ( excluding a permanently polarized atom ) . within first - order perturbation theory , both the dipole moment @xmath16 and the field @xmath17 can be split into fluctuating ( fl ) and induced ( in ) parts : the fluctuating part describes the intrinsic equilibrium fluctuation , while the induced part arises in perturbation theory from the dipole coupling @xcite . . becomes @xmath18 here , we assume the fluctuating parts of the dipole and of the field to be decorrelated at this order . this assumption would break down at higher orders of perturbation theory . note that while in eq.([eneq ] ) , the total dipole and field operators ( heisenberg picture ) commute at equal times , this is no longer true for their ` in ' and ` fl ' constituents in eq.([eq : f - split - in - fl ] ) . the induced quantities are given , in frequency space , by the retarded response functions @xcite @xmath19 where the frequency dependence allows for a temporal delay . the equilibrium fluctuations follow from the fluctuation - dissipation theorem @xcite @xmath20 ~ , \label{eq : ftd_field } \\ \langle \mu_i^{\mathrm{fl}}(\omega)\mu_j^{\mathrm{fl}}(\omega')\rangle_t & = & \frac{\hbar}{2\pi } \delta(\omega-\omega ' ) \coth\left(\frac{\hbar \omega}{2 k_b t}\right ) \nonumber\\ & & \times { \mathrm{im}\,\ ! } [ \beta_{i j}^{t}(\omega)]~. \label{eq : ftd_dipole}\end{aligned}\ ] ] combining eqs.([eq : f - split - in - fl][eq : ftd_dipole ] ) , we recover eq . , setting @xmath21 . one uses the fact that the imaginary part of both green s tensor and polarizability tensor are odd in @xmath22 ( retarded response functions ) . the field correlations are needed at the same position @xmath23 , and it is easy to remove the divergent free - space contribution ( lamb shift ) from @xmath24 , by keeping in the green s tensor only the reflected part of the field @xcite . in a planar geometry , it follows from symmetry that the result can only depend on the dipole - surface distance @xmath9 . note that the green s tensor can also depend on temperature through the surface reflectivity . in a two - level model for the atom , the thermal polarizability @xmath25 contains a stronger @xmath3-dependence because of a fermi - dirac - like statistics @xcite , see eq.([eq : two - level_polarizability ] ) below . eq . is often expressed in an equivalent form using the analyticity of @xmath26 and @xmath27 in the upper half of the complex frequency plane . performing a rotation onto the imaginary frequency axis yields the so - called matsubara expansion @xcite @xmath28 where @xmath29 are the matsubara frequencies and the prime in the sum indicates that the @xmath30 term must be weighted by a prefactor @xmath15 . both @xmath31 and @xmath32 are real expressions for @xmath33 . if the atom is in a well defined state @xmath34 rather than in a thermal mixture , we have the expression of wylie and sipe @xcite @xmath35~,\end{gathered}\ ] ] where @xmath36 is the state - specific polarizability @xcite @xmath37 here , @xmath38 are the dipole matrix elements , @xmath39 is the frequency of the virtual transition @xmath40 ( @xmath41 for a transition to a state of lower energy ) . finally , the thermal occupation of photon modes @xmath42 in the second term in eq.([ws ] ) is the bose - einstein distribution . at @xmath5 it occurs only for excited states , for which @xmath43 for @xmath41 ( see eqs.(4.3 , 4.4 ) of ref.@xcite ) . the real part of the green s tensor can be given an interpretation from the radiation reaction of a classical dipole oscillator @xcite . similarly , this term is practically absent for the electric casimir - polder interaction of ground - state atoms because of the higher transition frequencies , @xmath44 . in the following , we call the matsubara sum ( first line ) in eq.([ws ] ) the _ non - resonant _ contribution , and the second line the _ resonant _ one , because it involves the field response at the atomic transition frequency . the formalism presented in the previous section is quite general and @xmath45 [ @xmath8 ] could represent either the magnetic or electric polarizability [ green s tensor ] , respectively . we now give the specific forms of these quantities in the magnetic case , focusing on a planar surface and specific trapping scenarios . the green s tensor for a planar surface can be calculated analytically . let the atom be on the positive @xmath47-axis at a distance @xmath9 from a medium occupying the half - space below the @xmath48-plane . by symmetry , the magnetic green s tensor @xmath49 is diagonal and invariant under rotations in the @xmath48-plane : @xmath50 + 2\frac{k^{2}}{\kappa^{2}}r^{\rm te}(\omega , k ) \boldsymbol{\hat z\hat z}\right]e^{-2 \kappa l } ~ , \label{eq : magnetic_gf } \ ] ] where @xmath51 is the vacuum permeability and @xmath52 , @xmath53 and @xmath54 are the cartesian dyadic products . we consider here a local isotropic , nonmagnetic bulk medium [ @xmath55 , so that the fresnel formulae give the following reflection coefficients in the te- and tm - polarization ( also known as @xmath56- and @xmath57-polarization ) @xcite @xmath58 where @xmath59 , @xmath60 are the propagation constants in vacuum and in the medium , respectively ( roots with @xmath61 ) @xmath62 and @xmath63 is the modulus of the in - plane wavevector . note that the magnetic green s tensor can be obtained from the electric one @xmath64 by swapping the reflection coefficients @xcite @xmath65 all information about the optical properties of the surface is encoded in the dielectric function @xmath66 . we will use four different commonly established descriptions , each of which includes ohmic dissipation in a very characteristic way . as it turns out , the magnetic casimir - polder interaction is much more sensitive to dissipation than the electric one ( see sec.[dissipative quenching ] ) . this is due to the fact that the resonance frequencies in the magnetic polarizability @xmath67 are much lower ( see sec.[s : beta - and - alpha ] ) . the first model is a _ drude metal _ @xcite @xmath68 where @xmath69 is the plasma frequency and @xmath70 is a phenomenological dissipation rate . this is the simplest model for a metal with finite conductivity . if @xmath71 is constant ( independent of temperature ) , the conductivity can be attributed to impurity scattering in the medium . the second model is the dissipationless _ plasma model _ @xmath72 : here , one sets @xmath73 in the right - hand side of eq.([eq : def - drude - epsilon ] ) . this corresponds to a purely imaginary conductivity . in the context of atom chips , the case of a _ superconductor _ is particularly interesting because dissipation is suppressed as the temperature @xmath3 drops below the critical temperature @xmath74 . we adopt here ( third model ) a description in terms of the two - fluid model , a weighted sum of a dissipationless supercurrent response ( plasma model ) and a normal conductor response @xmath75 \varepsilon_\mathrm{dr}(\omega)~,\\ \eta(t ) & = & \left [ 1 - \left(\frac{t}{t_c}\right)^4\right ] \theta(t_c -t ) ~ , \label{eq : epsilon - two - fluid - model}\end{aligned}\ ] ] where the _ order parameter _ @xmath76 follows the gorter casimir rule @xcite . at @xmath5 , the superconductor coincides with the plasma model , as is known from the london theory of superconductivity @xcite . the plasma model is thus the simplest description of a superconductor at zero temperature rather than a model for a normal metal . more involved descriptions of superconductors ( including bcs theory ) also reproduce the plasma behavior at low frequencies ( @xmath22 well below the bcs gap ) and temperature close to absolute zero . the full bcs theory of superconductivity can be applied in this context , too , using its optical conductivity @xcite , as recently discussed in ref.@xcite . we shall see below ( sec.[sec : superconductor ] ) , however , that the two - fluid model and bcs theory give very close results for realistic choices of the physical parameters . our fourth model takes a look at the peculiar case of a very clean metal . here , rather than by impurity scattering , dissipation is dominated by electron - electron or electron - phonon scattering . in these cases , the dissipation rate in the drude formula ( [ eq : def - drude - epsilon ] ) follows a characteristic power law @xmath77 at small temperatures and saturates to a constant value at high temperatures ( bloch - grneisen law ) . it is reasonable to call this system the _ perfect crystal _ model . as in a superconductor , dissipation is turned on by temperature , but in a completely different manner . this can be distinguished in the atom - surface interaction potential . for the drude model , there are three different regimes for the atom - surface distance that are determined by physical length scales of the system ( see ref.@xcite for a review ) : the skin depth in the medium , @xmath78 where @xmath79 is the plasma wavelength , and the photon wavelength in vacuum , @xmath80 note that @xmath81 for frequencies @xmath82 ( hagen - rubens regime ) . this is the relevant regime for the relatively low magnetic resonance frequencies . we then have @xmath83 which leads to the following three domains : ( i ) the _ sub - skin - depth region _ , @xmath84 , ( ii ) the _ non - retarded region _ , @xmath85 , and ( iii ) the _ retarded region _ : @xmath86 . in zones ( i ) and ( ii ) , retardation can be neglected ( van - der - waals zone ) , while in zone ( iii ) , it leads to a different power law ( casimir - polder zone ) for the atom - surface interaction . since the boundaries of the three distance zones depend on frequency , the respective length scales differ by orders magnitude between the magnetic and the electric case . for electric dipole transitions , the hagen rubens regime can not be applied because the resonant photon wavelength is much smaller . the role of the skin depth is then taken by the plasma wavelength @xmath87 . this implies that the casimir - polder zone ( iii ) for the electric dipole interaction occurs in a range of distances where magnetic retardation is still negligible [ zones ( i ) and ( ii ) ] . in the three regimes , different approximations for the reflection coefficients that appear in the green s function ( [ eq : magnetic_gf ] ) can be made . we start with the drude model where in the _ sub - skin - depth zone _ @xcite , we have @xmath88 and @xmath89\frac{\omega^2}{4c^2k^2},\nonumber\\ r^{\rm tm}(\omega , k)&\approx & \frac{\epsilon(\omega)-1}{\epsilon(\omega)+1}\left[1 + \frac{\epsilon(\omega)}{\epsilon(\omega)+1}\frac{\omega^2}{c^2k^2}\right ] . \label{eq : r_subskindepth } \ ] ] at intermediate distances in the _ non - retarded zone _ , the wavevector is @xmath90 , hence , @xmath91 finally , in the _ retarded zone _ we can consider @xmath92 , so that @xmath93 a similar asymptotic analysis can be performed for the other model dielectric functions . it turns out that eqs.([eq : r_subskindepth][eq : epsilon_retarded ] ) can still be used , provided the assumption @xmath94 holds . this is indeed the case for a typical atomic magnetic dipole moment and a conducting surface . the asymptotics of the green s functions that correspond to these distance regimes are obtained by performing the @xmath95-integration in eq.([eq : magnetic_gf ] ) with the above approximations for the reflection coefficients . the results are collected in table [ magneticappr ] . one notes that the @xmath96-component is larger by a factor @xmath97 compared to the @xmath98- and @xmath99-components . the difference between the normal and parallel dipoles can be understood by the method of images @xcite . furthermore , the magnetic response for a normally conducting metal in the sub - skin - depth regime is purely imaginary and scales linearly with the frequency @xmath22 the reflected magnetic field is generated by induction . only the superconductor or the plasma model can reproduce a significant low - frequency magnetic response , via the meiner - ochsenfeld effect . in contrast , the electric response is strong for all conductors because surface charges screen the electric field efficiently . the imaginary part of the green s functions determines the local mode density ( per frequency ) for the magnetic or electric fields @xcite . these can be compared directly after multiplying by @xmath100 ( or @xmath101 ) , respectively . as is discussed in refs.@xcite , in the sub - skin - depth regime near a metallic surface , the field fluctuations are mainly of magnetic nature . this can be traced back to surface charge screening that efficiently decouples the interior of the metal and the vacuum above . magnetic fields , however , cross the surface much more easily as surface currents are absent ( except for superconductors ) . this reveals , in the vacuum outside the metal , the thermally excited currents from the bulk . [ cols="<,^,^,^,^,^,^ " , ] + the magnetic and electric polarizabilities are determined by the transition dipole matrix elements and the resonance frequencies . we are interested in the retarded response function , which for an arbitrary atomic state @xmath34 is given by eq.([eq : ws_polarizability ] ) above . when the atom is in thermal equilibrium , we have to sum the polarizability over the states @xmath34 with a boltzmann weight : @xmath102 where @xmath103 is the partition function . in the limit @xmath104 , we recover the polarizability for a ground state atom . for a two - level system with transition frequency @xmath105 , the previous expression takes a simple form and can be expressed in terms of the ground state polarizability [ eq . ( [ eq : ws_polarizability ] ) , where @xmath106 : @xmath107 let us now compare the electric and magnetic polarizabilities . the magnetic transition moment among states with zero orbital spin scales with @xmath108 where @xmath109 is the land factor for the electron spin and @xmath110 the bohr magneton . electric dipoles are on the order @xmath111 with @xmath112 the bohr radius . with the estimate that the resonance frequencies ( @xmath105 and @xmath113 ) determine the relevant range of frequencies , we have approximately @xmath114 where @xmath115 is the fine - structure constant . the magnetic interaction is thus expected to be a small correction . conversely , the narrower range of frequencies makes it much more sensitive to the influence of temperature . we have seen now that the polarizability of an atom takes a positive constant value at low frequency and the induced magnetic dipole is parallel to the magnetic field ( paramagnetism ) . let us consider for comparison a metallic nanosphere . if its radius @xmath116 is smaller than the penetration depth and the wavelength , the polarizability is given by @xcite @xmath117r^{3}~.\ ] ] this quantity vanishes at low frequencies and has a negative real part ( diamagnetism ) . for a qualitative comparison to an atom one can estimate , e.g. , the magnetic oscillator strength , defined by the integral over the imaginary part of the polarizability at real frequencies @xmath118 from the clausius - mossotti relation follows the electric counterpart : @xmath119 we find that the nanoparticle has a dominantly electric response , similar to an atom , but the ratio of the oscillator strengths depends on the material parameters and the radius . from the above expressions , we find that the absolute value of the nanosphere s magnetic oscillator strength is actually smaller than the one of an atom if the sphere s radius @xmath120 . the resonance frequencies relevant for the magnetic casimir - polder potential depend on the trapping scheme . we focus here on alkali atoms that are typically used in ultracold gases and distinguish between optical and magnetic traps . in an _ optical trap _ , we may consider the case that the magnetic sublevels are degenerate and subject to the same trapping potential ( proportional to the intensity of a far - detuned laser beam ) . magnetic dipole transitions can then occur between hyperfine levels whose splitting is on the order of @xmath121 , corresponding to temperatures of @xmath122 ( see appendix [ magnetic transition matrix elements ] for more details . ) in contrast , electric dipole transitions occur in the visible range @xmath123 or @xmath124 . if we average over the magnetic sublevels , we get an isotropic magnetic polarizability . this allows to write @xmath125 , so that in eq.([eq1 ] ) or ( [ eq : matsubara - series ] ) @xmath126 the setup we will consider in most of our examples is an atom in a _ magnetic trap_. in these traps , one uses the interaction of a permanent magnetic dipole with an inhomogeneous , static magnetic field @xmath10 . let us consider for simplicity a spin @xmath15 manifold : the zeeman effect then leads to a splitting of the magnetic sublevels by the larmor frequency , @xmath127 in weak fields . to give an order of magnitude , @xmath128 at @xmath129 . atoms in those magnetic sublevels where @xmath130 are weak - field seekers , and can be trapped in field minima . the magnetic trap we have in mind is a two - wire trap suspended below the surface of an atom chip . currents in the two wires , combined with a static field , create a field minimum below the chip surface , with gravity pulling the potential minimum into a position where the magnetic field is nonzero and perpendicular to the surface . magnetic dipole transitions are then generated by the parallel components @xmath131 , @xmath132 of the dipole moment ( see appendix [ magnetic transition matrix elements ] ) . in this anisotropic scenario , the components of the magnetic polarizability tensor are given by @xmath133 and @xmath134 . the relevant components of the green s tensor are , therefore , @xmath135 we should mention that many experiments do not realize a global equilibrium situation , as assumed in eq.([eq1 ] ) . in typical atom chip setups , atoms are laser cooled to @xmath136 temperatures or prepared in a well - defined state , while the surface is generally at a much higher temperature , even when superconducting . for the description of such situations , a more general approach _ la _ wylie and sipe [ eqs.([ws ] ) and ( [ eq : ws_polarizability ] ) ] is more suitable , and we discuss the results in sec.[non - thermal states ] . before addressing these , we start with thermal equilibrium free energies , however . this may be not an unrealistic assumption in spectroscopic experiments where casimir - polder energies are measured with atoms near the window of a vapor cell @xcite . from the theoretical viewpoint , thermal equilibrium provides an unambiguous definition of the entropy related to the atom - surface interaction . we shall see that this quantity shows remarkable features depending on the way dissipation and conductivity is included in the material response . this closely parallels the issue of the thermal correction to the macroscopic casimir interaction , a subject of much interest lately . to summarize , in an optical ( isotropic ) trap , the equilibrium casimir - polder free energy is given by the matsubara sum @xmath137 while in a magnetic ( anisotropic ) trap , we have @xmath138 the ( free ) energy vs. distance has been calculated numerically for an anisotropic magnetic dipole in front of a half - space filled with a normal conductor ( fig.[abb : l_polder_plasma&drude ] , top ) or described by the plasma model ( fig.[abb : l_polder_plasma&drude ] , bottom ) . the thick black curves give the zero - temperature result ( see the caption for parameters ) . the dashed asymptotes are discussed in this section . all energies are normalized to the @xmath139 power law of the non - retarded casimir - polder energy near a perfectly reflecting surface . the scale factor @xmath140 , given in the caption , is slightly smaller than eq.([eq : non - retarded - t=0 ] ) below . these and the following results have been obtained from the numerical procedure described in appendix [ a : numerics ] . for an anisotropic magnetic dipole ( transition moment parallel to the surface ) . drude model ( top ) and plasma model ( bottom ) , with plasma frequency @xmath141 ( @xmath142 ) and @xmath143 . the transition frequency is @xmath144 ( @xmath145 ) . in both plots , the free energy scale @xmath146 is the value at @xmath147 obtained in the plasma model . gray dashed lines indicate the asymptotic laws at zero temperature eqs.([eq : sub - skin - depth - regime - plasma][eq : retarded - limit ] ) , while colored dashed lines include thermal corrections according to eqs.([eq : thermal - regime - drude ] , [ eq : non - retarded - regime - finite - t - drude ] ) . , title="fig:",width=302 ] for an anisotropic magnetic dipole ( transition moment parallel to the surface ) . drude model ( top ) and plasma model ( bottom ) , with plasma frequency @xmath141 ( @xmath142 ) and @xmath143 . the transition frequency is @xmath144 ( @xmath145 ) . in both plots , the free energy scale @xmath146 is the value at @xmath147 obtained in the plasma model . gray dashed lines indicate the asymptotic laws at zero temperature eqs.([eq : sub - skin - depth - regime - plasma][eq : retarded - limit ] ) , while colored dashed lines include thermal corrections according to eqs.([eq : thermal - regime - drude ] , [ eq : non - retarded - regime - finite - t - drude ] ) . , title="fig:",width=302 ] the magnetic casimir - polder potential is always repulsive as expected from the interaction between an oscillating magnetic dipole and its image at the conducting surface . the sign is also consistent with the macroscopic casimir interaction between a conducting and a permeable surface ( ` mixed ' dirichlet - von neumann boundary conditions ) , see e.g. ref.@xcite . the curves in fig.[abb : l_polder_plasma&drude ] make manifest the crossovers between the distance regimes introduced in sec.[s : distance - regimes ] above . the relevant length scales are here the skin - depth @xmath148 , evaluated at the transition frequency ( for the normal conductor ) , the plasma wavelength @xmath149 ( for the plasma model ) , and the transition wavelength @xmath150 . the case of the superconductor is discussed in sec.[sec : superconductor ] below ( fig.[abb:2f - polder ] ) : within the two - fluid model adopted here , it is identical to the plasma model at zero temperature . the temperature dependence interpolates between the drude and plasma case , as discussed in secs.[s : plasma ] and [ s : dissipative - quenching ] . the zero - temperature ( black curves ) case for a drude model has been stated earlier in ref.@xcite ; we give details on the asymptotes . taking the limit @xmath151 in eq.([eq : matsubara - series ] ) recovers the well - known expression @xmath152 in the _ sub - skin - depth regime _ @xmath153 , the distance dependence in the anisotropic case ( [ eq : fe - aniso ] ) for the drude model becomes @xmath154 where @xmath155 is the magnetic transition dipole matrix element , cf . appendix [ magnetic transition matrix elements ] . this expression is obtained by using the sub - skin - depth asymptote of the magnetic green s tensor ( first column of table [ magneticappr ] ) under the @xmath156-integral ( [ fe_zero_t ] ) and cutting the integral off at the border of this regime , @xmath157 , i.e. , @xmath158 . the small - distance calculation for the plasma model can be done in a similar way . in both the sub - skin - depth and non - retarded regimes , the green s tensor becomes independent of @xmath156 ( see table [ magneticappr ] ) , and the frequency integral depends only on the polarizability . therefore , no logarithm appears as in the dissipative case , but @xmath159 in the _ non - retarded regime _ ( intermediate distances ) , the interaction energy in the drude model @xcite and in the plasma model behave alike @xmath160 this is calculated as outlined above for the plasma model . the energy ( [ eq : non - retarded - t=0 ] ) is identical to the interaction of the magnetic dipole @xmath161 with its image , calculated as for a perfectly conducting surface . indeed , the @xmath139 power law is consistent with the dipole field of a static ( image ) dipole . in the _ retarded region _ @xmath162 ( not discussed in ref.@xcite ) , the free energy of the drude is identical to the one of the plasma model . retardation effects lead to a change in the power law with respect to shorter distances , identical to the electric casimir - polder interaction : @xmath163 the calculation of this asymptote follows the same lines as in the electric dipole case , see ref.@xcite . comparing different transition wavelengths @xmath164 ( e.g. , zeeman vs hyperfine splitting ) : the smaller the transition energy , the larger the retarded interaction . the numerical data displayed in fig . [ abb : l_polder_plasma&drude ] agree very well with all three asymptotes . eqs.([eq : non - retarded - t=0 ] ) and ( [ eq : retarded - limit ] ) illustrate that the magnetic atom - surface interaction is reduced relative to the electric one by the fine - structure constant @xmath165 , as anticipated earlier in eq.([eq : magnitude ] ) . one should bear in mind , of course , that the length scales for the cross - overs into the retarded regime are very different . a crossing of the non - retarded magnetic and the retarded electric potentials would be expected for a distance of order @xmath166 , where it is clear that both energies are already extremely small . in addition , the temperature should be low enough so that the thermal wavelength ( @xmath167 is the first matsubara frequency ) @xmath168 satisfies @xmath169 . indeed , we shall see in the following section that a nonzero temperature can significantly reduce the magnetic casimir - polder potential . in this section , we consider the temperature dependence of the casimir - polder interaction at _ global _ equilibrium , in particular using the temperature - dependent polarizability @xmath170 ( [ eq : two - level_polarizability ] ) . this provides also a well - defined calculation of the atom - surface entropy , see sec.[s : entropy ] . scenarios with atoms prepared in specific magnetic sub - levels are discussed in sec.[non - thermal states ] . the set of curves in fig.[abb : l_polder_plasma&drude ] illustrates the strong impact of a nonzero temperature for the drude ( normally conducting ) metal : its magnitude is reduced for any distance @xmath9 . in the plasma model ( no dissipation ) , the main effect is the emergence of a different long - distance regime : the _ thermal regime _ @xmath171 [ eq.([eq : def - thermal - wavelength ] ) ] ) where the interaction becomes stronger than at @xmath172 . the latter kind of behavior could have been expected from the thermal occupation of photon modes within the thermal spectrum . the effect in the drude model is more striking and is explained in sec.[s : dissipative - quenching ] below . a significant difference with the electric dipole interaction is the fact that it is quite common to have temperatures much larger than the magnetic resonance energies , @xmath173 or @xmath174 . thermal effects thus start to play a role already in the non - retarded regime , and can be pronounced at all distances . the usual description of the high - temperature ( or keesom @xcite ) limit is based on the term @xmath175 in the matsubara sum ( [ eq : matsubara - series ] ) @xmath176 indeed , the higher terms are proportional to the small factor @xmath177 that appears in the green s function @xmath178 . this description is discussed in more detail in the following sections . in the plasma model and more generally , for all materials where the reflection coefficient @xmath179 goes to a nonzero static limit , the magnetic green s tensor @xmath180 is nonzero as well . the leading order potential in the thermal regime is then given by eq.([eq : fe_hight ] ) . for the anisotropic polarizability of eq.([eq : fe - aniso ] ) , and assuming @xmath173 , the temperature dependence drops out , and we find from a glance at table [ magneticappr ] @xmath181 ( assuming @xmath182 ) . this is identical to the zero - temperature result in the non - retarded regime ( [ eq : non - retarded - t=0 ] ) , as can also be seen in fig.[abb : l_polder_plasma&drude ] . if the temperature is lower , @xmath183 , but the distance still in the thermal regime , the factor @xmath184 in the static polarizability reduces the interaction slightly ( @xmath185 in fig.[abb : l_polder_plasma&drude ] , bottom ) . the atom - superconductor interaction shows a richer behavior compared to the plasma model , as illustrated in fig.[abb:2f - polder ] . , @xmath71 , @xmath105 and normalization @xmath186 as in fig.[abb : l_polder_plasma&drude ] , critical temperature @xmath187 . temperatures are @xmath188 . , width=302 ] at @xmath172 , it strictly coincides with the plasma model , as it must for the two - fluid description ( [ eq : epsilon - two - fluid - model ] ) adopted here . the large - distance ( thermal ) asymptotes are the same as in the plasma model for @xmath189 . the reasoning leading to eq.([eq : plasma - thermal - regime ] ) can be applied here as well : the response of the superconducting surface to a static magnetic field is characterized by a nonzero value for @xmath190 because of the meiner - ochsenfeld effect . although the superconducting fraction decreases to zero , proportional to the product @xmath191 , the interaction potential eq.([eq : plasma - thermal - regime ] ) stays constant because it does not depend on this ` effective plasma frequency ' . this picture also explains the lowering of the sub - skin - depth asymptotes in fig.[abb:2f - polder ] : from eq.([eq : sub - skin - depth - regime - plasma ] ) , the casimir - polder potential is proportional to @xmath192 . this gives scale factors @xmath193 for the cases @xmath194 , in quite good agreement with the numerical data . it is worth mentioning that the full bcs theory can give results in close agreement with the simple two - fluid model we use here . in fig.[abb : mag_polder_bcs ] , we show the temperature dependence of the casimir - polder potential ( at fixed distance @xmath9 ) for the two cases . we choose here a damping parameter @xmath195 in the same order of the zero - temperature gap @xmath196 . the bcs calculations have been performed using a recently developed efficient technique of calculating the optical conductivity at imaginary frequencies @xcite and using an approximative form of the gap equation @xcite . calculations over a larger parameter range , but restricted to @xmath172 , have been reported by skagerstam _ _ @xcite . , @xmath105 as in fig.[abb : l_polder_plasma&drude ] . scattering rate @xmath195 , critical temperature @xmath197 , atom - surface distance @xmath198 . energies normalized to @xmath199 . , width=302 ] going back to fig.[abb:2f - polder ] , note that at @xmath200 , the superconductor jumps to a completely different behavior , identical to the drude metal . this is expected from the two - fluid model ( [ eq : epsilon - two - fluid - model ] ) , but also in mattis - bardeen theory where the gap parameter @xmath201 vanishes above @xmath74 , and the optical conductivity @xmath202 coincides with the drude model , see refs.@xcite . [ dissipative quenching ] as mentioned above , the drude model and the superconductor around the critical temperature show an unusually strong temperature dependence in the magnetic casimir polder potential . the strong suppression at large distances ( figs.[abb : l_polder_plasma&drude ] and [ abb:2f - polder ] ) arises from the fact that the green s tensor @xmath203 at zero frequency in the normal conducting state . the leading order potential ( [ eq : fe_hight ] ) vanishes , and one has to consider the next term @xmath204 in the matsubara sum [ eq : matsubara - series ] , so that the exponentially small factor @xmath205 governs the thermal ( large distance ) regime . we call this the _ thermal decoupling _ of the atom from the ( normal ) metal . this phenomenon is related to low - frequency magnetic fields that penetrate the ( non - magnetic ) surface . indeed , the vanishing of @xmath206 could have been expected from the bohr - van - leeuwen theorem @xcite that states that for any classical system , the magnetization response to static fields must vanish . both conditions apply here : the zeroth term in the matsubara series involves static fields , and is also known as the classical limit . indeed , except for the material coupling constants , eq.([eq : fe_hight ] ) no longer involves @xmath207 , while the next matsubara terms do ( via @xmath4 ) . the bohr - van - leeuwen theorem does not apply to a superconductor whose response is a quantum effect ( illustrated , for example , by the macroscopic wave function of ginzburg - landau theory ) , and by extension , not to the plasma model , as recently discussed by bimonte @xcite . we now calculate the next order in the matsubara series to understand the temperature dependence of the casimir - polder shift near a metal . for simplicity , we consider again the limiting case @xmath208 which simplifies the polarizability to the keesom form , @xmath209 in the thermal regime , @xmath171 , we use the large - distance limit of the green s tensor ( cf . the retarded regime of table [ magneticappr ] ) . the matsubara frequency @xmath210 then yields the mentioned exponential suppression @xmath211 where @xmath164 is the magnetic resonance wavelength ( cf fig.[abb : l_polder_plasma&drude ] ) . at shorter distances , we have to perform the matsubara summation . in the regime @xmath212 , we consider the non - retarded approximation to the green s tensor and make the approximation @xmath213 . the sum over the polarizability @xmath214 can then be done with the approximation ( [ eq : keesom - polarisability ] ) , and we get @xmath215 the scaling @xmath216 is in good agreement with fig.[abb : l_polder_plasma&drude ] . the crossover into the sub - skin - depth regime is now temperature - dependent and occurs where the skin depth @xmath217 . this corresponds to a temperature @xmath218 . the involved frequency @xmath219 is characteristic for the diffusive transport of electromagnetic radiation in the metal at wavevectors @xmath220 . in the sub - skin - depth regime , the leading order approximation to the matsubara sum involves terms up to a frequency @xmath221 . this leads to an asymptote similar to eq.([eq : sub - skin - depth - drude ] ) , but with the ratio @xmath222 in the argument of the logarithm and an additional factor @xmath223 . it is well - known that at high temperatures where the free energy scales linearly in @xmath3 [ eq.([eq : fe_hight ] ) ] , the dispersion interaction is mainly of entropic origin @xcite . more precisely , the interaction is proportional to the change in entropy of the system `` atom plus field plus metallic surface '' , as the atom is brought from infinity to a distance @xmath9 from the surface . we calculate in this section the atom - surface entropy according to @xmath224 this entropy definition is unambiguous for the global equilibrium setting of the previous section . the behavior seen in the previous figures indicates significantly different entropies for the surface models , with a strong dependence on the presence of dissipation ( conductivity ) at low frequencies . this parallels the discussion of the macroscopic casimir entropy for the dispersion interaction between two plates , a subject of recent controversies , where the drude and plasma models give different answers @xcite . the results that follow indicate that the magnetic casimir - polder interaction may provide an alternative scenario to investigate this point . and [ abb:2f - polder ] . the unit of entropy @xmath225 is defined in eq .. the inset shows a the data for the drude model and its asymptotics obtained from eqs.([eq : thermal - regime - drude],[eq : non - retarded - regime - finite - t - drude ] ) in logarithmic scaling . , width=302 ] the atom - surface entropy ( [ eq : def - entropy ] ) is plotted in fig.[abb : mag_polder ] for surfaces of different material . in all models , the entropy vanishes at high temperatures because to leading order , the free energy ( [ eq : fe_hight ] ) becomes independent of @xmath3 , and higher orders vanish exponentially with @xmath3 . ( this feature is specific to the thermal polarizability of a two - level system . ) it is remarkable that the vanishing of the entropy happens at temperatures already much smaller than the ` geometrical scale ' @xmath226 [ i.e. @xmath227 with @xmath167 the first matsubara frequency ] . this points towards another characteristic energy scale in the atom - surface system , discussed below . one notes in fig.[abb : mag_polder ] very small values for a plasma and a superconductor , two cases where the dc conductivity diverges . the superconductor shows a narrow , pronounced entropy peak at @xmath74 : we interpret this as the participation of the atomic dipole in the phase transition . indeed , the electromagnetic waves near the surface are slightly shifted in phase due to the interaction with the magnetic dipole moment . the atom - surface interaction can be thought of a sum over all these phase shifts , similar to feynman s interpretation of the lamb shift . in the drude model , we observe a broad peak at temperatures where the thermal energy @xmath228 becomes comparable to the photon energy of the magnetic resonance , @xmath229 . comparable to this scale for our parameters is the diffusive energy @xmath230 , introduced after eq.([eq : non - retarded - regime - finite - t - drude ] ) . we thus attribute the atom - surface entropy to the participation of the atom in the thermally activated diffusive motion of charges and fields below the metal surface ( johnson - nyquist noise ) . this motion involves , at the relevant low energies , mainly eddy currents whose contribution to the casimir entropy ( in the plate - plate geometry ) has been recently discussed by two of us @xcite . as @xmath3 drops below the diffusive scale @xmath231 , the eddy currents ` freeze ' to their ground state and the entropy vanishes linearly in @xmath3 . the dotted curve in fig.[abb : mag_polder ] corresponds to the ` perfect crystal ' that has not been discussed so far . it gives rise to a nonzero atom - surface entropy in the limit @xmath151 which is an apparent violation of the nernst heat theorem ( third law of thermodynamics ) . this has also been discussed for the two - plate interaction @xcite , but the entropy defect here has a different sign ( it is negative for two plates ) . the sign can be attributed to our atomic polarizability being paramagnetic , while metallic plates show a diamagnetic response . using the technique exposed in ref.@xcite , the limit of the atom - surface entropy as @xmath151 can be calculated , with the result for an anisotropic dipole : @xmath232 the second line applies in the non - retarded limit @xmath233 . this expression is used to normalize the data in fig.[abb : mag_polder ] and provides good agreement for @xmath234 . one can argue along the same lines as in ref.@xcite that the nernst theorem is actually not applicable for this system , since the perfect crystal can not reach equilibrium over any finite time in the limit of vanishing dissipation , @xmath235 . the entropy @xmath225 then describes the modification that the atom imposes on the ensemble of field configurations that are ` frozen ' in the perfectly conducting material . in the two - plate scenario considered in refs.@xcite it has been shown that not only dissipation but also nonlocality of the response has strong implications for the entropy . in particular , the residual entropy @xmath236 vanishes , because at very low temperatures , the anomalous skin effect and landau damping take the role of a nonzero dissipation rate . though we have not considered nonlocality in this work explicitly , one can expect the same thing to happen in the magnetic casimir - polder interaction . it should be mentioned , that we have also found negative values for the atom - surface entropy , albeit very small , for temperatures between @xmath237 and @xmath238 and in the retarded regime ( see fig.[abb : mag_polder_plasma ] ) . the sign depends on the orientation of the dipole and is sensitive to a balance between the te- and tm - polarized parts of the magnetic green s tensor . , @xmath105 as in fig.[abb : l_polder_plasma&drude ] , atom - surface distance @xmath239 where the thermal effects are better visible . same entropy scale factor @xmath225 as in fig.[abb : mag_polder ] , defined in eq.([eq : delta_entropy ] ) . the anisotropic curve is blown up to become visible . , width=302 ] we have argued in the last section , that many realistic setups involve non - equilibrium situations . atom - chips are a typical example where two independent phenomenological temperatures can be introduced for the atom ( or a sample of atoms ) and the surface . this temperature gradient is metastable on experimentally relevant time scales because of the weak interaction between the subsystems . we now analyze the case where the surface is described by a temperature @xmath3 , and the atom prepared in a well - defined state @xmath34 . more complex configurations can be studied starting from this simple case . we first consider two - level atoms and then multilevel atoms , including hyperfine transitions as they occur for the alkali group . as before , there is a single resonance frequency @xmath105 for the two - level atom . depending on whether the atom is prepared in the ground - state @xmath240 or the excited state @xmath241 , the sign of the polarizability ( [ eq : ws_polarizability ] ) changes . referring to the first line of eq.([ws ] ) , @xmath36 does no longer depend on temperature . the resonant term of the second line involves the thermal occupation number that we approximate by its classical value @xmath242 . this is sufficiently accurate at room temperature and typical magnetic resonances . in this limit , we obtain a simple expression for the magnetic casimir - polder free energy ( anisotropic case , argument ` @xmath243 ' for ground - state atom ) @xmath244 - \mathcal{h}_{xx}(l,0 ) \right\ } \nonumber~.\end{aligned}\ ] ] the first line is similar to the result in the drude model because of the missing zeroth matsubara term . from this expression , we now discuss the differences between the drude and the plasma model . for the drude model , the resonant contribution involving @xmath245 becomes significant in the non - retarded regime . in particular , combined with the non - resonant contribution , it changes the sign of the casimir - polder potential already at short distances , resulting in an _ attractive interaction _ , as soon as @xmath246 , see figs.[abb : ooe_mag_polder ] and [ abb : ooe_mag_polder_2 ] . , @xmath71 , @xmath105 as in fig.[abb : l_polder_plasma&drude ] . note the scale factor @xmath247 to show the classical limit @xmath248 . the limiting curve at high temperatures ( gray dashed line ) can be inferred from the second line of eq.([ws2level ] ) . , width=302 ] , but vs. temperature . distance @xmath249 . the energy scale @xmath250 is the value at @xmath251 . , width=302 ] in contrast , the second line of eq.([ws2level ] ) nearly vanishes in the plasma model because the magnetic green s function @xmath252 is approximately independent of frequency , at least in the non - retarded regime . we thus get a situation where the zeroth matsubara term is nearly removed from the casimir - polder potential . the resonant term still gives the leading order contribution , once the expansion of the occupation number is pushed to the next order , @xmath253 . we then get @xmath254 where the last expression applies in the non - retarded regime and is identical to the @xmath172 case [ eq.([eq : non - retarded - t=0 ] ) ] , cf . fig.[abb : ooe_mag_polder_plasma ] . at larger distances ( retarded regime ) , the difference between the green s functions in the second line of eq.([ws2level ] ) is nonzero and becomes the leading term : @xmath255 \label{eq : retarded - n - eq - plasma}\end{aligned}\ ] ] note that this has a longer range than the @xmath256 power law ( [ eq : non - retarded - n - eq - plasma ] ) , see fig.[abb : ooe_mag_polder_plasma ] . this effect is well known from the electric - dipole interaction of excited atoms @xcite and consistent with the classical interpretation ( frequency shift of an antenna ) of the resonant term . , @xmath105 and scale factor @xmath257 as in fig.[abb : l_polder_plasma&drude](bottom ) . at high temperatures , the curve can be well approximated by the classical contribution ( thick gray dashed line ) . the oscillating part of eq.([eq : retarded - n - eq - plasma ] ) not shown here , sets in at distances @xmath258 . , width=302 ] the atom - surface potential now involves transitions @xmath259 to both higher and lower energy levels . eq.([ws ] ) yields the following form of the free energy @xmath260 - \mathcal{h}_{jj}(l,0 ) \right\}~ , \nonumber \ ] ] where we assume again that @xmath261 which is valid in many experiments . the sign of the interaction depends on the relative weight of virtual transitions to lower and higher energy levels . from eq.([eq : ws_polarizability ] ) , a virtual level @xmath262 gives a positive contribution to the polarizability and a positive prefactor for the second line in eq.([wsgeneral ] ) , these terms being negative for levels @xmath263 . we can interpret this sign change from the difference between stimulated emission into the thermal radiation field ( for the excited atom ) and photon absorption ( for the ground state ) . generally speaking , these contributions do not cancel each other because the matrix elements @xmath264 are not the same . let us consider the example of @xmath265 , prepared in the magnetically trappable hyperfine state @xmath266 of the @xmath267 ground state configuration ( fig.[fig : rb87 ] ) . this atom has vanishing orbital momentum @xmath268 , nuclear spin @xmath269 , and a single valence electron so that @xmath270 . the splitting between the hyperfine levels @xmath271 ( lower ) and @xmath272 is @xmath273 , to which the zeeman splitting in the magnetic trap must be added with the correct land factor . we use the same larmor frequency as before for the two - level atom ; because of @xmath274 , we are still in the weak - field regime . .,width=188 ] assuming a quantization axis perpendicular to the surface , see sec.[s : optical - magnetic - traps ] , we get an anisotropic polarizability . the necessary matrix elements are calculated in appendix [ magnetic transition matrix elements ] . numerical results for the casimir - polder interaction according to eqs . and , are shown in figs . [ abb : ooe_mag_rb_drude ] and [ abb : ooe_mag_rb_plasma ] for the drude and plasma models , respectively . atom in a given hyperfine state ( @xmath275 ) . parameters @xmath276 , @xmath71 as in fig.[abb : l_polder_plasma&drude ] . the larmor frequency ( for virtual transitions between neighboring zeeman levels ) has the same value @xmath277 as before , and the hyperfine splitting is @xmath278 . , width=302 ] , but for the plasma model . the numerical calculations for extremely low temperatures @xmath279 , where the potential changes sign , could not be performed with sufficient precision . , width=302 ] near a normal conducting surface described by the drude model , the interaction for @xmath172 is attractive at all distances . we associate the sign reversal ( compared to the absolute ground state considered so far ) to the coupling to the lower - lying zeeman levels . at high temperatures , the interaction becomes dominated by the resonant contribution that grows linearly with @xmath3 and is repulsive . again , we find the opposite sign as for the ground - state atom in fig.[abb : ooe_mag_polder ] . thus , the sum of both contributions leads to a maximum of the free energy at a nonzero , @xmath3-dependent distance . in the plasma model , the potential crosses over globally ( for all distances ) from attractive to repulsive . for all practical temperatures , @xmath280 , and the interaction will be repulsive as shown in fig.[abb : ooe_mag_rb_plasma ] . the results illustrate that the magnetic dipole interaction of an excited atom will be repulsive in all practical realizations . to summarize , the strong dependence of the thermal correction on dissipation in the surface occurs in both non - equilibrium situations considered here : two - level or multilevel atoms prepared in a given energy state . the magnetic casimir - polder potential thus offers an opportunity to distinguish between the two models on the basis of experimental data taken at low surface temperatures and small distances within the possibilities of today s experiments . we have considered the interaction of a magnetically polarizable particle with a metallic or superconducting surface including the effects of nonzero temperature and out - of - equilibrium situations . previous work had considered mostly the case of electric polarizability , e.g. , ref.@xcite , or was limited to a static magnetic dipole @xcite or zero temperature @xcite . the magnetic atom - surface interaction is repulsive over a large range of parameters and turns out to be highly sensitive to both thermal fluctuations and dissipation . in this respect , it shows similarities with the casimir interaction between metallic or magneto - dielectric plates . the @xmath5 results of ref.@xcite suggested that the magnetic interaction might be enhanced by raising the temperature , possibly creating a regime where it dominates over the electric contribution . in fact , thermal enhancement occurs only near a superconductor at distances beyond the thermal wavelength , where the material response is governed by the meissner effect . in normal conductors , field penetration prevents such a regime in accordance with the bohr - van leeuwen theorem and the casimir - polder energy is exponentially suppressed in global equilibrium . this behavior can be understood qualitatively from the competition between attractive and repulsive contributions to the force . repulsion arises from the fluctuations of the magnetic dipole , coupled to its mirror image . this is similar to the interaction between electric currents ( lenz rule ) . field fluctuations , on the other hand , produce attractive forces , due to the paramagnetic character of the atom polarizability . both contributions differ in their temperature dependence and depend on the state of the atom ( thermalized , spin polarized ) . for example , attractive forces arise between a ground - state atom and a normal conductor , as the temperature scale exceeds the magnetic transition energy . under realistic conditions , this flips the sign of the interaction in the regime accessible to experiments ( fig.[abb : ooe_mag_polder ] ) . considering the casimir - polder entropy , we found that atoms probe the fluctuations in the material : for instance , at the superconducting phase transition a pronounced peak appears . in most situations , the entropy vanishes at absolute zero in agreement with nernst s theorem . the only exception we found was the particular case of a ` perfect crystal ' [ drude dissipation rate @xmath281 , already discussed thoroughly in the context of the two - plate casimir interaction @xcite . here , the entropy takes a ( positive ) nonzero value at zero temperature and can be understood as the participation of the atom in the disorder entropy of currents frozen below the surface @xcite . indeed , the magnetic dipole moment mimics the current response of a second plate , except for the sign . if we compare with an electric dipole , the temperature dependence of the casimir - polder interaction is relatively weak there . this is due to the larger value of the transition frequency , so that for realistic temperatures there is no difference between the ground - state and the thermalized polarizability . this was assumed in ref.@xcite , where it was also stated that all good conductors behave very similarly . in fact , the electric dipole coupling is dominated by tm-,polarized modes which can not penetrate into the bulk due to screening by surface charges . we have shown here that the casimir - polder interaction between a metal and a fluctuating magnetic dipole resembles in many respects the casimir interaction between two metallic plates . in both scenarios , the thermal dependence is much more pronounced for a normally conducting surface compared to a description without ohmic losses . the role of dc conductivity in the two - plate scenario is still an open question at nonzero temperature . the atom - surface interaction may thus provide an alternative way to investigate the temperature dependence of the casimir effect , e.g. in atom chip experiments . the main challenge is the small value of the interaction energy as compared to the electric one . future measurements of the magnetic casimir - polder interaction may involve high precision spectroscopy on the shift in hyperfine or zeeman levels . in order to separate effects of the magnetic and the electric dipole coupling it will be important to find control parameters that affect exclusively the magnetic contribution . this may include the variation in external fields , isotope shifts , and highly polarizable atomic states like rydberg atoms . the work of h.h . and c.h . was supported by grants from the german - israeli foundation for scientific research and development ( gif ) . f.i . thanks the alexander von humboldt foundation for financial support . the potsdam group also received support from the european science foundation ( esf ) within the activity ` new trends and applications of the casimir effect ' . f.s . acknowledges support from micinn ( spain ) , through grant no . fis-2007 - 65723 , and the ramon areces foundation . r.p . and s.s . acknowledge partial financial support from ministero delluniversit e della ricerca scientifica and from comitato regionale di ricerche nucleari e di struttura della materia . the thermal free energies have then been calculated numerically in the matsubara - formalism by summing over a thermal spectral free energy density @xmath282 at discrete imaginary frequencies . now , the infinite sum is replaced by a finite one plus an integral to approximate the remaining infinite partial sum @xmath283 if the upper summation limit @xmath284 is chosen correctly , the remainder is a slowly varying function and the error is small , according to the euler - maclaurin formula . in all systems considered in this work , the exponential in the green s function ensures this property . numerical calculations require an automatic and fast estimate for the summation limit @xmath284 . in a similar scheme a fixed number of matsubara terms ( @xmath285 ) has recently been proposed @xcite . it should be pointed out , that since the matsubara frequencies are linear in @xmath3 , any such scheme with a fixed number of terms breaks down at low temperatures , where the integrand @xmath286 has not sufficiently decayed . a basic criterion for @xmath284 is that the partial sum is sufficiently large , so that the integral is only a small correction @xmath287 to avoid the evaluation of the integral in eq.([eq : n_numerik ] ) we have used an upper bound @xmath288 which exploits the exponential decay of @xmath243 introduced by the green s tensor . typically , @xmath289 . in our numerical calculations , we targeted errors @xmath290 $ ] and obtained sums over @xmath291 $ ] terms . in some cases like bcs theory , the calculation of the optical response is not trivial and the remainder integral was neglected completely . this requires a small enough value of @xmath292 and yields a systematic numeric error @xmath293 . this section gives the magnetic transition matrix elements used in the calculations of polarizabilities and free energies according to eqs.([ws ] , [ eq : ws_polarizability ] ) or . the simplest approach used is the spin @xmath294 system ( two - level atom ) with states @xmath295 . the dipole operator is @xmath296 and the quantization axis is @xmath297,so that @xmath298 according to eq . ( [ eq : two - level_polarizability ] ) we need only matrix elements connecting to the ground - state @xmath299 , the only nonvanishing of which yield @xmath300 transitions connecting identical states do not contribute and we obtain the result of eq . . let us now calculate the matrix elements for the @xmath301 atom prepared in a hyperfine s - state , @xmath302 the dipole operator @xmath303 , assuming vanishing orbital momentum and because of the smallness of the nuclear land g - factor @xmath304 . we express the states @xmath305 in the uncoupled basis of the spin and nuclear spin momenta with the help of the clebsch - gordan coefficients @xmath306 where the action of the components of the spin operator on a state is known [ eqs.([eq : spin_op_x]-[eq : spin_op_z ] ) ] . hence , we find @xmath307

a pde system consisting of the momentum balance , mass balance , and energy balance equations for displacement , capillary pressure , and temperature as a model for unsaturated fluid flow in a porous viscoelastoplastic solid is shown to admit a solution under appropriate assumptions on the constitutive behavior . the problem involves two hysteresis operators accounting for plastic and capillary hysteresis .

1601.02435

in a deformable porous solid filled with two immiscible fluids ( water and air , say ) , two sources of hysteresis are observed : the solid itself is subject to irreversible plastic deformations , and the fluid flow exhibits capillary hysteresis which is often explained by the surface tension on the interfaces between the two fluids . a lot of works have been devoted to this phenomenon , see , e.g. , @xcite . mathematical analysis of various mechanical porous media models with capillary hysteresis and without temperature effects has been carried out in @xcite . a pde system for elastoplastic porous media flow with thermal interaction was derived in @xcite , but the existence of solutions was only proved for the isothermal case . here , we focus on the qualitative analysis of the model derived in @xcite , assuming in addition that the heat conductivity depends in a controlled way on the temperature similarly as in the phase transition model considered in @xcite . indeed , we borrow here some techniques employed in @xcite and @xcite in order to prove existence of a weak solution for the initial boundary value problem associated with the pde system coupling the momentum balance ( cf . ) , featuring , in particular , a thermal expansion term depending on the temperature field , with a mass balance ( cf . ) ruling the evolution of the capillary pressure , and an energy balance ( cf . ) displaying , in particular , quadratic dissipative terms on the right hand side . the main mathematical difficulties are related to the low regularity of the temperature field mainly due to the presence of the high order dissipative terms in the internal energy balance . this is the reason why we need to employ here a key - estimate ( cf . ) , already exploited in @xcite and more recently in @xcite for the analysis of non - isothermal phase transition models . roughly speaking , since the test of the internal energy balance by the temperature @xmath0 is not allowed , we test by a suitable negative power of it and use the growth condition of the heat conductivity @xmath1 in hypothesis [ h1 ] ( ii ) . another key point in our proof is the @xmath2 estimate we get on the pressure which entails a bound in a proper negative sobolev space for the time derivative of the absolute temperature , which turns out to be another fundamental ingredient in order to pass to the limit in our approximation scheme . the structure of the paper is as follows . the model from @xcite is briefly summarized in section [ mod ] . in section [ hys ] we recall the definitions and main results of the theory of hysteresis operators that are used here . section [ mai ] contains the mathematical hypotheses and statements of the main results . in section [ reg ] we regularize the problem by adding a small parameter @xmath3 accounting for `` micro - movements '' and a large cut - off parameter @xmath4 to control the nonlinearities , and solve the regularized problem by the standard faedo - galerkin method . in section [ pt1 ] we let @xmath3 tend to @xmath5 and @xmath4 to @xmath6 and prove that in the limit , we obtain a solution to the original problem . consider a domain @xmath7 filled with a deformable solid matrix material with pores containing a mixture of liquid and gas . we state the balance laws in referential ( lagrangian ) coordinates , assume the deformations small , and denote for @xmath8 and time @xmath9 $ ] @xmath10 ... displacement vector of the referential particle @xmath11 at time @xmath12 ; @xmath13 ... linear strain tensor , @xmath14 ; @xmath15 ... stress tensor ; @xmath16 ... capillary pressure ; @xmath17 ... absolute temperature ; @xmath18 ... relative gas content . for the stress @xmath19 and gas content @xmath20 we assume the empirical constitutive relations [ e2 ] & = & _ t + p [ ] + ( p - ( - _ c ) ) , + [ e3 ] a & = & g[p ] , where @xmath21 is a hysteresis operator describing the elastoplastic response of the solid , see subsection [ plas ] , @xmath22 is a constant symmetric positive definite fourth order viscosity tensor , @xmath23 is the relative solid - liquid thermal expansion coefficient , @xmath24 is a fixed referential temperature , @xmath25 is the kronecker tensor , and @xmath26 is a hysteresis operator as on figure [ f3 ] , see subsection [ prei ] . we will see that both hysteresis operators @xmath21 and @xmath26 admit hysteresis potentials @xmath27 ( clockwise ) and @xmath28 ( counterclockwise ) and dissipation operators @xmath29 such that for all absolutely continuous inputs @xmath30 , the inequalities p[]:_t - v_p[]_t = d_p[]_t _ * , g[p]_t p - v_g[p]_t = |d_g[p]_t| hold almost everywhere , where @xmath31 is a seminorm in @xmath32 . we assume the heat conductivity @xmath33 depending on @xmath0 , and as in @xcite , we obtain the system of momentum balance , mass balance , and energy balance equations in the form [ e23 ] _ s u_tt & = & ( u_t + p[u ] ) + p - + g , + [ e24 ] g[p]_t & = & u_t + ( ( p)p ) , + c_0 _ t & = & ( ( ) ) + d_p[u]_t _ * + |d_g[p]_t| + u_t : u_t + ( p)|p|^2 + [ e25 ] & & -u_t , where @xmath34 is a constant specific heat , @xmath35 are the mass densities of the solid and liquid , respectively , @xmath22 is a positive definite viscosity matrix , @xmath36 is the relative thermal expansion coefficient , and @xmath37 is a given volume force ( gravity , e.g. ) . we complement the system with initial conditions u(x,0 ) = u^0(x),u_t(x,0 ) = u^1(x),p(x,0 ) = p^0(x ) , ( x,0 ) = ^0(x ) , and boundary conditions . rcl u & = & 0 + ( p)pn & = & _ p(x)(p^ * - p ) + ( ) n & = & _ ( x)(^ * - ) } , where @xmath38 are given smooth functions . we recall here the basic concepts of the theory of hysteresis operators that are needed in the sequel . in , @xmath21 stands for the elastoplastic part @xmath39 of the stress tensor @xmath19 . we proceed as in @xcite and assume that @xmath39 can be represented as the sum @xmath40 of an elastic component @xmath41 and plastic component @xmath42 . while @xmath41 obeys the classical linear elasticity law ^e = ^ewith a constant symmetric positive definite fourth order elasticity tensor @xmath43 , for the description of the behavior of @xmath42 , we split also the strain tensor @xmath44 into the sum @xmath45 of the elastic strain @xmath46 and plastic strain @xmath47 , and assume = ^p^e again with a constant symmetric positive definite fourth order elasticity tensor @xmath48 , and for a given time evolution @xmath49 of the strain tensor , @xmath50 $ ] , we require @xmath42 to satisfy the constraint ( t ) z t , where @xmath51 is the domain of admissible plastic stress components . we assume that it has the form z = z_0 \ { } , where @xmath52 is the 1d space spanned by the kronecker tensor @xmath25 and @xmath53 is a bounded convex closed subset with @xmath5 in its interior of the orthogonal complement @xmath54 of @xmath52 ( the _ deviatoric space _ ) . the boundary @xmath55 of @xmath56 is the _ yield surface_. the time evolution of @xmath47 is governed by the _ flow rule _ ^p_t : ( - ) 0 z , which implies that ^p_t : = m_z^*(^p_t ) , where @xmath57 is the minkowski functional of the polar set @xmath58 to @xmath56 . the physical interpretation of is the _ maximal dissipation principle_. geometrically , it states that the plastic strain rate @xmath59 points in the outward normal direction to the yield surface at the point @xmath42 . indeed , if @xmath42 is in the interior of @xmath56 , then @xmath60 . it follows from that there exist @xmath61 ( plastic stress deviator ) and @xmath62 ( pressure ) such that @xmath63 . on the other hand , putting in @xmath64 for an arbitrary @xmath65 , we obtain @xmath66 ( in other words , no volume changes occur during plastic deformation ) , so that m_z^*(^p_t ) = ^p_t : _ 0 ( z_0)|^p_t| . we can eliminate the internal variables @xmath67 and write in the form ( _ t - ( ^p)^-1_t ) : ( - ) 0 z. we now define a new scalar product @xmath68 in @xmath32 by the formula @xmath69 for generic tensors @xmath70 , and rewrite as 0 z. we prescribe a canonical initial condition for @xmath42 , namely ( 0 ) = ( ^p(0 ) ) , where @xmath71 is the orthogonal projection @xmath72 with respect to the scalar product @xmath68 and is characterized by the variational inequality x = ( u ) xz , 0 y z. we list here some properties of the variational problem , , . the proof can be found in ( * ? ? ? * chapter i ) . [ p3 ] for every @xmath73 there exists a unique @xmath74 satisfying , , . the solution mapping @xmath75 has the following properties . * for all @xmath76 we have @xmath77_t| \le |\e_t|$ ] a.e . , @xmath78 is strongly continuous , and admits an extension to a strongly continuous mapping @xmath79;\tens)\to c([0,t];\tens)$ ] . * there exists a constant @xmath80 such that for every @xmath81 and every @xmath50 $ ] we have |p_0[_1](t ) - p_0[_2](t)| c(|_1(0 ) - _ 2(0)| + _ 0^t|(_1)_t ( ) - ( _ 2)_t()| ) . * for all @xmath76 , the energy balance equation p_0[]:_t - 12 ( ( ^p)^-1p_0[]:p_0[])_t = m_z^*((- ( ^p)^-1p_0[])_t ) is satisfied almost everywhere in @xmath82 , where @xmath57 is the minkowski functional of the polar set @xmath58 to @xmath56 . it follows from proposition [ p3 ] that the operator @xmath21 in can be represented in the form p [ ] = ^e + p_0 [ ] , and the first energy identity in holds with the choice v_p [ ] = 12 ^e : + 12 ( ^p)^-1p_0[]:p_0 [ ] , d_p [ ] = - ( ^p)^-1p_0 [ ] , _ * = m_z^ * ( ) . similarly as in , the operator @xmath26 is considered as a sum g[p ] = f(p ) + g_0[p ] , where @xmath83 is a monotone function satisfying hypothesis [ h1](iii ) below , and @xmath84 is a preisach operator that we briefly describe here . the construction of the preisach operator @xmath84 is also based on a variational inequality of the type . more precisely , for a given input function @xmath85 and a memory parameter @xmath86 , we define the function @xmath87 as the solution of the variational inequality \ { ll |p(t ) - _ r(t)| r & t , + ( _ r)_t(p(t ) - _ r(t ) - z ) 0 & z , . with a prescribed initial condition @xmath88 $ ] . this is indeed a scalar version of with @xmath56 replaced by the interval @xmath89 $ ] , @xmath44 replaced by @xmath90 , and @xmath42 replaced by @xmath91 . here , we consider the whole continuous family of variational inequalities parameterized by @xmath86 . we introduce the memory state space = \{w^1,(0 , ) : |(r)| 1 } , and its subspace _ k = \ { : ( r ) = 0 r k}. we fix @xmath92 and an initial state @xmath93 , and choose the initial condition as _ r(0 ) = \{p(0 ) - r , \{_-1(r ) , p(0 ) + r}}. we have indeed for all @xmath86 the natural bound |_r(0)| \{|p(0)| , k}. the mapping @xmath94 which with each @xmath95 associates the solution @xmath96 \in w^{1,1}(0,t)$ ] of , is called the _ play_. this concept goes back to @xcite , and the proof of the following statements can be found , e. g. , in ( * ? ? ? * chapter ii ) . [ p5 ] for each @xmath86 , the mapping @xmath94 is lipschitz continuous and admits a lipschitz continuous extension to @xmath97 \to c[0,t]$ ] in the sense that for every @xmath98 $ ] and every @xmath50 $ ] we have |_r[p_1](t ) - _ r[p_2](t)| _ |p_1 ( ) - p_2()| . moreover , for each @xmath85 , the energy balance equation _ r[p]_t p - 12(_r^2[p])_t = |r _ r[p]_t| and the identity _ r[p]_t p_t = ( _ r[p]_t)^2 hold almost everywhere in @xmath82 . [ p6 ] let @xmath99 be given , and let @xmath100 be the family of play operators . then for every @xmath101 $ ] and every @xmath50 $ ] we have * @xmath102(t ) = 0 $ ] for @xmath103}|p(\tau)|\}$ ] ; * the function @xmath104(t)$ ] belongs to @xmath105 . given a nonnegative function @xmath106 ( the _ preisach density _ ) , we define the preisach operator @xmath84 as a mapping that with each @xmath101 $ ] associates the integral g_0[p](t ) = _ 0^_0^_r[p](t ) ( r , v)vr . for our purposes , we adopt the following hypothesis on the preisach density . [ h2 ] there exists a function @xmath107 such that for a. e. @xmath108 we have @xmath109 , and we put c_= _ 0^_-^(r , v)vr , c_^ * = _ 0^^*(r)r . for the reader who is more familiar with the original preisach construction in @xcite based on non - ideal relays , let us just point out that for integrable densities , the variational setting in is equivalent , as shown in @xcite . from , , and we immediately deduce the preisach energy identity g_0[p]_t p - v_0[p]_t = |d_0[p]_t| provided we define the preisach potential @xmath110 and the dissipation operator @xmath111 by the integrals v_0[p](t ) = _ 0^_0^_r[p](t ) v ( r , v)vr , d_0[p](t ) = _ 0^_0^_r[p](t ) r ( r , v)vr . the second identity in then holds with the choice v_g[p ] = p f(p ) - _ 0^p f(z)z + v_0[p ] , d_g[p ] = d_0[p ] . a straightforward computation shows that @xmath84 ( and , consequently , @xmath26 ) are lipschitz continuous in @xmath112 $ ] . indeed , using and hypothesis [ h2 ] , we obtain for @xmath98 $ ] and @xmath50 $ ] that @xmath113 , namely [ inig ] |g_0[p](0)| & = & [ inivg ] |v_0[p](0)| & = & |_0^_0^_r[p](0)v(v , r)vr| c_\{|p(0)| , k}. the preisach operator admits also a family of `` nonlinear '' energies . as a consequence of , we have for a. e. @xmath12 the inequality _ r[p]_t ( p - _ r[p])0 , hence _ r[p]_t ( h(p ) - h(_r[p]))0 for every nondecreasing function @xmath114 . hence , for every absolutely continuous input @xmath90 , a counterpart of in the form g_0[p]_t h(p ) - v_h[p]_t 0 holds with a modified potential v_h[p](t ) = _ 0^_0^_r[p](t ) h(v ) ( r , v)vr . this is related to the fact that for every absolutely continuous nondecreasing function @xmath115 , the mapping @xmath116 is also a preisach operator , see @xcite . we denote x_0 = \{w^1,2 ( ; ^3 ) : | _ = 0 } , x = w^1,2 ( ) , x_q = w^1,q ( ) for @xmath117 , and reformulate problem in variational form for all test functions @xmath118 , @xmath119 , and @xmath120 for a suitable @xmath121 as follows : [ e23v ] ( _ s u_tt+ ( u_t + p[u]):+ ( p - ) ) x & = & gx , + [ e24v ] ( ( g[p]_t - u_t)+ ( p)p ) x & = & _ p(x)(p^ * - p)s(x ) , + ( ( c_0 _ t - d_p[u]_t _ * - |d_g[p]_t| ) + ( ) ) x & & + [ e25v ] - ( u_t : u_t + ( p)|p|^2-u_t)x & = & _ ( x)(^ * - ) s(x ) . [ h1 ] we assume that @xmath122 is a bounded domain with @xmath123 boundary . we fix an arbitrary final time @xmath124 , a constant @xmath125 , and functions @xmath126 , @xmath127 such that @xmath128 , @xmath129 , @xmath130 , @xmath131 , @xmath132 , @xmath133 a. e. , @xmath134 , @xmath135 , @xmath34 . the coefficients @xmath35 are constant and positive , and @xmath22 is the isotropic symmetric positive definite fourth order tensor of the form _ ijkl = 2_ik_jl + _ kl_ij with constants @xmath136 , @xmath137 . the nonlinearities in satisfy the following conditions * @xmath138 $ ] is a @xmath139 function , @xmath140 are fixed constants , and we set m(p ) = _ 0^p ( p)p. * @xmath141 is a @xmath139 function , @xmath142 , and there exist constants @xmath143 such that @xmath144 * @xmath145 = f(p ) + g_0[p]$ ] , where @xmath84 is the preisach operator from subsection [ prei ] with an initial memory state @xmath99 for some @xmath146 . the dissipation operator @xmath147 associated with @xmath26 is defined in , and @xmath148 for some @xmath149 is a @xmath139 function such that there exist @xmath150 with the property f_2 f(p)(1+p^2 ) f_3p . * let @xmath32 denote the space of symmetric @xmath151 tensors . we assume that the operator @xmath152;\tens ) \to c([0,t];\tens)$ ] has the form with @xmath153 defined in proposition [ p3 ] , and with dissipation operator @xmath154 defined in . we prescribe initial conditions with @xmath155 , @xmath156 , @xmath157 , @xmath158 a. e. , @xmath159 , @xmath160 a. e. condition ( ii ) in hypothesis [ h1 ] is a slight generalization of hypothesis ( i ) of @xcite . we will see the role that it plays in the existence proof . the main result of this paper reads as follows . [ t1 ] let hypotheses [ h2 ] , [ h1 ] hold . then there exists @xmath161 depending on @xmath162 and @xmath163 and a solution @xmath164 to , and with the properties @xmath165 , @xmath166 , @xmath167 , @xmath168 with @xmath169 given by , @xmath170 , @xmath171 for every @xmath172 , @xmath173 , @xmath174 for some @xmath121 . we first regularize the problem , prove the existence of a solution for the regularized system , derive estimates independent of the regularization parameters , and pass to the limit . we choose regularizing parameters @xmath175 with @xmath176 from hypothesis [ h2 ] and @xmath177 with the intention to let @xmath178 and @xmath179 , and define mappings @xmath180 $ ] and @xmath181 by the formulas q_r(z ) = \{0 , \{z , r}},k_r(z ) = \{z - r , \{0 , z+r } } z . let @xmath182 denote the mapping v = -v it follows from a vector counterpart of ( * ? ? ? * lemma 9.17 ) , cf . also the methods proposed in ( * ? ? ? * lemma 3.2 , p. 260 ) , that @xmath183 is an equivalent norm for @xmath184 in @xmath185 , that is , there exist positive constants @xmath186 such that for every @xmath187 we have c_1 v_w^2,2 ( ) |v|_2 c_2 v_w^2,2 ( ) . we replace by the system ( _ s u_tt+ u_tt+ ( u_t + p[u]):)x & & + [ e23r ] + ( p -q_r ( ) ) x & = & gx , + [ e24r ] ( ( ( k_r(p)+g[p])_t - u_t)+ ( p)p ) x & = & _ p(x)(p^ * - p)s(x ) , + ( ( c_0 _ t - d_p[u]_t _ * - |d_g[p]_t| ) + ( ) ) x & & + [ e25r ] - ( u_t : u_t + ( p)q_r(|p|^2 ) -q_r ( ) u_t)x & = & _ ( x)(^ * - ) s(x ) with test functions @xmath188 , @xmath189 and initial conditions . [ p4 ] in addition to hypotheses [ h2 ] , [ h1 ] , assume that @xmath190 . then there exists a solution @xmath164 to with the properties @xmath191 , @xmath192 , @xmath168 with @xmath169 given by , @xmath170 , @xmath193 , @xmath173 , @xmath194 , where @xmath195 is the dual of @xmath196 . system for each fixed @xmath197 and @xmath177 will be solved by faedo - galerkin approximations . we choose @xmath198 in @xmath199 and @xmath200 in @xmath201 to be the complete orthonormal systems of eigenfunctions defined by e_k = _ k e_k , e_k| _ = 0 , -w_j = _ j w_j , w_jn| _ = 0 , with @xmath202 , @xmath203 for @xmath204 , and put for @xmath205 u(x , t ) = _ k=1^n u_k(t ) e_k(x ) , ( x , t ) = _ j=0^n _ j(t ) w_j(x ) with coefficients @xmath206 \to \real$ ] , @xmath207 \to \real$ ] which will be determined as the solution of the system ( _ s + _ k + _ k^2)u_k + p[u]:e_k x & & + [ e23fgk ] + ( p - q_r ( ) ) e_kx & = & ge_k x , + ( ( k_r(p)+g[p])_t - u_t)x & & + [ e24fgk ] + ( p)px & = & _ p(x)(p^ * - p)s(x ) , + c_0 _ j + ( - |d_g[p]_t| w_j + ( ) w_j ) x & & + + ( q_r ( ) u_t - d_p[u]_t_*)w_j x & & + [ e25fgk ] - ( u_t : u_t + ( p)q_r(|p|^2 ) ) w_j x & = & _ ( x)(^ * - ) w_j s(x ) for @xmath208 and @xmath209 , and for all @xmath119 . we prescribe initial conditions [ inidu ] u_k(0 ) = u^0(x)e_k(x ) x , & & u_k(0 ) = u^1(x)e_k(x ) x , + [ inidp ] _ j(0 ) = ^0(x ) w_j(x ) x , & & p(x,0 ) = p^0(x ) . this is an ode system , coupled with a standard pde with hysteresis , which has a strong solution in a maximal interval of existence , which coincides with the whole interval @xmath210 $ ] provided we prove that the solution remains bounded in the maximal interval of existence . put @xmath211 and @xmath212 . then can be equivalently written as ( _ s u_tt+ ( u_t + p[u]):)x & & + [ e23fg ] + u_ttx + ( p - q_r ( ) ) x & = & gx , + ( ( k_r(p)+g[p])_t - u_t)x & & + [ e24fg ] + ( p)px & = & _ p(x)(p^ * - p)s(x ) , + ( ( c_0 _ t - |d_g[p]_t| ) + ( ) ) x & & + + ( q_r()u_t - d_p[u]_t_*)x & & + [ e25fg ] - ( u_t : u_t + ( p)q_r(|p|^2))x & = & _ ( x)(^ * - ) s(x ) with test functions @xmath213 , @xmath214 , and @xmath215 . we now derive a series of estimates . by @xmath216 we denote any positive constant depending only on the data , by @xmath217 any constant depending on the data and on @xmath4 , and by @xmath218 any constant depending on the data , on @xmath4 , and on @xmath3 , all independent of the dimension @xmath219 of the galerkin approximation . to simplify the presentation , we introduce from now on the notation @xmath220 for the norm in @xmath221 , and by @xmath222 the norm in @xmath223 . we will systematically use the gagliardo - nirenberg inequality in the form |w|_q c(|w|_s + |w|_s^1- every @xmath224 . for the proof , see , e.g. , @xcite . we test with @xmath225 and with @xmath226 and sum up the results to obtain & & ( _ s u_ttu_t + u_ttu_t + ( u_t + p[u]):u_t + & & + ( k_r(p)+g[p])_t p+ ( p ) |p|^2)x + [ em7 ] & = & ( q_r ( ) u_t + gu_t)x + _ p(x)p(p^ * - p)s(x ) . integrating in time from @xmath5 to @xmath12 and using the energy identities we obtain for all @xmath227 the estimate & & |u_t(t)|_2 ^ 2 + [ em1 ] & & + |p(t)|_2 ^ 2 + p_2 ^ 2 + _ 0^t_p(x ) |p|^2s(x)t c_r . we choose in @xmath228 with @xmath229 given by . by proposition [ p5 ] and formula , we have @xmath230_t m(p\om)_t = g_0[p\om]_t p\om_t \mu(p\om ) \ge 0 $ ] . by hypothesis [ h1](i ) , ( iii ) , we thus have @xmath231_t ) m(p\om)_t \ge \mu_0 \min\left\{1 , \frac{f_2}{1+r^2}\right\ } |p\om_t|^2\ ] ] and we obtain for all @xmath227 the estimate p_t_2 ^ 2 + m(p)_l^2(0,t ; w^2,2 ( ) ) c_r . choosing in @xmath232 yields & & ( _ s |u_tt|^2 + [ em2 ] & = & ( p[u]_t : u_t + ( q_r()-p^(n ) ) u_tt + gu_tt)x . we now integrate in time again and use proposition [ p3 ] , estimate , as well as the gronwall argument , to conclude for all @xmath227 that u_tt_2 ^ 2 + u_tt_2 ^ 2 + |u_t(t)|_2 ^ 2 c_r , . we choose in @xmath233 . the only superlinear term in is @xmath234 which has to be estimated in the norm of @xmath235 , that is , @xmath236 has to be estimated in @xmath237 . this will be done using the gagliardo - nirenberg inequality , which yields for every @xmath238 that |u_t(t)|_4 c(|u_t(t)|_2 + |u_t(t)|_2 ^ 1/4 pointwise inequalities @xmath239_t| \le c |p\om_t|\ , , \quad \|d_p[\nas u\om]_t\| _ * \le c |\nas u\om_t|\ ] ] which follow from , , , , and proposition [ p3](i ) . we thus obtain @xmath238 . finally , let @xmath240 be arbitrary , @xmath241 . we test with @xmath242 and obtain using the previous estimates that _ 0^t_txt c_r,_l^2(0,t ; x ) , or , in other words , _ t_l^2(0,t ; x^ * ) c_r , . we keep for the moment the regularization parameters @xmath4 and @xmath3 fixed , and let @xmath219 tend to @xmath6 . by a standard argument based on compact anisotropic embeddings , see @xcite , we infer , passing to a subsequence , if necessary , that there exist functions @xmath164 such that the following convergences take place : @xmath244;\tens))\,,\\ \nas u\om & \to & \nas u & \mbox { strongly in } & l^4(\omega ; c([0,t];\tens))\,,\\ p[\nas u\om ] & \to & p[\nas u ] & \mbox { strongly in } & l^4(\omega ; c([0,t];\tens))\,,\\ \|d_p[\nas u\om]_t\| _ * & \to & \|d_p[\nas u]_t\| _ * & \mbox { strongly in } & l^2(\omega ; c[0,t])\,,\\ p\om & \to & p & \mbox { strongly in } & l^4(\omega ; c[0,t])\,,\\ p\om_t & \to & p_t & \mbox { weakly in } & l^2(\omega\times ( 0,t))\,,\\ k_r(p\om)_t & \to & k_r(p)_t & \mbox { weakly in } & l^2(\omega\times ( 0,t))\,,\\ g[p\om]_t & \to & g[p]_t & \mbox { weakly in } & l^2(\omega\times ( 0,t))\,,\\ \nabla p\om & \to & \nabla p & \mbox { strongly in } & l^2(\omega\times ( 0,t);\real^3)\,,\\ q_r(|\nabla p\om|^2 ) & \to & q_r(|\nabla p|^2 ) & \mbox { strongly in } & l^2(\omega\times ( 0,t))\,,\\ \gamma_p p\om & \to & \gamma_p p & \mbox { strongly in } & l^2(\partial\omega \times [ 0,t])\,,\\ \theta\om & \to & \theta & \mbox { strongly in } & l^2(\omega\times ( 0,t))\,,\\ \theta\om_t & \to & \theta_t & \mbox { weakly in } & l^2(0,t ; x^*)\,,\\ \nabla\theta\om & \to & \nabla\theta & \mbox { weakly in } & l^2(\omega\times ( 0,t);\real^3)\,,\\ \gamma_\theta \theta\om & \to & \gamma_\theta \theta & \mbox { strongly in } & l^2(\partial\omega \times [ 0,t])\ , . \end{array}\ ] ] the convergences of the hysteresis terms @xmath245 $ ] , @xmath246_t$ ] , @xmath247_t\|_*$ ] , @xmath248_t|$ ] follow indeed from , , , , and . we can therefore let @xmath219 tend to @xmath6 in and conclude that the limit @xmath249 satisfies the conditions of proposition [ p4 ] . in this section , we show that a sequence of solutions to converges to a solution to as @xmath250 and @xmath179 . to this end , we fix sequences @xmath251 for @xmath252 such that _ i r_i = , _ i _ i = 0 , and choose a sequence @xmath253 of initial conditions in @xmath254 such that _ i u^1_i - u^1_x_0 = 0 , _ i _ i u^1_i^2_w^3,2 ( ; ^3 ) = 0 . we further denote by @xmath255 solutions @xmath249 to problem corresponding to the choice @xmath256 , @xmath257 , @xmath258 . the next step is to derive some properties of the sequence @xmath255 independent of @xmath259 . we first observe that there exists a constant @xmath80 such that for every nonnegative test function @xmath260 we have by virtue of that ( c_0 _ t + ( ) ) x - cq_r_i()^2x + _ ( x)(^ * - ) s(x ) . let @xmath261 be the solution of the ode c_0 v(t ) = -cv^2(t ) , v(0 ) = | , that is , v(t ) = ( t + ) ^-1 . for every nonnegative test function @xmath260 we have in particular ( c_0 v_t + ( ) v ) x - cv^2x + _ ( x)(^ * - v)s(x ) . subtracting from we obtain ( c_0 ( v - ) _ t + ( ) ( v - ) ) x c(q_r_i^2 ( ) - v^2)x + _ ( x)(- v)s(x ) . we now choose any smooth convex function @xmath262 such that @xmath263 for @xmath264 , @xmath265 for @xmath266 , and test by @xmath267 . we have in all cases @xmath268 hence c_0 f(v - ) x 0 , and we conclude for every @xmath252 that ( x , t ) v(t ) we now pass to a series of estimates independent of @xmath259 . to simplify the presentation , we occasionally omit the indices @xmath269 in the computations in subsections [ est1][est7 ] below , and write simply @xmath249 instead of @xmath255 whenever there is no risk of confusion . the symbol @xmath216 denotes as before any constant independent of @xmath259 . & & + ( p)(|p|^2 - q_r_i(|p|^2))x + [ ee1 ] & = & gu_tx + ( _ p(x)(p^ * - p ) p + _ ( x)(^ * - ) ) s(x ) , where we set @xmath273 for @xmath274 and @xmath275 . integrating in time and using and , we get for every @xmath238 the estimate ( + we test by @xmath276 with @xmath162 from hypothesis [ h1](ii ) , and observe that & & ^-a(d_p[u]_t _ * + |d_g[p]_t| + u_t : u_t + ( p)q_r_i(|p|^2))x + & & + a ( ) ^-1-a ||^2x + [ e25a ] & = & q_r_i()^-a u_tx - _ ( x ) ^-a(^ * - ) s(x ) + ^1-ax . integrating in time and using , , and hypothesis [ h1](ii ) , we obtain in particular _ 0^t(^-a|u_t|^2 + ||^2)xt c(1 + _ 0^t^1-a |u_t|xt ) . the integral on the right hand side can be estimated by hlder s inequality @xmath277 which entails that _ 0^t||^2xt c(1 + _ 0^t^2-axt ) . applying the gagliardo - nirenberg inequality with @xmath278 , @xmath279 , and @xmath280 and using , we estimate the right hand side of from above by @xmath281 . thus , for @xmath282 , inequality has the form @xmath283 with @xmath284 , and @xmath285 . choosing , for example , @xmath286 , we obtain _ 2 c. using again with @xmath278 , @xmath279 , and @xmath287 , we obtain _ 8/3 c. & & + ( ( u_t + p[u]):u_t + g[p]_t p + ( p ) |p|^2)x + [ ee7 ] & & = ( q_r_i()u_t + gu_t)x + _ p(x)p ( p^ * - p)s(x ) . by and , we have @xmath145_t p \ge v_g[p]_t$ ] , @xmath288(x , t ) \ge 0 $ ] , and @xmath288(x,0 ) \le c|p(x,0)|^2 \le c$ ] a. e. integrating in time , taking into account and the previous estimates , we get u_t_2 + p_2 + _ i|u_t(t)|_2 + _ 0^t_p(x)|p|^2s(x ) c for all @xmath238 . consequently , as @xmath289 does not identically vanish on @xmath290 by hypothesis [ h1 ] , we also have p_l^2(0,t;w^1,2 ( ) ) c. test by @xmath291 . then & & ( _ s |u_tt|^2 + _ i |u_tt|^2)x + 12 ( u_t + 2p[u]):u_t x + [ ee9 ] & & c(|g| + |p| + ||)|u_tt|x + p[u]_t : u_t x. from proposition [ p3 ] , , korn s inequality , and it follows that u_tt_2 ^ 2 + _ i u_tt_2 ^ 2 + we rewrite in the form ( ( _ s u_tt+ u_t)+ _ iu_tt)x = ( f+h)x for all @xmath188 , where f = g - q_r ( ) + p , h = p[u ] . we have @xmath292 by , , and hypothesis [ h1 ] . to estimate @xmath293 in @xmath294 , we use and proceed as follows . let @xmath295 , @xmath296 , be the @xmath297-th coordinate vector , let @xmath298 be an arbitrary lebesgue point of @xmath299 $ ] , and let @xmath300 be sufficiently small such that @xmath301 for @xmath302 . by and we have & & |p[u](x+se_l ) - p[u](x , t)| c(|u^0(x+se_l ) - u^0(x)| + [ ep5 ] & & + _ 0^t |u_t(x+se_l , ) - u_t(x,)| ) , so that in the limit as @xmath303 we have |p[u](x , t)| c(|u^0(x)| + _ 0^t |u_t(x,)| ) and |h(t)|_2 c(1 + _ 0^t|u_t()| ) . consider now the fourier expansion of @xmath304 in the form u(x , t ) = _ k=1^u_k(t ) e_k(x ) similar to , with coefficients u_k(t ) = u(t)e_k(x ) x. it follows e.g. from that the series u_t(x , t ) = _ k=1^_k u_k(t ) e_k(x ) is strongly convergent in @xmath305 . we now test by @xmath306 , where @xmath307 is as in with coefficients @xmath308 given by . then & & ( u_t : u_t + u_t : u_t)x + |u_t(t)|_2 ^ 2 + [ ep7 ] & & c(1 + |f(t)|_2 + _ 0^t|u_t()|_2)|u_t(t)|_2 , hence , & & ( u_t : u_t + u_t : u_t)x + |u_t(t)|_2 ^ 2 + [ ep7a ] & & c(1 + |f(t)|^2_2 + _ 0^t|u_t()|^2_2 ) . by , we can integrate this inequality from @xmath5 to @xmath12 , pass to the limit as @xmath243 , and use gronwall s argument to obtain in particular that u_t_2= u_t_2 c. the next computation based on is to check that u_t_l^r(0,t ; c(| ; ^3 ) ) c r[1,4 ) . indeed , we choose any @xmath309 and put @xmath310 . by , , and we have |_x_j u_t(t)|_q c(|_x_j u_t(t)|_2 + |_x_j u_t(t)|_2 ^ 1- |_x_j u_t(t)|_2^),= . then @xmath311 is bounded in @xmath312 for @xmath313 , that is , |_x_j u_t|_l^p(0,t;l^q ( ; ^3 ) ) c q = , p = . by , @xmath314 is bounded in @xmath315 , and by for @xmath316 we have |u_t(t)|_c(|u_t(t)|_2 + |u_t(t)|_2 ^ 1- . we have @xmath325 by virtue of , hence this is an admissible test function . by we have @xmath326_t h_{{q},m}(p ) \ge v_{h_{{q},m}}[p]_t$ ] and @xmath327 , with @xmath328 . we thus have & & ( k_r , q , m(p)+ v_h_q , m[p])x + f(p)h_q , m(p ) p_tx + _ 0 ( 2m+1 ) |p|^2 \{|p| , q}^2mx + [ m2 ] & & -(2m+1)u_t \{|p|,q}^2mp x + ( x ) ( p^ * - p)h_q , m(p)s(x ) , together with @xmath329 by hypothesis [ h2 ] , and v_h_q , m[p](x,0)x ck^2 m as a consequence of and , with @xmath216 independent of @xmath330 and @xmath321 . we estimate the right hand side of as follows & & -(2m+1)u_t \{|p|,q}^2mp x + & & ( 2m+1)u(t)(\{|p|,q}^2mx)^1/2 ( |p|^2 \{|p|,q}^2mx)^1/2 + & & ( 2m+1)|p|^2 \{|p|,q}^2mx + u^2(t ) \{|p|,q}^2mx . for the boundary term , we have @xmath331 on the left hand side of , we have @xmath332 where we set @xmath333 . we claim that for every @xmath334 we have |p|^2 m f_q , m(p ) ( \{|p|,q}^2 m - 1 ) . the upper bound is easy . we have for @xmath335 that @xmath336 and similarly for @xmath337 , and it suffices to integrate . to get the lower bound , set @xmath338 then for @xmath339 we have @xmath340 . for @xmath341 we have @xmath342 hence the minimum of @xmath343 is attained at @xmath344 , with @xmath345 , which is exactly . the case @xmath346 is symmetric . summarizing the above estimates , we obtain by integrating from @xmath5 to @xmath12 that & & \{|p|,q}^2m(x , t)x + 2m(2m+1)_0^t |p|^2 \{|p|,q}^2mx + & & + ( x)\{|p|,q}^2m+2s(x ) + [ m4 ] & & c(1+m^2 ) _ 0^t u^2()(k^2 m + \{|p|,q}^2m(x,)x ) . in particular , the function @xmath347 satisfies the inequality @xmath348 and by gronwall s argument ( note that @xmath349 by ) , there exists a constant @xmath350 depending on @xmath321 and independent of @xmath330 such that @xmath351 } w_m(t ) \le c(m)$ ] . hence , we can let @xmath330 tend to @xmath6 in and obtain & & |p|^2m(x , t)x + 2m(2m+1)_0^t |p|^2 |p|^2mx + & & + ( x)|p|^2m+2s(x ) + [ m4a ] & & c(1+m^2 ) _ 0^t u^2()(k^2 m + |p|^2m(x,)x ) . in particular , p l^(0,t ; l^q())q 1 , but the norm of @xmath90 in this space still depends on @xmath352 . note that _ 0^t |p|^2 |p|^2mx & & _ 0^t |p|^2 set @xmath356 for @xmath319 . then reads |w(t)|_2 ^ 2 + c _ 0^t|w()|_2 ^ 2c(1+m^2 ) _ 0^t u^2()(k^2 m + |w()|_2 ^ 2 ) . we now choose @xmath357 , and invoke the gagliardo - nirenberg inequality in the form @xmath358 hlder s inequality enables us to rewrite as & & |w(t)|_2 ^ 2 + c _ 0^t|w()|_2 ^ 2 + & & c(1+m^2 ) _ 0^t u^2()(k^2 m + |w()|_2s^2)+ c(1+m^2)_0^t u^2()|w()|_2s^4/3|w()|_2 ^ 2/3 + & & c(1+m^2 ) _ 0^t u^2()(k^2 m + |w()|_2s^2 ) + & & + c(1+m^2)(_0^t u^3()|w()|_2s^2)^2/3 ( _ 0^t|w()|_2 ^ 2)^1/3 , and we conclude that |w(t)|_2 ^ 2 c(1+m^3 ) _ 0^t u^3()(k^2 m + |w()|_2s^2 ) , or , in terms of @xmath90 , |p(t)|_2m^2 m c(1+m^3 ) _ 0^t u^3()(k^2 m + @xmath361 , and put @xmath362 by we have for all @xmath363 that v_k ( c(1 + ( 43)^3k))^((3/4)^k)/2 v_k-1 , hence v_k - v_k-1 12(34)^k ( c(1 + ( 43)^3k ) ) . the right hand side of is a convergent series , and @xmath364 is finite by virtue of , so that we can conclude that the sequence @xmath365 is uniformly bounded independently of @xmath259 , which we wanted to prove . test by @xmath366 , with @xmath169 given by . from and proposition [ p2 ] , it follows that there exists a constant @xmath320 such that for every @xmath238 we have & & cp_t^2(x , t ) x + |m(p)|^2(x , t ) x + _ p(x ) m(p)(x , t)s(x ) + & & ( |u_t| |m(p)_t|)(x , t)x + _ p(x ) ( m(p ) p^*)(x , t)s(x ) + [ e61 ] & & - _ p(x ) ( m(p ) p^*_t)(x , t)s(x ) with @xmath367 . hence , using also , we have for all @xmath227 that p_t_2 ^ 2 + |p(t)|_2 ^ 2 + _ p(x ) |p|^2(x , t)s(x ) c. by comparison in eq . , we also obtain that m(p)_l^2(0,t ; w^2,2 ( ) ) c. by the same argument based on , we have similarly as in that p_l^p(0,t;l^q ( ) ) c q = , p = for every @xmath309 . in particular , for @xmath368 , we have @xmath369 . hence , eq . has the form ( c_0 _ t + ( ) ) x + _ ( x)(- ^*)s(x ) = ( a(x , t ) + b(x , t ) ) x with @xmath370 , @xmath371 bounded independently of @xmath259 . we test with @xmath372 ( with @xmath373 to be chosen later ) and obtain using hlder s inequality for every @xmath227 that [ e71a ] & & ^r+1(x , t)x + r_0^t^r-1()||^2(x,)x+ _ 0^t_(x)^r+1(x,)s(x ) + & & c(1 + _ 5r/2^r + _ 10(r+1)/7^r+1 ) with , by hypothesis [ h1]((ii ) , ^r-1()||^2^r+a||^2 . we already have the estimate . assume that for some @xmath374 we have proved _ z c_0(z ) with some @xmath375 . for this value of @xmath376 we choose in r = r(z ) = \ { . then @xmath377 , and we have by virtue of that ^r+1(x , t)x + r_0^t^r+a||^2(x,)xc(1 + c_0(z)^r+1 ) . set p = + 1 , s = , w = ^p . then can be written as w^s(x , t)x + r_0^t|w|^2(x,)xc(1 + c_0(z)^r+1 ) . for @xmath378 we have by virtue of the gagliardo - nirenberg inequality that |w()|_q c(|w()|_s + |w()|_s^1- |w()|_2^ ) , = . if @xmath352 is chosen in such a way that @xmath379 , that is , q = 23 s + 2 , then it follows from and young s inequality that w_q c(_|w()|_s + _ |w()|_s^1- w_2 ^ 2/q ) c(_|w()|_s + w_2 ) . by , we have _ |w()|_s & & c((1+r)(1 + c_0(z)^r+1))^1/s , + w_2 & & c(1 + c_0(z)^r+1)^1/2 c(1 + c_0(z)^r+1)^1/s ( note that @xmath380 ) , so that w_q c((1+r)(1 + c_0(z)^r+1))^1/s . putting @xmath381 we have by that _ z = w_q^1/p c((1+r)(1 + c_0(z)^r+1))^1/(ps ) , and using the identity @xmath382 , we obtain the implication _ z c_0(z ) _ z c_0(z),c_0(z ) = c(1 + c_0(z ) ) . the sequence @xmath383 converges to @xmath384 . after finitely many iterations we obtain _ z c z < 8 + 3a . consequently , by , we have for @xmath385 and @xmath227 that ^r+1(x , t)x + r_0^t^r-1()||^2(x,)x+ _ 0^t_^r+1(x,)s(x)c . this enables us to derive an upper bound for the integral @xmath386 which we need for getting an estimate for @xmath387 from the equation . we have by hlder s inequality and hypothesis [ h1](ii ) that | ( ) | x & = & |^(1-r)/2 ^ 1/2 ( ) ^(r-1)/2 ^ 1/2()| x + [ e715 ] & & c(^r-1 ( ) ||^2x)^1/2 ( ^2+b - r||^2x)^1/2 . we now want to choose @xmath388 and @xmath373 satisfying such that 2+b - r = , or , equivalently , r = = + . if @xmath163 satisfies the condition in hypothesis [ h1](ii ) , then there exists @xmath389 such that for @xmath373 given by the condition in holds . hence , by , | ( ) | x c(^r-1 ( ) ||^2x)^1/2 ( ^r+1 x)^1/(1+)(||^q^*x)^1/q^ * with @xmath390 . hence , by , _ 0^t| ( ) | x t c _ l^2(0,t;w^1,q^ * ( ) ) . by and for @xmath391 , the functions @xmath20 and @xmath392 in satisfy uniform bounds @xmath393 , @xmath394 independent of @xmath259 , so that testing with @xmath395 is admissible . we thus obtain from that _ 0^t_t x t c _ l^2(0,t;w^1,q^ * ( ) ) . in the system with @xmath257 , @xmath256 , and @xmath397 , we fix test functions @xmath398 , @xmath119 , and @xmath395 with @xmath161 from . the term @xmath399 in converges to @xmath5 in @xmath294 by , the regularization @xmath400 vanishes for @xmath401 by proposition [ p2 ] . by and , the sequence @xmath402 is precompact in @xmath403 for every @xmath404 . similarly , by , @xmath405 is precompact in @xmath406 for every @xmath404 , and @xmath407 is precompact in @xmath408)$ ] for every @xmath117 by virtue of proposition [ p2 ] . finally , by , , and , @xmath409 is precompact e.g. in @xmath410 . hence , using also , we select a subsequence and pass to the weak limit in the linear terms in , and to the strong limit in all nonlinear non - hysteretic terms . obviously , if @xmath411 converge strongly to @xmath0 in @xmath410 , then @xmath412 strongly in @xmath410 as well , and if @xmath413 in @xmath414 strongly , then @xmath415 in @xmath414 strongly . by the same argument as in subsection [ ninf ] , we show that the hysteresis terms @xmath416_t$ ] , @xmath417_t\|_*$ ] , @xmath418_t|$ ] converge weakly in @xmath235 , and that the limit as @xmath396 yields a solution to with @xmath419 . by density we conclude that @xmath118 is an admissible test function , which completes the proof of theorem [ t1 ] . albers b. on the influence of the hysteretic behavior of the capillary pressure on the wave propagation in partially saturated soils . submitted to _ journal of physics : conference series , proceedings of the 7th international workshop on multi - rate processes and hysteresis_. 2014 . albers b. _ modeling and numerical analysis of wave propagation in saturated and partially saturated porous media_. habilitation thesis . verffentlichungen des grundbauinstitutes der technischen universitt berlin , shaker : aachen ; 2010 . besov ov , ilin vp , nikolski sm . _ integral representations of functions and imbedding theorems_. scripta series in mathematics . halsted press ( john wiley & sons ) : new york - toronto , ont .- london ; 1978 ( vol . i ) , 1979 ( vol . russian version nauka : moscow ; 1975 . flynn d , mcnamara h , okane jp , pokrovski av . application of the preisach model to soil - moisture hysteresis . _ the science of hysteresis , volume 3 _ , eds . bertotti g , mayergoyz i. academic press : oxford ; 2006 : 689744 . haverkamp r , reggiani p , ross pj , parlange j - y . soil water hysteresis prediction model based on theory and geometric scaling . _ environmental mechanics , water , mass and engergy transfer in the biosphere_. eds . raats pac , smiles d , warrick aw . american geophysical union : 2002 ; 213246 .

the reactionsi +ta has been studied at e=125 and 135 mev . coincidences between high energy particles and various x- and-rays from abnormally long - lived states were observed . e.g. 7.8 - 8.6 mev-particles with-rays of a superdeformed band , 5.1 - 5.5 mev-particles with x- and-rays of w , re , and pt , and 3.88 mev particles ( interpreted as protons ) with 185.8 kev-rays . the data are interpreted in terms of the production of long - lived ( t of several months ) high spin isomeric states in the second well of the potential in the parent nuclei , which decay to the normal states in the daughters , and in the third well of the potential , which decay to the second well . pacs numbers : 23.60.+e , 21.10.re , 21.10.tg , 27.80.+w

nucl-ex0006012

the first evidence for a new kind of long - lived isomeric states was obtained in actinide fractions @xcite produced via secondary reactions in a cern w target which had been irradiated with 24-gev protons @xcite . isomeric states with t@xmath5 @xmath6 0.6 y and @xmath7 30 d ( 10@xmath8 - 10@xmath9 times longer than the expected half - lives of the corresponding ground states ) were found in neutron - deficient @xmath10am and @xmath10bk nuclei , respectively . about 3x10@xmath9 atoms of @xmath10am and 4x10@xmath8 atoms of @xmath10bk were produced in the isomeric states , and decayed by the @xmath11 or electron capture processes . the character of these states was not clear : they are far from closed shells where high spin isomers are usually found , and they have very long lifetimes as compared to the known shape isomers . in addition @xcite , some new particle groups were found in the decay of various actinide fractions separated from the same w target . for example , an unambiguous 5.14 mev @xmath4-particle group with t@xmath5 @xmath6 4 y was seen in the bk source @xcite . this half - life is a factor of 2x10@xmath12 3x10@xmath13 too short @xcite for the normal @xmath4-decay of bk pu nuclei . it was very difficult to understand such an enhancement . furthermore , 3.0 and 4.0 mev particle groups were seen from the am source in coincidence with l@xmath14 x - rays in the am region @xcite . since the relationship between the particle energies and their lifetimes deviate by about 23 and 12 orders @xcite from the systematics @xcite of @xmath4-particles , it was assumed @xcite that they were protons of unknown origin . the clue to the understanding of the above mentioned findings has been obtained recently in several studies @xcite of the @xmath15o + @xmath16au reaction at e@xmath2 = 80 mev , where similar isomeric states and particle decays were found . an isomeric state which decays by emitting a 5.20 mev @xmath4-particle with t@xmath5 @xmath6 90 m has been found in @xmath17fr . since this half - life is longer than the known half - life of the ground state of @xmath17fr , it was concluded that a long - lived isomeric state had been formed in this nucleus . a t@xmath5 of 90 m for 5.20 mev @xmath4-particles in @xmath17fr is enhanced by a factor of 3x10@xmath9 as compared to normal transitions @xcite . however , this group was observed in coincidence with @xmath3-rays which fit predictions for a super - deformed band @xcite . therefore the effect of large deformations of the nucleus on the @xmath4-particle decay was calculated and found @xcite to be consistent with the observed enhancement . it was argued @xcite that since the isomeric state decays to a high spin state , it should also have high spin , and since it decays by enhanced @xmath4-particle emission to state(s ) in the second well of the potential , it should be in the second well itself . the predicted @xcite excitation energies of the second minima in the evaporation residue nuclei and their daughters produced by the @xmath15o+@xmath16au reaction are above the proton separation energies . therefore a search for long - lived proton decays has been performed @xcite using the same reaction . two long - lived proton activities with half - lives of about 6 h and 70 h were found @xcite with proton energies of 1.5 4.8 mev with a sharp line at 2.19 mev . this energy may correspond to a predicted transition @xcite from the second minimum in @xmath18tl to the ground state of @xmath16hg with e@xmath19 = 2.15 mev . the superdeformed state in @xmath18tl may be produced in an alpha decay chain of transitions from shape isomeric states to shape isomeric states starting from @xmath17fr . it should be noted that a long - lived high - spin isomeric state in the second minimum of the potential in @xmath20pu has been predicted by s. g. nilsson et al . @xcite back in 1969 . the present work reports on the discovery of new long - lived isomeric states , produced by the @xmath0si + @xmath1ta reaction at bombarding energies of 125 and 135 mev , which decay by strongly hindered @xmath4-particle and proton transitions . the lower energy is about @xmath21 below the coulomb barrier . a fusion cross section of about 10 mb is predicted for this energy using a coupled - channel deformation code @xcite with deformation parameters @xmath22=0.41 for @xmath0si and @xmath22=0.26 for @xmath1ta @xcite , value for @xmath1ta was taken as the average from the corresponding values of @xmath23hf and @xmath24w . ] and allowing for 2@xmath25 and 3@xmath26 excitations in @xmath0si . only 2 @xmath27b is predicted when no deformations are included in the calculations . for 135 mev the corresponding predicted fusion cross sections are 95 mb with deformations and 40 mb without . at bombarding energies of 125 and 135 mev , the compound nucleus is formed at excitation energies of 42.1 and 50.8 mev , respectively . preliminary results of this work have been published before @xcite . in three irradiations , 2.5 mg/@xmath28 ta targets followed by stacks of c catcher foils were bombarded with a @xmath0si beam obtained from the pelletron accelerator in rehovot . irradiations i and ii were performed with a 125 mev @xmath0si beam and irradiation iii with 135 mev . carbon catcher foils of 200 @xmath27g/@xmath28 and 60 @xmath27g/@xmath28 were used in order to catch the evaporation residue nuclei and their daughters . in all experiments the average beam intensity was about 11 pna and the total dose about 1x10@xmath15 particles . long period off - line measurements were carried out in the laboratory in jerusalem , using the irradiated catcher foils as sources . in the present paper we summarize the results of the particle-@xmath3 coincidence measurements.-@xmath4 correlation measurements were performed with 60 @xmath27g/@xmath28 c catcher foils situated in between two 300 @xmath27g/@xmath28 si detectors , but no @xmath4 correlation events with @xmath29t@xmath30 10 s could be significantly established . the preliminarily claimed correlations @xcite can not presently be ruled out to be due to other physical or electronic background effects . at longer correlation times the number of random events was too large . ] a 450 mm@xmath31 , 300 @xmath27 m thick , si surface barrier detector and a 500 mm@xmath31 , 10 mm thick , thin window ge(li ) detector were used for these measurements . the si detector was calibrated using the 3.18 mev @xmath4-particle group of a @xmath32gd source and an accurate pulse generator . the ge(li ) detector was calibrated using the x- and @xmath3-rays of a @xmath33co source . the source was sandwiched between the si and the ge detector . two 0.01 inch thick be foils separated the source from the ge detector . the first foil was the window for the vacuum chamber which included the source and the si detector , and the second one the window of the ge detector . the transparency of the two be windows together was close to 100% for gamma - rays with e@xmath34 @xmath7 10 kev . the solid angle of the si detector was about @xmath35 of 4@xmath36 sr and of the ge(li ) detector @xmath37 . the peak - to - total ratio of the ge(li ) detector was 100% up to about 120 kev and decreased gradually to 22% at 250 kev . its fwhm energy resolution was about 900 ev . the intrinsic resolution of the si detector was about 25 kev . the full line widths for @xmath4-particles of 5.0 to 8.6 mev were 0.52 to 0.34 mev , respectively , with a 200 @xmath27g/@xmath28 c foil , and the corresponding values for the 60 @xmath27g/@xmath28 foil were 0.16 to 0.10 mev . the resolving time of the coincidence system was 1 @xmath27s . the coincidence events together with the singles particle events were recorded event by event with a time accuracy of 1 ms . 1 and 2 show typical singles particle and @xmath3-ray spectra obtained at 125 mev bombarding energy . figs . 3 and 4 show similar spectra obtained at 135 mev beam energy . from the measured @xmath4-particle and @xmath3-ray energies and lifetimes , production cross sections for various evaporation residue nuclei , with estimated errors of @xmath3825% , were deduced and are summarized in table 1 . ( cross sections for some isomeric states ( see below ) , are also given in table 1 . ) figs . 5 and 6 give respective @xmath4-@xmath3 coincidence plots from measurements i and ii , corresponding to irradiations i and ii . . 7 and 8 are one dimensional projection plots on the @xmath4-particle axis of figs . 5 and 6 , respectively , for e@xmath3 @xmath7 25 kev . 9 presents a particle-@xmath3 coincidence spectrum from measurement iii , corresponding to irradiation iii.e - e system of ref . @xcite was performed and gave negative results with an upper limit of about 0.5 nb for half - lives of between 20 hours to 70 days . on the other hand , in the 135 mev experiment ( fig . 9 ) we most probably saw protons in the particle-@xmath3 coincidences . ( see further below ) . ] the estimated numbers of random coincidences in the 3 10 mev particle energy and 10 250 kev @xmath3-energy are 5.4 x 10@xmath39 , 4.4 x 10@xmath39 , and 0.4 in figs . 5 , 6 , and 9 , respectively . some of the coincidence events in figs . 5 , 6 and 9 , between 5 - 6 mev @xmath4-particles and various @xmath3-rays , may perhaps be due to a contamination from an emanated @xmath40pb source from @xmath41th , which was used in the same chamber about 4 months earlier @xcite . despite the long time delay to our present long term measurements of 77 to 235 days , and careful cleaning of the chamber , we might have picked up some residual contamination . also , indication for k x - rays of rn is seen in this region . while in principle their origin may be from some reaction products , they also may be from the decay of @xmath42ra , which belongs to the @xmath43u chain . although it is not likely , but because of this ambiguity , we will not try to make a claim about this region . a two - dimensional background measurement , taken for 8 days , before the @xmath40pb source was used in the chamber , gave zero events in the whole region of 3.5 10 mev particles and 0 250 kev @xmath3 s . it is seen in figs . 5 and 6 , and figs . 7 and 8 , that while the region of @xmath4-particles of 6 8 mev is relatively empty , coincidence events are seen between various @xmath3-rays and @xmath4-particles of 8 9 mev . in fig . 5 , 13 coincidence events between 7.99 - 8.61 mev @xmath4-particles and various @xmath3-rays of e @xmath7 20 kev are observed , where 8 of them fall within a narrow range of 190 kev particle energy , between 8.42 - 8.61 mev . ( the estimated @xmath4-particle full line width for the 200 @xmath27g/@xmath44 c foil is around 340 kev ) . from the intensities of the singles in the @xmath4- and @xmath3-ray spectra ( 339 @xmath4-particles and 2.14x10@xmath45 gamma - rays in 76.8 days ) , and the resolving time of 1 @xmath27s of the coincidence system , the total number of random coincidences in the 8 - 9 mev region was estimated to be 2.2x10@xmath46 . in fig . 6 one finds 8 coincidence events between 8.19 9.01 mev , where 5 of them fall within 70 kev , between 8.55 8.62 mev , and one at 9.01 mev . ( the @xmath4-particle full line width for a 60 @xmath27g/@xmath44 c foil is about 100 kev ) . the corresponding estimated total number of random coincidences between 8 to 9 mev is 1.4x10@xmath46 . one does not see such concentration of events in the same energy region in fig . 9 which was taken for about 2.7 times longer period than figs . 5 and 6 . 3 events are seen in the 8 - 9 mev region in fig . 9 which sets an upper limit on the background in figs . 5 and 6 to be aroud 1 count as compared to the respective 13 and 8 counts seen experimentally . it was estimated that in the 8 9 mev range of @xmath4-particles , at most 0.8 events in fig . 5 and 0.3 events in fig . 6 may be due to the contamination mentioned above of coincidences between @xmath47 and @xmath3-rays from @xmath40bi and the 8.78 mev @xmath4-particles from @xmath40po ( t@xmath5 = 0.3 @xmath27s ) . shows time sequence plots for the @xmath4-@xmath3 coincidences with @xmath4-energies of 89 mev obtained in measurements i and ii and seen in figs . 5 and 6 , respectively . in measurement ii a growth in the intensity of the 89 mev group was found from the beginning up to 97 days after the end of irradiation . in measurement i no significant change in decay rate was observed between 77 to 154 days and , by binning the data , t@xmath5 @xmath7 40 days was estimated for the lower limit of the decaying half - life . an additional measurement , taken about 4.5 y after the one presented in fig . 5 and measurement i of fig . 10 , gave 0 counts in 22.0 d. an upper limit for the half - life of t@xmath5 @xmath30 2.1 y is deduced from this measurement . table 2(a ) and fig . 11 show that almost all the coincidence events in figs . 5 and 6 with 7.8 - 8.61 mev @xmath4-particles ( the encircled ones ) fit very nicely with a j(j+1 ) law assuming e@xmath48 = 4.42xj(j+1 ) kev and @xmath29j = 1 . this fit is very significant from a statistical point of view @xcite . all the gamma - rays in fig . 5 ( 7 out of 7 events ) , and almost all of them in fig . 6 ( 6 out of 7 events ) , which are above 90 kev , fit , to within @xmath380.5 kev on the average , with this formula . the probability for 13 out of 14 events which are distributed evenly ( for instance due to compton effect from higher energy gamma - rays ) , to fall into 13 specified energy positions , within 1 kev is : @xmath49 . p is equal to 36x1.0/(250 - 90 ) . 36 is the number of possible gamma - transitions between states of both integer and half integer spins with e@xmath48 = 4.42 x j(j+1 ) , in the range of 90 to 250 kev . ( the four low energy events of e@xmath50 @xmath30 60 kev in fig . 5 and the one event in fig . 6 which do not fit with the j(j+1 ) rule may be due to compton events ) . it is also seen in fig . 6 and in table 2(b ) that the @xmath3-rays of the 3 coincidence events in the region of 6 - 7 mev @xmath4-energy of 6.16 mev - 176.1 kev , 6.94 mev - 207.4 kev and 6.41 mev - 242.3 kev , fit with 20 @xmath51 19 ( 176.4 ) , 47/2 @xmath51 45/2 ( 207.3 ) and 55/2 @xmath51 53/2 ( 242.6 ) j @xmath51 j-1 transitions , assuming e@xmath48 = 4.41xj(j+1 ) kev . the probability in this case that this fit is accidental due to a chance coincidence of evenly distributed events is 1.2x10@xmath46 , taking into account that p = 18x0.47/(250 - 170 ) . 18 is the number of possible gamma - transitions in the range of 170 to 250 kev , and 0.47 kev is twice the average deviation from the specified energies . ( this probability increases to 1.1x10@xmath39 if , instead of 0.47 kev , the value which was used before of 1.0 kev is assumed in the calculation ) . it should be mentioned that rotational constants around 4.4 kev correspond to superdeformed bands [ sdb ] in this region of nuclei @xcite . in fig . 9 one sees about 10 scattered events in the 6.5 9 mev region . some of them may also fit with a j(j+1 ) law . however , the statistical significance of the fit in this case is not so good and we will not discuss them further . the appearance in table 2a of both integer and half - integer spins at low @xmath3-ray energies indicates that bands in both even and odd nuclei were formed . ( at high energies the observed transitions are probably sum events . see the following paragraphs . ) @xmath52rn , produced via pn evaporation , and its daughters , are , from the kinematic point of view , the best candidates for the half - integer transitions . the 8.42 mev @xmath4-particle in coincidence with a photon of 67.1 kev ( seen in fig . 5 and table 3 below ) , which fits with k@xmath14 x - ray of pt ( known energy 66.832 kev ) , may indicate that the @xmath4-transition is from @xmath53hg to @xmath54pt . @xmath53hg is the daughter after 3 @xmath4-particle decays from @xmath52rn . it should be mentioned that the high energy @xmath3-rays ( above about 120 kev , table 2 ) are most probably sum events rather than photo - peak events . if , for instance , the three coincidence events in fig . 6 at e@xmath34 of 247.0 ( 2 events ) and 225.3 kev are photo - peak events , then [email protected] compton events below 120 kev ( the position of the compton edge ) should have been seen in coincidence with @xmath4-particles of 8.55 - 8.61 mev , but only one event is seen . similarly if the @xmath3-rays which are in coincidence with 6 - 7 mev @xmath4-particles , seen in the same figure and mentioned above , are photo - peak events , then 7.6@xmath55 compton events should have been seen in this @xmath4-energy range at e@xmath50 @xmath30 120 kev , while only one event is seen . on the other hand , for a de - excitation of a superdeformed band with @xmath29j = 1 , a measured energy which fits a transition between high - spins , may be due to many sum combinations of lower energy transitions between lower spins , for which low - energy compton events should not be observed . for instance , an energy of 225.4 kev which fits a 51/2 @xmath51 49/2 transition ( assuming e@xmath48 = 4.42@xmath56j ( j+1 ) kev , table 2 ) may be due to one of 17 different combinations of 3-fold events , if only the first 14 levels of the band up to maximum spin of 27/2 , and maximum transition energy of 119.3 kev are considered , for which the photo - peak to total ratio is 100% . ( 119.3 ( 27/2 @xmath51 25/2 ) + 92.8 ( 21/2 @xmath51 19/2 ) + 13.3 ( 3/2 @xmath51 1/2 ) = 225.4 , or 84.0 ( 19/2 @xmath51 17/2 ) + 75.1 ( 17/2 @xmath51 15/2 ) + 66.3 ( 15/2 @xmath51 13/2 ) = 225.4 are two examples out of these 17 possible combinations ) . because of the large solid angle of the @xmath3-detector ( 0.16 ) and the large number of possible 3-fold combinations , the probability p@xmath57 to see a 3-fold event is larger than the probability p@xmath58 to see a 1-fold event . ( these probabilities for the above example are : p@xmath58 = @xmath59(1-@xmath59)@xmath60 = 0.042(1 - 0.16)@xmath61 = 6.4 x 10@xmath62 , and p@xmath57 = n@xmath57@xmath59@xmath13(1-@xmath59)@xmath63 = 17x0.16@xmath13(1 - 0.16)@xmath64 = 1.2x10@xmath39 , about 20 times larger . @xmath59 is the photo - peak efficiency of the detector for the particular energy , @xmath59 is the geometrical efficiency , n is the number of transitions in the band counting from the level to which the @xmath4-particle is decaying , and n@xmath57 is the number of 3-fold combinations ) . many combinations of 2- or 4- fold events of a half - integer spin band will give energies which fit with those of a band of an integer spin . for instance , the two events at 247.0 kev ( table 2 ) which fit in energy to a 28 @xmath51 27 transition , and the 186.5 kev event which fits with a 21 @xmath51 20 transition , may be due to such combinations , since the @xmath4-particle energy is the same as that of the half - integer transitions of 225.3 and 171.8 kev , and therefore they presumably belong to the same band . ( in an integer - spin band of @xmath29j = 1 various combinations of 2- , 3- and 4-fold events of low energy transitions have large efficiencies and give energies which fit with those of transitions between higher spins . this property , both for half - integer and integer spin bands , does not exist for @xmath29j = 2 transitions ) . as shown above the detection of sum events which are energetically degenerate with the diagram line leads to a substantially increased effective detection efficiency ( dependent , however , on the maximum available spin ) for this energy . the minimum spin which is consistent with the data is 19/2 . the largest efficiency of the detector ( which corresponds to the smallest production cross section ) , taking into account 2- and 4-fold sum events of transitions between half - integer spins in order to get the 247.0 kev transition , is obtained with a spin of 35/2 . the lower limit for the production cross section estimated from irradiation ii and taking into account the detector efficiency for various possible combinations of sum events , as described above , is about 130 nb ( table 1).-particles , are photo - peak events due to known transitions in @xmath53hg and @xmath16tl . however , in this case [email protected] compton events below 120 kev should have been seen in fig . 6 in coincidence with @xmath4-particles of 8.55 - 8.62 mev , but only one event is seen.]-energies are compton events due to higher energy @xmath3-rays . but then the probability @xcite to have two events in the same position ( within the energy resolution ) is only a few percent , and this occurred twice , at 247.0 ( fig . 6 ) and 225.3 kev ( figs . 5 and 6 ) . ] as shown above , the observation in table 2 of gamma rays which fit in energies with transitions between high spin sdb states is consistent with various sums of gamma rays due to transitions between lower spin states with @xmath29j = 1 . since the intensity of the characteristic x - rays seen in figs . 5 and 6 in coincidence with 8 - 9 and 6 - 7 mev @xmath4-particles is very low , the observed @xmath3-rays should be due to e1 transitions , with quite low conversion factors , rather than m1 ones . such transitions have been seen before @xcite and were predicted for sd wells in nuclei around z = 86 and n = 116 @xcite . in order to identify some of the coincidence events seen in figs . 5 , 6 , and 9 at lower particle energies below 6 mev , we first looked on groups of events . the existence of groups of 3 or 2 coincidence events in clean regions indicates that they are probably not background . secondly , one looks on their accurately measured photon energies . at low energies the observed photon may be either an x - ray or a @xmath3-ray . the energies of the characteristic x - rays of the elements are known very well @xcite . the k@xmath14 x - rays of the various elements of the evaporation residue nuclei and their daughters are separated one from another by about 2 kev , where the value of a k@xmath14-line of an element with z protons is about the same as the value of the k@xmath65-line of an element with z+1 protons . if the measured energy turns out to be very close to a known x - ray energy , then it is reasonable to assume that the observed photon is an x - ray rather then a @xmath3-ray . ( in the cases found by us and discussed below the deviations from the known values are from 0.03 to 0.28 kev , while the width of one channel in the @xmath3-spectrum is about 0.24 kev and the fwhm is about four channels ) . the x - ray may be emitted if the @xmath4-particle is decaying to an excited state which decays by conversion electrons . such a process is followed by emitting characteristic x - rays of the daughter nucleus . the z - value of the daughter is thus determined . in order to determine its a - value we tried to identify the @xmath3-rays which are in coincidence with @xmath4-particles of about the same energy . a consistency is obtained when the @xmath3-ray fit in energy with the measured value , belongs to an isotope of the determined element , and it is consistent with known or expected values of the conversion factors , and also with the decay scheme . in several cases , as seen below , some of the observed events have been identified as due to characteristic x - rays ( the events surrounded with squares in figs . 5 , 6 and 9 ) and as known @xmath3-transitions in various nuclei ( the events surrounded with triangles in figs . 5 , 6 and 9 ) . in fig . 6 two coincidences between about 5.50 mev @xmath4-particles and 61.1 and 59.4 kev photons are seen , and fit with k@xmath14 and k@xmath65 x - rays of re of 61.140 and 59.718 kev . at about the same @xmath4-energy two 144.0 kev events are observed and can be identified with a known transition in @xmath66re . this identification is supported by the observation in fig . 5 , in the same @xmath4-energy range , of coincidence events with respective 59.7 kev and 140 kev x- and @xmath3-rays . the first fits with k@xmath65 of re , and the second with a transition in @xmath66re , which precedes the above 144.0 kev transition . a coincidence between 5.45 mev @xmath4-particles and 59.4 or 59.7 kev photons may in principle be also due to the known decay of @xmath20am of 5.486 mev @xmath4 and 59.54 kev @xmath3 . however , we did not use and never had in our laboratory a @xmath20am calibration source . under these circumstances the observation of the 61.1 kev photon , which fits very nicely with k@xmath14 of re , and which is very far off ( 1.56 kev ) from the @xmath3-rays of @xmath20am , and of the two consecutive @xmath3-rays of @xmath66re , suggests that the two events at about 59.6 kev are due to k@xmath65 of re and not due to @xmath20am . a 59.6 kev x - ray may in principle be also due to k@xmath14 of w of 59.3182 kev . here also the observation of the other x - rays and @xmath3-rays suggest that it is k@xmath65 of re rather than k@xmath14 of w. out of total number of 3 x - rays , one is expecting to see 1.9 k@xmath14 events and 1.1 k@xmath65 events . the observation of 1 event in the first case and 2 in the second , is well within the statistical error . in fig . 9 , for @xmath4-particles between 5.18 and 5.53 mev , groups of two and three events with photon energies of 66.8 kev and 59.6 kev , respectively , are seen . they may correspond to the known k@xmath14 x - rays of pt of 66.832 kev and of w of 59.318 kev of w. at the same corresponding @xmath4-energies known @xmath3-rays of @xmath67pt and @xmath68w are observed . here the three 59.6 kev events are more likely to be due to k@xmath14 of w , rather than k@xmath65 of re . if they were k@xmath65 of re then 5.2 events of k@xmath14 at 61.140 kev are expected . the observation of zero events when 5 are expected is unlikely . on the other hand , if the 3 observed events are k@xmath14 x - rays of w , then 1.7 events of k@xmath65 of w at 57.98 kev are expected . the observation of zero events when 1.7 are expected is well within the statistical errors . as mentioned above and seen in table 3 , about the same @xmath4-energy corresponds to two different decays . one leads to @xmath66re and was obtained in the e@xmath2 = 125 mev experiment , and the other presumably leads to @xmath68w and was obtained in the e@xmath2 = 135 mev experiment . as seen below ( section iva ) , these data are interpreted as due to transitions from the second well of the potential in the parent nuclei to the normal states in the daughters . there is nothing against having about the same @xmath4-particle energy in two different transitions . because of the low statistics we consider the identifications mentioned above as tentative only , where the identification of @xmath66re is better than of the other . the results are summarized in table 3 together with possible reaction channels and corresponding decay chains which lead to the observed transitions . the deduced production cross sections ( with an accuracy of about a factor of 2 ) are given in table 1 , columns 3 and 5 . in general lower limits were deduced since the branching ratios along the decay chains are not known . some coincidence events between 6.17 9.01 mev @xmath4-particles and identified x - rays with no corresponding @xmath3-rays were seen . they are also given in table 3 . a very well defined coincidence group of three events is seen in fig . 9 at an average particle energy of 3.88 mev and @xmath3-energy of 185.8 kev . because of the low energy of the particles and the very narrow total width of 40 kev , one may conclude that the particles are protons . ( the total estimated widths at this energy are 55 kev for protons and 630 kev for @xmath4-particles . the estimated half - life for 3.88 mev @xmath4-particles decaying to , for instance , @xmath69hg is about 1x10@xmath70 y @xcite ) . the 185.8 kev @xmath3-rays fit with a known transition in @xmath71rn ( table 3 ) . the production cross section of this group is given in table 1 . the e@xmath72 values for the g.s . to g.s . transitions for @xmath73ir @xmath51 @xmath66re ; @xmath74os @xmath51 @xmath68w and @xmath75hg @xmath51 @xmath67pt mentioned in table 3 are 2.74 , 2.662 and 2.927 mev @xcite , respectively . the corresponding observed @xmath4-energies of 5.43 , 5.53 and 5.18 mev ( table 3 ) are clearly due to decay of isomeric states in the parent nuclei . the estimated half - lives for normal @xmath4-particles of these energies @xcite are 2.9x10@xmath13 ; 3.5x10@xmath31 and 1.2x10@xmath12 s. the observed lifetimes of several months , which may be due to combined lifetimes along the long decay chains , are retarded by 3 and 4 orders of magnitude in the first and second case , respectively . from the above mentioned measured @xmath4-energies the lower limits for the excitation energies of the isomeric states in the parent nuclei @xmath73ir , @xmath74os and @xmath75hg ( see table 3 ) are deduced to be 3.1 ; 3.0 and 4.5 mev , respectively . ( lower limits are deduced since the @xmath4-decay may proceed through a higher excitation energy than the one seen experimentally ) . extrapolated and interpolated predicted energies @xcite for the second minima of 4.1 @xcite or 4.2 @xcite ; 3.4 @xcite or 3.6 @xcite ; and 4.2 @xcite or 4.6 @xcite mev , respectively , are also given in table 3 . the observed energies of the isomeric states and the predicted positions of the second minima seem to be in the same range of excitation energies . this suggests that the isomeric states in these cases are in the second well of the potential , and that the @xmath4 transitions are from the superdeformed well in the parent nuclei to less deformed or normal states in the daughters . these are the last transitions after several decays , presumably from isomeric state to isomeric state , all within the second wells along the decay chain . for instance , @xmath73ir in the isomeric state may be produced after 6 decays , 4@xmath4 and 2@xmath76@xmath77 or ec decays , all from superdeformed to superdeformed states ( see table 3).-particle or even by proton emission , is in principle not impossible . ] since the last step of the chain is a decay to a relatively high spin state , it seems that the originally produced isomeric state in the evaporation residue nucleus has high spin , and that the transitions from mothers to daughters along the decay chain are between high spin states . we now discuss the three coincidence events seen in fig . 6 and mentioned above ( secion iii.b . and table 2b ) of 6.16 mev 176.1 kev ; 6.94 mev 207.4 kev and 6.41 242.3 kev . these events appeared 96 , 68 and 78 days respectively , after the end of irradiation , and their lifetimes therefore are of the order of several months . as shown above , the energies of the first two @xmath3-rays fit very nicely with sdb transitions between half - integer spins , where the most probable candidates are @xmath52rn , produced by the pn evaporation process , and its daughters . the coincidence event of a 6.17 mev @xmath4-particle with 79.1 kev photon ( fig . 5 and table 3 ) which fits with po k@xmath14 x - rays of 79.290 kev , suggests that the transitions mentioned above with 6 7 mev @xmath4-particles might be from isomeric state(s ) in @xmath52rn to the sdb states in the second well of the potential in @xmath78po . ( the half - life of the ground state of @xmath52rn is 9.3 m ) . the measured @xmath4-particle energies are in the predicted range for the transition from the second minimum in @xmath52rn to the second minimum in @xmath78po of 6.2 @xcite or 6.9 @xcite mev . similarly the energies of 7.16 mev ( fig . 6 ; e@xmath2 = 125 mev ) and 7.05 mev ( fig . 9 ; e@xmath2 = 135 mev ) @xmath4-particles in coincidence with l@xmath79(at ) ( see table 3 ) may correspond to the predicted @xmath4-particle transitions between the second minima of @xmath80fr@xmath81 and @xmath82at@xmath81 ( 3n reaction ) of 6.9 @xcite or 7.3 @xcite mev , and of @xmath83fr@xmath81 to @xmath84at@xmath81 ( 4n reaction ) of 7.0 @xcite and 7.2 @xcite mev , respectively . in both cases discussed in the present subsection ( iv.b . ) , while the @xmath4-energies are consistent with this picture , the very long lifetime may indicate that the internal structure of the parent isomeric states and the final rotational sdb states are different . for the 3.88 proton group ( section iii.d . above ) , if the identification given in table 3 is correct , the excitation energy of the isomeric state in @xmath83fr is @xmath7 6.7 mev ( see table 3 ) . this is quite high as compared to 3.9 @xcite or 5.0 @xcite mev predicted for the second minimum , and may indicate that the origin of the isomeric state in this case is different . the most striking result is the @xmath4-particle group around 8.6 mev which was found in coincidence with sdb transitions ( figs . 5 , 6 , 7 , 8 , and table 2a ) . as mentioned above the appearance of @xmath3-transitions between half - integer spins shows that most of these transitions are in an odd a nucleus , where @xmath52rn and its daughters are most probable candidates from the reaction kinematic point of view . the observation of a coincidence event between a 8.42 mev @xmath4-particle ( which is , within the experimental spread for @xmath4-particles , consistent with 8.6 mev for 200 @xmath27g/@xmath28 c catcher foil ) and a k@xmath14 x - ray of pt ( fig . 5 and table 3 ) indicates that the transition may be from @xmath53hg ( produced after 3 @xmath4-decays from @xmath52rn ) to @xmath54pt . however , transitions from other hg isotopes produced by fewer @xmath4-decays , which are followed by several @xmath76@xmath77 or ec decays can not be excluded . the predicted @xcite half - life for normal 8.6 mev @xmath4-particles is below 1 @xmath27s . therefore in principle the @xmath3-rays of the sdb transitions may either precede or follow the 8.6 mev @xmath4-particles .- particles is always much longer than 1@xmath27s , and they could not be detected in coincidence with @xmath3-rays which precede them . ] in the case where the @xmath3-rays come first , the @xmath4-particles are due to either a transition from the second minimum in the parent nucleus to the second minimum in the daughter , or from the second minimum in the parent to normal states in the daughter . however in the first case the energy of 8.6 mev is much larger than the predicted 3.3 mev in @xmath53hg@xmath81 @xmath51 @xmath54pt@xmath81 transition @xcite , and 6.3 @xcite or 6.9@xcite in @xmath52rn@xmath81 @xmath51 @xmath78po@xmath81 transition . on the other hand in the second case of an @xmath4-transition from the second minimum to normal states , the @xmath4-particle lifetime will most probably be retarded , and the 8.6 mev @xmath4-particles will not be in coincidence within 1 @xmath27s with the sdb transitions . ( furthermore , the predicted transition energy from the second minimum to the ground state in @xmath53hg@xmath81 @xmath51 @xmath54pt@xmath85 is about 7.2 mev @xcite and in @xmath52rn@xmath81 @xmath51 @xmath78po@xmath85 is 6.3 @xcite or 6.9 mev @xcite . these values are also not in accord with the experimental value of 8.6 mev ) . it is therefore reasonable to assume that the 8.6 mev @xmath4-particles decay first and then are followed by the sdb @xmath3-rays , similar to the situation in all the other cases discussed above . let us first consider the case where the isomeric state is in the second well of the parent nucleus which decays by the 8.6 mev @xmath4-particles to the sdb in the daughter . one faces here two problems , namely the very large hindrance factor of the @xmath4-particles , and their high energy . the hindrance factor of the @xmath4-particles is in the range of 10@xmath15 ( t@xmath5 @xmath6 40 d as compared to the predicted value of 1.4 x 10@xmath86 s assuming @xmath76@xmath87 = 0.7 @xcite ) . as for the energy , if , for instance , the 8.6 mev @xmath4-particles decay to a sdb state of spin 27/2 in @xmath54pt at e@xmath48 = 865 kev ( e@xmath88 = 4.42 kev ) , and taking into account the predicted excitation energy of the second minimum in @xmath54pt at about 4.1 mev @xcite , and the q@xmath89([email protected] . ) value for @xmath53hg of 2.190 mev @xcite , the isomeric state in @xmath53hg turns out to be at about 11.6 mev above the g.s . assuming an excitation energy of the second minimum in @xmath53hg of about 5.2 mev @xcite , the above energy of 11.6 mev corresponds to an excitation energy of about 6.4 mev above the second minimum , which seems unlikely . ( in the same way , about 5.0 and 3.0 mev excitation energies above the second minimum are respectively deduced for more neutron - rich hg isotopes and for @xmath52rn , which is unlikely as well . ) it seems reasonable to conclude that the large hindrance factor might be due to another barrier transition , as , for instance , from the third well ( the hyper - deformed well ) @xcite to the second one.cm at @xmath90 = 0.63 and e@xmath48 = 8.2 mev , using hartree - fock calculations @xcite . ] it should be mentioned that an extrapolated value of about 11.5 mev is obtained for the excitation energy of the third minimum in @xmath53hg from the predictions of ref . this value is in accord with the deduced value mentioned above of about 11.6 mev .- events of 6 - 7 mev discussed in section iv.b . above is unlikely to be due to transitions from the third minimum , as its predicted excitation energy of about 18.8 mev ( extrapolated from ref.@xcite ) comes out far from the data of 7.8 or 10.3 mev ( table 3 ) . ] the large production cross section of the isomeric state which decays by the 8.6 mev @xmath4-particles of about 130 nb , as compared to the production of the isomeric states in the second minima of a few nb ( see table 1 ) , may also indicate that its origin is from the third minimum , which can presumably be produced more easily at bombarding energies below the coulomb barrier . it is also seen in table 1 that while the highest cross section for the production of the normal evaporation residue nuclei is about a factor of 60 larger at 135 mev as compared to 125 mev , the production of the isomeric states in the second minima is about the same at these two bombarding energies , whereas there is an exceptionally large cross section at 125 mev for the 8.6 mev @xmath4-particle group which is in coincidence with the sdb transitions . in summary , evidence for long - lived isomeric states with lifetimes of up to several months , with abnormal decay properties , was obtained . relatively high energy @xmath4-particles of 5.1 - 5.5 mev were tentatively found in the usually non-@xmath4-emitting nuclei , @xmath74os , @xmath73ir and @xmath75hg , and interpreted as due to transitions from the second well of the potential in the parent nuclei to less deformed or normal states in the daughters . the observed transitions are the last in long chains of up to 6 steps of @xmath4-particle and @xmath11 or ec transitions , presumably within the second well itself . long - lived 6 7 mev @xmath4-particles , in coincidence with sdb @xmath3-rays , were found with energies consistent with transitions from the second minimum in the parent nucleus to the second minimum in the daughter . the reason for their long lifetimes though is not clear . a 3.88 mev long - lived proton group was found . its state of origin is not entirely clear . by far the most exciting observation is the very high energy and strongly retarded ( t@xmath5 @xmath7 40 d ) 8.6 mev @xmath4-particle group in coincidence with sdb transitions , which is interpreted as due to a long - lived isomeric state in @xmath53hg at e@xmath48 @xmath91 11.6 mev , and is consistent with a transition from the third ( hyperdeformed ) well of the parent nucleus to the second ( superdeformed ) well in the daughter . the production cross sections of the isomeric states in the second well were about the same at e@xmath2 = 125 ( 10% below the barrier ) and at e@xmath2 = 135 mev , while on the average a factor of 15 60 larger cross section was found for the 8.6 mev group at 125 mev . the existence of such long - lived isomeric states , with lifetimes much longer than those of their corresponding ground states , and with preferential production cross sections at relatively low bombarding energies , may add new considerations regarding the stability and the production mechanism of heavy and superheavy elements @xcite . in particular it should be mentioned that the extra - push energies needed for the production of such nuclei in their superdeformed and hyperdeformed wells are much smaller than expected for producing them in their normal states @xcite . we acknowledge the support of the accelerator crew of the weizmann institute at rehovot , and the technical assistance of s. gorni , o. skala and the electronic team of the racah institute . d.k . acknowledges the financial support of the dfg . we are grateful to n. zeldes and j. l. weil for very valuable discussions . a. marinov , s. eshhar and d. kolb , phys . b191 * , 36 ( 1987 ) . a. marinov , s. eshhar and j. l. weil , _ proc . inter . symp . on superheavy elements , lubbock , texas , 1978 _ , edited by m. h. k. lodhi ( pergamon , new york , 1978 ) , p. 72 . a. marinov , c. j. batty , a. i. kilvington , g. w. a. newton , v. j. robinson and j. d. hemingway , nature * 229 * , 464 ( 1971 ) . a. marinov , s. eshhar and d. kolb , fizika * 19 * , supplement 1 , 67 ( 1987 ) . a. marinov , 6th adriatic int . , _ frontiers of heavy - ion physics , 1987 _ eds . n. cindro , w. greiner and r. aplan , ( world scientific 1987 ) , p. 179 . a. marinov , s. gelberg , and d. kolb , int . _ exotic nuclear states _ , debrecen , hungary , 1997 , eds dombrdi , z. gsci and a. krasznahorkay ( akadmiai kiad , budapest , aph n.s . , heavy ion physics * 7 * , 47 ( 1998 ) . v. e. viola jr . and g. t. seaborg , j. inorg . nucl . chem . * 28 * , 741 ( 1966 ) . a. marinov , s. gelberg and d. kolb , mod . a11 * , 861 ( 1996 ) . a. marinov , s. gelberg and d. kolb , mod . * a11 * , 949 ( 1996 ) . w. satula , s. wiok , w. nazarewicz , r. wyss and a. johnson , nucl . phys . * a529 * , 289 ( 1991 ) . s. j. krieger , p. bonche , m. s. weiss , j. meyer , h. flocard and p .- h . heenen , nucl . phys . * a542 * , 43 ( 1992 ) . s. g. nilsson , g. ohln , c. gustafson and p. mller , phys . 30b * , 437 ( 1969 ) . j. fernndez - neillo , c. h. dasso and s. landowne , code ccdef , comp . . comm . * 54 * , 409 ( 1985 ) . s. raman , c. h. malarkey , w. t. milner , c. w. nestor , jr . and p. h. stelson , adndt * 36 * , 1 ( 1987 ) . a. marinov , s. gelberg and d. kolb , _ physics of unstable nuclear beams , so paulo , brazil , 1996 , _ eds . c. a. bertulani , l. f. canto and m. s. hussein , ( world scientific , 1996 ) , p. 181 . a. marinov , s. gelberg and d. kolb , @xmath92 , dubna , russia,1997 , eds . oganessian r. kalpakchieva world scientific , 1998 , p. 437 . f. s. stephens et al . lett . * 65 * , 301 ( 1990 ) . s. @xmath93berg , nucl . phys . * a520 * , 35c ( 1990 ) . j. hller , and s. @xmath93berg , z. phys . * a336 * , 363 ( 1990 ) . w. nazarewicz , p. olanders , i. ragnarsson , j. dudek and g. a. leander , phys lett . * 52 * , 1272 ( 1984 ) . k .- h . schmidt , c .- c . sahm , k. pielenz and h .- clerc , z. phys . * a316 * , 19 ( 1984 ) . g. audi , o. bersillon , j. blachot and a. h. wapstra , nucl . * a624 * , 1 ( 1997 ) . r. b. firestone , v. s. shirley , c. m. baglin , s. y. f. chu and j. ziplin , table of isotopes , john wiley and sons , 1996 . w. m. howard and p. mller , adndt * 25 * , 219 ( 1980 ) . w. nazarewicz , phys . lett . * b 305 * , 195 ( 1993 ) . s. wiok , w. nazarewicz , j. x. saladin , w. plciennik and a. johnson , phys . lett . * b 322 * , 304 ( 1994 ) . d. kolb and a. marinov , proc . florence , vol . 1 , p. 92 a. marinov , s. eshhar , j. l. weil and d. kolb , phys . . lett . * 52 * , 2209 ( 1984 ) ; * 53 * , 1120(e ) ( 1984 ) . a. marinov , s. gelberg and d. kolb , in _ 6th int . conf . on nuclei far from stability and 9th int . conf . on atomic masses and fundamental constants _ , bernkastel - kues , germany , inst . 132 , p. 437 k. kumar , superheavy elements , adam hilger , bristol and new york , 1989 . j. p. blocki , h. feldmeier and w. j. swiatecki , nucl . * a459 * , 145 ( 1986 ) . lllll reaction & @xmath94@xmath85(@xmath27b ) & & @xmath94@xmath85(@xmath27b ) & + & 125 mev & * 125 mev * & 135 mev & * 135 mev * + + 3n & @xmath95 & & @xmath96@xmath97 & + p2n & 4.5 & & @xmath98@xmath97 & + 4n & & & 274 & + p3n & & & 101 & + 5n & & & @xmath99@xmath100 & + p4n & & & @xmath99@xmath100 & + @xmath42n & & & & + pn & & & & + @xmath101 lower limit is given since the branching ratios along the decay chain ( table 3 ) are not known . + @xmath97 it was impossible to distinguish between the 3n and the p2n reactions . the value given was deduced assuming that the relevant observed activity is due to this reaction channel only . + @xmath100 it was impossible to distinguish between the 5n and the p4n reactions . the value given was deduced assuming that the relevant observed activity is due to this reaction channel only . + @xmath102 based on the coincidence group of the 3.88 mev protons with 185.8 kev @xmath3-rays and assuming that the @xmath3-rays are from @xmath71rn ( see text ) . + lllll e@xmath72 & e@xmath50(expt.)@xmath101 & e@xmath50(cal . ) & @xmath103e & transition@xmath97 + ( mev ) & ( kev ) & ( kev ) & ( kev ) + + + 7.81 & 25.9 & 26.5 & 0.6 & 3@xmath512 + 8.52 & 40.1 & 39.8 & + 0.3 & 9/2@xmath517/2 + 8.21 & 52.9 & 53.0 & 0.1 & 6@xmath515 + 8.53@xmath100 & 97.9@xmath100 & 97.2 & + 0.7 & 11@xmath5110 + 8.19 & 118.4 & 119.3 & 0.9 & 27/2@xmath5125/2 + 7.88 & 123.4 & 123.8 & 0.4 & 14@xmath5113 + 8.37 & 141.3 & 141.4 & 0.1 & 16@xmath5115 + 7.99 & 145.6 & 145.9 & 0.3 & 33/2@xmath5131/2 + 8.51 & 171.8 & 172.4 & 0.6 & 39/2@xmath5137/2 + 8.60 & 186.5 & 185.6 & + 0.9 & 21@xmath5120 + 8.07 & 212.9 & 212.2 & + 0.7 & 24@xmath5123 + 8.59@xmath102 & 225.3@xmath102 & 225.4 & 0.1 & 51/2@xmath5149/2 + 8.61@xmath104 & 247.0@xmath104 & 247.5 & 0.5 & 28@xmath5127 + + 6.16 & 176.1 & 176.4 & -0.3 & 20@xmath5119 + 6.94 & 207.4 & 207.3 & + 0.1 & 47/2@xmath5145/2 + 6.41 & 242.3 & 242.6 & -0.3 & 55/2@xmath5153/2 + @xmath101 the peak to total ratio was 100% up to about 120 kev and reduced gradually to 22% at 250 kev . + @xmath97 the transitions between the maximum possible spins are given . at relatively high energies the observed @xmath3-rays are most probably sum events due to various combinations of lower energy transitions in the band which , because of the j(j+1 ) law , fit in energy to calculated transitions between higher energies . ( see text ) . + @xmath100 average of 2 events : 8.47 mev @xmath4 in coincidence with 97.4 kev @xmath3 in the first measurement and 8.58 mev @xmath4 in coincidence with 98.3 kev @xmath3 in the second measurement . + @xmath102 average of 2 events : 8.61 mev @xmath4 in coincidence with 224.9 kev @xmath3 in the first measurement and 8.57 mev @xmath4 in coincidence with 225.7 kev @xmath3 in the second measurement . + @xmath104 average of 2 events in the second measurement of 8.62 mev @xmath4 in coincidence with 246.5 kev @xmath3 and 8.60 mev @xmath4 in coincidence with 247.5 kev @xmath3 . + e@xmath50(meas . no . ) & e@xmath105 & isotope & transition & reaction & e@xmath106(@xmath107z ) & e@xmath108 + kev & mev & & & * decay * & mev & mev + * time / d*@xmath101 & & & & * chain * & & s ; k + + 61.1(ii ) & 5.48 & & k@xmath14(re ) & & & + * 31.1 * & & & & & & + 59.6(x2)(i ; ii ) & 5.45 & & k@xmath65(re ) & & & + * 7.4 ; 107.2 * & & & & & & + 144.0(x2)(ii ) & 5.43 & @xmath66re & 330(?)@xmath97@xmath51186 & 3n & @xmath73.1(@xmath73ir ) & 4.1@xmath102 ; 4.2@xmath102 + * 60.2 ; 88.8 * & & & ( 5@xmath77)@xmath51(6)@xmath109 & * 4 , * & & + & & & e1 ; 100% & * 2(ec)@xmath100 * & & + 141.0(i ) & 5.29 & @xmath66re & 471(?)@xmath97@xmath51330(?)@xmath97 & 3n & @xmath73.2(@xmath73ir ) & 4.1@xmath102 ; 4.2@xmath102 + * 130.2 * & & & ( 4@xmath77)@xmath51(5@xmath77 ) & * 4 , * & & + & & & m1+e2 & * 2(ec)@xmath100 * & & + 66.8(x2)(iii ) & 5.18 & & k@xmath14(pt ) & & & + * 13.9 ; 58.6 * & & & & & & + 236.2(iii ) & 5.12 & @xmath67pt & [email protected] & p3n ; 4n & @xmath74.53 ( @xmath75hg ) & 4.2@xmath104 ; 4.6@xmath104 + * 139.5 * & & & ( @xmath110@xmath77)@xmath51(@xmath111@xmath109 ) & & & + & & & ( e1 ) & * ; * @xmath100 & & + 59.6(x3)(iii ) & 5.53 & & k@xmath14(w ) & & & + * t@xmath112 = 90 * & & & & & & + 210.3(iii ) & 5.43 & @xmath68w & [email protected] & @xmath42n & @xmath73.04(@xmath74os ) & 3.4@xmath102 ; 3.6@xmath102 + * 168.4 * & & & @xmath113@xmath109@xmath51@xmath114@xmath109 & & & + & & & e2 ; 100% & @xmath100 & & + 70.9(ii ) & 9.01 & & k@xmath115(hg ) & & & + * 89.7 * & & & & & & + 67.1(i ) & 8.42 & & k@xmath115(pt ) & & & + * 81.7 * & & & & & & + 79.1(i ) & 6.17 & & k@xmath115(po ) & & & + * 133.5 * & & & & & & + 13.9(ii ) & 7.16 & & l@xmath116(at ) & & & + * 64.6 * & & & & & & + 13.9(iii ) & 7.05 & & l@xmath116(at ) & & & + * 127.8 * & & & & & & + 185.8(x3)(iii ) & 3.88(p ) & @xmath71rn & [email protected]@xmath117 & 4n & @xmath76.7(@xmath83fr ) & 3.9@xmath104 ; 5.0@xmath104 + * t@xmath112 = 120 * & & & ( 9@xmath109)@xmath51(8@xmath77 ) & & & + & & & e1 ; 55% & & & + sdb(ii)@xmath118 & 6 - 7 & @xmath78po & sdb & pn & 7.8@xmath119;10.3@xmath120(@xmath52rn ) & 6.8@xmath102 ; 9.9@xmath104 + * t@xmath112 = 53 * & & & table 2(b ) & & + sdb(i ; ii)@xmath121 & 8.6 & @xmath54pt@xmath122 & sdb & pn & @xmath611.6(@xmath53hg)@xmath122 & 11.5(@xmath123)@xmath124 + * t 40 * & & & table 2(a ) & * 3 * & & + @xmath101 for a group of three events the half - lives as estimated according to the formulas of k .- h . schmidt et al . [ 21 ] are given . + @xmath97 the excitation energy in @xmath66re is not certain [ 23 ] . + @xmath100 the order of the decay is not known . + @xmath102 extrapolated value . + @xmath104 interpolated value . + @xmath117 the identification given in the table is not certain as explained in the text . + @xmath118 see text and table 2(b ) for the @xmath3-eneries of this band . + @xmath119 estimated excitation energy in @xmath52rn assuming e@xmath72=6.5 mev , a predicted e@xmath48 of the second minimum in @xmath78po of 6.6 mev [ 10 ] , and decay to a sdb state with a spin of 27/2 at excitation energy above the second minimum of 863 kev . + @xmath120 as comment ( h ) above except that e@xmath48 of the second minimum in @xmath78po was taken at 9.1 mev as predicted in ref . + @xmath121 see table 2(a ) for the @xmath3-energies of this band . + @xmath122 see text . + @xmath124 extrapolated value for the excitation energy of the third minimum from ref . [ 25 ] is given .

we study the topological and differentiable singularities of the configuration space of a mechanical linkage in , defining an inductive sufficient condition to determine when a configuration is singular . we show that this condition holds for generic singularities , provide a mechanical interpretation , and give an example of a type of mechanism for which this criterion identifies all singularities .

1112.2334

the mathematical theory of robotics is based on the notion of a mechanism consisting of links , joints , and rigid platforms . the _ mechanism type _ is a simplicial ( or polyhedral ) complex @xmath3 , where the parts of dimension @xmath4 correspond to the platforms , and the complementary one - dimensional graph corresponds to the links ( = edges ) and joints ( = vertices ) . the _ linkage _ ( or mechanism ) @xmath1 itself is determined by assigning fixed lengths to each of the links of @xmath3 . see @xcite and @xcite for surveys of the mechanical and topological aspects , respectively . configuration spaces[scspace ] here we concentrate on the most prevalent type of mechanism @xmath3 : namely , a finite @xmath5-dimensional simplicial complex ( undirected graph ) , with @xmath6 vertices and @xmath7 edges . note that a rigid platform is completely specified by listing the lengths of all its diagonals ( i.e. , the distance between any two vertices ) , so we need not list the platforms explicitly . our results actually hold also for the case when some links of @xmath1 are _ prismatic _ ( or telescopic ) i.e. , have variable length but for simplicity we deal here with the fixed - length case only . a length - preserving embedding of the vertices of the linkage @xmath1 in a fixed ambient euclidean space @xmath2 is called a _ configuration _ of @xmath1 . in applications , @xmath8 is most commonly @xmath9 or @xmath10 . the set of all such embeddings , with the natural topology ( and differentiable structure ) , is called the configuration space of @xmath1 , denoted by @xmath0 . such configuration spaces have been studied intensively , with the hope of extracting useful mechanical information from their topological or geometric properties . much of the mathematical literature has been devoted to the special case when @xmath1 is a closed chain ( polygon ) : see , e.g. , @xcite . however , the general case has also been treated ( cf.@xcite ) . singularities[ssing ] there are two main types of singularities which arise in robotics . singularities of a mechanism , which appear as singularities of work and actuation maps defined on @xmath0 ( [ dwork ] ) , have obvious mechanical interpretations , and have been studied intensively ( see , e.g. , @xcite , @xcite , and @xcite ) . on the other hand , the _ topological _ or differentiable singularities of the configuration space @xmath0 itself have not received much attention in the literature since @xcite , aside from some special examples ( see , e.g. , @xcite and @xcite ) . for any linkage @xmath1 , the configuration space @xmath0 is the zero set of a smooth function @xmath11 ( see [ dcspace ] below ) , so that @xmath0 is typically a smooth manifold ( when @xmath12 is a regular value of @xmath13 ) , and even if not , `` most '' points of @xmath0 are smooth , since a simple _ necessary _ condition for a point @xmath14 in @xmath0 to be singular is that @xmath15 . thus we are in the common situation where it is relatively straightforward to identify configurations which are _ possibly _ singular , but not so easy to pinpoint when this is in fact so . our goal in this paper is threefold : 1 . to provide a straightforward inductive description of a _ sufficient _ condition for a configuration @xmath14 to be differentiably singular ( in fact , this will imply that @xmath14 is even a topological singularity ) see proposition [ pgensing ] and theorem [ tgensing ] . 2 . to show that this condition applies generically ( that is , to all but a positive - codimension subset of the singular locus @xmath16 ) see remarks [ rgensin ] and [ rgensnt ] . 3 . to obtain a mechanical interpretation for all singularities in the configuration space of a linkage @xmath1 as a tangential conjunction of two kinematic singularities of type i ( cf.@xcite ) for complementary sub - mechanisms of @xmath1 see remark [ rmechanism ] . the third goal is completely achieved only in the plane ( for @xmath17 ) , since the model we use for configuration spaces is not completely realistic for rigid rods in @xmath18 . see remark [ rccs ] below for an explanation of the difficulties involved . [ ralgvar ] since the function @xmath19 defining the configuration space is a quadratic polynomial ( cf . [ dcspace ] ) , @xmath0 is actually a real algebraic variety . thus any topological or differentiable singularity @xmath14 is in particular an algebraic singularity ( cf . ii , 1.4 ) ) . somewhat more surprisingly , every real algebraic variety is a union of components of the configuration spaceof some planar linkage ( @xmath17 ) see @xcite . thus our results here appear to be statements about any real algebraic variety . however , the point we wish to make here is not that the cone singularities are the most common ones in algebraic varieties ; it is rather the mechanical interpretation of the generic singularities , and the mechanical underpinnings of the inductive process described in section [ ciccspaces ] . in fact , while the topological , differentiable , and geometric structures on configuration spaces of linkages can be used to study their mechanics ( cf.@xcite ) , the algebraic structure usually plays no role ( but see @xcite ) . organization in section [ cbcspaces ] we briefly review some of the basic notions used in this paper . in section [ clecspaces ] , various concepts of local equivalences of configuration spaces are defined ; these help to simplify the study of singular points . in section [ cpbc ] we explain the role played by pullbacks of configuration spaces . this is applied in section [ ciccspaces ] to provide an inductive construction , which is used both to describe the sufficient condition mentioned in [ ssing](b ) , and to show that they are indeed singular points . an example is studied in detail in section [ ctriang ] . we wish to thank the referee for his or her comments . we first recall some general background material on the construction and basic properties of configuration spaces . this also serves to fix notation , which is not always consistent in the literature . [ dcspace ] consider an abstract graph @xmath3 with vertices @xmath20 and edges @xmath21 . a _ linkage _ ( or _ mechanism _ ) @xmath1 of type @xmath3 is determined by a function @xmath22 specifying the length @xmath23 of each edge @xmath24 in @xmath25 ( subject to the triangle inequality as needed ) . we write @xmath26 for the vector of squared lengths . the set of all embeddings of @xmath20 in an ambient euclidean space @xmath2 is an open metric subspace of @xmath27 , denoted by @xmath28 . we have a _ squared length map _ @xmath29 with @xmath30 , and the _ configuration space _ of the linkage @xmath31 is the metric subspace @xmath32 of @xmath28 . a point @xmath33 is called a _ configuration _ of @xmath1 . note that @xmath13 is an algebraic function of @xmath34 ( which is why the lengths were squared ) , so @xmath0 is a real algebraic variety . [ rsubmer ] by ( * ? ? ? * i , theorem 3.2 ) , we know that @xmath0 is a smooth manifold if @xmath35 is a regular value of @xmath13 : that is , if its differential @xmath36 is of maximal rank for every @xmath37 with @xmath38 . however , for some mechanism types @xmath3 , this condition may not be generic : there exist mechanism types @xmath3 and an open set @xmath39 in @xmath40 consisting of non - regular values of @xmath41 . this means that for each @xmath42 , the configuration space @xmath43 has at least one configuration @xmath44 such that @xmath13 not a submersion at @xmath14 . see @xcite for an example . isometries of configuration spaces [ srestr ] the group @xmath45 of isometries of the euclidean space @xmath2 acts on the space @xmath0 . when @xmath1 has a rigid `` base platform '' @xmath46 of dimension @xmath47 , this action is free . in this case we can work with the `` restricted configuration space '' @xmath48 , and the quotient map has a continuous section ( equivalent to choosing a fixed location in @xmath2 for @xmath46 ) . see [ sppm ] for an example of such a @xmath1 . in general , certain configurations ( e.g. , those contained in a proper linear subspace @xmath49 of @xmath2 ) may be fixed by certain transformations ( those fixing @xmath49 ) , so the action of @xmath45 is not free . [ dfixlink ] choose a fixed vertex @xmath50 of @xmath1 as its _ base - point _ : the action of the translation subgroup @xmath51 of @xmath45 on @xmath50 _ is _ free , so its action on @xmath0 is free , too , and we call the quotient space @xmath52 the _ pointed configuration space _ for @xmath1 . thus @xmath53 , and a pointed configuration ( i.e. , an element of @xmath54 ) is simply an ordinary configuration expressed in terms of a coordinate frame for @xmath2 with the origin at @xmath50 . if we also choose a fixed link @xmath55 in @xmath1 starting at @xmath50 , we obtain a smooth map @xmath56 which assigns to a configuration @xmath14 the direction of @xmath55 . the fiber @xmath57 of @xmath58 at @xmath59 will be called the _ reduced configuration space _ of @xmath1 . note that the bundle @xmath60 is locally trivial . [ dwork ] a mechanism @xmath1 may be equipped with a special point @xmath61 in engineering terms this is the `` end - effector '' of @xmath1 , whose manipulation is the goal of the mechanism . we think of @xmath62 as a sub - mechanism of @xmath1 ( more generally , we could choose any rigid sub - mechanism ) . assuming that the base - point @xmath50 of @xmath1 is not @xmath61 , the inclusion @xmath63 induces a map of configuration spaces @xmath64 , whose image @xmath65 is called the _ work space _ of the mechanism . the _ work map _ @xmath66 of @xmath1 is the factorization of @xmath67 through @xmath65 ( which is not always a smooth manifold ) . [ eglwork ] now consider a closed @xmath68-chain @xmath69 , as in figure [ ffivebar ] , with end - effector @xmath70 . here the direction of @xmath71 is fixed . = 3.5 cm the work space of each of the two open sub - chains of @xmath69 starting at @xmath72 and ending at @xmath73 is a closed annulus . therefore , @xmath65 is the intersection of these two annuli ( see figure [ fworkfive ] ) , i.e. a curvilinear polygon in @xmath74 , whose combinatorial type depends on the lengths of the links . = 3.5 cm [ rccs ] the configuration spaces studied in this paper are mathematical models , which take into account only the locations of the vertices of @xmath1 , disregarding possible intersections of the edges . in the plane , there is some justification for this , since we can allow one link to slide over another . this is why this model is commonly used ( cf . @xcite ; but see @xcite ) . however , in @xmath18 the model is not very realistic , since it disregards the fact that rigid rods can not pass through each other . thus a proper treatment of configurations in @xmath18 must cut our `` naive '' version of @xmath28 ( and thus @xmath0 and @xmath54 ) along the subspace of configurations which are not embeddings of the full graph @xmath3 . the precise description of such a `` realistic '' configuration space @xmath75 is quite complicated , even at the combinatorial level , which is why we work here with @xmath28 , @xmath0 , and @xmath54 as defined in [ dcspace]-[dfixlink ] . note , however , that @xmath0 has a dense open subspace @xmath76 consisting of embeddings of the full graph ( including its edges ) , which may be identified with a dense open subset of @xmath75 . we observe that even such a model @xmath75 is not completely realistic , in that it disregards the thickness of the rigid rods . unfortunately , the generic singularities we identify here are not in @xmath76 . nevertheless , in some cases at least , our method of replacing one singular configuration by another ( see section [ clecspaces ] below ) allows us to replace the generic singularity in @xmath77 with a configuration in @xmath78 , for a suitable linkage @xmath79 . see section [ ctriang ] for an example of this phenomenon ( which also occurs in the @xmath10-dimensional version of the linkage described there ) . let @xmath1 and @xmath80 be two linkages . we would like to think of points in the respective configuration spaces as being equivalent if they are both smooth , or both have `` similar '' singularities . since these concepts are local , we make the following : [ dloce ] two configurations @xmath14 in @xmath0 and @xmath81 in @xmath82 are : 1 . _ locally equivalent _ if there are neighborhoods @xmath39 of @xmath14 in @xmath54 and @xmath83 of @xmath81 in @xmath84 , and a homeomorphism @xmath85 with @xmath86 . 2 . _ locally product - equivalent _ if there are neighborhoods @xmath49 of @xmath14 in @xmath54 and @xmath87 of @xmath81 in @xmath84 equipped with homeomorphisms @xmath88 ( taking @xmath14 to @xmath89 ) and @xmath90 ( taking @xmath81 to @xmath91 ) , as well as a homeomorphism @xmath85 with @xmath92 . see @xcite for other formulations of this and similar notions . evidently , any two smooth configurations in any two configuration spaces are locally product - equivalent . in the next section we decompose our configuration spaces into simpler factors ( locally ) , gluing them along appropriate work maps . the singularities of the configuration spaces translate into work singularities on the factors , so we need an analogous notion of work maps being locally equivalent ( at smooth configurations ) , or locally equivalent up to a euclidean factor : [ dlocew ] if @xmath93 and @xmath94 are inclusions of a common rigid sub - mechanism @xmath95 ( usually a single point ) in two distinct linkages , and @xmath96 , @xmath97 are two smooth configurations , we say that @xmath98 and @xmath99 are 1 . _ work - equivalent _ at @xmath100 if there are neighborhoods @xmath39 of @xmath14 , @xmath83 of @xmath81 , and @xmath49 of @xmath101 , and a diffeomorphism @xmath102 making the following diagram commute : @xmath103 \hspace*{1 mm } u \ar[dddr]_{i^{\ast}{\lvert_{u}}}\ar[rr]^{f}_{\cong } & & u'\hspace*{1mm}\ar@^{(->}[rd ] \ar[dddl]^{(i')^{\ast}{\lvert_{u ' } } } & \\ { { { { \euscript c}}_{\ast}}(\gamma)}\ar[d]_{i^{\ast } } & & & & { { { \euscript c}}_{\ast}}({\gamma ' } ) \ar[d]^{(i')^{\ast } } \\ { { { \euscript c}}_{\ast}}(\delta ) & & & & { { { \euscript c}}_{\ast}}(\delta ' ) \\ & & \ar@_{(->}[llu]\hspace*{1 mm } w \hspace*{1mm}\ar@^{(->}[rru ] & & } } \ ] ] 2 . _ @xmath104-equivalent _ at @xmath100 if there are neighborhoods @xmath88 of @xmath14 and @xmath105 of @xmath81 and a homeomorphism @xmath85 as in [ dloce](b ) above , such @xmath98 factors through the projection @xmath106 and @xmath99 factors through @xmath107 in such a way that the diagram analogous to commutes . an important example of these notions is provided by the following simple mechanism : [ dopen ] an _ open @xmath7-chain _ is a linkage @xmath108 , where @xmath3 is a connected linear graph with @xmath109 vertices ( where all but the endpoints @xmath72 and @xmath110 are of valency @xmath9 ) , with lengths @xmath111 . see figure [ fopenchain ] below . it is natural to choose the base - point @xmath112 ( fixed at the origin , say ) to define the pointed configuration space @xmath113 , and @xmath114 as end - effector . the resulting workspace @xmath65 is @xmath115 $ ] , for fixed @xmath116 , where @xmath117 and @xmath118 are respectively the minimal and maximal possible distances of @xmath61 from @xmath50 . the spherical ( or polar ) coordinate @xmath119 is the direction of the vector @xmath120 . a _ closed ( @xmath109)-chain _ is a linkage @xmath121 , where @xmath3 is a cycle with @xmath109 vertices ( of valency @xmath9 ) , having lengths @xmath122 ( see figure [ ffivebar ] ) . a _ prismatic closed ( @xmath109)-chain _ @xmath123 has the same @xmath3 , with lengths @xmath111 as for @xmath121 , but with the last link prismatic that is , the length @xmath124 varies in the range @xmath125 . [ lwork ] the work map @xmath126 of an open chain is a submersion , unless @xmath14 is aligned ( that is , all links have a common direction vector @xmath127 in @xmath2 at @xmath14 ) . in this case the @xmath128-dimensional subspace @xmath129 is orthogonal to @xmath127 . clearly the configuration spaces of an open @xmath7-chain and the corresponding prismatic closed ( @xmath109)-chain are isomorphic . however , the following result will be useful in understanding the work map singularities of an open chain , by allowing us to disregard its ( @xmath130)-dimensional non - singular direction . [ pzero ] if @xmath108 is an open @xmath7-chain with links @xmath111 , then the _ pointed _ configuration space @xmath113 is @xmath104-equivalent at any configuration @xmath14 to the _ reduced _ configuration space @xmath131 of a closed prismatic ( @xmath109)-chain . = 3.5 cm we may choose @xmath132 as local coordinates for the smooth configuration space @xmath113 near @xmath14 , where @xmath133 is the spherical angle between the vectors @xmath134 and @xmath135 ( see figure [ fopenchain ] ) , and @xmath136 is as in [ dopen ] ( for @xmath137 ) . thus in a coordinate neighborhood @xmath138 of @xmath14 the work map @xmath139 $ ] factors as @xmath140 , where @xmath141 is the projection , and @xmath142 $ ] . now for each @xmath143 $ ] , the fiber @xmath144 is diffeomorphic to the configuration space @xmath145 of a closed chain having @xmath109 links of lengths @xmath146 . as in [ dfixlink ] , we have @xmath147 , so @xmath145 is locally product - equivalent to @xmath148 , and in fact @xmath145 is @xmath104-equivalent to @xmath148 with respect to @xmath149 . as @xmath150 varies , we obtain the mechanism @xmath123 . if @xmath151 vanishes at @xmath14 , but @xmath14 is not aligned , then the work map @xmath126 is a submersion at @xmath14 , and the same holds for @xmath152 , so they are @xmath104-equivalent . if @xmath153 at @xmath14 and @xmath14 is aligned , choose the coordinate @xmath136 be the direction of the alignment vector @xmath127 . decomposing the work map consider an arbitrary mechanism @xmath1 with base point @xmath50 and work map @xmath154 for the end - effector @xmath61 . note that @xmath54 is locally diffeomorphic to the product @xmath155 ( [ dfixlink ] ) , since the bundle @xmath156 ( for @xmath157 ) is locally trivial ( assuming @xmath55 does not vanish ) . if we choose local spherical coordinates @xmath158 for the work space @xmath159 , the work map @xmath160 may be written locally in the form [ eqpsi ] = _ s^d-1:s^d-1 _ s^d-1r^_+ for some smooth function @xmath161 ( which is the work function for the associated reduced configuration space ) . note that the derivative of the work function @xmath126 may thus be written in the form : [ eqdpsi ] ( ) _ ( , ) = ( ll i_d-1 & + & ( ) _ ) . which shows that @xmath162 has rank @xmath8 or @xmath130 . [ pgensin ] if @xmath163 is a smooth configuration for a mechanism @xmath1 with work function @xmath164 as in with @xmath165 , and @xmath166 is a non - degenerate singular point of @xmath167 , then @xmath54 is @xmath104-equivalent at @xmath14 to an aligned configuration of an open @xmath168-chain for some @xmath169 . by the morse lemma ( cf . * theorem 2.16 ) ) we may choose local coordinates @xmath170 for @xmath57 near @xmath166 ( where @xmath171 ) , so that @xmath167 has the form [ eqmorse ] ( ) = a_0+_i=1^jt_i^2-_i = j+1^k - d+1t_i^2 . on the other hand , by proposition [ pzero ] the configuration space @xmath172 for an open @xmath168-chain at any configuration @xmath173 is @xmath104-equivalent to the reduced configuration space @xmath174 at some configuration @xmath175 , where @xmath176 is a prismatic closed ( @xmath177)-chain . the reduced work map @xmath178 assigns to each @xmath179 the length of the variable link ( with @xmath180 $ ] , the segment of possible lengths ) . as shown in ( * ? ? ? * theorem 5.4 ) , @xmath181 is a morse function , having ( non - degenerate ) singular points precisely at the aligned configurations @xmath175 of the closed chain @xmath176 . although milgram and trinkle do not calculate the index of @xmath181 at @xmath175 , their computation of the hessian of @xmath181 in ( * ? ? ? * key example , p. 255 ) , combined with farber s proof of ( * ? ? ? * lemma 1.4 ) for the planar case , show that this index is equal to @xmath182 , where @xmath7 is the number of forward - pointing links in the configuration @xmath175 . thus by the morse lemma again we may choose an aligned configuration @xmath175 and local coordinates in @xmath174 around it so that @xmath181 too has the form and thus @xmath54 is @xmath104-equivalent at @xmath14 to @xmath174 at @xmath175 . by proposition [ pzero ] it is then readily seen to be @xmath104-equivalent to @xmath172 at the corresponding aligned open - chain configuration @xmath173 . we now describe a procedure for viewing the configuration space of an arbitrary linkage @xmath1 as a pullback , obtained by decomposing @xmath1 into two simpler sub - mechanisms . the basic idea is a familiar one see , e.g. , @xcite . pullbacks[spb ] let @xmath108 denote an open chain which is a sub - mechanism of @xmath1 ( cf . [ dopen ] ) , and let @xmath80 denote the mechanism obtained from @xmath1 by omitting the @xmath7 links of @xmath108 ( and all vertices but @xmath72 and @xmath110 ) . for simplicity we choose @xmath112 as the common base - point of @xmath1 , @xmath108 , and @xmath80 , and @xmath114 as the common end - effector of @xmath108 and @xmath80 . see figure [ fdecompose ] . = 3.5 cm the work space of both mechanisms @xmath80 and @xmath108 ( i.e. , the set of possible locations for @xmath61 ) is contained in @xmath2 , and we have work maps @xmath183 and @xmath184 which associate to each configuration the location of @xmath61 . note that the pointed configuration space @xmath113 is a manifold ( diffeomorphic to @xmath185 ) with a natural embedding @xmath186 , and similarly @xmath187 for a suitable euclidean space @xmath188 . this can be done , for example , by using the position coordinates in @xmath2 for every vertex in @xmath1 . let @xmath189 and @xmath190 , and define @xmath191 to be the product map @xmath192 and @xmath193 to be @xmath194 , so that @xmath195 is an embedding of @xmath84 as a submanifold in @xmath196 . since we have a pullback square : @xmath197\ar[r ] & { { { \euscript c}}_{\ast}}({{\gamma_{{\operatorname{op}}}^{k } } } ) \ar[d]^{\phi}\\ { { { \euscript c}}_{\ast}}({\gamma'})\ar[r]^{\psi } & { { \mathcal w}}\subseteq{{{\mathbb r}}^{d}}~ , } } \ ] ] @xmath54 may be identified with the preimage of the subspace @xmath198 under @xmath199 . let @xmath97 and @xmath200 be matching configurations with @xmath201 , and let @xmath202 be the configuration @xmath203 , so that @xmath204 : @xmath205_{h } & = & { { { \euscript c}}_{\ast}}({{\gamma_{{\operatorname{op}}}^{k}}})\ar@<-2.8ex>[d]_{\phi}\ni{{{{\mathcal v}}{\sp{({k } ) } } } } & \times & { { { \mathbb r}}^{m}}\ar@<-3.7ex>[d]_{{\operatorname{id}}}\ni j({{{\mathcal v}}\,'})\\ h({\mathbf{x}})\in y & = & \hspace*{8mm}{{{\mathbb r}}^{d}}\ni\phi({{{{\mathcal v}}{\sp{({k } ) } } } } ) & \times & { { { \mathbb r}}^{m}}\ni j({{{\mathcal v}}\ , ' } ) \\ \hspace*{0.5mm}{{{\mathcal v}}\,'}\in{{{\euscript c}}_{\ast}}({\gamma ' } ) \ar@^{(->}[u]<-2.7ex>^{g } \ar[urr]^{\psi}\ar@_{(->}[urrrr]_{j } & & & & } } \ ] ] we want to know if the point @xmath96 defined by @xmath206 is singular . by ( * ? * i , theorem 3.3 ) , @xmath14 is smooth if @xmath207 i.e. , @xmath199 is locally transverse to @xmath84 at the points @xmath202 and @xmath97 , which means that @xmath208 . since @xmath209 is onto , this is equivalent to : [ etransv ] ( ) _ v + ( ) _ v = r^d generic singularities in pullbacks [ sgensin ] clearly , the failure of is a _ necessary _ condition for @xmath210 to be singular in @xmath54 . note that if does not hold , then neither @xmath211 nor @xmath212 is onto @xmath2 . by lemma [ lwork ] , the first implies that the configuration @xmath213 for the open chain @xmath214 must be aligned , while the second implies that @xmath212 is of rank @xmath215 . [ dgensing ] given a pullback diagram as in a configuration @xmath216 will be called _ generically non - transverse _ if @xmath217 is a non - degenerate singular point of @xmath167 , and @xmath218 . [ rgensin ] note that since @xmath219 is an algebraic function , generically it will be a morse function , so any singular point @xmath217 is non - degenerate . likewise , in the moduli space @xmath220 for open @xmath7-chains , the subspace of moduli @xmath13 for which @xmath108 has no aligned configurations with @xmath221 is zariski open in @xmath222 . thus among the potentially singular configurations of @xmath54 ( i.e. , those for which fails ) , the generically non - transverse ones are indeed generic . [ pgensing ] given a pullback diagram any generically non - transverse configuration @xmath206 is the product of a euclidean space with a cone on a homogeneous quadratic hypersurface , so in particular it is a topological singularity of @xmath54 . since @xmath217 is a non - degenerate singular point of @xmath167 , by proposition [ pgensin ] the work map @xmath223 is work - equivalent to the work map @xmath224 of an open chain @xmath225 at some aligned configuration @xmath173 . thus the pullback diagram may be replaced by one of the form @xmath226\ar[r ] & { { { \euscript c}}_{\ast}}({{\gamma_{{\operatorname{op}}}^{k } } } ) \ar[d]^{\phi}\\ { { { \euscript c}}_{\ast}}({\gamma_{{\operatorname{op}}}^{n}})\ar[r]^{\eta } & { { \mathcal w}}\subseteq{{{\mathbb r}}^{d}}~ , } } \ ] ] so that @xmath227 itself is @xmath104-equivalent at @xmath228 to the configuration space of a closed chain with @xmath229 links at an aligned configuration ( since @xmath230 and @xmath224 were non - transverse ) . this is known to be the cone on a homogeneous quadratic hypersurface , by ( * ? ? ? * theorem 1.6 ) and ( * ? ? ? * theorem 2.6 ) , so it is topologically singular . we now define an inductive process for studying the local behavior of a configuration @xmath14 of a linkage @xmath1 . this consists of successively discarding open chains of @xmath1 while preserving the local structure . the inductive procedure[sip ] we saw in [ spb ] how removing an open chain sub - mechanism from @xmath1 allows one to describe the configuration space @xmath54 as a pullback of two configuration spaces @xmath113 and @xmath84 , where the first is completely understood , and the second is simpler than the original @xmath54 . this idea may now be applied again to @xmath84 : by repeatedly discarding ( or adding ) open chain sub - mechanisms , we construct a sequence of pullbacks @xmath231\ar[r ] & { { \euscript c}}(\lambda_{n } ) \ar[d]^{\phi_{n}}\\ { { \euscript c}}(\gamma_{n})\ar[r]^{\psi_{n } } & { { { \mathbb r}}^{d}}~ , } } \ ] ] for @xmath232 , where each @xmath233 is a sub - mechanism of @xmath234 , with @xmath235 , and @xmath236 is an open chain in @xmath2 ( so @xmath237 is a product of @xmath128-spheres ) . the maps @xmath238 and @xmath239 are work maps for the common endpoint of @xmath234 and @xmath236 . each configuration @xmath14 for @xmath1 determines a sequence of pairs @xmath240 in @xmath241 , as in where @xmath213 is necessarily a smooth point of @xmath237 . evidently , if @xmath242 is a smooth point of @xmath243 , @xmath244 will be , too , if holds . [ rip ] note that there is usually more than one way to decompose a given linkage @xmath1 as in [ spb ] , so the full inductive process described above is actually encoded by an ( inverted ) rooted tree , with varying degrees at each node ( and the root at @xmath1 itself ) . any rooted branch @xmath245 ( @xmath246 ) of this tree will be called a _ decomposition _ of @xmath1 . this flexibility can be very useful in applying the inductive procedure ( see [ egsing ] below for an example ) . generic singularities in @xmath54[sgensinc ] our goal is to use this procedure to study singular configurations of @xmath0 . here we start with the simplest case , which is also the _ generic _ form of singularities in configuration spaces , as we shall see below . thus we assume by induction that @xmath242 is a smooth configuration , but _ fails_. our goal is to analyze this failure in the generic case , and then show that in this case @xmath244 is a singular point . eventually , we would like to use this to deduce that the original configuration @xmath14 is singular , too . in [ sgensin ] , we saw how to identify positively the generic singularities appearing in one step in the inductive process of [ sip ] , defined by a pullback diagram namely , if @xmath247 is defined by a pair of smooth configurations @xmath248 , but fails , then generically at least , @xmath244 is a topological singularity . however , this does not yet guarantee that the corresponding configuration @xmath14 in @xmath54 itself is singular ( unless @xmath249 , of course ) . [ egsing ] let @xmath250 be a planar closed @xmath251-chain with links of lengths @xmath252 , and @xmath253 . see figure [ ffourbar ] . = 4.0 cm generically , @xmath254 is a smooth @xmath5-dimensional manifold , with local parameter given by @xmath136 ( the angle between @xmath255 and @xmath256 , say ) . however , if @xmath257 , then @xmath254 has a topological singularity a node at the aligned configuration @xmath166 where the links @xmath255 and @xmath256 face right , say , and @xmath258 and @xmath259 face left ( see ( * ? ? ? * theorem 1.6 ) ) . in fact , if there are no further relations among @xmath260 , this is the only singularity , and @xmath254 is a figure eight ( the one point union of two circles ) . we can think of @xmath250 as being decomposed into two sub - mechanisms @xmath80 and @xmath261 , each an open @xmath9-chain : @xmath80 consisting of @xmath255 and @xmath258 , and @xmath261 of @xmath256 and @xmath259 . note that @xmath262 , where @xmath81 and @xmath263 are both aligned . in this case we can describe @xmath264 explicitly in terms of the work map @xmath265 ( for the vertex @xmath266 ) , which is a four - fold covering map at all points but @xmath14 : in a punctured neighborhood of @xmath14 , neither @xmath81 nor @xmath263 can be aligned , and each independently can have an `` elbow up '' ( @xmath267 ) or `` elbow down '' ( @xmath268 ) position , which together provide the four discrete configurations corresponding to a single value of @xmath230 . in @xmath254 , taken together , these give four different branches of the curve ( parameterized by @xmath136 ) which coincide at @xmath14 . see figure [ fbranch ] . now assume given a linkage @xmath1 in which @xmath269 as above ( with @xmath257 ) . assume that to obtain @xmath270 we add an open @xmath9-chain @xmath271 , having vertices @xmath72 , @xmath272 , and @xmath273 , with @xmath274 and @xmath275 . we therefore now have a rigid triangle @xmath276 ( with @xmath273 in `` elbow up '' or `` elbow down '' position relative to the edge @xmath277 ) . thus @xmath278 , and the singularity at @xmath279 is unaffected . in the last stage @xmath280 is obtained by adding another open @xmath9-chain @xmath281 with one new vertex @xmath282 , @xmath283 and @xmath284 . we require the configuration @xmath285 of @xmath286 in which @xmath287 , @xmath273 , and @xmath282 are aligned to coincide with the aligned configuration @xmath288 of @xmath289 ( and thus @xmath290 of @xmath270 ) . the effect of adding @xmath286 is to prevent the open chain @xmath291 from ever being in an `` elbow down '' position , thus eliminating two of the four branches of @xmath292 ( see figure [ fbranch ] ) , so @xmath293 ( which reduces to @xmath166 in @xmath294 ) is _ not _ singular in @xmath54 . to show that this is indeed so , consider an alternative decomposition of @xmath1 ( see remark [ rip ] above ) , in which we start with the closed @xmath68-chain @xmath295 , with base point @xmath272 . see figure [ faltdec ] . note that @xmath296 corresponding to @xmath166 is non - singular in @xmath297 . when we add the open @xmath9-chain @xmath298 , we see that the configuration @xmath299 corresponding to @xmath14 _ is _ aligned , but since the work map @xmath300 determined by the work point @xmath287 is a submersion at @xmath296 , condition holds at @xmath301 , so @xmath14 is smooth . = 5.0 cm singularities in the inductive process[ssip ] in example [ egsing ] we saw that a singularity appearing at one stage in the inductive process described in [ sip ] can disappear at a later stage . however , in that case the reason was that the aligned configuration @xmath285 of @xmath302 matched up in with the aligned configuration @xmath303 of @xmath270 . [ dgennt ] for any linkage @xmath1 , a configuration @xmath96 will be called _ generically non - transversive _ if for some decomposition @xmath304 of @xmath235 ( see [ rip ] ) , the pair @xmath305 is generically non - transverse in the sense of definition [ dgensing ] , and the open chain configurations @xmath306 are not aligned for @xmath307 . [ rgensnt ] as noted in remark [ rgensin ] , the condition that the original pair @xmath308 is generically non - transverse is indeed generic , in the sense that it occurs in a subvariety of @xmath309 of positive codimension . since the work maps each open chain @xmath310 are algebraic for each @xmath311 , the subvariety of @xmath312 consisting of pairs @xmath248 for which @xmath242 corresponds to @xmath313 ( and eventually to @xmath314 ) and @xmath213 is aligned form a subvariety of positive codimension , so the condition that @xmath14 is generically non - transversive in the sense of definition [ dgennt ] is indeed generic among the singular points of @xmath54 . [ tgensing ] for any linkage @xmath1 , a generically non - transversive configuration @xmath14 is a topological singular point of @xmath54 in fact , the product of a cone on a homogeneous quadratic hypersurface by a euclidean space . let @xmath308 be a generically non - transverse configuration of @xmath315 , so by proposition [ pgensing ] it is the cone on a homogeneous quadratic hypersurface . by induction on the decomposition @xmath304 , we may assume that at the @xmath168-th stage the configuration @xmath316 has a neighborhood @xmath39 of the stated form . by definition [ dgennt ] we know that the work map @xmath310 is a submersion at @xmath213 , so it is work - equivalent at @xmath213 ( definition [ dlocew ] ) to a projection @xmath317 ( see ( * ? ? ? * theorem 7.8 ) ) . therefore , in the pullback @xmath318 the configuration @xmath240 has a neighborhood @xmath319 which is again of the required form . [ rmechanism ] note that if @xmath235 has a decomposition @xmath304 as in [ rip ] and holds at @xmath240 for each @xmath168 , then the configuration @xmath320 is smooth , of course . thus we obtain a mechanical interpretation of _ all _ differentiable singularities in any configuration space : namely , they must occur at a kinematic singularity of type i for some sub - mechanism @xmath234 of @xmath1 that is , a ( smooth ) configuration @xmath316 at which the work map @xmath321 is not a submersion ( see @xcite ) . in fact , more than this is required , since at the same point @xmath14 another sub - mechanism namely , the open chain @xmath236 must be aligned , and it must be `` co - aligned '' with @xmath242 in the sense that together they are @xmath104-equivalent to an aligned closed chain ( see proof of proposition [ pgensing ] ) . we call this situation a _ conjunction _ of two kinematic singularities . we now consider an explicit example , which shows how all singular configurations of a certain type of planar linkage can be identified , by making use of a non - trivial @xmath104-equivalence . parallel polygonal linkages [ sppm ] in @xcite , a certain class of mechanisms were studied , called _ parallel polygonal linkages_. these consist of two polygonal _ platforms_. the first is the _ fixed _ platform , which is equivalent to fixing in @xmath2 the initial point @xmath322 of each of @xmath7 open chains ( called _ branches _ ) ( @xmath323 ) , of lengths @xmath324 , respectively . the terminal point @xmath325 of the @xmath326-th branch is attached to the @xmath326-th vertex of a rigid planar @xmath7-polygon @xmath327 , called the _ moving _ platform . see figure [ fpentagon ] . 1 . two of its branch configurations @xmath328 and @xmath329 are aligned , with coinciding direction lines @xmath330 . 2 . three of its branch configurations are aligned , with direction lines in the same plane meeting in a single point @xmath46 ( see figure [ ftriangle ] ) . [ rtypea ] in the type ( a ) singularity there is obviously a sub - mechanism @xmath80 which is isomorphic to an aligned closed chain , so the corresponding configuration @xmath81 is singular . evidently , the caveat exemplified in [ egsing ] does not apply here , so in fact @xmath14 is singular in @xmath54 . consider the sub - mechanism @xmath80 of @xmath1 obtained by omitting the third branch ( but retaining the fixed platform ) , with base point at @xmath332 ( the fixed endpoint of the omitted branch ) , and work point at @xmath333 ( the moving endpoint of this branch ) . let @xmath81 be the configuration of @xmath80 corresponding to @xmath14 of case ( b ) above ( so in particular the remaining two branches are aligned ) . assume that the first branch has @xmath334 links , and the second has @xmath335 links . we may then choose `` internal '' parameters @xmath336 for the first branch , and @xmath337 for the second branch ( as in the proof of proposition [ pzero ] ) . we can then express the lengths @xmath338 and @xmath339 as functions of @xmath336 and @xmath337 , respectively . note that @xmath80 has @xmath340 degrees of freedom , so one additional parameter is needed . two obvious choices are one of the `` base angles '' @xmath341 or @xmath342 for the two branches ( see figure [ ftrisub ] ) . let @xmath343 be the meeting point of the direction lines @xmath344 and @xmath345 for the two branches ( this is the point @xmath46 of figure [ ftriangle ] ) . as our additional parameter we take the angle @xmath136 between the direction line @xmath346 for the ( missing ) third branch and the line @xmath347 ( see figure [ ftheta ] ) . note that @xmath348 or @xmath349 in our special configuration @xmath81 . letting @xmath350 , the standard parametrization for the open @xmath6-chain @xmath351 defines a ( local ) diffeomorphism @xmath352 . in order to show that @xmath353 is a work - equivalent at @xmath81 to an aligned configuration @xmath354 of @xmath351 ( definition [ dlocew ] ) , we must show that @xmath81 is a generic singularity for @xmath80 that is , that the reduced work map @xmath355 has an ( isolated ) singularity at @xmath81 , where @xmath167 assigns to any configuration @xmath14 of @xmath80 the length @xmath356 . it is difficult to write @xmath167 explicitly as a function of @xmath136 : for this purpose it is simpler to use @xmath230 or @xmath357 as above . however , if we fix the lengths @xmath358 and @xmath359 of the direction vectors for the two branches , the resulting linkage @xmath360 is a planar closed @xmath251-chain with one degree of freedom ( parameterized by @xmath230 , say ) , and the third vertex @xmath361 of the moving triangle traces out a curve in @xmath74 , called the _ coupler curve _ for @xmath360 ( cf . * ch . 4 ) ) . therefore , the infinitesimal effect of a change in @xmath230 is the rotation of @xmath361 about the point @xmath343 described above ( called the instantaneous point of rotation for @xmath360 ) . in particular , the angle @xmath136 also changes , so we deduce that @xmath362 at the aligned configuration @xmath81 . this allows us to investigate the vanishing of @xmath363 instead of @xmath364 . this is the point where we are assuming genericity of @xmath81 : it might happen that the coupler curve is singular precisely at this point , in which case @xmath365 may vanish , so we are no longer guaranteed that @xmath136 is a suitable local parameter . but such instances of case ( b ) are not generic . since in the reduced configuration space @xmath366 we do not allow rotation of @xmath80 about the base - point @xmath367 , we may assume that @xmath368 and @xmath369 . write @xmath370 and @xmath371 for the ( fixed ) sides of the moving triangle ( with fixed angle @xmath372 between them ) , as in figure [ fparsub ] . we find that the following identities hold : @xmath373 ( where @xmath126 is the angle between side @xmath374 and the @xmath375-axis ) , and : @xmath376 after differentiating we find : @xmath377 and we deduce that @xmath363 vanishes if and only if : @xmath378 this formula expresses the fact that the area of the triangle @xmath379 is the sum of the areas of the quadrangle @xmath380 and @xmath381 , which holds if and only if @xmath382 are aligned . from the formulas for @xmath358 and @xmath359 we see that @xmath383 and @xmath384 all vanish at @xmath81 ( as for any aligned open chain ) , so in case ( b ) @xmath385 , taken with respect to @xmath386 . since all but one of the parameters are the standard internal angles for open chains , we can check that the morse indices for the reduced work maps of @xmath80 and @xmath351 match up at @xmath81 and @xmath387 , showing that @xmath353 is indeed a work - equivalence ( see proof of proposition [ pgensin ] ) . thus we may apply proposition [ pgensing ] to deduce that @xmath81 is a cone singularity . [ sexample ] since the caveat of [ egsing ] does not apply to case ( b ) , either ( cf . [ rtypea ] ) , we have shown that for a generic triangular planar linkage @xmath1 , any configuration @xmath96 satisfying one of the _ necessary _ conditions ( a ) and ( b ) of ( * ? ? ? * proposition 2.4 ) is ( @xmath104-equivalent to ) a generically non - transverse configuration ( definition [ dgensing ] ) . by theorem [ tgensing ] we can therefore deduce that it is indeed a topological singularity that is , conditions ( a ) and ( b ) are also _ sufficient_. abcd l. caporaso , `` geometry of tropical moduli spaces and linkage of graphs'',preprint , 2010 , arxiv:1001.2815 . r. connelly , e.d . demaine , & g. rote , `` straightening polygonal arcs and convexifying polygonal cycles '' , _ discrete comput . geom . _ * 30 * ( 2003 ) , pp . 205 - 239 . farber , _ invitation to topological robotics_,european mathematical society , zurich , 2008 . farber , s. tabachnikov , & s.a . yuzvinski , `` topological robotics : motion planning in projective spaces'',_int . . notices _ * 34 * ( 2003 ) , 1853 - 1870 . c. gosselin & j. angeles , `` singularity analysis of closed loop kinematic chains'',_ieee trans . on robotics & automation _ * 6 * ( 1990 ) , pp . 281 - 290 . gottlieb , `` topology and the robot arm'',_acta applied math . _ * 11 * ( 1988 ) , pp . 117 - 121 . a.s . hall , jr . , _ kinematics and linkage design_,prentice - hall , englewood cliffs , nj , 1961 . hausmann , `` sur la topologie des bras articuls'',in s. jackowski , r. oliver , & k. pawaowski , eds . , _ algebraic topology - poznn 1989 _ , lect . notes math.*1474 * , springer verlag , berlin - new - york , 1991 , pp . 146 - 159 . j .- hausmann & a. knutson , `` the cohomology ring of polygon spaces'',_ann . fourier ( grenoble ) _ * 48 * ( 1998 ) , pp . 281 - 321 . hirsch , _ differential topology_,springer - verlag , berlin - new york , 1976 . m. holcomb , `` on the moduli space of multipolygonal linkages in the plane'',_topology & applic . _ * 154 * ( 2007 ) , pp . 124 - 143 . hunt , _ kinematic geometry of mechanisms_,oxford university press , new york , 1978 . d. jordan & m. steiner , `` compact surfaces as configuration spaces of mechanical linkages'',_israel j. math . _ * 122 * ( 2001 ) , pp . 175 - 187 . y. kamiyama , `` the homology of the configuration space of a singular arachnoid mechanism'',_jp j. geom . _ * 7 * ( 2007 ) , pp . . y. kamiyama & m. tezuka , `` symplectic volume of the moduli space of spatial polygons'',_j . kyoto u. _ * 39 * ( 1999 ) , pp . 557 - 575 . y. kamiyama & s. tsukuda , `` the configuration space of the @xmath168-arms machine in the euclidean space '' , _ topology & applic . _ * 154 * ( 2007 ) , pp . 1447 - 1464 . m. kapovich & j. millson , `` on moduli space of polygons in the euclidean plane'',_j . * 42 * ( 1995 ) , pp . 430 - 464 . m. kapovich & j. millson , `` the symplectic geometry of polygons in euclidean space '' , _ j. diff . geom . _ * 44 * ( 1996 ) , pp . 479 - 513 . m. kapovich & j. millson , `` universality theorems for configuration spaces of planar linkages'',_topology _ * 41 * ( 2002 ) , pp . 1051 - 1107 . h.c . king , `` planar linkages and algebraic sets'',_turkish j. math . _ * 23 * ( 1999 ) , pp . lee , _ introduction to smooth manifolds_,springer - verlag , berlin - new york , 2003 . y. matsumoto , _ an introduction to morse theory_,american math . society , providence , ri , 2002 . j. p. merlet , _ parallel robots_,kluwer academic publishers , dordrecht , 2000 . r. j. milgram & j. c. trinkle , the geometry of configuration spaces for closed chains in two and three dimensions,_homology , homotopy & applic . _ * 6 * ( 2004 ) , pp . 237 - 267 . j. ohara , `` the configuration space of planar spidery linkages'',_topology & applic . _ * 154 * ( 2007 ) , pp . 502 - 526 . selig , _ geometric fundamentals of robotics_,springer - verlag _ mono . comp . _ , berlin - new york , 2005 . shafarevich , _ basic algebraic geometry , 1 . varieties in projective space_,springer - verlag , berlin - new york , 1994 . n. shvalb , m. shoham , h. bamberger , & d. blanc , `` topological and kinematic singularities for a class of parallel mechanisms'',_math . prob . in eng . _ * 2009 * ( 2009 ) , art . 249349 , pp . 1 - 12 . n. shvalb , m. shoham , & d. blanc , `` the configuration space of arachnoid mechanisms'',_fund . math . _ * 17 * ( 2005 ) , pp . 1033 - 1042 . n. shvalb , m. shoham and d. blanc , `` the configuration space of a parallel polygonal mechanism'',_jp j. geom . * 9 * ( 2009 ) , pp . 137 - 167 . l. w. tsai , _ robot analysis - the mechanics of serial and parallel manipulators_,wiley interscience publication - john wiley & sons , new york , 1999 . zlatanov , r.g . felton , & b. benhabib , `` singularity analysis of mechanisms and robots via a motion space model of the instantaneous kinematics '' , _ proceedings ieee conf . in robotics & automation _ ( 1994 ) , pp . 980 - 985 . zlatanov , i.a . bonev , & c.m . gosselin , `` constraint singularities as c - space singularities,'',_8th international symposium on advances in robot kinematics ( ark 2002 ) _ , caldes de malavella , spain , 2002 .

massive quiescent galaxies have much smaller physical sizes at high redshift than today . the strong evolution of galaxy size may be caused by progenitor bias , major and minor mergers , adiabatic expansion , and/or renewed star formation , but it is difficult to test these theories observationally . herein , we select a sample of 35 massive , compact galaxies ( m , m/kpc ) at in the cosmological hydrodynamical simulation illustris and trace them forward to to uncover their evolution and identify their descendants . by , the original factor of 3 difference in stellar mass spreads to a factor of 20 . the dark matter halo masses similarly spread from a factor of 5 to 40 . the galaxies evolutionary paths are diverse : about half acquire an ex - situ envelope and are the core of a more massive descendant , a third survive undisturbed and gain very little mass , 15% are consumed in a merger with a more massive galaxy , and a small remainder are thoroughly mixed by major mergers . the galaxies grow in size as well as mass , and only% remain compact by . the majority of the size growth is driven by the acquisition of ex - situ mass . the most massive galaxies at are the most likely to have compact progenitors , but this trend possesses significant dispersion which precludes a direct linkage to compact galaxies at . the compact galaxies merger rates are influenced by their environments , so that isolated or satellite compact galaxies ( which are protected from mergers ) are the most likely to survive to the present day . [ firstpage ] galaxies : high - redshift , galaxies : evolution

1507.02291

massive galaxies at high redshift have been observed to have properties which can be quite different from their counterparts at similar mass in the local universe . one of the most striking differences is the much smaller sizes observed for high - redshift galaxies . for example , a quiescent galaxy with a stellar mass of @xmath7 m@xmath1 , which has a typical effective radius of 8 kpc in the local universe , has a typical size of only 1.7 kpc at @xmath4 @xcite . early reports of these extremely small sizes based on their spatial extent @xcite have been confirmed by dynamical measurements of their stars @xcite and , more recently , gas @xcite , in addition to higher - resolution hst imaging @xcite . several observational studies have tracked the strong evolution of galaxy size with redshift @xcite , which is present in both star - forming and quiescent galaxies but is particularly dramatic in the latter case . part of the observed trend in median size is due to the continual addition of new , larger galaxies to the massive , quiescent population . star - forming disks have larger sizes at later times @xcite . as these galaxies quench , their appearance in the quiescent population can shift the median size without changing the number density of compact galaxies , a phenomenon known as progenitor bias " @xcite . careful accounting suggests that progenitor bias alone is insufficient to explain the observed size growth , however @xcite , so that the growth of individual galaxies within the quiescent population is also required . recently , some authors have found that compact galaxies may not be as rare in the local universe as previously thought @xcite , suggesting that they must continue to be produced to replace any which size out " of the compact selection . several authors have proposed mechanisms for increasing galaxies sizes after quenching . dry mergers between galaxies of similar masses increase size as @xmath8 @xcite , while minor mergers may increase the sizes even more efficiently , @xmath9 @xcite . in the latter case , it has been suggested that the massive , compact galaxies at high redshift now lie at the centers of the most massive galaxies in the local universe , having accreted a substantial amount of stellar mass since their initial formation @xcite or regrown a star - forming disk @xcite . this idea is supported by the observational fact that massive galaxies interior densities evolve much less strongly than their overall size @xcite . the observed minor merger rate , however , may be too low to be entirely responsible for galaxies growth @xcite . in the absence of mergers and accretion , galaxies may also puff up " due to adiabatic expansion in response to the expulsion of gas from the inner regions from stellar winds and black hole feedback as @xmath10 @xcite . these attempts to explain what happens to high - redshift compact galaxies are hindered by the long timescales of galactic evolution , which limit observations of any individual system to a single evolutionary state . as a result , galaxy populations can only be connected at different epochs in a statistical sense . simulations of galaxy formation , however , offer the opportunity to directly observe " the evolution of individual systems over time . large - volume simulations , moreover , can provide a cosmological context for that evolution and produce populations of galaxies which can be compared against observations . in @xcite , we identified a set of galaxies in the cosmological hydrodynamical simulation illustris which are analogs to observed massive compact ellipticals , with stellar masses of @xmath11 m@xmath1 and stellar half - mass radii of 2 kpc or smaller at @xmath4 . these compact galaxies formed naturally within the cosmological volume alongside non - compact galaxies at similar stellar masses , and we traced their progenitors back to higher redshift to discover how their formation paths differed . we found that the galaxies all reached their compact state through some combination of assembling their stellar mass at very early times when the universe was extremely dense , and/or intense , centralized bursts of star formation usually triggered by gas - rich mergers . both in our work and in that of other simulators ( see e.g. zoom - in simulations by @xcite and @xcite ) , gas - rich environments are required to dissipate angular momentum and produce such compact objects . we now turn our attention to the low - redshift outcomes of these massive , compact galaxies . our aim in this work is to show how a specific population of galaxies which were massive and compact at @xmath4 evolve forward to the present , describe their descendants , and characterize their evolutionary paths . as before , we select a massive , compact population at @xmath4 in illustris , this time tracing the galaxies forward to @xmath5 . in section [ sec : illustris ] , we describe the illustris simulation suite and the merger trees used to connect galaxies at different redshifts . in section [ sec : general ] , we select a population of 35 compact galaxies at @xmath4 and follow them forward to their @xmath5 descendants . in section [ sec : descendants ] , we take a detailed look at four distinct types of descendants and show how common each type of evolutionary track is , and in section [ sec : environment ] we discuss how a galaxy s @xmath4 environment might influence its eventual outcome . finally , in section [ sec : discuss ] we address the likelihood of massive galaxies in the local universe hosting compact cores , explore the different mechanisms for size growth , and discuss the implications for the number density of compact galaxies at low redshift . we summarize the results and conclude in section [ sec : conclusion ] . illustris @xcite is a cosmological hydrodynamical simulation with a periodic volume of ( 106.5 mpc)@xmath12 that is initialized at @xmath13 using cosmological intial conditions and evolved forward to the present day . the simulation suite consists of 6 runs in total : 3 different levels of resolution , each run once with dark matter particles only and once with full baryonic physics , including gas cells , stellar and wind particles , and black hole particles . the results presented in this paper are drawn from the highest - resolution full - physics run , illustris-1 , which has a baryonic mass resolution element of about @xmath14 m@xmath1 . the gravitational forces exerted by dm and star particles are softened on a comoving scale of @xmath15 = 1.4 kpc . for gas cells , the force softening is spatially adaptive with cell volume . for all baryonic resolution elements ( including both star particles and gas ) , @xmath15 is limited to a maximum physical size of 0.7 kpc at @xmath16 . in addition to gravity , the full - physics runs of illustris include a hydrodynamical treatment of gas using the moving - mesh code arepo @xcite and models for the baryonic physical processes that regulate the growth of stellar mass . these include primordial @xcite and metal - line gas cooling @xcite with self - shielding corrections @xcite , star formation and evolution , supernova feedback @xcite , chemical enrichment of the ism @xcite , and supermassive black hole growth with quasar - mode @xcite and radio - mode feedback @xcite . these models are described in detail by @xcite and @xcite , who show that they produce realistic galaxy populations and reproduce key observables across cosmological time . of particular relevance to the results presented herein are the models for star formation , stellar feedback , and agn feedback , which affect the sizes and quenching of galaxies . star - forming gas is modeled using a @xcite effective equation of state which forms stars stochastically once above a certain density threshold , according to a @xcite initial mass function . as stars evolve via winds and supernovae , mass and metals are returned to the gas as wind particles launched in a bipolar outflow with velocities scaled to the local dm velocity dispersion . supermassive black holes accrete mass from the surrounding gas at a rate that scales with the gas density and black hole mass and limited by the sound speed of the gas . at high accretion rates , black hole feedback operates in a quasar mode which thermally returns a portion of the accreted energy to the nearby gas , and at low accretion rates black hole feedback occurs mechanically by the injection of hot radio bubbles whenever the black hole has accreted a small additional fraction of its own mass . for more details , we refer the reader to @xcite . these physical models , in combination with those mentioned above , create galaxy populations which reproduce the stellar mass function , cosmic star formation rate , black hole mass scalings , and a mixture of galaxy morphologies and colors as observed in the real universe . the simulation does deviate from observations in a few ways that are relevant to this work in particular , the size - mass relation is too shallow at stellar masses @xmath17 m@xmath1 ( i.e. , galaxies below this mass have sizes which are too large ) , and the quenching of star formation appears to be incomplete at the high - mass end ( i.e. , the color bimodality is not as strong as observed , which is due to a low level of residual star formation post - quenching ) , see @xcite for details . the galaxies we focus on in this paper , however , are only lightly affected by these limitations as they lie in a mass regime where the sizes are in reasonable agreement with observations , and in most cases do not undergo much star formation at low @xmath18 . additionally , the softened gravity employed in illustris may result in lower central gas densities and lower central star formation rates , hence the stellar masses quoted herein may be lower , and the stellar half - mass radii larger , than they would be otherwise ( see @xcite and the convergence study in @xcite ) . the subfind algorithm @xcite is employed to identify gravitationally - bound structures ( i.e. dm halos and subhalos ) within the simulation volume . their baryonic components ( gas , stars , and black holes ) comprise the galaxies that inhabit each subhalo . subhalos are tracked from snapshot to snapshot using the sublink merger trees developed by @xcite . a subhalo s potential descendants in the following snapshot are identified by shared ownership of individual particles ( dm , stars , and gas ) . the particles are ranked according to gravitational binding energy and weighted according to ( rank)@xmath19 . the subhalo with the highest weighted sum over all shared particles is then assigned as the descendant . a merger occurs when two or more subhalos share a common descendant . the main ( or primary ) progenitor of this descendant is chosen according to which subhalo has the most massive history " @xcite , which protects against the misallocation of particles that can occur when subhalos are close to merging . we traverse this merger tree forward in time to find the low - redshift descendants of our compact galaxies . we quantify compactness " using the observationally - motivated ( see e.g. * ? ? ? * ) quantity @xmath20 where @xmath21 is a galaxy s stellar mass and @xmath22 its size . this constitutes a diagonal cut across the mass - radius plane so that lower - mass galaxies must have smaller sizes in order to qualify as compact . we use subhalo stellar masses and stellar half - mass radii to select a population of galaxies at @xmath4 with @xmath23 and stellar masses of @xmath11 m@xmath1 , consisting of 35 galaxies in total . this selection differs from ( and is more generous than ) the strict 2 kpc size threshold imposed in @xcite . with this @xmath24 criterion , a galaxy of @xmath7 m@xmath1 must still have a size of approximately 2 kpc or below , but we now allow larger sizes at higher masses so the total size of the compact population is more than doubled from the previous work . ( the original 14 galaxies are a subsample of the 35 galaxies we examine here . ) the selection criteria and resulting compact population are shown on the @xmath4 size - mass plane in the upper left - hand panel of figure [ fig : generalevol ] . note that this mass cut selects galaxies which already had stellar masses of @xmath7 m@xmath1 at @xmath4 , so that the results herein pertain only to high - mass compact galaxies . two - thirds of the compact galaxies thus selected are quiescent ( ssfr @xmath25/yr ) at @xmath4 , which is a broadly similar fraction as is found in observations @xcite . note that our definition of @xmath24 is not directly comparable to that used by observers , since we measure mass and radius in different ways . stellar masses are computed as a sum of the masses of all star particles bound to the subhalo rather than through sed fitting , though these masses are generally the same to within a factor of two @xcite . the radius used here is the stellar half - mass radius ( a three - dimensional quantity ) , rather than a projected two - dimensional half - light or effective radius ( which will be smaller by about 25% on average , see e.g. * ? ? ? * ) . in @xcite , we measured effective radii from sersic fits to mock observations for a small number of compact galaxies and found that the measured effective radii were 0.5 - 1 times the physical 3d radii , with an average difference of 0.74 as expected from projection . the use of 3d radii therefore has the effect of making our values of @xmath24 smaller by @xmath26 dex . the threshold compactness value of 10.5 that we employ here is thus more restrictive than the cutoff of 10.3 using 2d sizes imposed by @xcite by roughly @xmath27 dex , selecting for the very most compact galaxies in the simulation volume . ( see section [ ssec : generalize ] for a discussion of how our results depend on the choice of threshold . ) the descendants of these compact galaxies are traced forward in time using the merger trees described in section [ sec : illustris ] . figure [ fig : generalevol ] shows the properties of these descendants at @xmath28 and 0 . satellite galaxies at each redshift are marked with triangles , and a square denotes that the compact galaxy is not in the main progenitor branch of its descendant s merger tree at that redshift , i.e. that it merged with a more massive galaxy . comparing the @xmath5 and @xmath4 panels , a couple of conclusions can immediately be drawn . first , the compact galaxies @xmath5 descendants are dispersed widely among the massive galaxy population by a factor of @xmath620 in stellar mass despite having been clustered to within a factor of 3 at @xmath4 . some galaxies maintain a similar mass all the way to the present day , while others grow by a factor of 10 or more . this order of magnitude in the spread of galaxies descendant masses implies that rank order can not be preserved , a topic which we discuss in more detail in @xcite . additionally , when comparing the numbers of galaxies that fall below the @xmath29 line , it is apparent that there are many fewer compact galaxies by @xmath5 than there were at @xmath4 . as we will show in the following sections , processes that galaxies undergo at low redshift tend to increase their sizes as well as their masses . as a result , we predict that the comoving number density of massive , compact galaxies should decrease in the local universe ( see section [ ssec : numberdensities ] ) . the half - mass radius , by its mathematical definition , can be increased by ( i ) physically inflating the profile itself ( i.e. shifting stellar mass to larger radii and decreasing the density everywhere ) ( ii ) expelling mass from the inner regions , or ( iii ) adding mass to the outer regions . in the latter case , the half - mass radius will change while the core density remains the same , while the former two will also decrease the core density . in the middle row of panels in figure [ fig : generalevol ] , we show the stellar mass within 2 physical kpc of the galaxy center ( the core mass ) at the same set of snapshots . by definition , the compact galaxies sit on the high end of the distribution of core masses at @xmath4 . as time goes on , the typical core mass decreases for the compact galaxies as well as for the overall galaxy population . by @xmath5 , the compact galaxies descendants cores are less dense than their progenitors by about a factor of two on average , but they are still among the densest stellar cores in the simulation . ( the two obvious exceptions to this rule are massive galaxies which experienced strong interactions between @xmath30 that disrupted their cores . ) the evolution in core density is less pronounced than the evolution in overall size , indicating that while the first two mechanisms must be at work , the third is driving much of the total size growth . the masses of the dark matter halos in which the galaxies are embedded are shown in the lower set of panels of figure [ fig : generalevol ] , with the stellar mass halo mass relation appearing across the diagonal in dark pink points . the light grey points which lie above correspond to satellite galaxies . at @xmath4 , the compact galaxies are spread evenly across the scatter in the relation . at lower redshifts , the compact galaxies which still remain in the original stellar mass bin become skewed toward higher halo masses , suggesting that while the galaxies themselves are quiescent , their dark matter halos do continue to accrete and grow in mass . similar to the stellar masses , the original factor of @xmath31 spread in halo mass at @xmath4 has increased to a factor of 40 by @xmath5 . one of the few quantities that does _ not _ experience much dispersion from @xmath32 is the mass of the galaxies supermassive black holes , @xmath33 . at @xmath4 , the compact galaxies host black holes with @xmath34 m@xmath1 , which is overmassive by about a factor of 2.5 in comparison to the median @xmath33 in the @xmath0 m@xmath1 range in the simulation . by @xmath5 , the descendants bh masses have approximately doubled to @xmath35 m@xmath1 , with a shallower than linear relation between the descendants stellar and bh masses . at all descendant stellar masses , these bhs are on the high edge of the @xmath21 - @xmath33 relation , and at lower stellar masses they become more extreme outliers . at the lowest descendant stellar masses in our study of 35 galaxies , the descendants that still lie in the same range @xmath0 m@xmath1 host bhs with masses of @xmath36 m@xmath1 , which is a factor of @xmath37 higher than the median @xmath33 in that range at @xmath5 in the simulation . as discussed in @xcite , in the models employed by illustris both the black hole accretion and central star formation rate are functions of the density and temperature of the central gas , and so the black hole accretion rate tends to track the central star formation rate . thus , galaxies which are dominated by stars formed close to the center will also tend to have high black hole masses . these results echo observations by @xcite , who find that ber - massive " smbhs in the local universe are hosted by massive , compact relic galaxies with old stellar populations . we have seen in figure [ fig : generalevol ] that the compact galaxies can grow in stellar mass by varying amounts between @xmath4 and @xmath5 , some gaining no mass at all while others grow by a factor of 10 or more . in figure [ fig : deltam ] , we show a detailed accounting of what drives this change for each galaxy . in each panel , the galaxies are ordered by their total change in mass relative to their @xmath4 mass ( the quantity shown in the top panel ) . the panels below break down that total change into the contributions from _ ( i ) _ ex - situ stellar mass formed elsewhere and brought in by mergers or accretion since @xmath4 , _ ( ii ) _ stars formed in situ since @xmath4 , _ ( iii ) _ the decrease in stellar mass over time due to the death of massive stars , and _ ( iv ) _ the loss of stellar mass due to stripping . ( at the top of the figure , the numbers show the index of each galaxy when they are ordered by absolute descendant mass , corresponding to figures elsewhere in the paper . colors correspond to the type of descendant , which will be discussed in section [ ssec : categorize ] . ) the classification of star particles as in - situ or ex - situ is identical to that of rodriguez - gomez et al . ( to be submitted ) : a particle is considered to have formed in situ if the galaxy in which it formed lies along the main branch of the descendant s merger tree . otherwise , it is considered ex - situ . in cases where the compact progenitor itself does not lie along the main branch of its descendant at @xmath5 , we identify stars formed in situ using the classification from the last snapshot when the compact progenitor _ did _ lie along the main branch . a star particle is considered to have been stripped if it was present in the compact galaxy at @xmath4 but absent at @xmath5 . note that the first and second panels for the total change in stellar mass and the mass of ex - situ stars are scaled to 15 while the remaining three panels are scaled to 1 , and that they appear very similar except at the far right of the figure . when the stellar mass changes by more than about 50% ( which is the case for 75% of the compact galaxies ) , the change is completely dominated by the acquisition of ex - situ mass . insets in the top right of these panels zoom in on the low - mass end on the same scale as the bottom three panels . below , the middle panel shows the relative stellar mass growth due to in - situ star formation after @xmath4 . several of the compact galaxies were caught in the act " at @xmath4 and were still star - forming at that time , though they quench ( i.e. , reach specific star formation rates below @xmath38 gyr@xmath19 and continue to fall ) by @xmath39 . this residual star formation between @xmath4 and @xmath39 is marked by hashing , and a few galaxies ( e.g. 11 , 26 , and 32 ) continued to form up to 20% additional stellar mass during this time . in most cases the galaxies remain quiescent after quenching , with a couple noticeable cases of rejuvenation ( e.g. 19 and 28 ) which also gain them up to 20% more stellar mass . the final two panels show means of losing stellar mass rather than gaining it . the fourth panel shows the change in stellar mass due to stellar evolution , i.e. massive stars post - main - sequence evolution and death via supernovae . this results in a decrease in stellar mass of about 15% , which is fairly constant across the compact population . finally , the bottom panel shows how much stellar mass the galaxies have lost since @xmath4 through stripping ( i.e. , the total mass of all star particles which were present at @xmath4 but are missing at @xmath5 ) . for most of the compact galaxies , stripping is inconsequential , but for the three galaxies at the low - mass end it is actually significant enough that they have a net loss of stellar mass . in the most extreme case , galaxy 35 dove through the middle of another halo and was dramatically stripped of 60% of its stars , leaving only the most bound part of the core behind as it emerges from the other side at @xmath5 . galaxies 28 and 34 are both satellites at @xmath5 and have been more gently stripped through interactions with their central galaxy . in general , the galaxies change in stellar mass between @xmath4 and @xmath5 can be largely attributed to ex - situ mass acquired through mergers and interactions , with other effects being sub - dominant . in the previous section , we have seen that the compact galaxies , originally tightly grouped at @xmath4 , can accrue widely differing amounts of stellar mass by @xmath5 . what events in a galaxy s evolutionary history drive that assembly of stellar mass , and where do the stars from the compact progenitor lie within the resulting stellar distribution ? figure [ fig : imagegrid ] displays images of the @xmath5 descendant of each of the 35 compact galaxies , ordered by decreasing descendant mass . the color in a given pixel is determined by two quantities , stellar surface density and compact fraction , as shown in the legend in the bottom right panel . the color s intensity ( how light or dark it is ) indicates the stellar surface density ( the total mass of all the star particles along the line of sight ) so that denser regions are darker . the hue indicates how many stars in that pixel belong to the compact progenitor ( the compact fraction " ) so that regions dominated by stars from the original compact galaxy are red and regions dominated by other stars are blue . ( in detail , the compact fraction is calculated as the fraction of stars in that pixel which were either present in the compact progenitor at @xmath4 or formed in - situ before @xmath39 . ) in some cases ( e.g. 5/6 and 17/18 ) , two compact galaxies have the same descendant . in the top row , the galaxies 5 and 6 were both consumed by the same massive galaxy , and in the third row galaxies 17 and 18 merged together . in such cases , the two panels will then have identical surface densities but differing compact fractions , since they refer to different progenitors . seeing the descendants in this way again highlights the variety of experiences they undergo . near the top where the most massive descendants reside , in many cases the compact progenitor galaxy can not be readily identified as a single object , indicating that it was tidally shredded through an interaction with a more massive galaxy . in several of these , a band of green highlights the plane on which the compact galaxy was disrupted . in others , the core of the galaxy is orange or red while the outskirts are blue , indicating that the compact galaxy was able to maintain coherence throughout the merger process(es ) and exists as the core of its descendant . moving further down in mass , the panels which are primarily red in color are compact galaxies which experienced little merger or star formation activity between @xmath4 and @xmath5 . a more quantitative view of the galaxies structure can be found in the appendix figure [ fig : profgridz0 ] , which shows the cumulative stellar mass profiles of the @xmath5 descendants broken down into its original , ex - situ , and in - situ components . to classify the variety of descendants quantitatively , we define @xmath40 the fraction of stellar mass interior to a radius @xmath41 belonging to the compact progenitor . ( this compact fraction is measured in the same way as in figure [ fig : imagegrid ] , but includes all stellar mass enclosed within @xmath41 rather than the stellar mass in a single pixel . ) we then measure @xmath42 at radii of 2 kpc and 20 kpc ( chosen to be representative of the core and envelope respectively ) and use these quantities to inform our classification . a galaxy whose core is compact - dominated , for example , will have a high @xmath43 . if it also has a high @xmath44 , it is essentially the same galaxy , while a low @xmath44 means that it has acquired an ex - situ envelope . we thus define four general classes of descendant , as depicted at @xmath5 in the top - left panel of figure [ fig : categories ] : * * consumed : * @xmath45 , the compact galaxy s identity was erased by accretion onto a more massive descendant . * * mixed : * @xmath46 , the compact galaxy experienced major merger activity and is partially disrupted . * * cores : * @xmath47 and @xmath48 , the compact galaxy exists as the core of its more massive descendant . * * undisturbed : * @xmath47 and @xmath49 , the compact galaxy has entirely preserved its identity . the specific choice of radii as well as the dividing lines between classes are somewhat arbitrary , particularly between the core and undisturbed classes . however , we find that the classification is not very sensitive to our specific adopted radii as long as one radius is below 3 kpc and the other is at least 8 kpc . moreover , upon visual inspection of the mass profiles the appropriate class is obvious in most cases , so the boundaries can be chosen accordingly . ( see figure [ fig : indivexamples ] for archetypal examples in each class and figure [ fig : profgridz0 ] for the full set of profiles . ) at @xmath5 , of the original 35 compact galaxies , 5 have been consumed , " 2 are mixed , " about half are cores , " and 11 are undisturbed . " the top - right panel of figure [ fig : categories ] shows how the classes developed over time . all the galaxies are initially undisturbed ( by definition ) , and over time the fraction of compact galaxies which develop outer stellar envelopes increases , leveling off at about 50% . compact galaxies undergo mixing or consumption at a lower rate . note that galaxies may transition between classes more than once one galaxy , for example , is mixed " in a major merger and then later consumed . " we have found that using radii which scale with @xmath4 mass as @xmath50 produces the same results , so that within this small initial mass range the use of constant radii does not impose a mass - dependent bias . the bottom panels in figure [ fig : categories ] show again the compact galaxies at @xmath4 and their descendants at @xmath5 on the mass - radius plane , this time colored according to the class to which the descendant belongs . at the low - mass end at @xmath5 , the descendants all belong to the undisturbed " class in red , since they have gained very little stellar mass . at the high - mass end , those descendants which have consumed a compact galaxy are shown in blue . between those extremes lie compact galaxies which are cores ( in orange ) and which were thoroughly mixed by merging ( in green ) . ( note that there is not always a one - to - one correspondence between @xmath4 and 0 because some of the compact galaxies have the same descendant . ) examples showing the evolution of an individual galaxy from each of these classes are shown in figure [ fig : indivexamples ] and discussed in the following subsection . each of the four sets of panels in figure [ fig : indivexamples ] is composed of three items demonstrating how a single compact galaxy from @xmath4 developed the characteristics that determined its descendant group at @xmath5 . the bottom left panel in each frame shows the evolution in stellar mass , with major mergers marked with a vertical dotted grey line and the stellar mass ratio listed at the top of the panel . large grey points correspond to the six images above , which are colored using the same scheme as in figure [ fig : imagegrid ] where a color s intensity indicates stellar surface density and hue indicates the fraction of stars belonging to the compact progenitor . the image farthest to the left depicts the compact progenitor at @xmath4 , and the image farthest to the right depicts its final @xmath5 descendant . the cumulative stellar mass profile of this descendant is plotted in the panel in the lower right of each frame with a black line . the red line , for contrast , shows the cumulative stellar mass profile of only those stars belonging to the compact progenitor . the ratio of the red to black line is the quantity @xmath42 described in section [ ssec : categorize ] . grey dashed lines mark the locations of @xmath43 and @xmath44 which were used to classify the descendants . each of these examples demonstrates the kind of event ( or lack thereof ) that determines the final descendant type , and is representative of that class . in the first ( blue ) frame in figure [ fig : indivexamples ] which shows the evolution of galaxy 4 , the stellar mass shows a sudden jump around 9 gyr which marks the point in time when the compact galaxy merges into a much more massive galaxy . as it is falling in , its outer stellar envelope is slowly stripped away before the core finally falls into the central object . the images above show that the compact galaxy is completely torn apart and its identity erased , though it is still faintly visible at @xmath5 as a band of green tracing the orbital plane along which it was disrupted . five of the 35 compact galaxies ( 14% ) fall into this class at @xmath5 . their @xmath4 progenitors tend to fall into the lower - mass end of the selection range , which is not surprising since the less massive a galaxy is , the more likely that a merger partner will be capable of disrupting it . interestingly ( and perhaps unexpectedly ) , the regrowth of a star - forming disk is visible in the second image of the sequence . this is not an uncommon event among the objects in our sample , with disk regrowth occuring at some point for about a quarter of them . these disks are very minor in terms of their stellar mass when compared to the mass of the core , but would undoubtedly make up a more significant portion of the light during their star - forming period . the second ( green ) frame shows a galaxy ( number 17 ) which was primarily undisturbed until about 12 gyr . at that time , it experienced a merger with another galaxy of similar mass . as the two galaxies spiraled in together , their stars became thoroughly mixed . as is visible in the profiles on the right , stellar material from the compact progenitor is distributed uniformly throughout the final descendant and makes up a large fraction of its total mass the two galaxies are able to mix together so efficiently because they have not only similar masses , but similar binding energies in fact , the galaxy which merges is also in the compact sample ( number 18 ) ! as mergers between galaxies of nearly the same mass and binding energy are rare , galaxies 17 and 18 are the only two of the sample of 35 ( 6% ) which fall into this class of descendant . an example of the third and most common type of descendant is shown in the third ( orange ) frame . this galaxy ( number 19 ) experienced two major mergers between @xmath4 and 0 , along with several more minor mergers . at 6 gyr , it disrupts another galaxy of 75% its mass and gains a thin stellar envelope . later , around 11 gyr , it does the same to another loosely bound massive galaxy at a similar mass ratio , ultimately residing at the core of a final descendant many times its original mass . these 3:4 merger events are not so different in terms of stellar mass ratio from the 1:1 merger in the previous ( mixed ) case . the crucial difference between the two cases is not necessarily mass , but gravitational binding energy . in dissipationless mergers , it has been shown that particles tend to preserve their rank order in binding energy , with the most tightly bound particles of the merger progenitors remaining the most tightly bound particles of the descendant @xcite . the compact galaxies , with their large masses and small sizes , are very tightly bound . thus , they are more likely to survive a merger intact even in cases of near unity mass ratios , and it requires a galaxy of similar or greater mass to unbind these objects ( as in the previous two examples ) . 17 of the original 35 compact galaxies ( 49% ) survive as the cores of more massive descendants in this manner . the evolution of the final object ( galaxy 29 , red frame ) is remarkable in its uneventfulness . this galaxy maintains the same stellar mass throughout its evolution , gaining stars neither through accretion , mergers , nor in - situ star formation . eleven objects from our sample of 35 ( 31% ) are similarly able to avoid further action and thus completely maintain their same identity from @xmath4 ( though not all are quite so pristine ) . though these galaxies do not grow in stellar mass , they do grow in size , a topic which is further discussed in section [ ssec : sizeevol ] . what determines which evolutionary path a compact galaxy will take ? clearly , mergers are a very important component of their growth . the compact galaxies are primarily quiescent at @xmath4 and retain low specific star formation rates throughout their lifetimes , so most of their mass growth is due to the acquisition of ex - situ stars through mergers and interactions . figure [ fig : environment ] shows the relative growth in both mass ( @xmath51-axis ) and size ( @xmath52-axis ) for the compact sample between @xmath4 and 0 . in the first panel , galaxies are colored according to their descendant type , as in figure [ fig : categories ] . the bottom left region of this panel is dominated by the undisturbed ( red ) compact galaxies which have experienced little size and mass growth , while the rightmost galaxies are those which were consumed ( blue ) by their very massive descendants . ( the outlying red symbol which has grown significantly in size but not much in mass is galaxy 20 , which has some ex - situ mass which was deposited at radii of tens of kpc or greater . the steepness of the density profile at intermediate radii then forces the half - mass radius to expand substantially to enclose enough mass to compensate . ) in the middle panel , the symbol color indicates the number of mergers with stellar mass ratios of at least 1:10 that the galaxies have experienced since @xmath4 . the gradient in color from red to blue with increasing mass growth shows that , as expected , the more merger activity a galaxy experiences , the more stellar mass it gains and the more likely it is to fall into the core " or consumed " groups . for illustris galaxies . color indicates the maximum compactness ( @xmath24 ) of all the galaxy s progenitors along the main branch of its merger tree above @xmath53 m@xmath1 . solid colored lines mark a constant value of @xmath24 . galaxies which have satisfied @xmath54 ( the compactness criterion employed in this paper ) _ at any redshift _ are marked with black circles . ( over half of these are the same 35 which were compact at precisely @xmath4 . ) _ middle panel : _ same as above , but with the y - axis indicating the galaxies formation redshifts , defined as the median formation redshift over all in - situ stars . _ lower panel : _ cumulative distribution of @xmath55 for all galaxies , binned by stellar mass and normalized against the number of galaxies in each bin . thus , given a value of @xmath24 ( e.g. @xmath56 , marked with a vertical dashed line ) , one can determine what fraction of galaxies at a given stellar mass have ever achieved that level of compactness . there is a clear relationship between stellar mass and past ] compactness more massive galaxies have more compact progenitors but the trend has significant dispersion which is partially due to galaxies differing formation times . [ fig : maxpastcompact ] in turn , a compact galaxy s merger rate is a function of its environment . the rightmost panel of fig . [ fig : environment ] demonstrates this correlation by coloring the same points according to the density of galaxies in the compact progenitors @xmath4 environments , quantified as the distance to the 5th nearest neighbor with an r - band magnitude of at least -19.5 ( corresponding to a stellar mass of about @xmath57 m@xmath1 ) as in @xcite . the same trend appears in this panel as in the middle one , indicating that a compact galaxy s environment has some power to predict how it will evolve . for this subsample of galaxies , it appears that in denser , more crowded environments , the galaxies are more likely to merge with their neighbors of comparable mass . the more isolated ( red ) a compact galaxy is , the fewer mergers it will undergo , and the less mass growth it will experience . the descendants of the galaxies which were in isolation at @xmath4 and experienced little merger activity in the interim are also preferentially isolated at @xmath5 . the connection between galaxy merger rate and environment , which was investigated here only for this particular set of galaxies which were massive , compact , and primarily quiescent at @xmath4 , will be explored in more detail for all illustris galaxies by rodriguez - gomez et al . ( in prep ) . two interesting exceptions to the merger rate / environment correlation are the white triangles in the lower left corner , which were in moderately dense environments at @xmath4 but experienced very little mass growth . at @xmath5 , these two are satellite galaxies which have fallen into a nearby massive dark matter halo in the intervening 10 gyr but have not ( yet ) merged with the central galaxy . a similar dm - only result is reported by @xcite , who find that the most compact dm structures in the simulation are substructures embedded in larger halos . based on this evidence , we predict that massive compact relics in the local universe should preferentially live in regions where they are protected from mergers , either in isolation or as satellites in a larger group . the latter option also has some observational support from @xcite , who have found that nearby massive compact galaxies tend to live in cluster environments . given that about half of the compact galaxies from @xmath4 exist at @xmath5 as the cores of very massive galaxies , one can ask the reverse question : how many massive galaxies have cores which were once compact galaxies ? we address this question by tracing all the @xmath5 massive galaxies through the merger tree and calculating @xmath24 at each point on its main progenitor branch . we then define the maximum past compactness " @xmath55 to be the maximum value . ] of @xmath24 when the progenitor had a stellar mass of at least @xmath53 m@xmath1 . in contrast to the analysis thus far , this approach permits a galaxy to be compact at any redshift , rather than specifically tying them to @xmath4 . this allows , for example , a galaxy which forms very early at @xmath58 in a compact state but grows out of the compact population by @xmath4 , or a galaxy which forms via a compact starburst at @xmath39 , both of which would be missed by the preceding analysis . the colors of the points on the mass - size plane in the upper panel of figure [ fig : maxpastcompact ] indicate @xmath55 for all of the massive galaxies at @xmath5 . lines of constant @xmath24 are drawn in the same color scheme . any galaxy whose primary progenitor met the @xmath29 compactness criterion at any point in time is outlined in black . about half of these were compact at exactly @xmath4 and overlap with the sample discussed elsewhere in this paper , while the other half experienced their compact phases at other redshifts . from this scatter plot , there is a clear trend between mass and maximum past compactness : more massive galaxies tend to have more compact progenitors . ( see also @xcite for a discussion of the relationship between mass and compactness in illustris . ) thus , many of the most massive galaxies were indeed once compact galaxies according to our definition but it is interesting to note that this is not universally the case , since there is also considerable scatter in the distribution of @xmath55 at this mass ( and , in fact , at all masses ) . one cause of the spread in @xmath55 at a given mass can be seen in the second panel , which shows the galaxies formation times . here , we have defined formation time " as the median formation redshift of all star particles in the galaxy which formed _ in - situ_. ( this excludes stars acquired via mergers . ) as before , color indicates @xmath55 . in addition to the trend with stellar mass , a trend between formation time and past compactness is clearly visible this is a direct reflection of the fact that galaxies form with smaller sizes at earlier times . thus , the distribution of formation times for galaxies of a given mass also manifests as a distribution of past compactness . the relationship between mass and @xmath55 is further quantified in the lower panel of figure [ fig : maxpastcompact ] . each line shows the cumulative fractional distribution of @xmath55 for the galaxies in a given stellar mass bin . as was suggested by the upper panel , moving to lower masses shifts the distribution to lower past compactness values . for any @xmath24 , we can now see what fraction of galaxies at that mass have ever met that threshold . the orange line denotes the @xmath29 criterion used elsewhere in this paper . approximately 55% of galaxies with @xmath59 m@xmath1 have met this definition of compact , " along with 30% of 6 - 10 @xmath60 m@xmath1 galaxies , 15% of 3 - 6 @xmath60 m@xmath1 galaxies , etc . although a substantial fraction ( 55% ) of very massive galaxies are former compact galaxies , it is important to note that few of these were compact at exactly @xmath4 . rather , their compact phases are spread over @xmath61 , so that only about @xmath62 of galaxies above @xmath63 m@xmath1 at @xmath5 have compact progenitors at @xmath4 . similarly , of the 35 @xmath4 compact galaxies , only 3 ( 9% ) have direct descendants in that mass range at @xmath5 . these fractions demonstrate once again that while there is indeed a strong trend between final stellar mass and past compactness , there is also a large dispersion in galaxies evolutionary tracks that can not be neglected . nearly all of the compact galaxies have grown in size as well as mass since @xmath4 , with very few still remaining in the compact regime at @xmath5 . potential origins of this growth include physical effects such as major mergers @xcite , minor mergers and accretion @xcite , star formation @xcite , and adiabatic expansion due to the expulsion of gas from the galaxy through stellar winds and bh feedback @xcite , as well as numerical effects stemming from the non - constant physical size of the gravitational softening length in illustris . to examine the importance of each of these mechanisms for the size growth of the different descendant types , let us return to the same four example galaxies we discussed in section [ ssec : individs ] . the left - hand panels of figure [ fig : sizeexamples ] show the evolution of their stellar half - mass radii with time . in addition to the black line for total stellar half - mass radius ( which includes all stars bound to the object ) , we also show a red line for the the compact half - mass radius " that is , the half - mass radius including only those stars which belonged to the compact progenitor at @xmath4 . this allows us to disentangle the growth of the compact core from the growth of the galaxy as a whole . in the first ( consumed ) case , both the overall and compact half - mass radii jump up when the compact galaxy falls into the more massive central galaxy . this indicates , as we have seen , that the compact galaxy s stellar mass has been dispersed throughout the descendant . in a similar ( though less extreme ) way , the second ( mixed ) galaxy s compact half - mass radius increases abruptly when it undergoes a disruptive major merger . in these cases , the size growth is primarily driven by merging . conversely , in the third ( core ) case the compact half - mass radius hardly changes despite plenty of mergers and stellar mass growth , while the total half - mass radius greatly increases . the very small change in compact half - mass radius signifies that the core has remained intact , and that the vast majority of the overall size growth is driven by the acquisition of outer layers of stellar material from less - bound galaxies . finally , the galaxy which is undisturbed grows only slightly in size from 2 to 3 kpc over this 10 gyr period . in this case , the stellar mass has changed very little as a small amount of new stars ( @xmath610% , both in - situ and ex - situ ) offset the mass loss from preexisting stars . here , size growth mechanisms such as adiabatic expansion from mass loss and numerical effects start to become significant . in treating the gravitational forces between baryons , illustris uses a gravitational softening length @xmath64 which is a fixed comoving size at redshifts @xmath65 . the increase of the physical scale on which the potential is softened will decrease the depth of the gravitational well in dense regions , artificially inflating the stellar mass profiles as the star particles become less gravitationally bound . it is unlikely that the increasing @xmath64 is the sole culprit , however , since it is capped at a fixed physical size at @xmath16 but the size growth does not completely halt at this time . in the appendix , we estimate how these compact stellar systems will expand adiabatically in response to the changing softening length , stellar mass loss , and gas expulsion . we find that numerical relaxation " of the galaxies @xmath4 profiles to the @xmath5 softening length should only change the half - mass radius by about @xmath62 . adiabatic expansion as a response to stellar mass loss and gas expulsion should produce another @xmath66% growth , and the remaining @xmath67% growth is owed to the small amount of in - situ stars formed and ex - situ stars deposited at larger radii . returning to the compact population as a whole , the right - hand panel of figure [ fig : sizeexamples ] shows the growth in compact half - mass radius since @xmath4 ( @xmath51-axis ) against the ratio between the overall and compact half - mass radii ( @xmath52-axis ) for each @xmath5 descendant . the @xmath51-axis quantity is identical to the ratio between the value of the red lines in the left - hand panel at @xmath5 and @xmath4 , while the @xmath52-axis quantity is the ratio between the black and red lines at @xmath5 . the 30% growth in the compact half - mass radius from the changing softening length , stellar mass loss , and expulsion of gas between @xmath4 and @xmath5 that we estimate in the appendix is marked by the vertical dotted line . @xmath51-values significantly higher than this indicate some disruption of the compact core , like the mergers that caused jumps in the compact half - mass radius in the first and second examples on the left . all of the consumed " galaxies , whose stellar contents have been scattered far and wide , are therefore located on the far right of this panel . a high @xmath52-value indicates the presence of significant ex - situ stellar mass at large radii , so that the core "- type descendants separate from the undisturbed " descendants despite having similar @xmath51-values in many cases . in general , the growth in compact half - mass radius is much smaller than that of the overall half - mass radius . this implies that a majority of the size growth is due to the addition of stellar mass in the outer parts of the galaxies . galaxies which gained little stellar mass typically grow by about 50% or less in compact half - mass radius , while those which experienced some merger events are slightly more perturbed , and those which experienced majorly disruptive mergers grow even more . as we have seen in the previous subsection , nearly all of the compact galaxies grow in size between @xmath4 and @xmath5 even if they experience little growth in mass . of the original 35 compact galaxies , only two remain below the compactness threshold at @xmath5 . shifting the threshold up slightly to account for the 15% change in half - mass radius from numerical effects yields one additional galaxy , for a final survival fraction " of 9% for compact galaxies from @xmath4 to the present day . ( and of these three , one is an extreme case which dives through a massive halo , is stripped of its stellar envelope along the way , and is just escaping out the other side at @xmath5 , having managed to _ decrease _ in both size and mass ! ) the number density of compact galaxies depends not only on the rate at which galaxies _ leave _ the compact population , but also the rate at which they _ enter_. our previous study @xcite found mechanisms for producing compact galaxies which are most efficient in the high - redshift universe when cold , dense gas is abundant . as time increases and the universe becomes more rarefied , we expect that the production rate of compact galaxies should also decrease . in section [ ssec : cores ] , we identified all the galaxies in the simulation which had a massive ( @xmath68 m@xmath1 ) , compact progenitor at any redshift . for each of these galaxies , we have measured the median redshift of the compact phase as well as its total duration . the median redshifts of the compact phases are broadly distributed from @xmath61 , falling off at higher and lower @xmath18 and implying that the compact formation mechanisms indeed no longer operate as efficiently at low redshift . half of the compact phases lasted less than 2 gyr , and only 10% were longer than 5 gyr , indicating that we should expect few massive , compact galaxies to remain in the local universe . both of these exercises ( choosing compact galaxies at @xmath4 and tracing them forward , and finding ex - compact " galaxies at @xmath5 and examining the time and duration of their compact phases ) indicate that the absolute number density of high - mass compact galaxies should decrease at low redshift . physically , this is a reflection of the growing dominance of dissipationless processes ( e.g. dry mergers ) at late times over processes such as disk instability or wet mergers , more common at high redshift , which provide a way to dissipate angular momentum and thus produce more compact stellar populations . studies of compact massive high - redshift galaxies adopt varying initial selection methods , including differences in the compactness criterion ( e.g. the slope and normalization of the power - law cut across the mass - radius plane ) , minimum stellar mass , quiescence ( e.g. cuts in ssfr , uvj , or morphology ) , size estimation ( e.g. major - axis vs. circularized half - light radius , bulge - disk decomposition vs. single - sersic fits ) , and parent surveys ( e.g. sdss , candels , cosmos ) . these differences can make direct comparisons between observational surveys that infer the the mass or size evolution of massive compact galaxies difficult . in this paper , we have primarily focused on the _ physical _ mass and size evolution of compact massive galaxies as realized in a cosmological simulation , which introduces an additional level of complexity when making comparisons . in this subsection , we discuss some of the specific definitions used in this paper and their impact on our conclusions , which may help the reader to place these results in the context of other work , particularly observational work . one question which plagues both simulators and observers is , what is the most meaningful way to talk about size ? in this paper we have opted to use the stellar half - mass radius unambiguously defined as the radius of the sphere containing half of the stellar mass which is gravitationally bound to this object " as our fiducial measurement of size . using the stellar half - mass radius as a proxy for galaxy size has the disadvantages that it is not a directly observable quantity , and that it can be perturbed by the addition of stellar mass at some larger radius when the density profile is steep . however , the half - mass radius has clear physical meaning , can be easily and robustly calculated directly from subfind data , and is related to the circularized half - light radius ( modulo variations in the mass - to - light ratios for differently - aged stellar populations ) . in principle we could have adopted galaxy size estimates by fitting profiles to the stellar light distributions of every galaxy in the simulation and measuring an effective or scale radius ( e.g. , as done in * ? ? ? however , given the complexity and variety of the simulation data the resulting radii would likely be sensitive to several details of the fitting procedure and would likely not yield characteristically different trends from those found and presented in this paper . nevertheless , it could be worthwhile to consider in the future a more even - handed comparison of galaxy size evolution for compact massive galaxies . another choice we have made in the above analysis is a particular selection of compactness threshold value , @xmath29 . we mentioned in section [ sec : general ] that this selection is more severe than that employed by e.g. @xcite . how sensitive are the results to this particular choice of compactness " ? to explore this question , we repeated the analysis with a more comparable threshold of @xmath69 , which yielded a sample of 103 galaxies in the same @xmath11 m@xmath1 mass range . these galaxies are more likely to be star - forming than the original sample of 35 ( only 42% of them have specific star formation rates lower than @xmath70/yr , as opposed to 66% in the original sample ) , so more of them quickly leave the compact regime due to star formation at large radii . when this population is traced forward , we find that they are more likely than the original sample to have descendants in the mixed " or consumed " groups , since they are larger ( less tightly bound ) and hence easier to disrupt via merger . overall , however , we find that the same trends hold as have been described above , and therefore it seems that the qualitative conclusions drawn about the diversity of growth paths for massive compact galaxies is not highly sensitive to how they are initially selected . a more implicit set of assumptions adopted in this paper is the physical / feedback model employed by the illustris simulation . the masses and sizes of the simulated galaxy populations examined in this paper are specifically subject to influence from the finite mass resolution and spatial softening , pressurization of the ism , and agn feedback model . we discuss in some detail the potential effects of gravitational softening ( and its redshift evolution ) in the appendix of this paper . the gravitational softening and ism pressurization adopted in this paper can potentially prevent galaxies from fully collapsing during dissipational events such as gas - rich galaxy mergers . in general , both gravitational softening and our method of ism pressurization risk overestimating the sizes of galaxies and it is therefore possible that some galaxies that would at one time be compact would be missed . thus , our conclusions about the fractions of galaxies which pass through a compact phase might best be viewed as a lower limit . however , the basic convergence study presented in @xcite indicated that the sizes of galaxies are well - converged at the @xmath4 gravitational softening level , so the results should not dramatically change . our ability to further investigate the impact of these simulation traits on our results is limited since we can not easily re - run the illustris simulation volume . reconsidering the conclusions presented in this paper with other simulations ( e.g. eagle , * ? ? ? * ; * ? ? ? * ) or with zoom - in re - simulations of the specific objects studied in this paper or objects selected from other large full - volume simulations may provide additional context for these results in the future . finally , we would like to emphasize that the results herein are primarily focused on very massive galaxies with stellar masses of at least @xmath7 m@xmath1 at @xmath4 , and that caution is warranted when comparing to results in other mass regimes where different physical processes may govern the evolution . herein , we have identified a population of 35 massive ( @xmath11 m@xmath1 ) galaxies at @xmath4 in the cosmological hydrodynamical simulation illustris which satisfy the compactness criterion @xmath71 . we have traced these massive , compact galaxies forward in time through the simulation to their @xmath5 descendants at the present day . our most important findings are summarized below : * though the compact galaxies all originated at @xmath4 with the same stellar mass to within a factor of 3 , by @xmath5 their stellar masses differ up to a factor of 20 . similarly , their dm halo masses , originally within a factor of 5 , spread to a factor of 40 . rather than remaining tightly grouped , the descendants stellar masses are dispersed throughout the entire massive galaxy population at @xmath5 . * the compact galaxies can undertake a variety of evolutionary paths to @xmath5 . approximately half survive as the core of a more massive descendant , about 15% are consumed and destroyed in a merger with a much more massive galaxy , about a third are generally undisturbed by mergers , accretion , or star formation activity , and the remaining few are thoroughly mixed by major mergers . these percentages are a weak function of initial stellar mass , with less massive galaxies more likely to be consumed and more massive galaxies more likely to be cores . * very few ( about 10% ) of the galaxies from the @xmath4 compact sample still satisfy our compactness criterion at @xmath5 , suggesting a decrease in the number density of massive , compact galaxies at low redshift in the absence of any replenishing compact formation mechanism . * a compact galaxy s @xmath4 environment has some predictive power for its final evolutionary outcome . compact galaxies in denser environments ( with the exception of satellite galaxies ) are more likely to undergo mergers and gain stellar mass at larger radii , losing their compact identity in the process . those which are isolated at @xmath4 and experience little merger activity are also preferentially isolated at @xmath5 . thus we predict that massive , compact relics in the local universe should predominantly live in underdense environments or as satellites in larger groups . * a galaxy s @xmath5 stellar mass is related to the compactness of its progenitors , such that more massive galaxies are more likely to have once been compact . however , this trend possesses significant dispersion ( partially driven by a spread in formation times ) and it is _ not _ the case that the most massive galaxies at @xmath5 can be directly linked to massive compact galaxies at @xmath4 . while more than half the galaxies with stellar mass @xmath72 m@xmath1 at @xmath5 have a direct compact progenitor at @xmath73 , only 15% have a compact progenitor at exactly @xmath4 . * the dominant source of size growth for the compact galaxies is the addition of ex - situ stellar mass at larger radii through mergers and accretion into the stellar halo , followed by disruptive major mergers , with smaller contributions from ongoing or renewed star formation , adiabatic expansion due to mass loss , and/or numerical effects . sw is supported by the national science foundation graduate research fellowship under grant number dge1144152 . pt acknowledges support from nasa atp grant nnx14ah35 g . cpm acknowledges support from nasa grant nnx11ai97 g and nsf grant ast-1411945 . ap acknowledges support from the hst grant hst - ar-13897 . sg acknowledges support provided by nasa through hubble fellowship grant hst - hf2 - 51341.001-a awarded by the stsci , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas5 - 26555 . lh acknowledges support from nasa grant nnx12ac67 g and nsf grant ast-1312095 . to estimate the effect of a changing gravitational softening length @xmath15 on the matter distributions of galaxies which are undisturbed between @xmath32 , we mimic the process of adiabatic expansion that is invoked as a response to mass loss from the center of the potential on timescales longer than a dynamical time , @xmath74 . expansion occurs in such a way as to conserve particles specific angular momentum @xmath75 , which is an adiabatic invariant . if the distribution is spherically symmetric and particles move on circular orbits , @xmath76 . changing the gravitational softening length will have much the same effect not by physically removing mass , but by changing how much of it an external particle feels . " given a distribution of mass @xmath77 and the scale @xmath15 on which gravity is softened , we can define an effective mass enclosed " @xmath78 which decreases with increasing @xmath15 . consider the case where gravitational forces are plummer - softened , so that the gravitational force per unit mass @xmath79 exerted by a parcel of mass @xmath80 on a test particle is @xmath81 where @xmath82 is the distance between the mass and the test particle , and @xmath15 is the gravitational softening length . ( illustris employs a slightly different spline potential ( see @xcite ) but the plummer potential is much more analytically pliable and is similar enough for our purposes here . ) if the test particle lies at a distance @xmath22 from a spherically - symmetric distribution of mass with density profile @xmath77 , we can integrate over a series of concentric shells in @xmath41 to determine the total gravitational force . each shell consists of a series of rings around the line connecting the test particle and the center , so that on a given ring ( described by the angle @xmath83 from the connecting line ) , all points are the same distance @xmath82 from the test particle and the same distance @xmath41 from the center and hence exert the same force @xmath84 along the connecting line . by the law of cosines @xmath85 and by the law of sines @xmath86 so that the gravitational force integrated over the entire sphere interior to @xmath22 is @xmath87 the integral over @xmath83 has an analytical solution @xmath88 dr \\ & \equiv \frac{g}{r^2 } \int_0^r 4 \pi r^2 \rho(r ) \beta(r , r , \epsilon ) dr \\ & \equiv \frac{g m_{\rm eff}(r , \epsilon)}{r^2}\end{aligned}\ ] ] where the density correction function " @xmath89 is the ratio between the actual mass _ present _ in a shell at radius @xmath41 and the mass _ felt _ by a test particle at @xmath22 if gravity is plummer - softened on scales of @xmath15 . the form of @xmath90 will vary with the density profile @xmath77 and is in general not analytically solvable , so it must be evaluated numerically . we can directly apply the above to any arbitrary mass profile , including those we recover from the simulation . in practice , we would like to answer the following question : if a galaxy has a stellar mass profile @xmath77 when the softening length is @xmath91 , what will it be after the softening length is ( slowly ) changed to @xmath92 ? to relax the profile to a new softening length , we first measure the adiabatic invariant @xmath93 at shells of @xmath94 in the initial profile , as well as recording the masses @xmath95 between the shells . we then change the softening length to @xmath92 , remeasure @xmath96 and compare it against the original value . starting with the innermost shell , we iteratively change @xmath94 until the adiabatic invariant is within some error tolerance of its initial value ( placing @xmath95 halfway between the shells new locations ) , then move on to the next shell and work our way out . the results of performing this exercise on the compact galaxies @xmath4 profiles are shown in figure [ fig : profgridsoft ] . each galaxy actually has two components with differing @xmath92 : the ( dominant ) stellar component which reaches its maximum softening length of 0.7 kpc at @xmath16 , and the dark matter component which continues to soften to 1.4 kpc at @xmath5 . here we show the combined effect of both softenings on the stellar distribution , with the original @xmath4 profiles in red and the softened profiles in blue . on average , the changing softening length alone results in a change in half - mass radius of about @xmath62 for these galaxies . in addition to this numerical mass loss , " the galaxy will also physically lose mass as old stars evolve off the main sequence and explode as supernovae , and gas is expelled from the galactic center in outflows caused by stellar and black hole feedback . these processes will also cause the stellar system to adiabatically expand . in the green line in each panel , we repeat the above exercise with the additional step of decreasing the stellar mass in each shell by 15% ( the typical amount lost due to stellar evolution between @xmath4 and @xmath5 for this sample , see figure [ fig : deltam ] ) the stellar mass loss causes an additional inflation of the half - mass radius of about 12% . ( note that if there were no expansion in response to the mass loss , the half - mass radius would , by definition , not change , since the loss is applied equally at all radii . ) finally , in the black line we remove any gas that was present at @xmath4 , since these systems tend to be gas - poor at @xmath5 , and the radius increases slightly by another 3% . for comparison , the _ actual _ profiles of the same stars at @xmath5 are shown with dashed black lines . in many cases ( e.g. galaxies 21 - 26 and 31 - 34 ) , the profile constructed by adiabatic relaxation under these few simple assumptions does a reasonably good job of reproducing the actual profile . these cases correspond to evolutionary tracks where the compact core was not much perturbed , which includes both undisturbed " galaxies as well as core " galaxies . significant deviation from the estimated profile suggests events which were not taken into account in this simple model , such as major mergers .

we study general quantum waveguides and establish explicit effective hamiltonians for the laplacian on these spaces . a conventional quantum waveguide is an-tubular neighbourhood of a curve in and the object of interest is the dirichlet laplacian on this tube in the asymptotic limit . we generalise this by considering fibre bundles over a-dimensional submanifold with fibres diffeomorphic to , whose total space is embedded into an-neighbourhood of . from this point of view takes the role of the curve and that of the disc - shaped cross - section of a conventional quantum waveguide . our approach allows , among other things , for waveguides whose cross - sections are deformed along and also the study of the laplacian on the boundaries of such waveguides . by applying recent results on the adiabatic limit of schrdinger operators on fibre bundles we show , in particular , that for small energies the dynamics and the spectrum of the laplacian on are reflected by the adiabatic approximation associated to the ground state band of the normal laplacian . we give explicit formulas for the according effective operator on in various scenarios , thereby improving and extending many of the known results on quantum waveguides and quantum layers in . eberhard karls universitt tbingen , mathematisches institut , auf der morgenstelle 10 , 72076 tbingen , germany .

1402.1067

quantum waveguides have been studied by physicists , chemists and mathematicians for many years now and the rate at which new contributions appear is still high ( see @xcite and references therein ) . mathematically speaking , a conventional quantum waveguide corresponds to the study of the dirichlet laplacian on a thin tube around a smooth curve in @xmath1 . of particular interest are effects of the geometry of the tube on the spectrum of and the unitary group generated by the laplacian . similarly so - called quantum layers , i.e. the laplacian on a thin layer around a smooth surface , have been studied @xcite . the related problem of the constraining of a quantum particle to a neighbourhood of such a curve ( or surface ) by a steep potential rather than through the boundary condition was studied in @xcite . recently , progress has also been made on quantum waveguides and layers in magnetic fields @xcite . there are obvious geometric generalisations of these concepts . one can consider the dirichlet laplacian on small neighbourhoods of @xmath4-dimensional submanifolds of @xmath10 ( see e.g. @xcite ) , or of any @xmath11-dimensional riemannian manifold . another possibility is to look at the laplacian on the boundary of such a submanifold , which , in the case of a conventional waveguide , is a cylindrical surface around a curve in @xmath1 . beyond generalising to higher dimension and codimension , one could also ask for waveguides with cross - sections that change their shape and size along the curve , or more generally along the submanifold around which the waveguide is modelled . in the majority of mathematical works on quantum waveguides , with the exception of @xcite , such variations of the cross - section along the curve must be excluded . the reason is that , physically speaking , localizing a quantum particle to a thin domain leads to large kinetic energies in the constrained directions , i.e. in the directions normal to the curve for a conventional waveguide , and that variations of the cross - section lead to exchange of this kinetic energy between normal and tangent directions . however , the common approaches require that the laplacian acts only on functions that have much smaller derivatives in the tangent directions than in the normal directions . in this paper we show how to cope with several of the possible generalisations mentioned above : ( i ) we consider general dimension and codimension of the submanifold along which the waveguide is modelled . ( ii ) we allow for general variations of the cross - sections along the submanifold and thus necessarily for kinetic energies of the same order in all directions , with possible exchange of energy between the tangent and normal directions . ( iii ) we also include the case of `` hollow '' waveguides , i.e. the laplacian on the boundary of a general `` massive '' quantum waveguide . all of this is achieved by developing a suitable geometric framework for general quantum waveguides and by the subsequent application of recent results on the adiabatic limit of schrdinger operators on fibre bundles @xcite . as concrete applications we will mostly emphasise geometric effects and explain , in particular , how the known effects of `` bending '' and `` twisting '' of waveguides in @xmath1 manifest themselves in higher dimensional generalised waveguides . before going into a more detailed discussion of our results and of the vast literature , let us review the main concepts in the context of conventional quantum waveguides in a geometrical language that is already adapted to our subsequent generalisation . moreover , it will allow us to explain the adiabatic structure of the problem within a simple example . consider a smooth curve @xmath12 parametrised by arclength with bounded second derivative @xmath13 and its @xmath0-neighbourhood @xmath14 for some @xmath15 . by @xmath16 we denote the image of the curve in @xmath1 and we call @xmath17 the tube of a conventional waveguide . the aim is to understand the laplace operator @xmath18 on @xmath19 with dirichlet boundary conditions on @xmath17 in the asymptotic limit @xmath20 . as different metrics will appear in the course of the discussion , we make the euclidean metric @xmath21 explicit in the laplacian . for @xmath0 small enough one can map @xmath17 diffeomorphically onto the @xmath0-tube in the normal bundle of @xmath7 . in order to make the following compuations explicit , we pick an orthonormal frame along the curve . a natural choice is to start with an orthonormal basis @xmath22 at one point in @xmath7 such that @xmath23 is tangent and @xmath24 are normal to the curve . then one obtains a ( in this special case global ) unique frame by parallel transport of @xmath22 along the curve @xmath7 . this construction is sometimes called the relatively parallel adapted frame @xcite . the frame @xmath25 satisfies the differential equation @xmath26 with the components of the mean curvature vector @xmath27 ( @xmath28 ) given by @xmath29 the two normal vector fields @xmath30 form an orthonormal frame of @xmath7 s normal bundle @xmath31 . hence , for @xmath15 small enough , there is a canonical identification of the @xmath0-tube in the normal bundle denoted by @xmath32 with the original @xmath0-tube @xmath33 via the map @xmath34 we will refer to @xmath35 as the cross - section of @xmath36 and to @xmath37 as the cross - section of @xmath17 at @xmath38 . in order to give somewhat more substance to the simple example , let us generalise the concept of a conventional waveguide already at this point . for a smooth function @xmath39 $ ] with @xmath40 let @xmath41 be the tube with varying cross - section @xmath42 , a disc of radius @xmath43 . this gives rise to a corresponding tube @xmath44 in @xmath1 . to not overburden notation , we will drop the subscript @xmath45 in the following , i.e. put @xmath46 and @xmath47 . by equipping @xmath36 with the pullback metric @xmath48 , we can turn @xmath49 into an isometry . then the dirichlet laplacian @xmath50 on @xmath19 is unitarily equivalent to the dirichlet laplacian @xmath51 on @xmath52 . in order to obtain an explicit expression for @xmath51 with respect to the bundle coordinates @xmath53 associated with the orthonormal frame @xmath54 , we need to compute the pullback metric @xmath55 on the tube @xmath36 . for the coordinate vector fields @xmath56 and @xmath57 , @xmath58 , one finds @xmath59 here , we used @xmath60 and the differential equations . knowing that @xmath61 is an orthonormal frame of @xmath62 with respect to @xmath21 , this yields @xmath63 the laplace - beltrami operator on @xmath36 associated to @xmath55 is thus @xmath64 with @xmath65 . as the riemannian volume measure of @xmath55 on coordinate space reads @xmath66 , it is convenient to introduce the multiplication operator @xmath67 with the density @xmath68 as a unitary operator from @xmath52 to @xmath69 . here @xmath70 just denotes lebesgue measure on the coordinate space . a straightforward computation shows that @xmath71 with @xmath72 and @xmath73 the rescaling by @xmath74 thus leads to a simpler split up of the laplacian @xmath75 into a horizontal operator @xmath76 and a vertical operator @xmath77 given by the euclidean laplace operator with dirichlet boundary conditions on the cross - sections @xmath42 . this simplification is , however , at the expense of an additional potential @xmath78 . since @xmath42 is isometric to the disc of radius @xmath79 , the eigenvalues and eigenfunctions of the vertical operator @xmath77 are @xmath0-dependent and also functions of @xmath38 . in order to arrive at an @xmath0-independent vertical operator and an @xmath0-independent domain , one dilates the fibres of @xmath31 using @xmath80 and its associated lift to a unitary operator mapping @xmath81 to @xmath82 . one then arrives at @xmath83 with the second order differential operator @xmath84 and the bending potential @xmath85 these terms account for the bending of the curve in the ambient space , i.e. its extrinsic geometry . in order to make the asymptotic limit @xmath86 more transparent , we rescale units of energy in such a way that the transverse energies are of order one by multiplying the full laplacian with @xmath87 . in summary one finds that @xmath88 is unitarily equivalent to the operator @xmath89 acting on the domain @xmath90 with @xmath91 . the hamiltonian @xmath92 thus splits into the horizontal laplacian @xmath93 , the vertical laplacian @xmath94 and a additional differential operator @xmath95 that will be treated as a perturbation . this structure is reminiscent of the starting point for the born - oppenheimer approximation in molecular physics . there the @xmath96-coordinate(s ) describe heavy nuclei and @xmath87 equals the inverse mass of the nuclei . the @xmath97-coordinates describe the electrons with mass of order one . in both cases the vertical resp . electron operator depends on @xmath96 : the vertical laplacian @xmath98 in the quantum waveguide hamiltonian @xmath92 depends on @xmath38 through the domain @xmath99 and the electron operator in the molecular hamiltonian depends on @xmath96 through an interaction potential . this suggests to study the asymptotics @xmath20 for quantum waveguide hamiltonians by the same methods that have been successfully developed for molecular hamiltonians , namely by adiabatic perturbation theory . the latter allows to seperate slow and fast degrees of freedom in a systematic way . in the context of quantum waveguides the tangent dynamics are slow compared to the frequencies of the normal modes . to illustrate the adiabatic structure of the problem , let @xmath100 be the smallest eigenvalue of @xmath101 on @xmath102 and denote by @xmath103 the corresponding normalised non - negative eigenfunction , the so - called ground state wave function . let @xmath104 be the subspace of local product states and @xmath105 the orthogonal projection onto this space . now the restriction of @xmath92 to the subspace @xmath106 is called the adiabatic approximation of @xmath92 on the ground state band and the associated adiabatic operator is defined by @xmath107 a simple computation using @xmath108 and the unitary identification @xmath109 , @xmath110 , shows that the adiabatic operator @xmath111 can be seen as an operator acting only on functions on the curve @xmath7 , given by @xmath112 where @xmath113 and @xmath114 . as such it is a one - dimensional schrdinger - type operator with potential function @xmath115 and the asymptotic limit @xmath20 corresponds to the semi - classical limit . this analogy shows that , in general , @xmath116 can not be considered small compared to @xmath117 , despite the factor of @xmath87 . to see this , observe that all eigenfunctions @xmath118 ( and also all solutions of the corresponding time - dependent schrdinger equation ) are necessarily @xmath0-dependent with @xmath119 , unless @xmath120 for some constant @xmath121 . to be more explicit , assume that @xmath122 near a global minimum at @xmath123 . then the lowest eigenvalues of @xmath124 are @xmath125 for @xmath126 . while the level spacing of order @xmath0 is small compared to @xmath127 , it is large compared to the energy scale of order @xmath87 of the geometric potentials . and for states @xmath118 with @xmath128 the kinetic energy @xmath129 in the tangential direction is of order one . however , the majority of mathematical works on the subject considers the situation where @xmath130 is of order @xmath87 . clearly this only yields meaningful results if one assumes @xmath131 for some constant @xmath121 . but this , in turn , puts strong constraints on the possible geometries of the waveguide , which we avoid in the present paper . now the obvious mathematical question is : _ to what extent and in which sense do the properties of the adiabatic operator @xmath124 reflect the corresponding properties of @xmath92 ? _ this question was answered in great generality in @xcite and we will translate these results to our setting of generalised quantum waveguides in section [ sec : mainresults ] . roughly speaking , theorem [ thm : low spectrum ] states that the low - lying eigenvalues of @xmath124 approximate those of @xmath92 up to errors of order @xmath132 in general , and up to order @xmath133 in the special case of @xmath131 . in the latter case the order @xmath132 terms in @xmath124 turn out to be significant as well . our main new contribution in this work is to introduce the concept of generalised quantum waveguides in section [ chap : general ] and to compute explicitly the adiabatic operator for such generalised waveguides to all significant orders . for massive quantum waveguides , which are basically `` tubes '' with varying cross - sections modelled over submanifolds of arbitrary dimension and codimension , this is done in section [ chap : massiveqwg ] . there we follow basically the same strategy as in the simple example given in the present section . we obtain general expressions for the adiabatic operator , from which we determine the relevant terms for different energy scales . though the underlying calculations of geometric quantities have been long known @xcite , the contribution of @xmath134 has usually been neglected , because at the energy scale of @xmath87 it is of lower order than @xmath135 . this changes however on the natural energy scale of the example @xmath136 with non - constant @xmath45 , where they may be of the same order , as we see in section [ sect : h_a alpha ] . the contribution of the bending potential is known to be non - positive in dimensions @xmath137 or @xmath138 @xcite , while it has no definite sign in higher dimensions @xcite . it was stressed in @xcite that this leads to competing effects of bending and the non - negative `` twisting potential '' in quantum waveguides whose cross - sections @xmath102 are all isometric but not rotationally invariant and twist along the curve relative to the parallel frame . the generalisation of this twisting potential is the adiabatic potential @xmath139 , which is always non - negative and of the same order as @xmath135 . using this general framework we generalise the concept of `` twisted '' waveguides to arbitrary dimension and codimension in section [ sect : twist ] . in section [ chap : hollowqwg ] we finally consider hollow waveguides , which are the boundaries of massive waveguides . so far there seem to be no results on these waveguides in the literature and the adiabatic operator derived in section [ sect : h_a hollow ] is completely new . for hollow waveguides the vertical operator is essentially the laplacian on a compact manifold without boundary and thus its lowest eigenvalue vanishes identically , @xmath140 . the adiabatic operator on @xmath9 is quite different from the massive case . up to errors of order @xmath132 , it is the sum of the laplacian on @xmath7 and an effective potential given in . for the special case of the boundary of @xmath141 discussed above this potential is given by @xmath142 = { \varepsilon}^2 \left[\tfrac12 \tfrac{f''}{f } - \tfrac14 \big(\tfrac{f'}{f}\big)^2\right]\,,\ ] ] which , in contrast to massive waveguides , is independent of the curvature @xmath143 and depends only on the rate of change of vol@xmath144 . one can check for explicit examples that a local constriction in the tube , e.g. for @xmath145 , leads to an effective potential with wells . thus , constrictions can support bound states on the surface of a tube . in this part we give a precise definition of what we call generalised quantum waveguides . in view of the example discussed in the introduction , the ambient space @xmath1 is replaced by @xmath11-dimensional euclidean space and the role of the curve is played by an arbitrary smooth @xmath4-dimensional submanifold @xmath5 . the generalised waveguide @xmath3 is contained in a neighbourhood of the zero section in @xmath31 which can be diffeomorphically mapped to a tubular neighbourhood of @xmath146 . we will again call @xmath147 the cross - section of the quantum waveguide at the point @xmath38 and essentially assume that @xmath102 and @xmath148 are diffeomorphic for @xmath149 . this allows for general deformations of the cross - sections as one moves along the base , where in the introduction we only considered scaling by the function @xmath45 . in order to separate the laplacian into its horizontal and vertical parts , we follow the strategy of the previous section . however , it will be more convenient to adopt the equivalent viewpoint , where we implement the scaling within the metric @xmath150 on @xmath151 instead of shrinking @xmath152 and keeping @xmath153 fixed . for the following considerations , we assume that there exists a tubular neighbourhood @xmath154 with globally fixed diameter , i.e. there is @xmath155 such that normals to @xmath7 of length less than @xmath156 do not intersect . more precisely , we assume that the map @xmath157 restricted to @xmath158 is a diffeomorphism to its image @xmath159 . again , this mapping @xmath49 provides a metric @xmath160 on @xmath161 and the rescaled version @xmath162 , where @xmath163 is the dilatation of the fibres in @xmath31 . [ def : highdimqwg ] let @xmath5 be a smooth @xmath4-dimensional submanifold with tubular neighbourhood @xmath164 and @xmath8 be a compact manifold with smooth boundary and @xmath165 . suppose @xmath166 is a connected subset that is a fibre bundle with projection @xmath167 and typical fibre @xmath8 such that the diagram m & & b + ^_m & & _ _ b + b & _ _ b&b commutes . we then call the pair @xmath168 , with the scaled pullback metric @xmath169 a _ generalised quantum waveguide_. it immediately follows from the commutative diagram that the cross - sections @xmath102 coincide with the fibres @xmath170 given by the fibre bundle structure . from now on we will usually refer to this object simply as the fibre of @xmath3 over @xmath96 . although other geometries are conceivable , the most interesting examples of generalised waveguides are given by subsets @xmath171 of codimension zero and their boundaries . in the following we will only treat these two cases and distinguish them by the following terminology : [ def : massivehollow ] let @xmath172 be a generalised quantum waveguide as in definition [ def : highdimqwg ] . 1 . we call @xmath3 _ massive _ if @xmath8 is the closure of an open , bounded and connected subset of @xmath173 with smooth boundary . we call @xmath3 _ hollow _ if @xmath174 and there exists a massive quantum waveguide @xmath175 such that @xmath176 . this definition implies @xmath177 , i.e. each fibre @xmath178 of a hollow quantum waveguide is the boundary of @xmath179 , the fibre of the related massive waveguide @xmath180 . we denote by @xmath181 the vertical subbundle of @xmath182 . its elements are vectors that are tangent to the fibres of @xmath3 . we refer to the orthogonal complement of @xmath183 ( with respect to @xmath184 ) as the horizontal subbundle @xmath185 . clearly @xmath186 and this decomposition will turn out to be independent of @xmath0 . that is , the decomposition @xmath187 is orthogonal for every @xmath15 . furthermore , we will see ( lemma [ lem : pullback ] and equation for the massive case , equation for hollow waveguides ) that the scaled pullback metric is always of the form @xmath188 where * @xmath189 is the induced riemannian metric on the submanifold @xmath7 , * @xmath190 is a symmetric ( but not necessarily non - degenerate ) tensor with @xmath191 for any vertical vector field @xmath192 , * @xmath193 is the @xmath0-independent restriction of the scaled pullback metric to its vertical contribution . thus , if we define for any vector field @xmath194 its unique horizontal lift @xmath195 by the relation @xmath196 , we have @xmath197 for all @xmath198 and @xmath199 . [ ex : conv ] in the introduction we considered a massive waveguide @xmath200 with @xmath137 and @xmath201 . the typical fibre was given by @xmath202 . using the bundle coordinates @xmath53 induced by , one easily checks the the scaled pullback metric reads @xmath203 hence @xmath204 finally remark that @xmath205 is the typical fibre of the associated hollow waveguide . after having introduced the geometry of a generalised waveguide @xmath168 , we now analyse the laplace - beltrami operator @xmath206 with dirichlet boundary conditions . the boundary condition is of course vacuous if @xmath3 is hollow since @xmath207 in that case . similarly to the introduction , we apply a unitary @xmath208 ( which equals @xmath209 in the earlier notation ) . the according scaled density is given by @xmath210 we call the metric @xmath211 the ( scaled ) submersion metric on @xmath3 , since it turns @xmath212 into a riemannian submersion . the transformed laplacian then reads @xmath213 here , the horizontal laplacian is defined by its quadratic form ( with @xmath214 ) @xmath215 where @xmath216 denotes the orthogonal projection to @xmath217 , so integration by parts yields ( see also section [ sect : laplace ] ) @xmath218 the vertical operator is given on each fibre @xmath102 by the laplace - beltrami operator @xmath219 with dirichlet boundary conditions . the bending potential @xmath220 is a by - product of the unitary transformation @xmath221 and the second order differential operator @xmath222 accounts for the corrections to @xmath223 . in this section we show that the adiabatic operator @xmath111 approximates essential features of generalised quantum waveguide hamiltonians @xmath92 , such as its unitary group and its spectrum . this motivates the derivation of explicit expansions of @xmath111 in the subsequent sections . in this work we will only consider the ground state band @xmath224 and pay special attention to the behaviour of @xmath92 for small energies . this , as we will show , allows to view @xmath111 as an operator on @xmath9 . the results of this section were derived in @xcite in more generality . for a massive quantum waveguide set @xmath225 and for a hollow waveguide @xmath226 let @xmath227 be the smallest eigenvalue of the fibre operator @xmath228 acting on the fibre over @xmath96 . for hollow waveguides we have no boundary and @xmath229 with the eigenfunction @xmath230 in the massive case we have @xmath231 and denote by @xmath232 the uniquely determined positive normalised eigenfunction of @xmath233 with eigenvalue @xmath224 . let @xmath105 be the orthogonal projection in @xmath234 defined by @xmath235 the image of this projection is the subspace @xmath236 of @xmath234 . the function @xmath237 and its derivatives , both horizontal and vertical , are uniformly bounded in @xmath0 . thus , the action of the horizontal laplacian @xmath238 on @xmath237 gives a term of order @xmath0 and @xmath239p_0 = [ h^{\varepsilon}- h_f , p_0]p_0=\mathcal{o}({\varepsilon})\label{eq : p0comm}\ ] ] as an operator from @xmath240 to @xmath234 . since this expression equals @xmath241 , this justifies the adiabatic approximation for states in the image of @xmath105 . however , the error is of order @xmath0 , while interesting effects of the geometry , such as the potentials @xmath139 and @xmath242 discussed in the introduction , are of order @xmath87 . because of this it is desirable to construct also a super - adiabatic approximation , consisting of a modified projection @xmath243 and an intertwining unitary @xmath244 with @xmath245 , such that the effective operator @xmath246 provides a better approximation of @xmath92 than @xmath124 does . it then turns out that the approximation provided by @xmath247 can also be made more accurate than expected from using the unitary @xmath244 . such approximations can be constructed and justified if the geometry of @xmath168 satisfies some uniformity conditions . here we only spell out the conditions relevant to our case , for a comprehensive discussion see @xcite . the generalised quantum waveguide @xmath168 is a _ waveguide of bounded geometry _ if the following conditions are satisfied : 1 . the manifold @xmath248 is of bounded geometry . this means it has positive injectivity radius and for every @xmath249 there exists a constant @xmath250 such that @xmath251 where @xmath252 denotes the curvature tensor of @xmath7 and @xmath253 , @xmath254 are the connections and metrics induced on the tensor bundles over @xmath7 . the fibre bundle @xmath255 is uniformly locally trivial . that is , there exists a riemannian metric @xmath256 on @xmath8 such that for every @xmath38 and metric ball @xmath257 of radius @xmath258 there is a trivialisation @xmath259 , and the tensors @xmath260 and @xmath261 and all their covariant derivatives are bounded uniformly in @xmath96 . the embeddings @xmath262 and @xmath263 are bounded with all their derivatives . these conditions are trivially satisfied for compact manifolds @xmath3 and many examples such as `` asymptotically straight '' or periodic waveguides . the existence result for the super - adiabatic approximation can be formulated as follows . let @xmath3 be a waveguide of bounded geometry and set @xmath264 . for every @xmath265 there exist a projection @xmath266 and a unitary @xmath244 in @xmath267 , intertwining @xmath105 and @xmath266 , such that for every @xmath268)$ ] , satisfying @xmath269)$ ] for every @xmath270 , we have @xmath271 in particular the hausdorff distance between the spectra of @xmath92 and @xmath272 is small , i.e. for every @xmath273 @xmath274,\sigma({h_\mathrm{eff}})\cap ( -\infty , \lambda-\delta]\big)=\mathcal{o}({\varepsilon}^n)\,.\ ] ] for @xmath275 we can choose @xmath276 , so at first sight the approximation of @xmath92 by @xmath124 yields errors of order @xmath0 . more careful inspection shows that for @xmath277 we have @xmath278 as an operator from @xmath279 to @xmath234 , so the statement on the spectrum holds for @xmath124 with an error of order @xmath87 . this improvement over relies on the existence of @xmath244 for a better choice of trial states . close to the ground state the approximation is even more accurate . [ thm : low spectrum ] let @xmath3 be a waveguide of bounded geometry , @xmath280 and @xmath281 . then for every @xmath282 @xmath283,\sigma(h_\mathrm{a})\cap ( -\infty , \lambda_0 + c{\varepsilon}^\alpha]\big)=\mathcal{o}({\varepsilon}^{2+\alpha/2})\,.\ ] ] assume , in addition , that @xmath284 is strictly below the essential spectrum of @xmath111 in the sense that for some @xmath285 and @xmath0 small enough the spectral projection @xmath286}(h_{\rm a})$ ] has finite rank . then , if @xmath287 are all the eigenvalues of @xmath111 below @xmath288 , @xmath92 has at least @xmath289 eigenvalues @xmath290 below its essential spectrum and @xmath291 for @xmath292 . the natural energy scale @xmath293 to consider this theorem would be the spacing of eigenvalues of @xmath124 . this of course depends on the specific situation . if @xmath127 is constant we will see that @xmath294 is a natural choice . in the somewhat more generic case in which the eigenband @xmath224 has a global and non - degenerate minimum , as in the example of the waveguide @xmath136 in the introduction , the lowest eigenvalues of @xmath124 will behave like those of an harmonic oscillator and @xmath295 is the correct choice of scale . in this case the set @xmath296\cap \sigma(h_\mathrm{a})$ ] will just be empty for @xmath0 small enough and thus by the theorem there is no spectrum of @xmath92 in this interval . we remark that results can be obtained also for energies higher than @xmath297 and projections to other eigenbands than @xmath127 . the relevant condition is that they are separated from the rest of the spectrum of @xmath228 by a local gap . for @xmath127 this is a consequence of the bounded geometry of @xmath3 ( see ( * ? ? ? * proposition 4.1 ) ) . the approximation of spectra is not mutual as for low energies , but there is always spectrum of @xmath92 near that of @xmath272 ( see ( * ? ? ? * corollary 2.4 ) ) . from now on we will focus on analysing the adiabatic operator . in particular we will see how the geometry of the waveguide enters into this operator and its expansion up to order @xmath133 , which is relevant for small energies irrespective of the super - adiabatic corrections by theorem [ thm : low spectrum ] . we now give a general expression for @xmath124 from which we will derive the explicit form for various specific situations . first group the terms of @xmath92 in such a way that @xmath298 by taking @xmath299 & = -{\varepsilon}^2 s^{\varepsilon}+ { \varepsilon}{v_\mathrm{bend}}- { \varepsilon}^{-1}\bigl(\tfrac{1}{2 } \delta_{\mathsf{v}}(\log \rho_{\varepsilon } ) + \tfrac{1}{4 } g_f({\mathrm{d}}\log\rho_{\varepsilon},{\mathrm{d}}\log\rho_{\varepsilon})\big)\\ & = -{\varepsilon}^2 s^{\varepsilon}+ \underbrace{\tfrac{{\varepsilon}}{2 } \delta_{\mathsf{h}}(\log \rho_{\varepsilon } ) + \mathcal{o}({\varepsilon}^3)}_{=:{\varepsilon}\tilde v_{\rm{bend}}}\ , . \end{aligned}\end{aligned}\ ] ] projecting this expression with @xmath105 as in equation gives @xmath300 and @xmath301 where @xmath302 is the levi - civit connection of @xmath254 , @xmath303 is the one - form @xmath304 and @xmath305 is the mean curvature vector of the fibres ( see equation ) . the derivation for the projection of @xmath306 can be found in ( * ? ? ? * chapter 3 ) . altogether we have the expression @xmath307 for the adiabatic operator as an operator on @xmath9 . by analogy with the introduction , we view @xmath308 as an operator on @xmath7 via the identification @xmath309 . by the same procedure , projecting the potentials in @xmath310 amounts to averaging them over the fibres with the weight @xmath311 . the vast literature on quantum waveguides is , in our terminology , concerned with the case of massive waveguides . in this section we give a detailed derivation of the effects due to the extrinsic geometry of @xmath146 . the necessary calculations of the metric @xmath150 and the bending potential @xmath135 have been performed in all of the works on quantum waveguides for the respective special cases , and by tolar @xcite for the leading order @xmath312 in the general case . a generalisation to tubes in riemannian manifolds is due to wittich @xcite . we then discuss the explicit form of the adiabatic hamiltonian , calculating the adiabatic potential and the projection of @xmath310 . in particular we generalise the concept of a `` twisted '' waveguide ( c.f . @xcite ) to arbitrary dimension and codimension in section [ sect : twist ] . we also examine the role of the differential operator @xmath134 in @xmath124 , which is rarely discussed in the literature , and its relevance for the different energy scales @xmath313 . let @xmath314 be local coordinates on @xmath7 and @xmath315 a local orthonormal frame of @xmath3 with respect to @xmath316 such that every normal vector @xmath317 may be written as @xmath318 these bundle coordinates yield local coordinate vector fields @xmath319 on @xmath3 for @xmath320 and @xmath321 . the aim is to obtain formulas for the coefficients of the unscaled pullback metric @xmath322 with respect to these coordinate vector fields . let @xmath323 be an open neighbourhood of zero , @xmath324 , @xmath325 be a curve with @xmath326 and @xmath327 . it then holds that @xmath328 for the case @xmath329 , we choose the curve @xmath324 given by @xmath330 where @xmath331 is a smooth curve with @xmath332 and @xmath333 . we then have @xmath334 in order to relate the appearing derivative @xmath335 to the extrinsic curvature of @xmath7 , we project the latter onto its tangent and normal component , respectively . therefor , we introduce the weingarten map @xmath336 and the @xmath337-valued local connection one - form associated to the normal connection @xmath338 with respect to @xmath339 , i.e. @xmath340 with these objects we have @xmath341 for the case @xmath342 , one takes the curve @xmath324 with @xmath343 hence , @xmath344 combining the expressions for the tangent maps , we finally obtain the following expressions for the pullback metric @xmath55 : [ lem : pullback ] let @xmath345 denote the local bundle coordinates on @xmath3 introduced in and the associated coordinate vector fields . then the coefficients of the pullback metric are given by @xmath346 for @xmath347 and @xmath348 . here , @xmath349 stands for the second fundamental form defined by @xmath350 . let us now consider the scaled pullback metric @xmath351 . observe that for any @xmath352 @xmath353 and one immediately concludes from and that @xmath354 consequently , the coefficients of the scaled pullback metric are given by @xmath355\ , , \\ g_{i , d+\alpha}^{\varepsilon}(x , n ) & = g_b^\bot\bigl(\omega^{\mathsf{n}}(\partial_{x^i})\nu , e_\alpha\bigr)\ , , \\ g_{d+\alpha , d+\beta}^{\varepsilon}(x , n ) & = \updelta_{\alpha\beta}\end{aligned}\ ] ] for @xmath347 and @xmath348 . we see that @xmath356 is not orthogonal to @xmath357 with respect to @xmath150 . however , any vector @xmath358 can be orthogonalised by subtracting its vertical component . the resulting vector @xmath359 is the horizontal lift of @xmath360 . consequently , the orthogonal complement of @xmath361 with respect to @xmath150 is given by @xmath362 for all @xmath15 . finally , a short computation shows that @xmath363 . \nonumber\end{aligned}\ ] ] hence , the scaled pullback metric @xmath150 actually has the form with `` horizontal correction '' @xmath364 [ rem : complgeod ] the fibres @xmath102 of @xmath3 are completely geodesic for the pullback metric @xmath150 . in order to see this , we show that the second fundamental form of the fibres @xmath365 vanishes identically . since the latter is a symmetric tensor , it is sufficient to show that the diagonal elements @xmath366 are zero . using koszul s formula , four out of the six appearing terms obviously vanish and we are left with @xmath367,\partial_\alpha\bigr ) \\ & \stackrel{\eqref{eq : partialhm}}{= } g_f\bigl(\underbrace{[\partial_i,\partial_\alpha]}_{=0 } - \bigl[g_{i , d+\beta}(x , n)\partial_\beta,\partial_\alpha\bigr],\partial_\alpha\bigr ) \\ & \stackrel{\hphantom{(21)}}{= } \frac{g_{i , d+\beta}(x , n)}{\partial n^\alpha } \underbrace{g_f ( \partial_\beta,\partial_\alpha)}_{\updelta_{\beta\alpha } } \\ & \stackrel{\hphantom{(21)}}{= } g_b^\bot(\omega^{\mathsf{n}}(\partial_{x^i})e_\alpha , e_\alpha)\,.\end{aligned}\ ] ] but now , the last expression equals zero since @xmath368 is @xmath337-valued . consequently , the mean curvature vector @xmath305 defined by @xmath369 vanishes identically . finally note that the same considerations also hold for the submersion metric @xmath211 due to @xmath370 . now that we have a detailed description of the metric , we can explicitly express the horizontal laplacian by the vector fields @xmath371 . let @xmath372 be the dual basis ( note that in general @xmath373 since @xmath374 ) . then by definition @xmath375 and thus @xmath376 when acting on a horizontal vector field @xmath377 , the divergence takes the coordinate form @xmath378 since @xmath379 . for a horizontal lift @xmath380 we have the simple formula @xmath381 now for a massive waveguide @xmath382 and the horizontal laplacian takes the familiar form @xmath383 which is just @xmath384 with @xmath360 replaced by @xmath385 . in the introduction ( see equation ) we saw that the leading order of @xmath135 is attractive ( negative ) and proportional to the square of the curve s curvature @xmath386 . here we give a detailed derivation of @xmath135 for generalised massive waveguides and then discuss the sign of its leading part . therefor , let @xmath387 be a local orthonormal frame of @xmath388 with respect to @xmath254 and let @xmath389 be coordinates on @xmath31 as in equation . then @xmath390 for @xmath320 and @xmath391 form a local frame of @xmath182 . in this frame the scaled metrics have the form ( see also ) @xmath392 { \varepsilon}^{-2}(\operatorname{id}_{d\times d } -{\varepsilon}{\mathcal{w}}(\nu))^2 & 0\\ \hline 0 & \operatorname{id}_{k\times k } \end{pmatrix } \,,\quad g^{\varepsilon}_\mathrm{s } = \begin{pmatrix}[c|c ] { \varepsilon}^{-2}\operatorname{id}_{d\times d } & 0\\ \hline 0 & \operatorname{id}_{k\times k } \end{pmatrix}.\ ] ] from that and equation we easily conclude that @xmath393 using taylor s expansion for @xmath0 small enough , @xmath394 we have @xmath395 . next , we calculate the terms appearing in @xmath135 separately : @xmath396 @xmath397 + { \mathcal{o}}({\varepsilon}^2)\ , , \end{aligned}\ ] ] @xmath398 denoting by @xmath216 the adjoint of the original @xmath216 with respect to the pairing of @xmath399 and @xmath182 , and hence @xmath400 + { \mathcal{o}}({\varepsilon}^2)\ , .\end{aligned}\ ] ] putting all this together , we obtain the following expression for the bending potential in the case of massive quantum waveguides : @xmath401 \label{eq : vbend_0}\\ & \hphantom{=}\ + \frac{{\varepsilon}}{2 } \sum_{\alpha=1}^k \bigl[\operatorname{tr}{\mathcal{w}}(e_\alpha ) \operatorname{tr}\bigl({\mathcal{w}}(e_\alpha){\mathcal{w}}(\nu)\bigr ) - 2\operatorname{tr}\bigl({\mathcal{w}}(e_\alpha)^2{\mathcal{w}}(\nu)\bigr ) - \delta_{\mathsf{h}}\operatorname{tr}{\mathcal{w}}(\nu)\bigr ] \label{eq : vrhoeps3 } \\ & \hphantom{=}\ + { \mathcal{o}}({\varepsilon}^2)\ , . \nonumber \end{aligned}\ ] ] the leading term of this expression ( @xmath402 ) has been widely stressed in the literature concerning one - dimensional quantum waveguides ( see e.g. @xcite ) , where it has a purely attractive effect . its higher dimensional versions were discussed by tolar @xcite but are generally less known , so we will discuss their possible effects for the rest of this section . since @xmath403 is self - adjoint , we may choose for each @xmath391 the orthonormal frame @xmath387 such that it consists of the eigenvectors of @xmath403 with eigenvalues ( principal curvatures ) @xmath404 . in order to get an impression of @xmath312 s sign , we divide @xmath403 into a traceless part @xmath405 and a multiple of the identity : @xmath406 note that the prefactors @xmath407 equal the components of the mean curvature vector of @xmath7 in direction @xmath408 . with the notation @xmath409 for any @xmath410 , we get the relation @xmath411 for all @xmath391 since @xmath412 is traceless . this yields for the potential : @xmath413 \\ & = \frac{1}{4 } \sum_{\alpha=1}^k \left[h_\alpha^2 - 2 \left({\left \lvert { \mathcal{w}}_0(e_\alpha ) \right \rvert}^2 + \frac{h_\alpha^2}{d}\right)\right ] \\ & = \frac{1}{4 } \sum_{\alpha=1}^k \left[\left(1-\frac{2}{d}\right ) h_\alpha^2 - 2{\left \lvert { \mathcal{w}}_0(e_\alpha ) \right \rvert}^2\right].\end{aligned}\ ] ] the latter relation shows that for @xmath414 the leading order of the bending potential is non - positive . thus , the effect of bending has an attractive character ( @xmath415 ) for @xmath0 small enough , or is of lower order ( @xmath416 ) , independently of the codimension @xmath417 . for @xmath418 , the first term is non - negative and may overcompensate the second term leading to a positive contribution to @xmath135 . consequently , a repulsive bending effect is possible . we may rewrite expression in terms of principal curvatures as @xmath419.\ ] ] 1 . for a waveguide modelled around a curve @xmath420 , @xmath137 , one immediately sees that @xmath421 . we consider the case where @xmath422 is the @xmath4-dimensional standard sphere of radius @xmath252 . the principal curvatures in the direction of the outer - pointing normal are given by @xmath423 for all @xmath320 , hence the bending potential reads @xmath424 = \left(1-\frac{2}{d}\right)\frac{d^2}{4 r^2}\ , .\end{aligned}\ ] ] it follows that @xmath425 for @xmath137 , @xmath426 for @xmath138 and @xmath427 for @xmath418 , respectively . thus , depending on the dimension @xmath4 of the sphere , the effect of bending can be either attractive or repulsive . we are now ready to calculate the geometric terms in the adiabatic operator . in this we concentrate on the adiabatic operator and explicitly calculate all the relevant terms on the energy scale given by theorem [ thm : low spectrum ] . first we take care of the contribution of @xmath310 , then we turn to the potential @xmath139 and explain its connection to `` twisting '' of the quantum waveguide . the contribution of the bending potential , that was calculated in the previous section , is given by its adiabatic approximation @xmath428 since the leading part @xmath312 is independent of the fibre coordinate @xmath429 , it is unchanged by this projection . the next term in the expansion of @xmath135 is given by . the weingarten map is linear in @xmath429 and since @xmath430 is again linear in @xmath429 , @xmath431 is also linear in @xmath429 . consequently , the contribution of to @xmath432 is proportional to @xmath433 hence , this contribution vanishes if the centre of mass of the ground state @xmath237 lies exactly on the submanifold @xmath7 . this is a reasonable assumption to make and represents a `` correct '' choice of parametrisation of the waveguide . under this assumption we have @xmath434 from the expression for the horizontal block of the metric @xmath150 one obtains its expansion on horizontal one - forms by locally inverting the matrix @xmath435 ( see @xcite ) . the result is @xmath436 where @xmath437 denotes the second fundamental form of @xmath7 , defined on @xmath438 by @xmath439 . moreover , we extend the latter to @xmath399 , understanding @xmath440 as its lift to the horizontal part @xmath441 and extending to @xmath399 by zero . the vertical components of @xmath150 and @xmath211 coincide , hence as an operator on @xmath9 we have the expression @xmath442 with an error of order @xmath0 on @xmath279 . using the leibniz rule we can rewrite this as @xmath443 now @xmath237 vanishes on the boundary and @xmath444 is a horizontal vector field , so by we have @xmath445 if we assume again that @xmath237 is centred on @xmath7 , this term vanishes and we are left with the potential @xmath446 with an error bound in @xmath447 . since the fibres @xmath102 are completely geodesic with respect to @xmath448 for massive quantum waveguides ( cf . remark [ rem : complgeod ] ) , we have @xmath382 and the adiabatic potential defined in reduces to @xmath449 this is called the born - huang potential in the context of the born - oppenheimer approximation . this potential is always non - negative . it basically accounts for the alteration rate of @xmath237 in horizontal directions . in the literature , the adiabatic potential has been studied mainly for `` twisted '' quantum waveguides . these have two - dimensional fibres @xmath102 which are isometric but not invariant under rotations and twist as one moves along the one - dimensional base curve @xmath7 @xcite . the operators @xmath450 , @xmath38 , are isospectral and their non - trivial dependence on @xmath96 is captured by @xmath139 . we now generalise this concept to massive waveguides of arbitrary dimension and codimension and calculate the adiabatic potential for this class of examples . in this context , a massive quantum waveguide @xmath451 is said to be only twisted at @xmath452 , if there exist a geodesic ball @xmath453 around @xmath123 and a local orthonormal frame @xmath454 of @xmath455 such that @xmath456 this exactly describes the situation that the cross - sections @xmath457 are isometric to @xmath6 , but may vary from fibre to fibre by an @xmath458-transformation . moreover , it follows that @xmath127 is constant on @xmath459 and the associated eigenfunction @xmath237 is of the form @xmath460 , where @xmath461 is the solution of @xmath462 as for the calculation of @xmath139 at @xmath123 , we firstly compute for @xmath463 : @xmath464 \phi_0(n ) \nonumber \\ & \stackrel{\hphantom{\text{\eqref{eq : partialhm}}}}{= } - n^\alpha g_b^\bot \bigl(\nabla^{\mathsf{n}}_{\partial_{x^i } } f_\alpha , f_\beta\bigr)\bigr|_{x_0 } \frac{\partial\phi_0(n)}{\partial n^\beta}\ , . \label{eq : twist1}\end{aligned}\ ] ] in order to get a better understanding of @xmath465 , we introduce on @xmath459 a locally untwisted orthonormal frame @xmath315 of @xmath455 . it is obtained by taking the vectors @xmath466 and parallel transporting them along radial geodesics with respect to the normal connection @xmath467 . thus , twisting is always to be understood relative to the locally parallel frame @xmath315 . the induced map that transfers the reference frame @xmath315 into the twisting frame @xmath454 is denoted by @xmath468 . it is defined by the relation @xmath469 for @xmath470 and obeys @xmath471 due to the initial data of @xmath315 . consequently , using the differential equation of the parallel transport , we have @xmath472 and hence @xmath473 for @xmath474 , let @xmath475 defined by @xmath476 be a set of generators of the lie algebra @xmath337 . this induces generalised angle functions @xmath477 by the relation @xmath478 for @xmath470 . then a short calculation shows that @xmath479 for @xmath480 . combining , and , we obtain @xmath481 where @xmath482 defines the action of the @xmath483-component of the angular momentum operator in @xmath417 dimensions . from here , it is easy to see that the adiabatic potential at @xmath123 is given by @xmath484 the first matrix @xmath485 encodes the rate , at which the frame @xmath454 twists relatively to the parallel frame @xmath315 at @xmath123 . the second matrix @xmath486 measures the deviation of the eigenfunction @xmath461 from being rotationally invariant . it determines to which extent the twisting of the waveguide effects the states in the range of @xmath105 and it depends only on the set @xmath6 ( and not on the point @xmath123 of the submanifold @xmath7 ) . finally for the case of a twisted quantum waveguides with @xmath487 and @xmath201 , there exists only one angle function @xmath488 and one angular momentum operator @xmath489 . then formula yields the well - known result @xcite @xmath490 which clearly vanishes if @xmath8 is invariant under rotations . now that we have calculated all the relevant quantities , we can give an explicit expansion of @xmath124 . the correct norm for error bounds of course depends on the energy scale under consideration . for a constant eigenvalue @xmath127 and @xmath294 the graph - norm of @xmath491 is clearly equivalent ( with constants independent of @xmath0 ) to the usual norm of @xmath492 . in this situation the best approximation by @xmath124 given by theorem [ thm : low spectrum ] has errors of order @xmath133 , so the estimates just derived give @xmath493 if @xmath237 is centred . if this is not the case , the expansion can be read off from equations , and . if @xmath127 has a non - degenerate minimum and @xmath295 , the errors of our best approximation are of order @xmath132 . thus , the potentials of order @xmath132 can be disregarded in this case . note however that on the domain of @xmath494 we have @xmath495 , so the differential operator will be relevant . the error terms of equation , containing second order differential operators , are of order @xmath132 with respect to @xmath496 , so they are still negligible . thus , for @xmath497 with @xmath498 we have @xmath499 where the last term is of order @xmath87 in general and vanishes for centred @xmath237 . in this section we consider hollow quantum waveguides @xmath500 , which by definition [ def : massivehollow ] are the boundaries of massive waveguides . this underlying massive wavguide is denoted by @xmath501 in the following . the bundle structure is inherited from the massive waveguide as well , i.e. @xmath502 and the diagram hollow quantum waveguides have , to our knowledge , not been studied before . in fact , already the derivations of @xmath150 and @xmath135 constitute novel results . a slight generalisation of these calculations to objects that are not necessarily boundaries can be found in ( * ? ? ? * chapter 3 ) . note that @xmath503 and that we computed the unscaled pullback metric @xmath504 for the massive waveguide @xmath180 already in lemma [ lem : pullback ] . the latter reads @xmath505 where the `` horizontal correction '' @xmath506 vanishes on vertical vector fields and essentially depends on the extrinsic geometry of the embedding @xmath507 . if we restrict @xmath180 s tangent bundle to @xmath3 , one has the orthogonal decomposition @xmath508 with respect to @xmath509 . due to the commutativity of the above diagram , it follows that @xmath510 . this suggests to introduce the notation @xmath511 for the orthogonal complement of @xmath183 in @xmath512 with respect to @xmath513 , i.e. @xmath514 for any @xmath194 , let @xmath515 and @xmath195 be the respective unique horizontal lifts . it then holds that @xmath516 thus , the difference between @xmath380 and @xmath517 is a vertical field : @xmath518 with @xmath519 . moreover , @xmath520 since for arbitrary @xmath521 @xmath522 implies @xmath523 . for any @xmath38 . note that for any @xmath524 we have the canonical identification of @xmath525 and @xmath526 via the isomorphism . right : relationship between the horizontal lifts @xmath380 and @xmath527 . they are connected by the vertical field @xmath528.,title="fig:",scaledwidth=95.0% ] ( -10,36)@xmath3 ( -10,150)@xmath3 ( -96,72)@xmath529 ( -98,157)@xmath380 ( -98,119)@xmath517 ( -128,153)@xmath528 ( -119,136)@xmath429 ( -119,87)@xmath96 ( -109,17)@xmath530 ( -190,140)@xmath31 ( -187,60)@xmath180 ( -30,73)@xmath531 ( -153,137 ) ( -145,99 ) ( -109,95 ) ( -250,154 ) ( -305,160 ) ( -285,121)@xmath429 ( -336,87)@xmath532 ( -298,18)@xmath533 ( -360,50)@xmath534 ( -415,158)@xmath535 obviously , the relation @xmath536 does not shed light on the vertical part @xmath528 . the latter will be determined by the requirement @xmath380 to be a tangent vector field on @xmath3 , or equivalently by the condition @xmath537 , where @xmath538 denotes a unit normal field of @xmath3 in @xmath180 . in order to determine @xmath528 from this condition , we first need to show that the vertical component of @xmath97 is non - zero everywhere . decompose @xmath540 with @xmath541 . it then holds for any vector field @xmath542 : @xmath543 where we used for the second and fourth equality . this clearly implies @xmath544 by . now suppose there exists @xmath545 with @xmath546 . consider the space @xmath547 since @xmath548 is orthogonal to @xmath549 , one has @xmath550 . we will show that the kernel of @xmath551 is trivial . hence , @xmath552 clearly contradicts the fact that @xmath553 and finally the assumption that @xmath554 . therefor , let @xmath555 . on the one hand , since @xmath556 @xmath557 is an element of @xmath558 . but on the other hand , @xmath559 posseses a trivial kernel . together , this yields @xmath560 , i.e. @xmath561 . in view of equation , lemma [ lem : v ] suggests to define a function @xmath562 such that @xmath563 . thus , the requirement @xmath564 yields @xmath565 consequently @xmath566 note that @xmath567 is well - defined since @xmath568 by lemma [ lem : v ] . moreover , the latter equation shows that @xmath569 is actually a tensor . in summary , we just showed that the unscaled pullback metric on @xmath3 may be written as @xmath570 with `` horizontal block '' @xmath571 for @xmath198 . going over to the scaled pullback metric @xmath150 , we first show that the horizontal lift remains unchanged . it is sufficient to show that for any vector field @xmath194 its unique horizontal lift @xmath380 is given by the @xmath0-independent expression @xmath573 with @xmath562 and @xmath544 as before . we already know that @xmath380 is tangent to @xmath3 and satisfies @xmath574 . thus , the requirement that @xmath380 is orthogonal to any @xmath542 with respect @xmath150 is the only possible way for any @xmath0-dependence to come into play . therefore , we calculate @xmath575}_{\text{$=0 $ , since $ w\in\gamma({\mathsf{v}}m)\subset\gamma({\mathsf{v}}\mathring{m}|_m)$ } } + \gimel(x ) \underbrace{g_{\mathring{f}}(v_n , w)}_{\text{$=0 $ by~\eqref{eq : decompgf } } } \\ & = 0\ , .\end{aligned}\ ] ] in summary , if the scaled pullback metric of the massive waveguide @xmath180 has the form @xmath576 , the scaled pullback metric @xmath150 of the associated hollow waveguide @xmath3 reads @xmath577 with @xmath578 for @xmath198 . this shows that the scaled pullback metric @xmath150 is again of the form . let us consider a simple example of a hollow quantum waveguide with @xmath137 , @xmath201 . take @xmath579 as submanifold and parametrise the according massive quantum waveguide via @xmath580\bigr\ } , \ ] ] where @xmath581 $ ] with @xmath582 is a smooth function obeying the periodicity condition @xmath583 and @xmath584 stands for the `` radial unit vector '' . in view of example [ ex : conv ] with @xmath585 , the unscaled pullback metric on @xmath180 is given by @xmath586 furthermore , we immediately observe that @xmath587 with trivial horizontal lift @xmath588 . the hollow quantum waveguide associated to @xmath3 is obviously given by @xmath589 consequently , @xmath590 is given by @xmath591 , where @xmath592 with @xmath593 . one easily agrees that @xmath594 and @xmath595 are orthogonal to @xmath596 with respect to @xmath509 . @xmath597}}}_{=:v_n\in{\mathsf{v}}_{(x,\varphi)}m^\bot } + \underbrace{\frac{- r\frac{\partial r}{\partial x } e_x}{\sqrt{(\frac{\partial r}{\partial\varphi})^2 + r^2\left[1+(\frac{\partial r}{\partial x})^2\right]}}}_{=:h_n\in{\mathsf{h}}_{(x,\varphi)}\mathring{m}}\ ] ] is a unit normal vector of @xmath3 at @xmath598 for @xmath599 . noting that @xmath600}\ , , \ ] ] equation gives @xmath601 } } \left(\frac{(\frac{\partial r}{\partial\varphi})^2 + r^2}{(\frac{\partial r}{\partial\varphi})^2 + r^2\left[1+(\frac{\partial r}{\partial x})^2\right]}\right)^{-1 } \\ & = \frac{r \frac{\partial r}{\partial\varphi}\sqrt{(\frac{\partial r}{\partial\varphi})^2 + r^2\left[1+(\frac{\partial r}{\partial x})^2\right]}}{(\frac{\partial r}{\partial \varphi})^2 + r^2}\ , .\end{aligned}\ ] ] this yields the following expression for the `` horizontal block '' of the scaled pullback metric @xmath150 : @xmath602 with @xmath603 we now calculate the adiabatic operator for hollow waveguides . since in this case the fibre is a manifold without boundary , the ground state of @xmath228 is explicitly known : @xmath604 where @xmath605 is the @xmath606-norm of @xmath607 on the fibre @xmath102 . because of this we can express many of the terms appearing in @xmath124 , given in equation , through @xmath607 . let us begin with the sum of the modified bending potential @xmath608 appearing in @xmath310 and the adiabatic potential @xmath139 . first we obtain an expression for the one - form @xmath609 by observing that for any vector field @xmath529 on @xmath7 @xmath610 so we see that @xmath611 now to start with the first term of the adiabatic potential can be calculated as in @xmath612 for the the modified bending potential one has , using the shorthand @xmath613 , @xmath614 note that this expression is of order @xmath132 since @xmath615 . hence , bending does not contribute to the leading order of @xmath124 . now inserting the explicit form of @xmath237 an elementary calculation yields @xmath616 \int_{f_x } & - \tfrac{1}{2}\rho_{\varepsilon}\delta_{\mathsf{h}}{\left \lvert \rho_{\varepsilon}\right \rvert}_1^{-1 } - \tfrac{1}{2 } g_b\bigl(\operatorname{grad}{\left \lvert f_x \right \rvert}^{-1 } , \pi_{m * } \operatorname{grad}\rho_{\varepsilon}\bigr ) \\ & + \tfrac{1}{4 } \rho_{\varepsilon}{\left \lvert \rho_{\varepsilon}\right \rvert}_1 g_b\bigl(\pi_{m * } \operatorname{grad}{\left \lvert \rho_{\varepsilon}\right \rvert}_1^{-1 } , \pi_{m * } \operatorname{grad}{\left \lvert \rho_{\varepsilon}\right \rvert}_1^{-1}\bigr)\ { \mathrm{d}}g_f + \mathcal{o}({\varepsilon}^2)\ , . \end{aligned}\ ] ] with @xmath615 and @xmath617 one easily checks that up to order @xmath0 this expression equals @xmath618 as far as the remaining terms of @xmath310 are concerned , note that the scaled pullback metric @xmath150 of the hollow waveguide has the same expansion on horizontal one - forms up to errors of order @xmath133 as in the case of the massive waveguide , i.e. @xmath619 hence , we can calculate these terms starting from expression . since the latter all carry a prefactor @xmath132 , we may replace any @xmath237 by @xmath620 , obtaining for @xmath621 @xmath622 and @xmath623 as in equation we have @xmath624 again , this term vanishes if the barycentre of the fibres @xmath625 is zero , that is @xmath626 since @xmath229 , the adiabatic operator is of the form @xmath627 with an error in @xmath628 . thus , the adiabatic operator at leading order is just the laplacian on the base @xmath7 plus an effective potential depending solely on the relative change of the volume of the fibres . going one order further in the approximation , we have @xmath629 in the same norm . hence , @xmath124 also contains the second order differential operator if the barycentre of @xmath102 is different from zero . let us also remark that the leading order of the adiabatic potential can also be calculated by applying a unitary transformation @xmath630 that rescales fibre volume to one , in the spirit of @xmath631 ( cf . equation ) . in this way a similar potential was derived by kleine @xcite , in a slightly different context , for a special case with one - dimensional base and without bending . g. bouchitt , m. l. mascarenhas and l. trabucho , _ on the curvature and torsion effects in one dimensional waveguides _ , esaim : control , optimisation and calculus of variations * 13 * ( 2007 ) , no . 4 , pp . 793808 . c. r. de oliveira and a. a. verri , _ on the spectrum and weakly effective operator for dirichlet laplacian in thin deformed tubes _ , journal of mathematical analysis and applications * 381 * ( 2011 ) , no . 1 ,

this paper addresses the structure - and - motionproblem , that requires to find camera motion and 3d structure from point matches . a new pipeline , dubbed samantha , is presented , that departs from the prevailing sequential paradigm and embraces instead a hierarchical approach . this method has several advantages , like a provably lower computational complexity , which is necessary to achieve true scalability , and better error containment , leading to more stability and less drift . moreover , a practical autocalibration procedure allows to process images without ancillary information . experiments with real data assess the accuracy and the computational efficiency of the method . structure and motion ; image orientation ; bundle adjustment ; autocalibration ; 3d

1506.00395

the progress in three - dimensional ( 3d ) modeling research has been rapid and hectic , fueled by recent breakthroughs in keypoint detection and matching , the advances in computational power of desktop and mobile devices , the advent of digital photography and the subsequent availability of large datasets of public images . today , the goal of definitively bridging the gap between physical reality and the digital world seems within reach given the magnitude , breadth and scope of current 3d modeling systems . three dimensional modeling is the process of recovering the properties of the environment and optionally of the sensing instrument from a series of measures . this generic definition is wide enough to accommodate very diverse methodologies , such as time - of - flight laser scanning , photometric stereo or satellite triangulation . the structure - and - motion(a.k.a . structure - from - motion ) field of research is concerned with the recovery of the three dimensional geometry of the scene ( the structure ) when observed through a moving camera ( the motion ) . sensor data is either a video or a set of exposures ; additional informations , such as the calibration parameters , can be used if available . this paper describes our contributions to the problem of structure - and - motionrecovery from unordered , uncalibrated images i.e , the problem of building a three dimensional model of a scene given a set of exposures . the sought result ( the `` model '' ) is generally a 3d point - cloud consisting of the points whose projection was identified and matched in the images and a set of camera matrices , identifying position and attitude of each image with respect to an arbitrary reference frame . the main challenges to be solved are computational efficiency ( in order to be able to deal with more and more images ) and generality , i.e. , the amount of ancillary information that is required . to address the efficiency issue we propose to describe the entire structure - and - motionprocess as a binary tree ( called _ dendrogram _ ) constructed by agglomerative clustering over the set of images . each leaf corresponds to a single image , while internal nodes represent partial models obtained by merging the left and right sub - nodes . computation proceeds from bottom to top , starting from several seed couples and eventually reaching the root node , corresponding to the complete model . this scheme provably cuts the computational complexity by one order of magnitude ( provided that the dendrogram is well balanced ) , and it is less sensitive to typical problems of sequential approaches , namely sensitivity to initialization @xcite and drift @xcite . it is also scalable and efficient , since it partitions the problem into smaller instances and combines them hierarchically , making it inherently parallelizable . on the side of generality , our aim is to push the `` pure '' structure - and - motiontechnique as far as possible , to investigate what can be achieved without including any auxiliary information . current structure - and - motionresearch has partly sidestepped this issue using ancillary data such as exif tags embedded in conventional image formats . their presence or consistency , however , is not guaranteed . we describe our approach to autocalibration , which is the process of automatic estimation from images of the internal parameters of the cameras that captured them , and we therefore demonstrate the first structure - and - motionpipeline capable of using unordered , uncalibrated images . the main issue to address in structure - and - motionis the computational complexity , which is dominated by the bundle adjustment phase , followed by feature extraction and matching . a class of solutions that have been proposed are the so - called partitioning methods @xcite . they reduce the structure - and - motionproblem into smaller and better conditioned subproblems which can be effectively optimized . within this approach , two main strategies can be distinguished . the first one is to tackle directly the bundle adjustment algorithm , exploiting its properties and regularities . the idea is to split the optimization problem into smaller , more tractable components . the subproblems can be selected analytically as in @xcite , where spectral partitioning has been applied to structure - and - motion , or they can emerge from the underlying 3d structure of the problem , as described in @xcite . the computational gain of such methods is obtained by limiting the combinatorial explosion of the algorithm complexity as the number of images and points increases . the second strategy is to select a subset of the input images and points that subsumes the entire solution . hierarchical sub - sampling was pioneered by @xcite , using a balanced tree of trifocal tensors over a video sequence . the approach was subsequently refined by @xcite , adding heuristics for redundant frames suppression and tensor triplet selection . in @xcite the sequence is divided into segments , which are resolved locally . they are subsequently merged hierarchically , eventually using a representative subset of the segment frames . a similar approach is followed in @xcite , focusing on obtaining a well behaved segment subdivision and on the robustness of the following merging step . the advantage of these methods over their sequential counterparts lies in the fact that they improve error distribution on the entire dataset and bridge over degenerate configurations . in any case , they work for video sequences , so they can not be applied to unordered , sparse images . the approach of @xcite works with sparse datasets and is based on selecting a subset of images whose model provably approximates the one obtained using the entire set . this considerably lowers the computational requirements by controllably removing redundancy from the dataset . even in this case , however , the images selected are processed incrementally . moreover , this method does not avoid computing the epipolar geometry between all pairs of images . within the solutions aimed at reducing the impact of the bundle adjustment phase , hierarchical approaches include @xcite and this paper . the first can be considered as the first paper where the idea has been set forth : a spanning tree is built to establish in which order the images must be processed . after that , however , the images are processed in a standard incremental way . the approach described in @xcite is based on recursive partitioning of the problem into fully - constrained sub - problems , exploiting the bipartite structure of the visibility graph . the partitioning operates on the problem variables , whereas our approach works on the input images . orthogonally to the aforementioned approaches , a solution to the the computational complexity of structure - and - motionis to throw additional computational power at the problem @xcite . within such a framework , the former algorithmic challenges are substituted by load balancing and subdivision of tasks . such a direction of research strongly suggest that the current monolithic pipelines should be modified to accommodate ways to parallelize and optimally split the workflow of structure - and - motiontasks . in @xcite image selection ( via clustering ) is combined with a highly parallel implementation that exploits graphic processors and multi - core architectures . the impact of the bundle adjustment phase can also be reduced by adopting a different paradigm in which _ first _ the motion is recovered and _ then _ the structure is computed . all these methods start from the relative orientation of a subset of camera pairs ( or triplets ) , computed from point correspondences , then solve for the absolute orientation of all the cameras ( _ globally _ ) , reconstruct 3d points by intersection , and finally run a single bundle adjustment to refine the reconstruction . camera internal parameters are required . the method described in @xcite solves a homogeneous linear system based on a novel decomposition of the essential matrix that involves the absolute parameters only . in @xcite nonlinear optimization is performed to recover camera translations given a network of both noisy relative translation directions and 3d point observations . this step is preceded by outlier removal among relative translations by solving simpler low - dimensional subproblems . the authors of @xcite propose a discrete markov random field formulation in combination with levenberg - marquardt minimization . this technique requires additional information as input , such as geotag locations and vanishing points . other approaches ( e.g. @xcite ) compute translations together with the structure , involving a significant number of unknowns . the method presented in @xcite proposes a fast spectral solution by casting translation recovery in a graph embedding problem . govindu in @xcite derives a homogeneous linear system of equations in which the unknown epipolar scaling factors are eliminated by using cross products , and this solution is refined through iterative reweighted least squares . the authors of @xcite propose a linear algorithm based on an approximate geometric error in camera triplets . moulon et al . @xcite extract accurate relative translations by using an a - contrario trifocal tensor estimation method , and then recover simultaneously camera positions and scaling factors by using an @xmath0-norm approach . similarly to @xcite , this method requires a graph covered by contiguous camera triplets . the authors of @xcite propose a two - stage method in which relative translation directions are extracted from point correspondences by using robust subspace optimization , and then absolute translations are recovered through semidefinite programming . another relevant issue in structure - and - motionis the level of generality , i.e. , the number of assumption that are made concerning the input images , or , equivalently the amount of extra information that is required in addition to pixel values . existing pipelines either assume known internal parameters @xcite , or constant internal parameters @xcite , or rely on exif data plus external information ( camera ccd dimensions ) @xcite . methods working in large scale environments usually rely on a lot of additional information , such as camera calibration and gps / ins navigation systems @xcite or geotags @xcite . autocalibration ( a.k.a . self - calibration ) has generated a lot of theoretical interest since its introduction in the seminal paper by maybank and faugeras @xcite . the attention created by the problem however is inherently practical , since it eliminates the need for off - line calibration and enables the use of content acquired in an uncontrolled setting . modern computer vision has partly sidestepped the issue by using ancillary information , such as exif tags embedded in some image formats . unfortunately it is not always guaranteed that such data will be present or consistent with its medium , and do not eliminate the need for reliable autocalibration procedures . a great deal of published methods rely on equations involving the dual image of the absolute quadric ( diaq ) , introduced by triggs in @xcite . earlier approaches for variable focal lengths were based on linear , weighted systems @xcite , solved directly or iteratively @xcite . their reliability has been improved by more recent algorithms , such as @xcite , solving super - linear systems while directly forcing the positive definiteness of the diaq . such enhancements were necessary because of the structural non - linearity of the task : for this reason the problem has also been approached using branch and bound schemes , based either on the kruppa equations @xcite , dual linear autocalibration @xcite or the modulus constraint @xcite . the algorithm described in @xcite shares , with the branch and bound approaches , the guarantee of convergence ; the non - linear part , corresponding to the localization of the plane at infinity , is solved exhaustively after having used the cheiral inequalities to compute explicit bounds on its location . this paper describes a hierarchical and parallelizable scheme for structure - and - motion ; please refer to fig . [ fig : schema ] for a graphical overview . the front end of the pipeline is keypoint extraction and matching ( sec . [ sec : match ] ) , where the latter is subdivided into two stages : the first ( `` broad phase '' ) is devoted to discovering the tentative topology of the epipolar graph , while the second ( `` narrow phase '' ) performs the fine matching and computes the epipolar geometry . images are then organized into a dendrogram by clustering them according to their overlap ( sec . [ sec : dendro ] ) . a new clustering strategy , derived from the simple linkage , is introduced ( sec . [ sec : balance ] ) that makes the dendrogram more balanced , thereby approaching the best - case complexity of the method . the structure - and - motioncomputation proceeds hierarchically along this tree , from the leaves to the root ( sec . [ sec : hierarchical ] ) . images are stored in the leaves , whereas partial models correspond to internal nodes . according to the type of node , three operations are possible : stereo modeling ( image - image ) , resection - intersection ( image - model ) or merging ( model - model ) . bundle adjustment is run at every node , possibly in its `` local '' form ( sec . [ sec : localba ] ) . we demonstrate that this paradigm has several advantages over the sequential one , both in terms of computational performance ( which improves by one order of magnitude on average ) and overall error containment . autocalibration ( sec . [ sec : urec ] ) is performed on partial models during the dendrogram traversal . first , the location of the plane at infinity is derived given two perspective projection matrices and a guess on their internal parameters , and subsequently this procedure is used to iterate through the space of internal parameters looking for the best collineation that makes the remaining cameras euclidean . this approach has several advantages : it is fast , easy to implement and reliable , since a reasonable solution can always be found in non - degenerate configurations , even in extreme cases such as when autocalibrating just two cameras . being conscious that `` the devil is in the detail '' , section [ sec : devil ] reports implementation details and heuristics for setting the parameters of the pipeline . the experimental results reported in sec . [ sec : exp ] are exhaustive and analyze the output of the pipeline in terms of accuracy , convergence and speed . we report here on the latest version of our pipeline , called samantha . previous variants have been described in @xcite and @xcite respectively . the main improvements are in the matching phase and in the autocalibration that now integrates the method described in @xcite . the geometric stage has been carefully revised to make it more robust , to the point where in some cases bundle adjustment is not needed any more except at the end of the process . most efforts have been made in the direction of a robust and automatic approach , avoiding unnecessary parameter tuning and user intervention . in this section we describe the stage of samanthathat is devoted to the automatic extraction and matching of keypoints among all the @xmath1 available images . its output is to be fed into the geometric stage , that will perform the actual structure - and - motionrecovery . good and reliable matches are the key for any geometric computation . although most of the building blocks of this stage are fairly standard techniques , we carefully assembled a procedure that is fully automatic , robust ( matches are pruned to discard as many outliers as possible ) and computationally efficient . the procedure for recovering the epipolar graph is indeed new . first of all , keypoints are extracted in all @xmath1 images . we implemented the keypoint detector proposed by @xcite , where blobs with associated scale levels are detected from scale - space extrema of the scale - normalized laplacian : @xmath2 we used a 12-level scale - space and in each level the laplacian is computed by convolution ( in cuda ) with a @xmath3 kernel . as for the descriptor , we implemented a 128-dimensional radial descriptor ( similar to the log - polar grid of gloh @xcite ) , based on the accumulated response of steerable derivative filters . this combination of detector / descriptor performs in a comparable way to sift and avoids patent issues . only a given number of keypoints with the strongest response overall are retained . this number is a multiple of @xmath1 , so as to fix the average quota of keypoints per image ( details in sec . [ sec : devil ] ) . as the images are unordered , the first objective is to recover the _ epipolar graph _ , i.e. , the graph that tells which images overlap ( or can be matched ) with each other . this must be done in a computationally efficient way , without trying to match keypoints between every image pair . as a matter of fact , keypoint matching is one of the most expensive stages , so one would like to reduce the number of images to be matched from @xmath4 to @xmath5 . in this broad phase we consider only a small constant number of descriptors for each image . in particular , we consider the keypoints with the highest scales , since their descriptors are more representative of the whole image content . then , each keypoint descriptor is matched to its approximate nearest neighbors in feature space , using the ann library @xcite ( with @xmath6 ) . a 2d histogram is then built that registers in each bin the number of matches between the corresponding images . consider the complete weighted graph @xmath7 where @xmath8 are images and the weighted adjacency matrix is the 2d histogram . this graph is in general dense , having @xmath9 . the objective is to extract a subgraph @xmath10 with a number of edges that is linear in @xmath1 . in the approach of @xcite , also followed in @xcite , every image is connected ( in the epipolar graph ) to the @xmath11 images that have the greatest number of keypoint matches with it . this creates a graph with @xmath12 edges , where the average degree is @xmath13 ( by the handshaking lemma ) . when the number of images is large , however , it tends to create cliques of very similar images with weak ( or no ) inter - clique connections . on the other hand , one would like to get an epipolar graph that is strongly connected , to avoid over - fragmentation in the subsequent clustering phase . this idea is captured by the notion of _ @xmath14-edge - connectedness _ : in graph theory , a graph is @xmath14-edge - connected if it remains connected whenever fewer than @xmath14 edges are removed . so , the graph produced by the original approach has a low @xmath14 , while one would like to have @xmath14 as high as possible ( ideally @xmath15 ) , with same edge budget . we devised a strategy that builds a subgraph @xmath10 of @xmath16 which is `` almost '' @xmath11-edge - connected by construction . 1 . build the maximum spanning tree of @xmath16 : the tree is composed of @xmath17 edges ; 2 . remove them from @xmath16 and add them to @xmath10 ; 3 . repeat @xmath11 times . in the hypothesis that @xmath11 spanning trees can be extracted from @xmath16 , the algorithm produces a subgraph @xmath10 that is @xmath11-edge - connected ( a simple proof of this is given in [ sec : app2 ] ) . please note that by taking the _ maximum _ spanning tree we favor edges with high weight . so this strategy can be seen as a compromise between picking pairs with the highest score in the histogram , as in the original approach , and creating a strongly connected epipolar graph . if the hypothesis about @xmath16 is not verified , a spanning forest will be obtained at a certain iteration , and @xmath10 will not be @xmath11-edge - connected . however , when @xmath18 one could expect that `` almost '' @xmath11 spanning trees can be extracted from @xmath16 without disconnecting it . keypoint matching follows a nearest neighbor approach @xcite , with rejection of those keypoints for which the ratio of the nearest neighbor distance to the second nearest neighbor distance is greater than a threshold ( see sec . [ sec : devil ] ) . matches that are not injective are discarded . in order to speed up the matching phase we employ a keypoint clustering technique similar to @xcite . every keypoint is associated with a different cluster according to its dominant angle , as recorded in the descriptor . only keypoints belonging to the same cluster are matched together ( in our implementation we used eight equidistant angular clusters ) : this breaks down the quadratic complexity of the matching phase at the expense of loosing some matches at the border of the clusters . homographies and fundamental matrices between pairs of matching images are then computed using m - estimator sample consensus ( msac ) @xcite , a variation of ransac that gives outliers a fixed penalty but scores inliers on how well they fit the data . this makes the output less sensitive to a higher inlier threshold , thereby rendering less critical the choice of the threshold , at no extra computational cost with respect to ransac . the random sampling is done with a _ bucketing _ technique @xcite , which forces keypoints in the sample to be spatially separated . this helps to reduce the number of iterations and provide more stable estimates . since ransac and its variants ( like msac ) have a low statistical efficiency , the model must finally be re - estimated on a refined set of inliers . let @xmath19 be the residuals of _ all _ the @xmath20 keypoints after msac , and let @xmath21 be the sample that attained the best score ; following @xcite , a robust estimator of the scale is : @xmath22 the resulting set of inliers are those points such that @xmath23 where @xmath24 is a constant ( we used 2.5 ) . the model parameters are re - estimated on this set of inliers via least - squares minimization of the ( first - order approximation of the ) geometric error @xcite . the more likely model ( homography or fundamental matrix ) is selected according to the geometric robust information criterion ( gric ) @xcite : @xmath25 where @xmath26 is the standard deviation of the measurement error , @xmath14 is the number of parameters of the model , @xmath27 is the dimension of the fitted manifold , and @xmath28 is the dimension of the measurements . in our case , @xmath29 for fundamental matrices and @xmath30 for homographies . the model with the lower gric is the more likely . in the end , if the number of remaining matches @xmath31 between two images is less than 20% of the total number of matches before msac , then they are discarded . the rationale is that if an excessive fraction of oultliers have been detected , the original matches are altogether unreliable . a similar formula is derived in @xcite on the basis of a probabilistic model . as a safeguard , a threshold on the minimum number of matches is also set ( details in sec . [ sec : devil ] ) . after that , keypoint matching in multiple images are connected into _ tracks _ ( see figure [ fig : tracce ] ) : consider the undirected graph @xmath32 where @xmath8 are the keypoints and @xmath33 represents matches ; a track is a connected component of @xmath16 . vertices are labeled with the image the keypoints belong to : an inconsistency arises when in a track a label occurs more than once . inconsistent tracks and those shorter than three frames are discarded . a track represents the projection of a single 3d point imaged to multiple exposures ; such a 3d point is called a _ tie - point_. the images are organized into a tree with agglomerative clustering , using a measure of overlap as the distance . the structure - and - motioncomputation then follows this tree from the leaves to the root . as a result , the problem is broken into smaller instances , which are then separately solved and combined . algorithms for image clustering have been proposed in the literature in the context of structure - and - motion@xcite , panoramas @xcite , image mining @xcite and scene summarization @xcite . the distance being used and the clustering algorithm are application - specific . in this paper we deploy an image affinity measure that befits the structure - and - motiontask . it is computed by taking into account the number of tie - points visible in both images and how well their projections ( the keypoints ) spread over the images . in formulae , let @xmath34 and @xmath35 be the set of visible tie - points in image @xmath36 and @xmath37 respectively : @xmath38 where @xmath39 is the area of the convex hull of a set of image points and @xmath40 ( @xmath41 ) is the total area of image @xmath36 ( @xmath37 ) . the first term is an affinity index between sets , also known as the jaccard index . the distance is @xmath42 , as @xmath43 ranges in @xmath44 $ ] . the general agglomerative clustering algorithm proceeds in a bottom - up manner : starting from all singletons , each sweep of the algorithm merges the two clusters with the smallest distance between them . the way the distance between clusters is computed produces different flavors of the algorithm , namely the simple linkage , complete linkage and average linkage @xcite . we selected the _ simple linkage _ rule : the distance between two clusters is determined by the distance of the two closest objects ( nearest neighbors ) in the different clusters . simple linkage clustering is appropriate to our case because : i ) the clustering problem _ per se _ is fairly easy , ii ) nearest neighbors information is readily available with ann and iii ) it produces `` elongated '' or `` stringy '' clusters which fits very well with the typical spatial arrangement of images sweeping a certain area or building . before describing our hierarchical approach , let us set the notation and review the geometry tools that are needed . a _ model _ is a set of cameras and 3d points expressed in a local reference frame ( _ stereo - model _ with two cameras ) . the procedure of computing 3d point coordinates from corresponding points in multiple images is called _ intersection _ ( a.k.a . triangulation ) . recovering the camera matrix ( fully or limited to the external parameters ) from known 3d-2d correspondences is called _ resection_. the task of retrieving the relative position and attitude of two cameras from corresponding points in the two images is called _ relative orientation_. the task of computing the rigid ( or similarity ) transform that brings two models that share some tie - points into a common reference frame is called _ absolute orientation_. let us assume _ pro tempore _ that the internal parameters are known ; this constraint is removed in sec . [ sec : urec ] . images are grouped together by agglomerative clustering , which produces a hierarchical , binary cluster tree , called a _ dendrogram_. every node in the tree represents a partial , independent model . from the processing point of view , at every node in the dendrogram an action is taken that augments the model , as shown in figure [ fig:2.1 ] . three operations are possible : when two images are merged a stereo - model is built ( relative orientation + intersection ) . when an image is added to a cluster a resection - intersection step is taken ( as in the standard sequential pipeline ) . when two non - trivial clusters are merged , the respective models must be conflated by solving an absolute orientation problem ( followed by intersection ) . each of these steps is detailed in the following . corresponds to the creation of a stereo - model , the triangle corresponds to a resection - intersection , the diamond corresponds to a fusion of two partial independent models . ] while it is useful to conceptually separate the clustering from the modeling , the two phases actually occur simultaneously : during the simple linkage iteration , every time a merge is attempted the corresponding modeling action is taken . if it fails , the merge is discarded and the next possible merge is considered . the parameters of the relative orientation of two given cameras are obtained by factorization of the essential matrix @xcite . this is equivalent to knowing the external parameters of the two cameras in a local reference frame , and since the internal parameters are already known , the two camera matrices are readily set up . then tie - points are obtained by _ intersection _ ( see sec . [ sec : inter ] ahead ) from the tracks involving the two images , and the model is refined with bundle adjustment @xcite . it is worth noting that in order for this stereo - modeling to be successful the two images must satisfy two conflicting requirements : to have both a large number of tie - points in common and a baseline sufficiently large so as to allow a well - conditioned solution . the first requirement is implemented by the affinity defined in , but the second is not considered ; as a result , the pairing determined by image clustering is not always the best choice as far as the relative orientation problem is concerned . since in our pipeline the clustering and the structure - and - motionprocessing occurs simultaneously , these pairs will be discarded by simple sanity - checks before and after attempting to perform the stereo - modeling . the a - priori check requires that the relationship between the two images is described by a fundamental matrix ( instead of a homography ) , according to the gric score . the a - posteriori check considers the residual error and the cheirality check of the points before and after the bundle adjustment . intersection ( a.k.a . triangulation ) is performed by the iterated linear ls method @xcite . points are pruned by analyzing the condition number of the linear system and the reprojection error . the first test discards ill - conditioned intersections , using a threshold on the condition number of the linear system ( @xmath45 , in our experiments ) . the second test applies the so - called x84 rule @xcite , that establishes that , if @xmath19 are the residuals , the inliers are those points such that @xmath46 a safeguard threshold on the reprojection error is also set ( details in sec . [ sec : devil ] ) . in general , the intersection module obeys the following strategy . as soon as one track reaches length two in a given model ( i.e. at least two images of the track belongs to the model ) , the coordinates of the corresponding tie - point are computed by intersection . if the operation fails ( because of one of the sanity checks described above ) the 3d point is provisionally discarded but the track is kept . an attempt to compute the tie - point coordinates is undertaken every time the length of the track increases within the model . the tie - points belonging to the model that are also visible in the image to be added provides a set of 3d-2d correspondences , that are exploited to glue the image to the partial model . this is done by resection , where only the external parameters of the camera are to be computed ( a.k.a . external orientation problem ) . we used the ppnp algorithm described in @xcite inside msac , followed by non - linear minimization of the reprojection error at the end . after resection , which adds one image to the model , tie - points are updated by intersection , and bundle adjustment is run on the resulting model . when two partial independent ( i.e. , with different reference systems ) models are are to be conflated into one , the first step is to register one onto the other with a similarity transformation . the common tie - points are used to solve an absolute orientation ( with scale ) problem with msac . given the scale ambiguity , the inlier threshold for msac is hard to set . in @xcite a complex technique for the automatic estimation of the inlier threshold in 3d is proposed . we take a simpler but effective approach : instead of considering the length of the 3d segments that connect corresponding points as the residuals , we look at the average length of their 2d projections in the images ; in this way a meaningful inlier threshold in pixels can be set easily . the final transformation , computed with the orthogonal procrustean ( op ) method @xcite , minimizes the proper geometric residual , i.e. the sum of squared distances of 3d points . once the models are registered , tie - points are updated by intersection , and the new model is refined with bundle adjustment . this hierarchical algorithm can be summarized as follows : 1 . solve many independent relative orientation problems at the leaves of the tree , producing many independent stereo - models . 2 . traverse the tree ; in each node one of these operations takes place : 1 . update one model by adding one image with resection followed by intersection ; 2 . merge two independent models with absolute orientation . steps 1 . and 2.(a ) are the resection - intersection steps of classical sequential pipelines , as bundler . step 2.(b ) summons up the photogrammetric independent models block adjustment ( imba ) @xcite , where for each pair of overlapping photographs a stereo - model is built and then all these independent models are simultaneously transformed into a common reference frame with absolute orientation . if the tree reduces to a chain , the algorithm is the sequential one , whereas if the tree is perfectly balanced , only steps 2.(b ) are taken , and the resulting procedure resembles the imba , besides the fact that our models are disjoint and that models are recursively merged in pairs . compared to the standard sequential approach , this framework has a lower computational complexity , is independent of the initial pair of images , and copes better with drift problems , typical of sequential schemes . the hierarchical approach that has been outlined above allows us to decrease the computational complexity with respect to the sequential structure - and - motionpipeline . indeed , if the number of images is @xmath1 and every image adds a constant number of tie - points @xmath47 to the model , the computational complexity in time of sequential structure - and - motionis @xmath48 , whereas the complexity of samantha(in the best case ) is @xmath49 . the cost of bundle adjustment with @xmath11 tie - points and @xmath1 images is @xmath50 @xcite , hence it is @xmath49 if @xmath51 . in sequential structure - and - motion , adding image @xmath52 requires a constant number of bundle adjustments ( typically one or two ) with @xmath52 images , hence the complexity is @xmath53 in the case of the hierarchical approach , consider a node of the dendrogram where two models are merged into a model with @xmath1 images . the cost @xmath54 of adjusting that model is given by @xmath49 plus the cost of doing the same onto the left and right subtrees . in the hypothesis that the dendrogram is well balanced , i.e. , the two models have the same number of images , this cost is given by @xmath55 . hence the asymptotic time complexity @xmath56 in the best case is given by the solution of the following recurrence : @xmath57 that is @xmath58 by the third branch of the master s theorem @xcite . the worst case is when a single model is built up by adding one image at a time . in this case , which corresponds to the sequential case , the dendrogram is extremely unbalanced and the complexity drops to @xmath48 . as demonstrated in precedence , the hierarchical framework can provide a provable computational gain , provided that the resulting tree is well - balanced . the worst case complexity , corresponding to a sequence of single image additions , is no better than the standard sequential approach . it is therefore crucial to ensure a good balance during the clustering phase . our solution is to employ a novel clustering procedure , which promotes the creation of better balanced dendrograms . the image clustering procedure proposed in the previous section allows us to organize the available images into a hierarchical cluster structure ( a tree ) that will guide the structure - and - motionprocess . this approach decreases the computational complexity with respect to sequential structure - and - motionpipelines , from @xmath48 to @xmath49 in the best case , i.e. when the tree is well balanced ( @xmath1 is the number of images ) . if the tree is unbalanced this computational gains vanishes . it is therefore crucial to enforce the balancing of the tree . the preceding solution , which used the simple rule , specified that the distance between two clusters is to be determined by the distance of the two closest objects ( nearest neighbors ) in the different clusters . in order to produce better balanced trees , we modified the agglomerative clustering strategy as follows : starting from all singletons , each sweep of the algorithm merges the pair with the smallest cardinality among the @xmath47 closest pair of clusters . the distance is computed according to the simple linkage rule . the cardinality of a pair is the sum of the cardinality of the two clusters . in this way we are softening the `` closest first '' agglomerative criterion by introducing a competing `` smallest first '' principle that tends to produce better balanced dendrograms . the amount of balancing is regulated by the parameter @xmath47 : when @xmath59 this is the standard agglomerative clustering with no balancing ; when @xmath60 ( @xmath1 is the number of images ) a perfect balanced tree is obtained , but the clustering is poor , since distance is largely disregarded . figure [ fig:2.9 ] shows an example of balancing achieved by our technique . the height of the tree is reduced from 14 to 9 and more initial pairs are present in the dendrogram on the right . in the pursuit of further complexity reduction , we adopted a strategy that consists in reducing the number of images to be used in the bundle adjustment in place of the whole model . this strategy is an instance of local bundle adjustment @xcite , which is often used for video sequences , where the active images are the most recent ones . let us concentrate on the model merging step , as the resection is a special case of the latter . consider two models @xmath61 and @xmath62 , where @xmath61 has fewer images than @xmath62 . we always transform the smallest onto the largest ( if one is projective it is always the smallest ) . the bundle adjustment involves all the images of @xmath61 and the subset of images of @xmath62 that share some tracks with @xmath61 ( tie - points that are visible in images in both models ) . let us call this subset @xmath63 . all the tie - points linking @xmath63 and @xmath61 are considered in the bundle adjustment . images in @xmath64 are not moved by bundle adjustment but their tie - points are still considered in the minimization in order to anchor @xmath63 through their common tie - points . the tie - points linking only cameras in @xmath64 are not considered . this strategy is sub - optimal because in a proper bundle adjustment all the images in @xmath62 should be involved , even those that do not share any tie - point with @xmath61 . however , a bundle adjustment with all the images and all the tie - points can be run at the end to obtain the optimal solution . a model that differs from the true one by a projectivity is called _ projective_. a model that differs from the true one by a similarity is called euclidean . the latter can be achieved when calibration parameters are known , the former can be obtained if images are uncalibrated . in this section we relax the hypothesis that images are calibrated and integrate the autocalibration algorithm in our pipeline , so that the resulting model is still euclidean . the main difference from the procedure described in sec . [ sec : hierarchical ] is that now leaf nodes do not have proper calibration right from the start of the structure - and - motionprocess . the models is projective at the beginning , and as soon as one reaches a sufficient number of images , the euclidean upgrade procedure ( described in section [ sec : autocal ] ) is triggered . moreover , each step of hierarchical structure - and - motionmust be modified to accommodate for projective models , as described in sections [ sec:2vu ] , [ sec : res - int ] , and [ sec : merge ] . autocalibration starts from a projective model and seeks the collineation of space @xmath65 so as to transforms the model into a euclidean one . without loss of generality , the first camera of the euclidean model can be assumed to be @xmath66 $ ] , so that the euclidean upgrade @xmath65 has the following structure , since @xmath67 : @xmath68 \label{eq : heuc}\ ] ] where @xmath69 is the calibration matrix of the first camera , @xmath70 is a vector which determines the location of the plane at infinity and @xmath71 is a scale factor . our autocalibration technique is based on two stages : 1 . given a guess on the internal parameters of two cameras compute a consistent upgrading collineation . this yields an estimate of all cameras but the first . 2 . score the internal parameters of these @xmath17 cameras based on the likelihood of skew , aspect ratio and principal point . the space of the internal parameters of the two cameras is enumerated and the best solution is refined via non - linear least squares . this approach has been introduced in @xcite , where it is compared with several other algorithms obtaining favorable results . this section describes a closed - form solution for the plane at infinity ( i.e. , the vector @xmath70 ) given two perspective projection matrices and their internal parameters . while the first camera is @xmath72 $ ] , the second projective camera can be written as @xmath73 $ ] , and its euclidean upgrade is : @xmath74 \simeq p_2 h = \left [ a_2 k_1 + \mathbf{e}_2 \mathbf{r}^{\top } \vert \lambda \mathbf{e}_2 \right ] .\end{aligned}\ ] ] the rotation @xmath75 can therefore be equated to : @xmath76 using the constraints on orthogonality between rows or columns of a rotation matrix , one can solve for @xmath70 finding the value that makes the right hand side of equal to a rotation , up to a scale . the solution can be obtained in closed form by noting that there always exists a rotation matrix @xmath77 such as : @xmath78^{\top},$ ] where @xmath79 . left multiplying it to yields : @xmath80^{\top } \mathbf{r}^{\top}\ ] ] calling @xmath81 and its rows @xmath82 , we arrive at the following : @xmath83 / \arrowvert \mathbf{w}_3 \arrowvert \label{eq:8}\ ] ] in which the last two rows are independent of the value of @xmath70 and the correct scale has been recovered normalizing each side of the equation to unit norm . since the rows of @xmath84 are orthonormal , we can recover the first one taking the cross product of the other two . vector @xmath70 is therefore equal to : @xmath85 the upgrading collineation @xmath65 can be computed using ; the term @xmath71 can be arbitrarily chosen , as it will just influence the overall scale . when the calibration parameters are known only approximately , the right hand side of is no more a rotation matrix . however , will still yield the value of @xmath70 that will produce an _ approximate _ euclidean model . in the preceding section we showed how to compute the euclidean upgrade @xmath65 given the calibration parameters of two cameras of the projective model . the autocalibration algorithm loops through all possible internal parameter matrices of two cameras @xmath86 and @xmath87 , checking whether the entire upgraded model has the desired properties in terms of @xmath88 . the process is well - defined , since the search space is naturally bounded by the finiteness of the acquisition devices . in order to sample the space of calibration parameters we can safely assume , as customary , null skew and unit aspect ratio : this leaves the focal length and the principal point location as free parameters . however , as expected , the value of the plane at infinity is in general far more sensitive to errors in the estimation of focal length values rather than the image center . thus , we can iterate just over focal lengths @xmath89 and @xmath90 assuming the principal point to be centered on the image ; the error introduced with this approximation is normally well within the radius of convergence of the subsequent non - linear optimization . the search space is therefore reduced to a bounded region of @xmath91 . to score each sample @xmath92 , we consider the aspect ratio , skew and principal point location of the upgraded ( i.e. , transformed with @xmath65 ) camera matrices and aggregate their respective value into a single cost function : @xmath93 where @xmath94 is the internal parameters matrix of the @xmath47-th camera after the euclidean upgrade determined by @xmath92 , and @xmath95 reflects the degree to which @xmath96 meets a - priori expectations . let us consider the _ viewport matrices _ of the cameras , defined as : @xmath97 \label{eq : viewport}\ ] ] where @xmath98 and @xmath99 are respectively the width and height of each image . camera matrices are normalized with @xmath100 . in this way , the principal point expected value is @xmath101 and the focal range is @xmath102 $ ] . therefore , the term of the cost function writes : @xmath103 where @xmath104 denotes the entry @xmath105 of @xmath96 and @xmath98 are suitable weights , computed as in @xcite . the first term takes into account the skew , which is expected to be 0 , the second one penalizes cameras with aspect ratio different from 1 and the last two weigh down cameras where the principal point is away from @xmath101 . finally , the solution selected is refined by non - linear minimization of eq . . since it is usually very close to a minimum , just a few iterations of a levenberg - marquardt solver are necessary for convergence . ( ) @xmath106 _ compute @xmath107 _ _ build @xmath65 from _ ( ) @xmath108 @xmath109 @xmath110 _ internal of @xmath111 _ _ compute @xmath112 from _ _ aggregate cost and select minimum _ _ refine non - linearly _ the entire procedure is presented as pseudo - code in algorithm [ alg : selfcal ] . the algorithm shows remarkable convergence properties ; it has been observed to fail only when the sampling of the focal space was not sufficiently dense ( in practice , less than twenty focal values in each direction ) , and therefore all the tested infinity planes were not close enough to the correct one . such problems are easy to detect , since they usually take the final , refined solution outside the legal search space . in principle , autocalibration requires a minimum number of images to work , according to the autocalibration `` counting argument '' @xcite ( e.g. @xmath113 images with known skew and aspect ratio ) . however , as we strive to maintain an `` almost '' euclidean reference frame from the beginning , to better condition subsequent processing , autocalibration is triggered for models starting from two images . the result is an approximate euclidean upgrade ; in fact these models are still regarded as projective , until they reach a sufficient cardinality . after that point autocalibration is not performed any more and the internal parameters of each camera are refined further only with bundle adjustment , as the computation proceeds . in order not to hamper the process too much , the internal parameters of a camera becomes fixed after they have been bundle - adjusted together with a given number of cameras . the model that can be obtained from two uncalibrated images is always projective . the following two camera matrices are used : @xmath114 \quad \text{and } \quad p_2 = [ [ \mathbf{e}_2]_{\times}f \ \vert \ \mathbf{e}_2],\ ] ] this canonical pair yields a projective model with the plane at infinity passing through the centre of the second camera , which is very unnatural . therefore , the method of section [ sec : autocal1 ] is applied for guessing a better location for the plane at infinity compatible with rough focal estimates , obtained from the magnitude of the image diagonal . even when the true focal lengths are far from the estimates , this procedure will provide a useful , well conditioned starting point for the subsequent steps . cheirality is then tested and enforced on the model . in practice only a reflection ( switches all points from in front to behind the cameras ) may be necessary , as the twisted pair case never occurs . in fact , the twisting corresponds to the infinity plane crossing the baseline , which would imply that our guess for the infinity plane is indeed very poor . the 3d coordinates of the tie - points are then obtained by intersection as before . finally bundle adjustment is run to improve the model . the procedure is the same as in the calibrated case , taking care of using the direct linear transform ( dlt ) algorithm @xcite for resection , as the the single image is always uncalibrated . while ppnp computes only the external parameters of the camera , the dlt computes the full camera matrix . partial models live in two different reference frames , that are related by a similarity if both are euclidean or by a projectivity if one is projective . in this case the projectivity that brings the projective model onto the euclidean one is sought , thereby recovering its correct euclidean reference frame . the procedure is the same as in the calibrated case , with the only difference that when computing the projectivity the dlt algorithm should be used instead of op . the new model is refined with bundle adjustment ( either euclidean or projective ) and upgraded to a euclidean frame when the conditions stated beforehand are met . samanthais a complex pipeline with many internal parameters . with respect to this issue our endeavor was : i ) to avoid free parameters at all ; ii ) to make them data - dependent ; iii ) to make user - specified parameters intelligible and subject to an educated guess . in the last case a default should be provided that works with most scenarios . this guarantees the processing to be completely automatic in the majority of cases . all the heuristic parameter settings used in the experiments have been reported and summarized in table [ tab : parameters ] . + number of log pyramid levels & 12 + average number of keypoints per image & 7500 + + number of keypoints per image & 300 + degree of edge - connectedness & 8 + + matching discriminative ratio & 1.5 + maximum number of msac iterations & 1000 + bucket size ( msac ) in pixels & @xmath115 + minimum number of matches & 10 + homography - to - fundamental gric ratio & 1.2 + minimum track length & 3 + + maximum bundle adjustment iterations & 100 + reprojection error & @xmath116 + + number of cameras for autocalibration & 4 + number of cameras to fix internal param.s & 25 + + final minimum track length & 2 + final maximum reprojection error & @xmath117 + [ [ keypoint - detection ] ] keypoint detection + + + + + + + + + + + + + + + + + + the keypoints extracted from all images are ordered by their response value and the ones with the highest response are retained , while the others are discarded . the total number of keypoints to be kept is a multiple of the number of images , so as to keep the average quota of keypoints for each image fixed . during the broad matching phase the goal is to compute the 2d histogram mentioned in sec . [ sec : broad ] . to this end , the keypoints in each image are ordered by scale and the 300 keypoints with the highest scale are considered . the number of neighbors in feature space is set to six , as in @xcite . given the nature of the broad phase , this value is not critical , to the point where only the number of keypoints is exposed as a parameter . the number @xmath11 in sec . [ sec : broad ] ( `` degree of edge - connectedness '' , in table [ tab : parameters ] ) , has been set to eight following @xcite . in our case , the role of the parameter is more `` global '' , as it does not set the exact number of images to be matched but the degree of edge - connectedness of the graph . however our experiments confirmed that that @xmath118 is a good default . the number of max iterations of msac is set to @xmath119 during the matching phase . this is only an upper bound , for the actual value is dynamically updated every time a new set of inliers is found . the `` matching discriminative ratio '' refers to the ratio of first to second closest keypoint descriptors , used to prune weak matches at the beginning of this phase . the `` minimum number of matches '' parameter refers to the last stage of the narrow phase , when poor matches between images are discarded based on the number of surviving inliers after msac . [ [ clustering ] ] clustering + + + + + + + + + + the parameter @xmath47 of sec . [ sec : balance ] has been set to @xmath120 based on the graph reported in fig . [ fig : bilancia ] , where the number of reconstructed tie - points / images and the computing time are plotted as the value of @xmath47 is increased . after @xmath120 , the computing time stabilizes at around 30% of the baseline case , without any significant difference in terms of number of reconstructed images and tie - points . in the balancing heuristics . the values on the ordinate are in percentage, where the baseline case @xmath121 . ] [ [ reconstruction ] ] reconstruction + + + + + + + + + + + + + + the safeguard threshold on the reprojection error ( sec . [ sec : inter ] ) is set to 2 pixels with reference to a 6 mpixel image , and scaled according to the actual image diagonal ( assuming an aspect ratio of 4:3 , the reference diagonal is @xmath122 pixels ) . [ [ autocalibration ] ] autocalibration + + + + + + + + + + + + + + + the euclidean upgrade is stopped as soon as a cluster reaches a sufficient cardinality @xmath14 ( `` number of cameras for autocalibration '' ) that satisfies the following inequality @xcite , giving the condition under which autocalibration is feasible : @xmath123 where @xmath124 internal parameters are known and @xmath125 internal parameters are constant . with known ( or guessed ) skew and aspect ratio ( @xmath126 and @xmath127 ) four cameras are sufficient . the reason for keeping this value to the minimum is because we observed experimentally that projective alignment is fairly unstable and it is beneficial to start using a similarity transformation as soon as possible . the cluster cardinality after which the internal parameters are kept fixed in the bundle adjustment is set to 25 , a fairly high value , that guarantees all the internals parameters , especially the radial distortion ones , are steady . [ [ local - bundle - adjustment ] ] local bundle adjustment + + + + + + + + + + + + + + + + + + + + + + + as discussed previously , the local bundle adjustment is generally to be preferred over the full one . however , since the autocalibration phase is crucial , our strategy is to run the full bundle adjustment until the clusters become euclidean . it should also be noted that the computational benefits of the local bundle adjustment are more evident with large clusters of cameras . [ [ prologue ] ] prologue + + + + + + + + the last bundle adjustment is always full ( not local ) and is run with a lower safeguard threshold on the reprojection error ( 1.5 pixel in the reference 6 mpixel image ) . a final intersection step is carried out using also the tracks of length two , in order to increase the density of the model . please note however that these weak points do not interfere with the bundle adjustment , as they are added only _ after _ it . we run samanthaon several real , challenging datasets , summarized in tab . [ tab : summary ] . all of them have some ground truth available , being either a point cloud ( from laser scanning ) , the camera internal parameters or measured `` ground '' control points ( gcp ) . the qualitative comparison against visualsfmfocuses on differences in terms of number of reconstructed cameras and manifest errors in the model . the quantitative evaluation examines the accuracy of the model against gcps and laser scans , and/or the internal orientation accuracy . [ cols="<,>,^ , < , < , < " , ] [ tab : runningtimes ] in this paper we have described several improvements to the current state of the art in the context of uncalibrated structure - and - motionfrom images . our proposal was a hierarchical framework for structure - and - motion(samantha ) , which was demonstrated to be an improvement over the sequential approach both in computational complexity and with respect to the overall error containment . samanthaconstitutes the first truly scalable approach to the problem of modeling from images , showing an almost linear complexity in the number of tie - points and images . moreover , we described a novel self - calibration approach , which coupled with our hierarchical pipeline ( samantha ) constitutes the first published example of uncalibrated structure - and - motionfor generic datasets not using external , ancillary information . the robustness of our approach has been demonstrated on 3d model datasets both qualitatively and quantitatively . this technology has now been transferred to a company ( 3dflow srl ) which produced an industry grade implementation of samanthathat can be freely downloaded . thesis : the subgraph @xmath10 produced by the method reported in sec . [ sec : broad ] is @xmath11-edge - connected , provided that @xmath11 independent spanning tree can be extracted from @xmath16 . to prove the thesis we rely upon the following observation . consider an undirected graph @xmath16 with the capacity of all edges set to one ; @xmath16 is @xmath14-edge - connected if and only if the maximum flow from @xmath128 to @xmath129 is at least @xmath14 for any node pair @xmath130 . since our @xmath10 is the union of @xmath11 independent ( disjoint ) 1-edge - connected graphs , each of them adds an independent path with unit capacity from every node pair @xmath130 , so the maximum flow from every pair @xmath130 in @xmath10 is @xmath11 . the `` ventimiglia '' images ( courtesy of fabio remondino ) were acquired by fbk trento ( 3dom.fbk.eu ) within a project funded by arcus spa and supported by direzione regionale per i beni culurali e paesaggistici della ligura ( mibac ) and soprintendenza per i beni archeologici della liguria . the laser scanning of `` br '' was conducted by gexcel s.r.l . with the eu jrc - ispra and the permission of the municipality of verona . the laser data of the `` duomo di pisa '' comes from the `` cattedrale digitale '' project , while the photo set is courtesy of the visual computing lab ( isti - cnr , pisa ) . domenico visintini ( universit di udine ) took the laser survey of `` s. giacomo '' . the images of the `` termica '' dataset have been captured by airmap ( curtesy of davide zorzetto ) . alberto beinat and marco fassetta ( universit di udine ) surveyed the gcps . the navona dataset was acquired by l. barazzetti ( politecnico di milano , italy ) and used for the evaluation of automated image orientation methods during the isprs 3d - arch workshop in 2011 . the `` herz - jesu - p25 '' images come from epfl dense multi - view stereo evaluation dataset . t. thormhlen , h. broszio , a. weissenfeld , keyframe selection for camera motion and structure estimation from multiple views , in : proceedings of the european conference on computer vision , 2004 , pp . 523535 . shum , q. ke , z. zhang , efficient bundle adjustment with virtual key frames : a hierarchical approach to multi - frame structure from motion , in : proceedings of the ieee conference on computer vision and pattern recognition , 1999 , pp . 25382543 . f. schaffalitzky , a. zisserman , multi - view matching for unordered image sets , or `` how do i organize my holiday snaps ? '' , in : proceedings of the 7th european conference on computer vision , 2002 , pp . 414431 . r. gherardi , m. farenzena , a. fusiello , improving the efficiency of hierarchical structure - and - motion , in : proceedings of the ieee conference on computer vision and pattern recognition , san francisco , ca , 2010 , pp . 1594 1600 . s. agarwal , y. furukawa , n. snavely , i. simon , b. curless , s. m. seitz , r. szeliski , building rome in a day , communications of the acm 54 ( 10 ) ( 2011 ) 105112 . frahm , p. fite - georgel , d. gallup , t. johnson , r. raguram , c. wu , y .- h . jen , e. dunn , b. clipp , s. lazebnik , m. pollefeys , building rome on a cloudless day , in : proceedings of the 11th european conference on computer vision : part iv , 2010 , pp . 368381 . m. arie - nachimson , s. z. kovalsky , i. kemelmacher - shlizerman , a. singer , r. basri , global motion estimation from point matches , in : international conference on 3d imaging , modeling , processing , visualization and transmission , 2012 , pp . 8188 . d. crandall , a. owens , n. snavely , d. p. huttenlocher , discrete - continuous optimization for large - scale structure from motion , in : proceedings of the ieee conf . on computer vision and pattern recognition , 2011 , pp . 30013008 . m. brand , m. antone , s. teller , spectral solution of large - scale extrinsic camera calibration as a graph embedding problem , in : proceedings of the ieee conference on computer vision and pattern recognition , 2004 , pp . 262273 . p. moulon , p. monasse , r. marlet , global fusion of relative motions for robust , accurate and scalable structure from motion , in : proceedings of the international conference on computer vision , 2013 , pp . 32483255 . o. ozyesil , a. singer , r. basri , http://arxiv.org/abs/1312.5047[camera motion estimation by convex programming ] , arxiv e - prints 1312.5047 , to appear in the siam journal on imaging sciences . g. kamberov , g. kamberova , o. chum , s. obdrzalek , d. martinec , j. kostkova , t. pajdla , j. matas , r. sara , 3d geometry from uncalibrated images , in : proceedings of the 2nd international symposium on visual computing , 2006 , pp . 802813 . m. pollefeys , d. nistr , j. m. frahm , a. akbarzadeh , p. mordohai , b. clipp , c. engels , d. gallup , s. j. kim , p. merrell , c. salmi , s. sinha , s. sinha , b. talton , l. wang , q. yang , h. stewnius , r. yang , g. welch , h. towles , detailed real - time urban 3d reconstruction from video , international journal of computer vision 78 ( 2 - 3 ) ( 2008 ) 143167 . m. pollefeys , r. koch , l. van gool , self - calibration and metric reconstruction in spite of varying and unknown internal camera parameters , in : proceedings of the international conference on computer vision , bombay , 1998 , pp . 9095 . y. seo , a. heyden , r. cipolla , a linear iterative method for auto - calibration using the dac equation , in : proceedings of the ieee conference on computer vision and pattern recognition , vol . 1 , 2001 , p. 880 . m. chandraker , s. agarwal , f. kahl , d. nister , d. kriegman , autocalibration via rank - constrained estimation of the absolute quadric , in : proceedings of the ieee conference on computer vision and pattern recognition , 2007 , pp . b. bocquillon , a. bartoli , p. gurdjos , a. crouzil , on constant focal length self - calibration from multiple views , in : proceedings of the ieee conference on computer vision and pattern recognition , 2007 . m. chandraker , s. agarwal , d. kriegman , s. belongie , globally optimal affine and metric upgrades in stratified autocalibration , in : proceedings of the international conference on computer vision , 2007 , pp . m. farenzena , a. fusiello , r. gherardi , structure - and - motion pipeline on a hierarchical cluster tree , in : ieee international workshop on 3-d digital imaging and modeling , kyoto , japan , 2009 , pp . 14891496 . z. zhang , r. deriche , o. faugeras , q .- t . luong , a robust technique for matching two uncalibrated images through the recovery of the unknown epipolar geometry , artificial intelligence 78 ( 1 - 2 ) ( 1995 ) 87 119 . t. quack , b. leibe , l. van gool , world - scale mining of objects and events from community photo collections , in : proceedings of the international conference on content - based image and video retrieval , 2008 , pp . 4756 . v. garro , f. crosilla , a. fusiello , solving the pnp problem with anisotropic orthogonal procrustes analysis , in : proceedings of the second joint 3dim/3dpvt conference : 3d imaging , modeling , processing , visualization and transmission , 2012 , pp . 262269 . e. mouragnon , m. lhuillier , m. dhome , f. dekeyser , p. sayd , real time localization and 3d reconstruction , in : proceedings of the international conference on computer vision and pattern recognition , 2006 , pp . 363370 . f. remondino , s. del pizzo , t. p. kersten , s. troisi , low - cost and open - source solutions for automated image orientation a critical overview , in : progress in cultural heritage preservation , springer , 2012 , pp . 4054 . c. strecha , w. von hansen , l. van gool , p. fua , u. thoennessen , on benchmarking camera calibration and multi - view stereo for high resolution imagery , in : ieee conference on computer vision and pattern recognition , 2008 , pp .

we show that charge fluctuation processes are crucial for the nonlinear heat conductance through an interacting nanostructure , even far from a resonance . we illustrate this for an anderson quantum dot accounting for the first two leading orders of the tunneling in a master equation . the often made assumption that off - resonant transport proceeds entirely by virtual occupation of charge states , underlying exchange - scattering models , can fail dramatically for heat transport . the identified energy - transport resonances in the coulomb blockade regime provide new qualitative information about relaxation processes , for instance by strong _ negative _ differential heat conductance relative to the heat current . these can go unnoticed in the charge current , making nonlinear _ heat - transport spectroscopy _ with energy - level control a promising experimental tool . recently the experimental investigation of heat transport on the nanoscale has become possible . these measurements are accurate enough to investigate the heat dissipation in molecular junctions with conductances as low as . additionally integrating energy - level control into thermoelectric junctions , e.g. , by mechanical gating , does not seem out of reach and just recently , electrically - gated thermoelectric nanojunctions have been demonstrated . here , by analysing the generic effects of coulomb interactions on the nonlinear heat transport in nanoscale systems , we will show that this is very promising . interaction effects have long been probed using gate controlled _ charge_-current spectroscopy , a well - developed experimental tool to access the discrete quantum levels of nanostructures . two prominent features in the charge current driven by a source - drain voltage underpin this successful method . the first is resonant or single - electron tunneling ( set ) which depends on the level position relative to the electro - chemical potential , in ( a ) : an electron jumps into or out of an orbital level directly leading to a real change of its occupancy . the current shows sharp steps as new resonant transport processes are switched on with increasing bias . these processes are routinely identified in a three - terminal setup by plotting the charge conductance as function of the applied bias and the gate voltage , as exemplified in ( a ) . two - terminal measurements , e.g. , using a scanning probe , correspond to line traces through such a plot . the second type of resonance is independent of the level position and appears as a horizontal line at since it originates in the inelastic excitation by an energy at fixed local electron number on the nanostructure . this off - resonant feature requires a second - order tunneling process in which an electron `` scatters through '' , other charge states being only visited virtually , see ( b ) . this is known as inelastic electron tunneling ( iets ) or inelastic cotunneling ( icot ) . . [ fig:0 ] ] this resonance develops into a nonequilibrium kondo resonance for low and low temperatures which is much sharper than the feature corresponding to , providing better access to a range of physical phenomena _ in situ _ : an electronic level splitting ( e.g. , in a semiconductor nanostructure , carbon nanotube , or a dopand atom ) , a quantized vibrational frequency , or a spin - splitting due to a magnetic field , exchange interaction , magnetic anisotropy ( e.g. , in molecules or ad - atoms ) , or spin - orbit coupling . thermoelectric transport has also been investigated within the two above mentioned physical transport pictures . theory mostly focused on the thermopower in the linear - response regime . this includes the study of , and kondo processes . works addressing the nonlinear regime have either applied effective single - particle descriptions or focused on thermoelectric devices close to resonance assuming weak tunneling or weak coulomb interaction . the heat current has received much less attention . a classification of _ nonlinear heat - transport _ features for a strongly interacting nanostructure going beyond weak tunneling , matching that of charge transport , still seems to be missing . this is important both for scanning probe setups as well as thermoelectric setups with energy - level control . in this rapid communication we address this problem and show that the heat current driven by a nonlinear electric and/or thermal bias contains new qualitative information and deviates in a striking way from the charge transport , both in sign and amplitude . its dependence on the level position reveals that relaxation processes of first order [ ( a ) ] in the tunneling can be very important for heat transport _ far from resonance _ ( i.e. , energy detuning larger than temperature ) . the crucial competition with an inelastic second - order excitation at finite voltage bias [ ( b ) ] leads to real occupation of more than one charge state and is missed by inelastic transport theories relying on effective exchange- and potential - scattering amplitudes . , inelastic excitation , and tunnel coupling . ( a ) charge conductance / [ \gamma^2/u^2 ] \right)$ ] and ( b ) energy conductance / [ \gamma^2/u ] \right)$ ] using the signed log , for with , linearized to for . labels ( i)(vii ) indicate the features discussed in the text but are not labeled at horizontally mirrored positions . inset to ( b ) : linear energy conductance [ a.u . ] versus around the right set resonance . [ fig:1 ] ] _ model and method._to illustrate the generic picture of nonlinear thermoelectric transport through an interacting nanoscale object we analyze a resonant level with strong coulomb interaction and a well - defined spin - flip excitation due to an external field . it is described by the anderson quantum dot model which also suffices to classify nonlinear thermoelectric transport features for more complex models . these features should generally be observable in a range of nanostructures , at least for large level spacings and quasi - regular electron filling of energy shells . the dot is described by , where with are the electron operators on the dot . here is the orbital energy level and denotes the energy of a local spin excitation for fixed due to a magnetic field , where is the electron number . furthermore , is the strong coulomb energy penalty paid when counting electrons in the shell . the electrodes , indexed by=l , r , are described as noninteracting reservoirs , , with electron operators . we allow for a nonlinear voltage bias between the reservoirs through their electrochemical potentials with temperatures . ( we comment on nonlinear thermal bias effects later on . ) the tunnel coupling has the generic form , the bare resonance width is given by the tunnel rate , with the density of states in the reservoirs and we set . we will only consider the stationary currents entering the right reservoir . the charge current is , where is the electron number operator of the reservoir . similarly , the energy current is defined via . the measurable heat current can be obtained via . since experimentally the way the voltage is applied is known ( here ) and the conserved charge current is available , the conversion from to amounts to a simple background subtraction . below we focus on the contribution since it contains all interesting physical features . also , and are more easily compared , highlighting the differences between charge and heat transport most directly , in particular the bias and gate voltage dependence on which we focus here . we use as unit of energy ; in experiments it is readily obtained from the height of the coulomb diamond ( cf . ) and ranges from 0.1 - 10 mev in semiconductor and carbon nanotube quantum dots to 10 - 100 mev in molecular and atomic quantum dots . the effects in the currents that we focus on below scale as for the parameter regime of interest when fixing relative to . estimations based on this indicate that the predicted energy currents may be in range of experimental resolution of tens of nw , in particular in molecular junctions . the currents and the underlying nonequilibrium dot - state occupations are calculated using a reduced density - operator transport theory accounting for the strong local interaction . this approach is perturbative in the tunneling rates and well - controlled in the regime . while keeping this restriction we recover for the corresponding results of the landauer approach . we go beyond standard approaches by including the competition of all tunneling rates and [ ] into the stationary master equation for the occupations , see for more details of the calculations of the transition rate matrix and the current . we focus on the dominant energy dependence introduced by the interacting quantum dot , assuming a flat spectral density in the wide - band limit for the electrodes . _ effects._already a first glance at the charge and energy conductance plotted in ( a ) and ( b ) , respectively , reveals that the energy transport spectrum is much richer : there is a significant gain in the contrast due to its many sign changes . in these plots the stage is set by features due to processes of [ ( a ) ] which are well understood . these occur when one of the four single - electron addition energies , and ( ) matches . this happens , e.g. , at the lines labeled ( i)(iii ) in . indicated by ( i ) are transitions between the ground states of subsequent electron numbers of the dot ( and ) . as shown in the inset in ( b ) , there are sawtooth - shaped resonances in as function of the level position corresponding to the coulomb peaks in associated with one - electron processes and , respectively . the sign change in reflects that excess energy is carried by electrons or holes . this basic energy transport feature reappears at several positions in ( b ) , e.g. , also when at ( ii ) a process additionally excites the dot or at ( iii ) such a process starts off in the excited state , the process of ( a ) . the broadening of these lines is determined by the temperature for . _ effects._qualitative differences show up inside the central off - resonant regime opened up by the coulomb interactionwhere the simple resonant picture just discussed breaks down . here coherent electron - hole processes of that leave fixed , such as ( b ) , become important as well . these give rise to _ qualitatively _ new effects not captured by master equations or even approaches that also include tunnel broadening and shifts . for voltages only elastic tunneling processes are possible which produce a smooth nonexponential background in both ( qualitatively similar to that found for metallic islands ) and . however , above the threshold line , indicated in red by ( iv ) in the schematic ( b ) , a new process sets in : electrons tunnel onto and off the dot , while depositing an energy as sketched in ( b ) . this yields the characteristic step in the charge conductance in ( a ) at all across the off - resonant regime . our calculations show that the energy conductance also shows such an feature at the corresponding line ( iv ) in ( b ) . as expected , it changes sign when electron and hole processes change roles , similar to the sawtooth - shaped resonances discussed above , but now when tuning the level position through the _ center _ of the off - resonant regime ( ) . inspection of the magnitude of the inelastic step at as function of the level position in ( a ) reveals a dramatic difference : whereas the charge conductance amplitude at is smooth and featureless as is varied , the energy conductance amplitude sharply drops at ( v ) when or . this big difference also shows up in ( b ) where the central part of the horizontal onset is completely missing , in contrast to ( a ) . the strong reduction of when entering the central region is remarkable : everywhere in ( a ) we are still far from resonance , i.e. , . there is a second regime where the behavior of the energy conductance radically deviates from that of the charge conductance : the bias dependence plotted in ( c ) shows at ( vi ) strong _ negative _ differential energy conductance relative to the energy current by far exceeding the feature ( v ) discussed below in magnitude . as the inset indicates , after encountering the large energy current change at ( v ) the energy current drops back at ( vi ) . also this shows up in ( b ) as sharp blue ( red ) boundaries of the diamond - shaped region containing the label ( vi ) on the red ( blue ) background . the charge conductance in ( c ) is again featureless there . we now explain in three steps ( a)-(c ) how physically these dramatic differences come about , by following the vertical line in the schematic ( b ) at . our discussion explains which processes lead to changes in the occupations and transport rates and were substantiated by numerical calculations . ( a ) starting in the state at and increasing the bias we first hit the red threshold ( iv ) at . beyond this line the excited state becomes occupied by inelastic tunneling [ ( b ) ] and it relaxes by similar inelastic processes . these transitions are indicated by red arrows in the lower inset in ( b ) . this gives an increased charge conductance as electrons find an additional path through the quantum dot while keeping the charge fixed to , only virtually visiting other charge states . in contrast , the energy current is relatively low due to two effects : first , real charge fluctuations are suppressed ( ) and only processes occur , the rates for both of which are , much smaller than the rates . second , there is a significant partial cancellation of the energy currents of these processes , namely of a positive contribution due the relaxation process , and a negative contribution due to the excitation process [ ( b ) ] . this relates to the generic electron - hole symmetry between two coulomb - split set resonances associated with filling a single orbital shell , captured by our model . \(b ) increasing further and crossing the green line ( v ) in ( b ) , a two - step relaxation path is switched on , indicated by the green arrows in the right inset . this sharply increases the magnitude of the energy current at ( v ) in ( c ) since it lifts the above cancellation of opposing energy currents of comparable magnitude : the relaxation process ( red dashed downward arrow ) is overridden by a much faster two - step relaxation ( green arrows ) , of ( a ) followed by . this composite mechanism is called cotunneling assisted set ( coset ) and involves _ real _ occupation of the state , despite the prevalence of states due to coulomb blockade . \(c ) when finally crosses the blue line ( vi ) in ( b ) the _ other _ green relaxation path via the state also becomes active as the upper inset shows . although still far from resonance , _ both _ the and charge state become occupied for real because both relaxation pathways are turned on . remarkably , this increased relaxation does not increase the energy current as above , but instead suppresses it . the reason is that the signed energy - current contributions from the two pathways now cancel each other in contrast to case ( b ) where only one such pathway is active thereby strongly reducing the energy current . _ discussion._we establish the complete classification of the nonlinear energy transport by noting the additional feature due to the tunneling of _ pairs _ of either electrons or holes , which is again more prominent in the energy conductance [ line ( vii ) in ] . much of the above remains qualitatively the same when including a junction and spin dependence of the tunneling constants , or a combined voltage ( ) and thermal bias ( ) : interestingly , in the latter case the resonances may be used experimentally to estimate the thermal gradient _ in situ _ . the above described nonequilibrium competition between real and virtual processes together with the sign of energy currents leads to an unexpectedly rich energy current spectrum . importantly , for more complex multi - level quantum dots the above identified elementary signatures are just repeated every time a new electronic orbital is filled when scanning the gate voltage . our model captures this generic pattern which is well attested experimentally for the charge current . however , we even find that several replicas of these features can appear in energy transport inside the coulomb blockade regime ( e.g. , as negative relative to ) , but also _ outside _ , at higher voltage , again in stark contrast to charge transport . combining a three - terminal setup with measurements of the energy conductance may thus reveal new qualitative information about relaxation processes . the much enhanced effect of at ( v ) and ( vi ) in the energy transport of ( b ) should be experimentally accessible since even the weaker features in the charge transport of ( a ) have been measured . our results also indicate that the analysis of two - terminal thermoelectric measurements requires extra care due to the lack of gate - spectroscopic information . often the two regimes of pure and are not distinguished . for the charge conductance this may not seem so important , but our results show that for the energy conductance this distinction is absolutely vital . theoretical descriptions used to model scanning - probe experiments and quantum dots in the coulomb blockade regime are often based on effective models including only effective exchange and potential scattering terms . these may fail badly for the energy current since they include only ( keeping fixed ) and eliminate the important real charge fluctuations ( to and ) involved in coset . in ( b ) such an approximation is suitable only in a limited regime , the triangle labeled ( iv ) . extending the model to include more inelastic excitations further narrows it down . nonlinear energy transport thus requires careful consideration : one needs a physical _ model _ allowing for charge fluctuations as well as a nonequilibrium transport theory that captures at least the first _ two _ leading - orders of tunneling processes for strong interaction . experimentally , even higher - order tunneling effects may be important and it is of interest to explore renormalization effects . we thank c. cuevas , d. divincenzo , f. haupt , m. hell , m. leijnse , p. reddy , and r. saptsov for valuable discussions . nmg , cbmh and ds thank the institute for theory of statistical physics , rwth aachen university , where substantial parts of this work have been performed . this work is part of the d - itp consortium , an nwo program funded by the dutch ministry of education , culture and science ( ocw ) . cbmh and ds were supported by the dfg through the emmy - noether program under schu 2333/2 - 1 . 10 p. reddy , s .- y . jang , r. a. segalman , and a. mujamdar , science * 315 * , 1568 ( 2007 ) . k. baheti , j. a. malen , p. doak , p. reddy , s .- y . jang , t. d. tilley , a. mujamdar , and r. a. segalman , nano lett . * 8 * , 715 ( 2008 ) . j. r. widawsky , p. darancet , j. b. neaton , and l. venkataraman , nano lett . * 12 * , 354 ( 2012 ) . w. lee , k. kim , w. jeong , l. a. zotti , f. pauly , j. c. cuevas , and p. reddy , nature * 498 * , 209 ( 2013 ) . r. temirov , a. lassise , f. b. anders , and f. s. tautz , nanotechnology * 19 * , 065401 ( 2008 ) . y. kim , w. jeong , k. kim , w. lee , and p. reddy , nature nanotechnology * 9 * , 881 ( 2014 ) . j. lambe and r. c. jaklevic , phys . rev . * 165 * , 821 ( 1968 ) . a. j. heinrich , j. a. gupta , c. p. lutz , and d. m. eigler , science * 306 * , 466 ( 2004 ) . l. i. glazman and k. a. matveev , jetp lett . * 51 * , 484 ( 1990 ) [ pisma zh . eksp . . fiz . * 51 * , 425 1990 ) ] . d. v. averin and y. v. nazarov , phys . rev . lett . * 65 * , 2446 ( 1990 ) . s. de franceschi , s. sasaki , j. m. elzerman , w. g. van der wiel , s. tarucha , and l. p. kouwenhoven , phys . rev . lett . * 86 * , 878 ( 2001 ) . j. paaske , a. rosch , p. wlfle , n. mason , c. m. marcus , and j. nygrd , nature phys . * 2 * , 460 ( 2006 ) . a. rosch , j. paaske , j. kroha , and p. wlfle , phys . rev . lett . * 90 * , 076804 ( 2003 ) . s. kehrein , phys . rev . lett . * 95 * , 056602 ( 2005 ) . h. schoeller and f. reininghaus , phys . rev . b * 80 * , 045117 ( 2009 ) ; _ ibid . _ * 80 * , 209901(e ) ( 2009 ) . d. m. zumbhl , c. m. marcus , m. p. hanson , and a. c. gossard , phys . rev . lett . * 93 * , 256801 ( 2004 ) . s. sapmaz , p. jarillo - herrero , j. kong , c. dekker , l. p. kouwenhoven , and h. s. j. van der zant , phys . rev . b * 71 * , 153402 ( 2005 ) . a. k. httel , b. witkamp , m. leijnse , m. r. wegewijs , and h. s. j. van der zant , phys . rev . lett . * 102 * , 225501 ( 2009 ) . f. a. zwanenburg , a. s. dzurak , a. morello , m. y. simmons , l. c. l. hollenberg , g. klimeck , s. rogge , s. n. coppersmith , and m. a. eriksson , rev . mod . phys . * 85 * , 961 ( 2013 ) . e. a. osorio , k. oneill , n. stuhr - hansen , o. f. nielsen , t. bjrnholm , and h. s. j. van der zant , adv . mater . * 19 * , 281 ( 2007 ) . j. e. grose , e. s. tam , c. timm , m. scheloske , b. ulgut , j. j. parks , h. d. abrua , w. harneit , and d. c. ralph , nature mat . * 7 * , 884 ( 2008 ) . j. j. parks , a. r. champagne , g. r. hutchison , s. flores - torres , h. d. abrua , and d. c. ralph , phys . rev . lett . * 99 * , 026601 ( 2007 ) . a. s. zyazin , j. w. g. van den berg , e. a. osorio , h. s. j. van der zant , n. p. konstantinidis , m. leijnse , m. r. wegewijs , f. may , w. hofstetter , c. danieli , and a. cornia , nano lett . * 10 * , 3307 ( 2010 ) . s. loth , k. von bergmann , m. ternes , a. f. otte , c. lutz , and a. heinrich , nature phys . * 6 * , 340 ( 2010 ) . t. jespersen , k. grove - rasmussen , j. paaske , k. muraki , t. fujisawa , j. nygrd , and k. flensberg , nature phys . * 7 * , 348 ( 2011 ) . c. w. j. beenakker and a. a. m. staring , phys . rev . b * 46 * , 9667 ( 1992 ) . m. turek and k. a. matveev , phys . rev . b * 65 * , 115332 ( 2002 ) . j. koch , f. von oppen , y. oreg , and e. sela , phys . rev . b * 70 * , 195107 ( 2004 ) . b. kubala and j. knig , phys . rev . b * 73 * , 195316 ( 2006 ) . t. a. costi and a. c. hewson , j. phys . : condens . matter * 5 * , l361 ( 1993 ) . t .- s . kim and s. hershfield , phys . rev . lett . * 88 * , 136601 ( 2002 ) . r. scheibner , h. buhmann , d. reuter , m. n. kiselev , and l. w. molenkamp , phys . rev . lett . * 95 * , 176602 ( 2005 ) . t. a. costi and v. zlati , phys . rev . b * 81 * , 235127 ( 2010 ) . m. galperin , m. a. ratner , and a. nitzan , j. phys . : condens . matter * 19 * , 103201 ( 2007 ) . r. lpez and d. snchez , phys . rev . b * 88 * , 045129 ( 2013 ) . r. s. whitney , phys . rev . b * 88 * , 064302 ( 2013 ) . s. fahlvik svensson , e. a. hoffmann , n. nakpathomkun , p.m. wu , h. q. xu , h. a. nilsson , d. snchez , v. kashcheyevs , and h. linke , new j. phys . * 15 * , 105011 ( 2013 ) . l. a. zotti , m. brkle , f. pauly , w. lee , w. j. k. kim , y. asai , p. reddy , and j. c. cuevas , new j. phys . * 16 * , 015004 ( 2014 ) . m. esposito , k. lindenberg , and c. van den broeck , europhys . lett . * 85 * , 60010 ( 2009 ) . m. leijnse , m. r. wegewijs , and k. flensberg , phys . rev . b * 82 * , 045412 ( 2010 ) . h. wang , g. wu , y. fu , and d. chen , j. appl . phys . * 111 * , 094318 ( 2012 ) . d. m. kennes , d. schuricht , and v. meden , europhys . lett . * 102 * , 57003 ( 2013 ) . r. schleser , t. ihn , e. ruh , k. ensslin , m. tews , d. pfannkuche , d. c. driscoll , and a. c. gossard , phys . rev . lett . * 94 * , 206805 ( 2005 ) . j. koch , f. von oppen , and a. v. andreev , phys . rev . b * 74 * , 205438 ( 2006 ) . m. leijnse and m. r. wegewijs , phys . rev . b * 78 * , 235424 ( 2008 ) . n. m. gergs , c. b. m. hrig , m. r. wegewijs , and d. schuricht , supplementary material ; for specific aspects see also references . f. giazotto , t. t. heikkil , a. m. s. a. luukanen , and j. p. pekola , rev . mod . phys . * 78 * , 217 ( 2006 ) . h. schoeller , eur . phys . j. special topics * 168 * , 179 ( 2009 ) . r. b. saptsov and m. r. wegewijs , phys . rev . b * 86 * , 235432 ( 2012 ) . m. leijnse , m. r. wegewijs , and m. h. hettler , phys . rev . lett . * 103 * , 156803 ( 2009 ) . s. koller , m. grifoni , m. leijnse , and m. r. wegewijs , phys . rev . b * 82 * , 235307 ( 2010 ) . v. n. golovach and d. loss , phys . rev . b * 69 * , 245327 ( 2004).d m. pustilnik , y. avishai , and k. kikoin , phys . rev . lett . * 84 * , 1756 ( 2000 ) . , j. martinek , j. barna , and g. schn , in _ cfn lectures on functional nanostructures _ , _ lecture notes in physics _ , edited by k. busch et al . ( springer , berlin , 2005 ) . j. paaske , a. andersen , and k. flensberg , phys . rev . b * 82 * , 081309(r ) ( 2010 ) . m. pletyukhov and d. schuricht , phys . rev . b * 84 * , 041309(r ) ( 2011 ) . m. pletyukhov and h. schoeller , phys . rev . lett . * 108 * , 260601 ( 2012 ) . a. v. kretinin , h. shtrikman , and d. mahalu , phys . rev . b * 85 * , 201301(r ) ( 2012 ) . r. b. saptsov and m. r. wegewijs , phys . rev . b * 90 * , 045407 ( 2014 ) ; h. schoeller , _ interactions and transport properties _ ( springer , berlin , 1999 ) , p. 137 ; h. schoeller , _ an introduction to real - time renormalization group _ , lect . notes phys . * 544 * , 137 ( 2000 ) ; e. bonet , m. m. deshmukh , and d. c. ralph , phys . rev . b * 65 * , 045317 ( 2002 ) ; c. romeike , m. r. wegewijs , and h. schoeller , phys . rev . lett . * 96 * , 196805 ( 2006 ) ; m. m. deshmukh , e. bonet , a. n. pasupathy , and d. c. ralph , phys . rev . b * 65 * , 073301 ( 2002 ) ; c. stevanato , m. leijnse , k. flensberg , and j. paaske , phys . rev . b * 86 * , 165427 ( 2012 ) ; m. hell , b. sothmann , m. leijnse , m. r. wegewijs , and j. knig , phys . rev . b * 91 * , 195404 ( 2015 ) ; b. muralidharan and m. grifoni phys . , rev . b * 88 * , 045402 ( 2013 ) ; b. wunsch , m. braun , j. knig , and d. pfannkuche , phys . rev . b * 72 * , 205319 ( 2005 ) ; a. donarini , b. siegert , s. sobczyk , and m. grifoni , phys . rev . b * 86 * , 15451 ( 2012 ) ; d. darau , g. begemann , a. donarini , and m. grifoni , phys . rev . b * 79 * , 235404 ( 2009 ) . * supplement to charge fluctuations in nonlinear heat transport " *

1407.8284

in this part of the supplementary material we provide the basic physical background as well the technical details of our density - operator approach to thermotransport through strongly interacting nanostructures . it combines the general reduced density - operator approach , accounting for the strong local interaction @xmath23 , with a real - time diagrammatic method for the calculation of self - energies that determine transition rates of a master equation . the approach is perturbative and well - controlled in the regime where the tunneling rates are well below the electrode temperatures , @xmath86 . we go beyond standard approaches by including the competition of all @xmath47 and @xmath48 transition rates into the master equation for the state occupations and what is new in particular calculating the full energy - transport rate matrix to @xmath48 . see refs . for more details and applications in the context of charge transport . before we present the detailed derivation of the theoretical method in the next section , we discuss the general form of the resulting master equation and current formulas and their physical significance . in particular , we summarize the dependence of the transition rates on the level position and bias voltage in support of the central discussion in the main article explaining fig . 3 . our master equation describes the nonequilibrium state of the quantum dot , in particular the occupations @xmath88 of the quantum states @xmath89 , @xmath90 , in a statistical mixture @xmath91 in the stationary long - time limit @xmath92 for clarity in the indexing of @xmath93 and @xmath51 we denoted the two - electron singlet by @xmath94 . the transition rates obey an exact sum rule @xmath95 which derives from the probability conservation @xmath96 . we can thus restrict our discussion of the _ transition rates _ @xmath97 to those between different states @xmath98 as sketched in ( a ) in the same manner as in the main article [ inset boxes to fig . 3(b ) ] : * _ resonant / single - electron tunneling ( set ) rates _ @xmath99 , @xmath100 , @xmath101 , @xmath102 _ with _ these rates are dominated by their @xmath47 contributions , which coincide with fermi s golden rule expressions @xmath104 where @xmath105 and @xmath106 and @xmath107 . whenever one of the fermi functions changes between 0 and 1 , a transport resonance is naively expected [ i.e. , without solving the master equation ] , i.e. , at one of the eight lines defined in the @xmath108 plane by @xmath109 these are sketched in ( b ) : in the indicated set regimes transport by @xmath47 transitions between two ground states is energetically possible , whereas outside these regimes a single ground state is stable . the set rates vary exponentially when going off - resonance , i.e. , @xmath110 . these rates have @xmath48 corrections which describe the shift and broadening of the resonance . however , the naive expectation for a resonance is not precise in general , even when staying in @xmath47 : in we will discuss anomalous temperature - dependent shifts [ cf . ] which can be understood only when actually solving the master - equation . still , the naive resonance positions are useful as they provide a framework for discussing of the effects on which the main article focuses on . * _ inelastic cotunneling ( icot ) rates _ @xmath111 , @xmath112 + these rates are due to @xmath48 processes and cause transitions between the @xmath21 spin states , see ( a ) . the inelastic excitation rate @xmath111 is exponentially suppressed for @xmath113 but starting from @xmath114 it steadily increases as one goes beyond the red onset line in ( b ) . this is directly reflected by the inelastic current as function of the bias plotted in the inset to fig . 3(c ) in the main article and also below in ( d ) . the inelastic relaxation rate @xmath112 , in contrast , is nonzero at zero bias and increases from there . + the explanation of the central results of the main article in fig . 3 revolves around the fact that at low bias the energy current contributions due to @xmath111 and @xmath112 cancel each other [ regime ( iv ) ] . this cancellation is undone once the relaxation rate @xmath115 gets `` overridden '' by _ either _ the set rate @xmath116 _ or _ @xmath117 of @xmath47 is switched on [ crossing either line ( v ) ] . once _ both @xmath116 and @xmath117 _ are switched on their energy contributions give rise to a new cancellation and the energy current is again reduced [ regime ( vi ) ] , see ( a ) . * _ electron pair / hole pair tunneling rates _ @xmath118 _ and _ @xmath119 + these transition rates are often overlooked and were first pointed out in ref . . they are generated in @xmath48 and lead to _ real _ occupation of the charge state 2 ( 0 ) in regimes where only charge states 0 and 1 ( 1 and 2 ) are accessible , respectively , by @xmath47 set processes , see ( a ) . this leads to a step resonance in @xmath60 , well below the set resonance @xcite [ not indicated in ( b ) to maintain clarity , but see line ( vii ) in fig . 2 of the main article ] that would lead to real occupation of this state when staying in @xmath47 . remarkably , in the energy conductance these pair - tunneling processes appear much more prominently , as mentioned in the discussion section of the main article . an important aspect of our method is that we consistently calculate the _ nonequilibrium _ populations of the quantum dot from the master equation instead of just combining some of the above transition rates with equilibrium occupations as is sometimes done in inelastic scattering theories . in our discussion of the processes responsible for the features in fig . 3 of the main article we have carefully verified our identification of the physics by comparing with the bias and energy - level dependence of the full numerically calculated rate matrices and the resulting occupations and current contributions . technically , the rates in the master equation can be determined from a superoperator kernel @xmath51 that we construct below [ cf . ] by letting it act on a state projector and taking diagonal matrix elements @xmath120 \ket{i } \label{eq : matrix - elements } .\end{aligned}\ ] ] the charge and energy currents flowing out of electrode @xmath121 can be calculated from formulas with the physically appealing form of rate for transporting a quantity in a transition @xmath122 @xmath123 the occupation of the initial state @xmath124 , summed over the initial ( @xmath124 ) and final ( @xmath125 ) states : @xmath126 note that in the main article we have evaluated everything for @xmath36 and dropped this superscript . as we shall see in the next section , the calculation of the charge transport poses no real additional challenge since the required charge - current rate matrix @xmath127 can be constructed from the reservoir - resolved contributions @xmath128 required to calculate the transition matrix @xmath129 for the master equation . this is due to the kirchhoff law ( charge conservation ) . for the heat or energy current no equivalent of the kirchhoff law holds , and in particular when going to @xmath48 the energy - current rate matrix @xmath130 needs to be calculated separately . this is the main technical achievement underlying the results reported in the main article . in this section we derive the master equation and the current formulas starting from a general microscopic hamiltonian @xmath131 where @xmath132 and @xmath133 act on the corresponding hilbert spaces of the electronic reservoirs and the quantum dot , respectively . these two subsystems are coupled by @xmath134 which is assumed to be bilinear in the electron field operators and the reservoirs are effectively noninteracting , i.e. , @xmath132 is quadratic in the fields . we first set up a perturbation theory in the tunnel coupling between the quantum dot and the reservoirs . this is formulated from the start in liouville - fock space and has as a main advantage that one does not need to introduce an imaginary time or a keldysh contour_even for non - equilibrium problems _ : the time evolution of the density operator is fully described on a _ single , real time axis_. this comes at the price of having to deal efficiently with superoperators , which is achieved by introducing second quantization techniques directly in liouville - fock space , as we now briefly describe following refs . , see also ref . . [ [ liouville - fock - space ] ] liouville - fock space : + + + + + + + + + + + + + + + + + + + + + liouville - fock space is the space spanned by the many - body operators acting on an underlying hilbert - fock space . it is convenient to introduce from the start multi - index notation @xmath135 for a field operator on the hilbert - fock space of the quantum dot , where @xmath136 and the extra index @xmath137 corresponds to a creation operator and @xmath138 to a destruction operator . here @xmath139 labels the internal quantum numbers of the quantum dot , which in our case is just the spin [ for multi - level systems considered in sec . [ sec : genereric ] below @xmath139 also contains the orbital index ] . to obtain field superoperators that generate the corresponding liouville - fock space we first consider the naive construction of field superoperators via the left action @xmath140 and the right action @xmath141 by a given field @xmath142 on an operator argument indicated by @xmath143 . it turns out that this ansatz spoils the fermionic commutation rules . one can circumvent this problem@xcite by taking linear combinations and involving the parity operator @xmath144 associated with the electron number @xmath22 by defining instead @xmath145 the superoperators @xmath146 indeed anticommute like fermionic fields : @xmath147_{+}=\delta_{q_{2},\overline{q}_{1}}\delta_{2,\overline{1}}\mathcal{i}$ ] , where @xmath148 , @xmath149 and @xmath150 is the unit superoperator . the reservoir field superoperators @xmath151 are defined analogously where now @xmath152 and @xmath153 again corresponds to the inversion of the @xmath154-index . assuming a grand - canonical density operator for the macroscopic reservoirs one finds that for odd @xmath155 @xmath156 , where @xmath157 , whereas for even @xmath155 one obtains a fermionic wick theorem in liouville - fock space @xmath158 where @xmath159 counts the number of permutations needed to join all pairs of contracted superoperators . the one - point correlation function is given by @xmath160 with @xmath161 and @xmath162 $ ] , which are up to a prefactor just the symmetric and antisymmetric parts of the fermi - dirac - distribution , respectively . [ [ perturbation - theory - in - the - tunnel - coupling ] ] perturbation theory in the tunnel coupling : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we first follow the standard procedure of the density - operator approach by solving the liouville - von - neumann equation @xmath163 , where @xmath164 is the liouvillian of the total system . by laplace transform its formal solution @xmath165 becomes @xmath166 . assuming the initial state factorizes as @xmath167 , where @xmath168 denotes the initial density matrix of the dot , we can trace out the reservoir degrees of freedom . here the real - time approach deviates from traditional approaches by using the liouville - space wick theorem _ right from the start_. from the resulting diagrammatic representation one readily identifies a dyson equation which propagates the initial density operator _ of the quantum dot only _ , @xmath168 , @xmath169 by an effective , energy - dependent liouvillian @xmath170 here @xmath171 denotes the bare dot liouvillian , @xmath172 , which contains the energy scales @xmath23 and @xmath4 . we stress that we denote the density matrix of the dot and the dot liouvillian by @xmath173 and @xmath171 , respectively , i.e. , we omit the subscripts d. with the mentioned assumptions this scheme is still exact so far , but we proceed by truncating this series . the superscript `` irred '' on the right hand side of indicates irreducibility of the expressions : this stipulates that after expanding in @xmath174 one can not move pairs of reservoir field superoperators apart from each other without changing any contraction - line crossing . the one - point correlation function in first nonvanishing order @xmath175 is obviously irreducible . for the second nonvanishing order , the pairs @xmath176{}{\mathcal{j}_{4}^{q_{4}}}{\mathcal{j}_{3}^{q_{3}}\mathcal{j}_{2}^{q_{2}}}{\mathcal{j}_{1}^{q_{1}}}\wick[1.5ex]{\mathcal{j}_{4}^{q_{4}}}{\mathcal{j}_{3}^{q_{3}}}{}{\mathcal{j}_{2}^{q_{2}}}\mathcal{j}_{4}^{q_{4}}\mathcal{j}_{3}^{q_{3}}\mathcal{j}_{2}^{q_{2}}\mathcal{j}_{1}^{q_{1}}$ ] and @xmath176{}{\mathcal{j}_{4}^{q_{4}}}{\mathcal{j}_{3}^{q_{3}}}{\mathcal{j}_{2}^{q_{2}}}\wick[1.5ex]{\mathcal{j}_{4}^{q_{4}}}{\mathcal{j}_{3}^{q_{3}}}{\mathcal{j}_{2}^{q_{2}}}{\mathcal{j}_{1}^{q_{1}}}\mathcal{j}_{4}^{q_{4}}\mathcal{j}_{3}^{q_{3}}\mathcal{j}_{2}^{q_{2}}\mathcal{j}_{1}^{q_{1}}$ ] are irreducible while @xmath177 is reducible and must be omitted from . to perform the expansion we need the reservoir and tunnel liouvillians : using the notation @xmath178 and a quite generic form of the tunnel and reservoir model we obtain @xmath179 since we want to focus on the strong energy dependence introduced by the quantum dot on the energy currents , we use a flat density of states @xmath180 with some large cutoff @xmath181 ( as compared to the energy scales @xmath182 and @xmath183 ) . this simplifies the present discussion and allows the energy integrations to be done analytically up to second order for @xmath29 in , but it presents no principal limitation of our method . to be able to apply the wick theorem to , we need to commute the reservoir fields in @xmath184 through the resolvents containing @xmath185 . since these do not commute , @xmath186_{-}=\overline{\omega}_{1}\mathcal{j}_{1}^{q}$ ] , this changes the denominators of the intermediate propagators in and we obtain for the effective liouvillian @xmath187 here @xmath188 denotes a summation over reservoir and spin - indices , combined with integration over the energies @xmath189 . the main challenge in going to higher orders evenatually becoming prohibitive is first the computation of the integrals over these energies and second the handling of the algebra of superoperators to efficiently evaluate the matrix elements . however , a rescaling of the integration variable @xmath190 by a factor @xmath191 ( to make the argument of the one - point correlation function dimensionless ) reveals that @xmath192 is the relevant perturbation parameter , where @xmath193 denotes the scale of the tunnel coupling constants @xmath194 . since in the main article we focus on the regime @xmath195 we truncate the liouvillian at the second order . the effective liouvillian in @xmath196 becomes @xmath197 to obtain a description of the inelastic transport in the coulomb blockade regime that is even _ qualitatively correct _ it is crucial to keep the _ first two leading orders _ , as demonstrated in the main article . it should be noted that the interaction @xmath23 and inelastic excitation energy @xmath4 ( zeeman energy for a spin - flip ) are treated non - perturbatively in this scheme , appearing in the denominators through @xmath171 . [ [ stationary - state - rate - matrix - w ] ] stationary - state rate matrix @xmath51 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + taking the stationary limit of using @xmath198 we obtain the stationary - state equation determining the density operator denoted by @xmath199 @xmath200 we write laplace transforms @xmath201 at @xmath202 as @xmath203 , without the risk of confusing them with @xmath204 evaluated at @xmath205 . using the charge- and spin - symmetry of the anderson model one proves @xcite that the probabilities @xmath206 decouple from all coherences @xmath207 with @xmath208 . the set of equations containing only the probabilities form a closed subset involving only the rates with @xmath209 [ the factor @xmath124 is important for consistency with the current rate expressions below ] . we thus obtain the stationary master equation announced earlier.@xcite by truncating the series to we compute the rates for all physical processes of orders @xmath210 consistently . however , one has to be aware that by solving for the stationary density operator one generates an expression containing higher - order terms , which are however negligible for @xmath192 , see ref . for a discussion . to check that the obtained features are consistently calculated , we changed @xmath193 and checked that the observed features scale consistently with orders to which the kernel is calculated . this important analysis is discussed in . [ [ stationary - state - charge - and - energy - current - rate - matrices - w_c - and - w_e ] ] stationary - state charge and energy current rate matrices @xmath211 and @xmath212 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to obtain the stationary charge current all relevant information is already encoded in the rate matrix @xmath51 computed for the stationary density operator above . to extract this information one uses charge conservation @xcite and obtains @xmath213_-}(t ) = \braket{\ , i\left[n , h_{\text{tun}}^{\alpha}\right]_{- } \,}(t ) = \underset{\text{dot}}{\operatorname{tr } } \left [ -\tfrac{1}{2 } l^{n,+ } \left [ \underset{\text{res}}{\operatorname{tr } } \left(-il_{\text{tun}}^{\alpha}\right ) \rho_{\text{res } } \right ] \rho(t ) \right],\ ] ] where @xmath214 , and we have used @xmath215 . the outer expression involves a simple action of @xmath216 , whereas in the complicated part involving the trace over the reservoirs one can identify the same perturbative expressions involved in . repeating the perturbative analysis described there one obtains for the _ stationary _ particle current @xmath217 where @xmath173 is the stationary state determined from and @xmath218 denotes the reservoir - resolved part of the effective liouvillian @xmath219 , i.e. , @xmath220 is given by the second part of by replacing @xmath221 with reservoir index fixed to @xmath25 . the corresponding explicit result is given by when fixing the left - most reservoir index to @xmath25 . in contrast , for the energy current one has to consider @xmath222_- \,}(t ) = \braket { i\sum_{2}\delta_{\alpha_{2}\alpha}\omega_{2 } t_{2 } d_{2 } c_{\overline{2 } } } .\ ] ] this term does look similar to terms appearing in the perturbative series for the density operator evolution as before and one can express the _ stationary _ energy current as @xmath223 its explicit from is obtained by replacing the left - most @xmath184 in by @xmath224_{+ } = \frac{i}{2}\sum_{2}\delta_{\alpha_{2}\alpha}t_{2}\left(\omega_{2}-\mu_\alpha\right)\mathcal{g}_{2}^{q_{2}}\mathcal{j}_{\overline{2}}^{q_{2}}.\end{aligned}\ ] ] the result again looks similar to . however , the energy factor @xmath225 now complicates the energy integrals , and this new energy - current kernel requires an independent calculation . as for the density - operator kernel , the charge- and spin - symmetry of the anderson model imply that @xcite in the current formulas and only terms involving the probabilities @xmath206 from the density operator @xmath173 contribute with matrix elements of the type of the superoperators @xmath226 and @xmath227 . as noted before , in the main article we have evaluated everything for @xmath36 and dropped this superscript . we thus obtain the current formulas announced in . in summary , we have to compute two kernels the kernel @xmath228 of the kinetic equation to determine @xmath173 and the energy - current kernel @xmath229 ; the charge current kernel requires no separate computation . moreover , because the density operator is self - adjoint and the currents are real , we only need the imaginary parts of both kernels @xmath230 and @xmath231 , which have much simpler energy integrals than their real parts . the subset of their matrix elements coupling probabilities defined by determines the rate matrices @xmath51 , @xmath232 and @xmath233 required in the master equation and the current formulas . this completes the derivation . [ [ mathcalogamma - results ] ] @xmath47 results : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to @xmath47 it is possible to give a compact general expression for the charge and energy current , valid also for tunneling rates with a dependence on the junction ( @xmath121 ) and the spin ( @xmath234 ) ( but collinear spin - polarization axes of the electrodes ) . we used the basis adapted to the symmetries of the problem from @xcite . the integrations in the self - energy @xmath230 are carried out immediately by inserting the sokhotski - plemelj relation @xmath235 , giving that all principal value parts cancel out . solving the master equation and inserting the stationary occupations into the current formula we obtain for the charge current @xmath236,\ ] ] where @xmath237\\ \frac{1}{2}\gamma_{\downarrow\alpha}[\tanh\left(\frac{\varepsilon+u-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)+\tanh\left(\frac{\varepsilon-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right ) ] \end{array}\right),\\ \xi^{\alpha } & = \left(\begin{array}{cc } \gamma_{\uparrow\alpha } & \frac{1}{2}\gamma_{\uparrow\alpha}[\tanh\left(\frac{\varepsilon+u+\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)-\tanh\left(\frac{\varepsilon+\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)]\\ \frac{1}{2}\gamma_{\downarrow\alpha}[\tanh\left(\frac{\varepsilon+u-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)-\tanh\left(\frac{\varepsilon-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right ) ] & \gamma_{\downarrow\alpha } \end{array}\right),\end{aligned}\ ] ] while @xmath238 and @xmath239 . for the energy current we obtain @xmath240 where we defined @xmath241\\ \sum_{\sigma}\frac{1}{2}\gamma_{\sigma\alpha}u\\ \frac{\gamma_{\downarrow\alpha}}{2}[\left(\varepsilon+u-\delta/2\right)\tanh\left(\frac{\varepsilon+u-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)-\left(\varepsilon-\frac{b}{2}\right)\tanh\left(\frac{\varepsilon-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)]+\gamma_{\uparrow\alpha}\left(\varepsilon+\frac{u}{2}+\delta/2\right)\\ \frac{\gamma_{\uparrow\alpha}}{2}[\left(\varepsilon+u+\delta/2\right)\tanh\left(\frac{\varepsilon+u+\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)-\left(\varepsilon+\delta/2\right)\tanh\left(\frac{\varepsilon+\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)]+\gamma_{\downarrow\alpha}\left(\varepsilon+\frac{u}{2}-\delta/2\right ) \end{array}\right)^{t},\\ \zeta & = \sum_{\sigma\alpha}\frac{1}{2}\gamma_{\downarrow\alpha}\left [ \tanh\left(\frac{\varepsilon+u-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right)+\tanh\left(\frac{\varepsilon-\delta/2-\mu_{\alpha}}{2t_{\alpha}}\right ) \right].\end{aligned}\ ] ] furthermore we define with @xmath242 a corresponding @xmath243 , where @xmath244 corresponds to two entries of the right - most vector above . [ [ mathcalogamma2-resultssecgamma2 ] ] @xmath48 results:[sec : gamma2 ] + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + when going to @xmath245 a number of things change : ( i ) additional rates @xmath48 has to be computed and added to the @xmath47 to obtain @xmath51 , @xmath232 and @xmath233 . ( ii ) the master equation has to be solved again for the occupations ( with these new rates ) . ( iii ) the currents need to be recomputed with the new rates and new occupations . for the effects considered here , analytical formulas for the resulting currents do not bring much insight and the equations have been implemented numerically . the main challenge lies in step ( i ) in the computation of the energy integrals listed below , after taking matrix elements . we have numerically implemented these integrations for arbitrary temperatures @xmath246 and @xmath247 , but at zero thermal bias @xmath29 , the case focused on in the main article , these integrals can be calculated analytically . in the latter case , they can be expressed in the standard complex digamma function @xmath248 and we list the results here . for the integrals for the time - evolution ( and charge current ) kernel we obtain in agreement with refs . @xmath249 & \frac{1}{2\pi}\text{im}\ , \int_{-d}^{d}\int_{-d}^{d}d\overline{\omega}_{1}d\overline{\omega}_{2}\frac{1}{\overline{\omega}_{1}+z_{3}}\frac{\gamma_{2}^{-}\gamma_{1}^{-}}{\sum\limits _ { i=1,2}\overline{\omega}_{i}+z_{2}}\frac{1}{\overline{\omega}_{1}+z_{1}}\nonumber\\ & \quad= \coth\left(\frac{z_{2}+\overline{\mu}_{1}+\overline{\mu}_{2}}{2t}\right)\frac{\psi\left(\frac{1}{2}-i\frac{z_{3}+\overline{\mu}_{1}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{3}-z_{2}-\overline{\mu}_{2}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{1}+\overline{\mu}_{1}}{2\pi t}\right)+\psi\left(\frac{1}{2}-i\frac{z_{1}-z_{2}-\overline{\mu}_{2}}{2\pi t}\right)}{z_{3}-z_{1}}\nonumber\\ & \quad\quad -\frac{\tanh\left(\frac{z_{3}+\overline{\mu}_{1}}{2t}\right)\phi\left(\frac{1}{2}-i\frac{z_{2}-z_{3}+\overline{\mu}_{2}}{2\pi t}\right)-\tanh\left(\frac{z_{1}+\overline{\mu}_{1}}{2t}\right)\phi\left(\frac{1}{2}-i\frac{z_{2}-z_{1}+\overline{\mu}_{2}}{2\pi t}\right)}{z_{3}-z_{1}},\\[2 mm ] & \frac{1}{2\pi}\text{im}\,\int_{-d}^{d}\int_{-d}^{d}d\overline{\omega}_{1}d\overline{\omega}_{2}\frac{1}{\overline{\omega}_{2}+z_{3}}\frac{\gamma_{2}^{-}\gamma_{1}^{-}}{\sum\limits _ { i=1,2}\overline{\omega}_{i}+z_{2}}\frac{1}{\overline{\omega}_{1}+z_{1}}\nonumber\\ & \quad= -\coth\left(\frac{z_{2}+\overline{\mu}_{1}+\overline{\mu}_{2}}{2t}\right)\frac{\psi\left(\frac{1}{2}-i\frac{z_{3}+\overline{\mu}_{2}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{3}-z_{2}-\overline{\mu}_{1}}{2\pi t}\right)+\psi\left(\frac{1}{2}-i\frac{z_{1}+\overline{\mu}_{1}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{1}-z_{2}-\overline{\mu}_{2}}{2\pi t}\right)}{z_{1}+z_{3}-z_{2}}\nonumber\\ & \quad\quad -\tanh\left(\frac{z_{3}+\overline{\mu}_{2}}{2t}\right)\frac{\psi\left(\frac{1}{2}-i\frac{z_{2}-z_{3}+\overline{\mu}_{1}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{1}+\overline{\mu}_{1}}{2\pi t}\right)}{z_{1}+z_{3}-z_{2}}\nonumber\\ & \quad\quad -\tanh\left(\frac{z_{1}+\overline{\mu}_{1}}{2t}\right)\frac{\psi\left(\frac{1}{2}-i\frac{z_{2}-z_{1}+\overline{\mu}_{2}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{3}+\overline{\mu}_{2}}{2\pi t}\right)}{z_{1}+z_{3}-z_{2}}.\end{aligned}\ ] ] the integrals for the energy current kernel have not been reported before . defining @xmath250 we obtain @xmath251 & & \frac{1}{2\pi}\text{im}\ , \int_{-d}^{d}\int_{-d}^{d}d\overline{\omega}_{1}d\overline{\omega}_{2}\frac{\overline{\omega}_{1}}{\overline{\omega}_{1}+z_{3}}\frac{\gamma_{2}^{-}\gamma_{1}^{-}}{\sum\limits _ { i=1,2}\overline{\omega}_{i}+z_{2}}\frac{1}{\overline{\omega}_{1}+z_{1 } } \nonumber\\ & & \quad= -\coth\left(\tfrac{z_{2}+\overline{\mu}_{1}+\overline{\mu}_{2}}{2t}\right ) \text{re}\frac{z_{3}\left[\psi\left(\frac{1}{2}-i\frac{z_{3}+\overline{\mu}_{1}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{3}-z_{2}-\overline{\mu}_{2}}{2\pi t}\right)\right]-z_{1}\left[\psi\left(\frac{1}{2}-i\frac{z_{1}+\overline{\mu}_{1}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{1}-z_{2}-\overline{\mu}_{2}}{2\pi t}\right)\right]}{z_{3}-z_{1 } } \nonumber\\ & & \quad\quad + \frac{z_{3}\tanh\left(\frac{z_{3}+\overline{\mu}_{1}}{2t}\right)\phi\left(\frac{1}{2}-i\frac{z_{2}-z_{3}+\overline{\mu}_{2}}{2\pi t}\right)-z_{1}\tanh\left(\frac{z_{1}+\overline{\mu}_{1}}{2t}\right)\phi\left(\frac{1}{2}-i\frac{z_{2}-z_{1}+\overline{\mu}_{2}}{2\pi t}\right)}{z_{3}-z_{1}},\\[2 mm ] & & \frac{1}{2\pi}\text{im}\ , \int_{-d}^{d}\int_{-d}^{d}d\overline{\omega}_{1}d\overline{\omega}_{2}\frac{\overline{\omega}_{2}}{\overline{\omega}_{2}+z_{3}}\frac{\gamma_{2}^{+}\gamma_{1}^{-}}{\sum\limits _ { i=1,2}\overline{\omega}_{i}+z_{2}}\frac{1}{\overline{\omega}_{1}+z_{1}}=-\phi\left(\frac{1}{2}-i\frac{z_{1}}{2\pi t_{1}}\right),\end{aligned}\ ] ] @xmath252-\left(z_{1}-z_{2}\right)\left[\psi\left(\frac{1}{2}-i\frac{z_{1}+\overline{\mu}_{1}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{1}-z_{2}-\overline{\mu}_{2}}{2\pi t}\right)\right]}{z_{1}+z_{3}-z_{2 } } \nonumber\\ & & \quad\quad + z_{3}\tanh\left(\frac{z_{3}+\overline{\mu}_{2}}{2t}\right)\frac{\psi\left(\frac{1}{2}-i\frac{z_{2}-z_{3}+\overline{\mu}_{1}}{2\pi t}\right)-\psi\left(\frac{1}{2}-i\frac{z_{1}+\overline{\mu}_{1}}{2\pi t}\right)}{z_{1}+z_{3}-z_{2}}\nonumber\\ & & \quad\quad -\tanh\left(\frac{z_{1}+\overline{\mu}_{1}}{2t}\right)\text{re}\frac{\left(z_{2}-z_{1}\right)\phi\left(\frac{1}{2}-i\frac{z_{2}-z_{1}+\overline{\mu}_{2}}{2\pi t_{1}}\right)-z_{3}\phi\left(\frac{1}{2}-i\frac{z_{3}+\overline{\mu}_{2}}{2\pi t_{1}}\right)}{z_{1}+z_{3}-z_{2}}.\end{aligned}\ ] ] effective single - particle descriptions have been used to analyze energy transport @xcite . a prominent one is the landauer approach , which in particular was used to explain the asymmetry of heat dissipation in quantum devices caused by the energy flow@xcite . a natural question is how our density - operator approach relates to this method and we provide two answers to this question : first , in general for @xmath46 the density - operator approach _ without perturbative expansion _ in @xmath193 exactly reduces to a landauer formula for any @xmath193 and @xmath253 . we show how this resummation can be simply performed by _ extending _ our equations for @xmath46 to include broadening nonperturbatively . second , explicitly setting @xmath46 in the perturbative equations for @xmath195 used in the present article explicitly reduces to landauer formula _ in the corresponding limit _ ( i.e. , expanded in @xmath193 ) . these two answers are discussed subsequently . it has been known since the beginning of the real - time diagrammatic formulation of the density - operator transport theory that the exact solution , taking a landauer form , is incorporated @xcite when performing the complete summation of the diagrammatic series for the self - energy @xmath230 or rate matrix @xmath51 in powers of @xmath193 . this seemed only explicitly doable for simple concrete examples such as the noninteracting resonant level model ( without spin ) @xcite . the fact that in the density - operator approach the @xmath46 limit is not easy to see _ in general _ may seem puzzling at first , since for , e.g. , green s functions approaches , based on an expansion in the interaction @xmath23 but exact in @xmath193 , this limit is trivially recovered . however , the density - operator approach is built around the picture of a locally strongly interacting system : it is entirely formulated in terms of many - body density operators and therefore carries a certain `` overhead '' in the @xmath46 limit . in the recently developed @xcite formulation of the same theory used in this article this noninteracting limit becomes entirely transparent by exploiting the field superoperators . this works for any number of orbitals including spin and allowing for arbitrary tunnel coupling matrix elements ( breaking any local orbital and spin symmetry ) . for this only the wide - band limit needs to be assumed for simplicity , which prevails anyhow in practical applications and which is not an essential restriction . this allows for a single renormalization step @xcite to be performed with the following very simple result : starting from our series 1 . one replaces all liouvillians @xmath171 by @xmath254 , which introduces an energy - independent level broadening . importantly , this is _ derived _ , not put in by hand . one leaves out all terms in the series containing @xmath255 superoperators since they are included via the first step . one obtains a new perturbation theory of the same form which is however much simpler . importantly , taking this renormalized perturbation theory to finite order @xmath85 ( @xmath256 ) for a 2 ( @xmath55 ) level quantum dot one recovers the _ exact _ density - operator evolution for @xmath46 . moreover , for one - particle observables such as the charge current a first order calculation is already exact for @xmath46 . thus , the @xmath210 perturbation theory used in the present article already has the correct form of the exact solution , except that the bare liouvillian is used ( i.e. , step 1 is missing ) . since we focus on the limit @xmath195 we instead incorporate the corresponding term with @xmath257 superoperators ( step 2 ) , which includes the leading broadening corrections , see also below . we thus neglect only the nonperturbative effect of broadening . for a detailed exposition and explanation of the physics underlying this formulation see @xcite [ cf . ( 79 ) there ] and also ref . . finally , we note that going beyond the perturbative limit in @xmath193 _ for finite _ @xmath23 is possible within the framework that we use and seamlessly connects to the real - time renormalization - group approach @xcite . such calculations are , however , very involved and for the anderson model this has only recently been worked out for charge transport @xcite . to concretely illustrate the above general conclusion we now simply insert @xmath46 into our equations , first restricting attention to the @xmath47 contributions . we then obtain the landauer result for the charge and energy currents , respectively ( @xmath258 notes the fermi - dirac - distribution ) @xmath259 , \quad\quad i_{\text{e } } = \frac{1}{\pi}\int d\omega\ , t\left(\omega\right)\cdot\omega\cdot\left[f\left(\omega-\mu_\text{l } \right)-f\left(\omega-\mu_\text{r } \right)\right],\end{aligned}\ ] ] with the transmission function @xmath260 since we are in the leading order in @xmath261 , the transmission function is a @xmath262-function of the energy @xmath263 . including just the @xmath48 corrections in our density - operator approach ( while maintaining @xmath46 ) only leads to broadening corrections . this is done numerically in and the corrections are negligibly small as expected for @xmath195 . performing the above mentioned renormalization step one obtains the full @xmath193-broadened lorentzian function of the landauer result valid for any relation between @xcite @xmath193 and @xmath253 . however , we emphasize that even in that case there are _ no _ inelastic cotunneling signatures in the transport since @xmath46 . this shows the importance of our density - operator approach which is capable of capturing these processes and which were found to be especially important for the energy transport . physically speaking , when not accounting for coulomb blockade in the off - resonant transport regime the system is not in a particular well - defined charge state with well - defined inelastic excitation , which can thus not appear as a gate - voltage independent excitation . in this second part of the supporting information we present additional results and discussion addressing specific issues raised in the main article . [ [ scaling - with - parametersorder - of - magnitude - of - charge - energy - and - heat - currentsecscaling - analysis ] ] scaling with parameters order of magnitude of charge , energy and heat current[sec : scaling - analysis ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ in the main article we mentioned that the effects in the currents that we focus on the and features in the coulomb blockade scale as ( cf . also @xcite ) : @xmath264 for the parameter regime of interest where @xmath44 . it is important that this refers to _ corresponding features _ , i.e. , at voltages and level positions expressed in units of the interaction @xmath23 . the prefactors depend on the relative position of the onset @xmath265 , @xmath266 , @xmath267 ; for the temperature @xmath253 dependence see below . to illustrate this , we plot in ( a ) and ( b ) the same data as in fig . 3 ( a ) and ( c ) of the main text , respectively , but now calculated for rescaled @xmath193-values . plotted in dimensionless units as function of @xmath267 and @xmath266 respectively , the data fall onto two single curves , confirming the respective _ conductance _ scaling behavior . we thus conclude that our perturbative treatment to @xmath48 is consistent . as can be seen in ( a)-(b ) , the only exception to the conductance scaling is the position and amplitude of the first coset resonance position shifts . ( c ) shows the corresponding scaling of the _ currents_. this reveals that the temperature dependence corresponds _ only _ to a shift of the position of the resonance ( v ) : one switches from low values of @xmath41 to high values , the curve before and after this switch being _ independent _ of @xmath253 . it is only the @xmath2-position of the transition between them that shifts with @xmath253 . for the case of the charge current this relates to a known effect that is not relevant for the scaling discussion and is addressed separately below in . with this in mind we can thus conclude that the magnitudes of the currents scale as @xmath268 for corresponding features ( as defined above ) . to estimate the orders of magnitude for the experimentally accessible _ heat current _ @xmath269 [ all quantities refer the right electrode as in the main article ] , we note that the sharp spectroscopic features in @xmath40 are due to the energy current @xmath41 whereas the background is due to the convective part @xmath270 . importantly , @xmath41 and @xmath270 scale in the same way , their relative magnitude being controlled by the relative excitation energy @xmath271 . we illustrate this in ( d ) : the energy current is able to cause pronounced negative differential _ heat conductance _ , i.e. , a notable sharp drop in the heat current magnitude at the second coset resonance [ marked by ( vi ) as in the main article ] . we emphasize that from the experimentally measured heat current @xmath40 one can obtain in a _ model - free way _ the interesting energy current @xmath41 by simply subtracting the known background formed by the convective contribution @xmath270 . thus , when estimating orders of magnitude we can ignore the difference between @xmath40 and @xmath41 . based on a first quick experimental estimate for the order of magnitude is obtained as @xmath272 taking interaction energies @xmath273 10@xmath274 - 10@xmath275 mev typical for molecular junctions and inelastic cotunneling currents @xmath276 1 - 10 na one obtains @xmath277 0.1 - 10 nw . for comparison , nowadays heat currents of 10s of nw can already be measured @xcite . however , these measurements apply a larger bias @xmath278 to achieve ( nearly ) resonant set transport @xmath279 [ such that @xmath280 , cf . ] . in this case we can estimate using the conductance quantum @xmath281 @xmath282 \text{~na } , \quad \quad \quad i_\text{e}^\text{set } \sim \frac{e^2}{h } \gamma v^\text{set } \sim \left [ \frac{i_\text{c}^\text{set}}{\text{1 na } } \right ] \cdot \left [ \frac{v^\text{set}}{\text{1 mev } } \right ] \text{~nw}.\end{aligned}\ ] ] to gain more insight into where the smaller heat current values in the inelastic transport regime come from , we now specify the scaling of the heat dissipation for the inelastic energy current more precisely : the factor @xmath23 in comes from the voltage being equal to the inelastic excitation @xmath4 which is always a fraction of @xmath23 inside the coulomb blockade regime [ e.g. , @xmath283 at ( iv ) in fig . [ fig : sup-4](c ) ] . using the scaling of the inelastic heat current relative to that of the resonant case is found to be @xmath284 the first suppression factor expresses that when staying in the coulomb blockade regime one applies a smaller voltage than when going on resonance ( leaving the coulomb blockade regime ) , @xmath285 , thus giving a smaller energy current . the second factor is the effect of the coulomb interaction @xmath286 suppressing the magnitude of the inelastic current relative to the resonant one . equation indicates that for devices where @xmath287 10 @xmath288w one may expect the inelastic currents to be of the order of @xmath289 10 nw , which shows good prospects for coming within range of heat current measurements . we immediately emphasize that our estimates have been conservative in the following sense : * the above scaling is basically those around the center of the coulomb blockade regime where all currents are minimal . away from the center all the effects are enhanced . * we used strictly perturbative formulas , assuming regimes where @xmath290 is small . for larger values of @xmath193 and lower temperatures renormalization effects can greatly enhance the inelastic tunneling relative to the resonant tunneling , as is well known for charge transport , and expected also for energy transport , see for a simple formula . we thus underestimate the heat currents that one should experimentally be able to resolve . a more detailed analysis is of interest but relies on details of a given experimental setup as well as goes beyond the scope of the present article and the methods employed here , which mainly serve to identify the basic processes of energy transport that were overlooked so far . as mentioned in the outlook of the main article , the second point is an interesting topic for future studies and requires a nonequilibrium renormalization group analysis to treat stronger tunnel couplings in the presence of interactions . [ [ anomalous - temperature - dependence - of - coset - resonance - positionssect - dependence ] ] anomalous temperature dependence of coset resonance positions[sec : t - dependence ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ in fig . 3(a ) and ( c ) of the main article , as well as , the first resonance [ labeled ( v ) as in the main article ] do not appear exactly at the vertical dotted lines indicating the coset resonance positions [ those parts of the resonance lines that lie within the coulomb blockade regime ] . for the charge transport at the resonance this effect has been first discussed in @xcite . the mechanism causing the effect is however quite general and already plays a role in regimes where _ only set transport occurs_. understanding this mechanism is relevant for the coset thermometry discussed in . in general , transport through a quantum dot shows a resonance whenever a new transition rate is `` switched on '' by changing the energy levels relative to the chemical potentials of the electrodes ( either by gate or bias voltage or both ) . this changes the rate matrices @xmath51 , @xmath127 , @xmath130 causing the occupation probabilities to change [ cf . ] and adding a new contribution to the currents . roughly speaking this happens when the relevant addition energy falls above or below the electrochemical potential of one of the electrodes . although this is often ignored , the _ relevant criterion for resonance _ is a significant change in the occupations and _ not this energy condition_. already for the symmetric anderson model there is a difference between the two conditions @xcite . the idea is most clearly understood from a simple case @xcite : assume that a rate @xmath291 is being `` switched on '' by tuning an energy level @xmath59 through @xmath288 of some electrode in the function @xmath292 . we lump together all other rates depopulating this state into @xmath293 and both @xmath293 and @xmath294 are assumed roughly constant [ i.e. , much weaker @xmath59 dependence than @xmath295 . naively one expects the resonance position to be @xmath296 , but this ignores interaction effects that come in through the stationary master equation . in this case the stationary master equation contains a line of the form @xmath297 with solution @xmath298 . the occupation @xmath93 of our state becomes comparable to the occupation @xmath159 of the other state ( already populated ) `` feeding '' it when @xmath299 . when the tunneling rate constants @xmath294 and @xmath293 are identical , this reduces to the naive resonance condition @xmath300 . however , when the `` feeding '' rate is larger , @xmath301 , the ratio already reaches 1 while still in the tail of the fermi - distribution , @xmath302 . solving this modified condition , the naive expectation is modified to @xmath303 physically speaking , while still far from resonance tunneling with a relatively large constant rate @xmath294 is sufficient to alter the occupations of a state that relaxes slowly ( @xmath304 ) . the resonance position thus _ shifts linearly with temperature_. for significantly differing rates @xmath294 and @xmath293 this shift may drastically alter the excitation spectrum from the expected `` bare '' spectrum , but even small factors are noticeable . this fact must already be taken into account in charge transport spectroscopy even when only @xmath47 processes are taken into account and in this case has been experimentally observed @xcite and modeled in detail in in @xcite . returning to resonance of interest in the main article , both @xmath47 and @xmath48 processes are competing . the above sketched situation of having largely differing tunneling rates naturally arises @xcite . in this case one considers the _ depopulation _ of the excited spin state @xmath305 by an @xmath47 process which is switching on while a much smaller populating rate @xmath306 is already present . this however , leads to the same scenario and the logarithmic prefactor scales as @xmath307 in this case , making temperature shifts significant . this again underlines the importance of solving a full nonequilibrium master equation for the occupations including interaction effects , cf . . [ [ energy - transport - under - combined - voltage - and - thermal - biascoset - thermometrysecthermal - bias ] ] energy transport under combined voltage and thermal bias coset thermometry[sec : thermal - bias ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ we briefly discuss the case of a joint voltage ( @xmath308 ) and thermal bias ( @xmath309 ) mentioned in the discussion section of the main article . so far this has been studied in prior works@xcite only to @xmath47 . the charge and energy conductance for a single - level anderson dot are shown in . besides the expected asymmetry in the thermal broadening of the resonant tunneling lines associated with the different reservoirs , there are temperature - dependent offsets in both coset lines that are highlighted in the frame in fig . [ fig : sup-3](b ) . these are caused by the mechanism discussed in the previous . in a thermoelectric setup this mechanism offers the interesting opportunity of _ converting a thermal bias into an energy splitting_. indeed , the largest offset between the blue lines splits proportional to the temperature @xmath247 and can be used experimentally to detect a temperature gradient _ in situ_. this can be done more directly and therefore more accurately than by taking the difference of resonant tunneling line broadenings alone . in this case , such _ coset - thermometry _ can not be performed with the charge conductance . in the main article we analyzed the nonlinear thermoelectric transport properties of a resonant level with strong coulomb interaction and a well - defined spin - flip excitation @xmath4 described by the anderson model . this suffices to classify nonlinear thermoelectric transport through an interacting nanoscale object , in the following sense : these should generally be observable in a range of nanostructures with quasi - regular filling of well - separated electronic energy shells . here we first provide support for this conclusion from a general point of view and then illustrate and further strengthen it with concrete multi - level calculations . we close by discussing where we expect deviations from the generic behavior . the general argument for the generic nature of the predicted effects is the following : the features that we considered arose from the first two leading orders , @xmath47 and @xmath48 , that we took into account . as explained in , the rate matrices in the master equation due to these processes also couple charge states differing by _ two _ electrons by pair tunneling processes , see . however , these are typically not important in the coulomb blockade regime on which we focus , but instead lead to qualitatively new features in the set regime between subsequent resonances @xcite . with this insight , it is clear that the features depend mostly on neighboring charge states and will be repeated for every next orbital state being filled . this is strongly supported by experimental data on quantum dot systems for charge transport . there is no general reason why for energy currents this argument does not apply and the concrete multi - level calculations in the next section bear this out . in fact they show that the features in cases may proliferate rather than simply replicate . [ [ example - multi - level - quantum - dotsreplication - and - proliferation - of - energy - current - effects ] ] example multi - level quantum dots replication and proliferation of energy - current effects ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ we now present results for a more complex model @xmath15 , where @xmath310 describes a quantum dot with _ two orbitals _ : @xmath311 + u ' n_1 n_2 , \label{eq:2levelqd}\ ] ] where we sum over @xmath312 corresponding to @xmath313 and @xmath314 is the occupation operator for orbital @xmath315 . the junction hamiltonian is generalized to include an amplitude for tunneling between each dot and each electrode @xmath316 but with otherwise the same assumptions as in the main article . also we again focus on the voltage - bias energy currents , i.e. , @xmath317 . this model accommodates for two qualitatively new details that are experimentally relevant and well - known from charge transport studies : + first , inelastic excitations can now be of two types : * _ spin splitting _ @xmath318 : as before , the dots are assumed to have spin splittings and these are assumed the same . * _ orbital splitting _ @xmath319 : the orbital splitting is a new feature of the multi - orbital models . thus @xmath320 as before and @xmath321 for @xmath103 . we assume that as one varies the gate voltage , @xmath59 linearly varies while the orbital splitting @xmath322 remains constant . + second , the shell filling pattern in this model can be more complicated due to the presence of an additional _ interdot interaction _ @xmath323 and the orbital splittings @xmath322 . * for sufficiently large orbital splitting @xmath322 we first fill orbital 1 and then orbital 2 . once orbital 1 is filled , it provides a `` gating '' of orbital 2 by the interdot interaction @xmath324 , shifting the latter s addition energies . in this case we have a very precise electron - hole symmetry between the two coulomb - split resonances associated with the filling each orbital . * in contrast , for lower @xmath322 we may distribute electrons over the two orbitals ( since @xmath325 ) and then fill them up completely . in this case the electron - hole symmetry is not exact but only approximate . before we discuss the example calculations , we formulate the key questions they should answer about the generic nature of the predictions in the main article . for the anderson model we identified two pronounced effects in the nonlinear energy conductance : 1 . [ triangle ] _ the suppression triangle at low bias _ above the onset of an inelastic excitation [ labeled ( iv ) in the main article ] . + a large part of the inelastic onset seems `` missing '' in the energy transport . the first question is whether this prediction holds also for _ inelastic excitations of different types _ , i.e. , other than the spin - splitting we studied . 2 . [ diamond ] _ the suppression diamond at high bias _ inside the coulomb blockade regime [ labeled ( vi ) in the main article ] . + at the lower boundaries of this regime we found pronounced negative ( positive ) @xmath58 for positive ( negative ) @xmath41 . the second question is whether the suppression diamonds [ diamond ] can still be identified in the qualitatively new coulomb regimes that can appear due to the different shell - filing possible in the model . in the noninteracting limit @xmath326 , which we mention here for completeness ( no results shown ) , our approach also includes the landauer results for model . in fact , as pointed out in this is the case for _ any multi - orbital model _ : both concretely in the limit @xmath195 , as well as in general for any @xmath193 and @xmath253 when we would renormalize the perturbation theory by appropriately including broadening ( not `` by hand '' ! ) while staying in second order . such a model does however completely miss the qualitatively new effects due to predicted in the main article which are generated by the intradot coulomb interaction @xmath23 , see our remark in . when now each dot is interacting with large intradot charging energy @xmath327 , but there are is negligible interdot interaction , @xmath328 , then the two - orbital model reduces to a sum @xmath329 of commuting anderson level models , @xmath330=0 $ ] . in principle , this does not yet imply that the currents add up : since each quantum dot still is interacting , the two dots may effectively start to interact _ via _ the tunneling to the common electrode . however , this is negligible except for special resonant level configurations ( @xmath331 ) , see . in the generic nonresonant case , the current will thus be well approximated by a sum of the currents through the two quantum dots , each strongly interacting , each current as calculated in the main article but with only an orbital offset @xmath322 . this is confirmed by explicitly calculated results shown in . in this case , all the predicted features [ triangle ] and [ diamond ] are _ replicated precisely once for each orbital_. now we switch on the _ interdot _ interaction @xmath324 . the first example of this kind is shown in . one clearly sees a _ single inelastic excitation _ relative to the ground state _ for each charge state _ in three subsequent coulomb blockade regimes due to the spin splitting @xmath4 ( assumed equal on both dots ) . in the central coulomb blockade regime the ground state is a spin triplet ( in our simple model we did not include exchange interaction ) . for the charge transport this leads to the standard spectrum , very similar to that calculated @xcite and observed experimentally in a variety of systems , qualitatively repeating itself . the coset features are weakly visible and one notices in the coulomb blockade regimes adjacent to the central one an asymmetry in the intensities due to the breaking of exact electron - hole symmetry . features in the energy conductance appear exactly at the corresponding locations as expected on the basis of the anderson model calculations of the main article : [ triangle ] the inelastic tunneling threshold is missing in the energy conductance , in particular in the central diamond , but also in the adjacent ones ; [ diamond ] the suppression diamond is most clearly seen in the central coulomb blockade regimes . the latter feature also appears in the adjacent coulomb blockade regimes but only with sharp energy conductance _ dips _ at its lower boundaries , not negative resonances . what is more , the inelastic contributions to the energy current are even pronounced _ outside _ the coulomb blockade regimes , in particular in the set regimes between the coulomb blockade regimes . thus the predicted features even tend to _ proliferate _ in the energy transport spectrum . in the second example of a fully interacting two - orbital model , shown in , we now see _ several inelastic excitations in one charge state _ , due to both _ spin _ and _ orbital _ excitations . it can be clearly seen for each of these that the inelastic threshold is missing in the energy conductance , i.e. , the predicted suppression triangle [ triangle ] is _ also replicated within one charge state_. moreover , the suppression diamond [ diamond ] is also replicated : in the central coulomb blockade regime two such diamonds are clearly visible with pronounced negative @xmath58 at their boundaries . in the left and right coulomb blockade regimes single suppression diamonds [ diamond ] with clear negative @xmath58 are seen . however , in addition negative @xmath58 is seen even at several coset features at lower bias , i.e. , this feature can also _ proliferate_. finally , this figure also illustrates the statement in the conclusion of the main article that the regime where approximations that neglect the charge fluctuations involved in coset are valid rapidly shrinks when more inelastic excitations are present : a large part of the coulomb blockade regime shows the coset energy transport of the type predicted in the article . [ [ expected - deviations - from - generic - energy - current - effectssecnongeneric ] ] expected deviations from generic energy current effects[sec : nongeneric ] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the above examples confirm that the effects predicted in the main article are quite generic for _ multi - orbital _ models in the sense that they are replicated for each orbital level as the gate voltage is scanned . in fact , the observation that the energy transport effects tends to proliferate while the charge transport remains featureless further strengthens a central point of the main paper : the energy conductance is far more sensitive and thus a promising new experimental tool . although we studied only two orbitals , adding further orbitals does not bring anything qualitatively new . only the intensities of inelastic excitations may vary , which are governed by the many - body quantum state of the dot in combination with the orbital specific tunneling matrix elements . finally , within our general approach it also is straightforward to identify situations in which we can expect interesting deviations from the generic behavior : * as mentioned in the discussion of the main article and in , at low temperature processes are significantly enhanced by renormalization effects and are generally expected to make the inelastic energy transport effects more pronounced . charge fluctuations already enhance the low - temperature as shown in ref . and most importantly spin - fluctuations will further renormalize the . extending the considerations in using naive poor - man scaling arguments the conductance at @xmath3 becomes logarithmically enhanced to @xmath332 as @xmath253 approaches the kondo scale @xmath333 from above ( @xmath334 ) and we expect that at this voltage the energy current also experiences an enhancement . it is an interesting question how the cancellations of energy current contributions described in the main article develop under this renormalization . * when a quantum dot has degenerate levels that are not protected by symmetries of the total system ( i.e. , dot plus electrodes ) the master equation must be extended to a _ quantum master equation _ which includes the coupling of occupations to coherences of the density matrix in the energy basis . this is , however , not a standard quantum master equation since it requires the coherences to be calculated to @xmath210 . such equations can be derived systematically within our general approach , see @xcite . to @xmath47 thermoelectric effects involving quantum dots with spin degeneracies coupled to spin - polarized electrodes ( spin - valve junctions ) have been studied in the regime @xcite . degeneracy effect are important also for orbital degeneracies and have been studied without energy transport , e.g. , in double quantum dots @xcite , single - atom @xcite or single - molecule junctions @xcite .

intrinsic josephson - junction stacks realized in mesas fabricated out of high - temperature superconductors may be used as sources of coherent electromagnetic radiation in the terahertz range . the major challenge is to synchronize josephson oscillations in all junctions in the stack to get significant radiation out of the crystal edge parallel to the axis . we suggest a simple way to solve this problem via artificially prepared lateral modulation of the josephson critical current identical in all junctions . in such a stack phase oscillations excite the in - phase fiske mode when the josephson frequency matches the fiske - resonance frequency which is set by the stack lateral size . the powerful almost standing electromagnetic wave is excited inside the crystal in the resonance . this wave is homogeneous across the layers meaning that the oscillations are synchronized in all junctions in the stack . we evaluate behavior of the i - v characteristics and radiated power near the resonance for arbitrary modulation and find exact solutions for several special cases corresponding to symmetric and asymmetric modulations of the critical current .

0708.3269

josephson junctions are natural voltage - to - frequency converters , since a finite voltage drop across the junction always leads to oscillating current with frequency proportional to the voltage ( _ ac _ josephson effect @xcite ) . this fundamental property suggested an attractive possibility to use the ac josephson effect for developing of voltage - tunable generators of electromagnetic waves . radiation from a josephson junction directly into the waveguide in the microwave frequency range has been indeed detected a long time ago @xcite . however , the typical detected radiated power @xmath11pw occurred to be too small for practical applications . a natural route to enhance this power is to use large arrays of junctions . this possibility has been extensively explored by several experimental groups , see reviews @xcite . when all junctions oscillate in phase , the total emitted power is expected to be proportional to the square of the total number of junctions in the array . inevitable variations of junction parameters , however , may cause variations of the oscillating frequencies leading to desynchronization and dramatic drop in emission power . therefore , the major challenge is to synchronize oscillations in all junctions @xcite . one way to solve this problem is to couple the junctions with a resonant cavity . the efficient synchronization in such systems has been demonstrated experimentally @xcite and has been extensively studied in several simulation papers @xcite . demonstration of intrinsic josephson effect in high - temperature superconductors @xcite opened a completely new route to developing electromagnetic sources . layered high - temperature superconducting materials , such as bi@xmath2sr@xmath2cacu@xmath2o@xmath3 ( bscco ) , are composed of superconducting cuo@xmath2 layers coupled by josephson interaction . intrinsic josephson effect has been extensively investigated in the past decade and most `` classical '' _ ac _ and _ dc _ josephson phenomena have been observed , including fraunhofer magnetic oscillations of critical current in small - size samples @xcite , josephson plasma resonance @xcite , shapiro steps in current - voltage characteristics induced by external microwave irradiation @xcite , fiske resonances @xcite , etc . therefore a small - size mesa fabricated out of this material represents a natural one - dimensional array of josephson junctions . a large value of superconducting gap , up to 60 mev , allows to bring the josephson frequency into the practically important terahertz range . due to atomic nature of the intrinsic junctions , it is much easier to prepare large arrays of virtually identical junctions than in the case of artificially fabricated arrays . nevertheless , the same challenge to synchronize oscillations in all junctions also remains for this system . electromagnetic waves propagate inside large - size layered superconductor in the form of plasma modes . in zero magnetic field the minimum frequency of these waves corresponding to homogeneous oscillations is given by the josephson plasma frequency . the in - plane velocity of the mode strongly depends on the c - axis wave vector , @xmath4 . the highest velocity corresponds to the in - phase mode , @xmath5 , and the lowest velocity corresponds to the antiphase mode , @xmath6 . a stack of the intrinsic josephson junctions with lateral size smaller than the decay length of electromagnetic wave behaves as a cavity . it is characterized by a spectrum of fiske resonant modes corresponding to excitation of almost standing electromagnetic waves @xcite . frequencies of these modes strongly depend on the wave vector perpendicular to the layer direction , with the maximum frequency corresponding to the homogeneous in - phase mode in all junctions and the minimum frequency corresponding to the antiphase mode . at the stack edge the electromagnetic waves excited inside the intrinsic junctions are converted into electromagnetic waves propagating into dielectric media outside the crystal . to use the stack as a coherent source of such radiation , it would be desirable to excite the in - phase resonance mode . then the radiation power is proportional to the square of the total number of junctions positioned at distances smaller than the wavelength of radiation . however , to synchronize the intrinsic junctions one needs to have strong enough coupling between them . in the simplest case of a homogeneous stack at zero magnetic field , the josephson oscillations typically interact very weakly inside the crystal . a very popular way to achieve coherent josephson oscillations in the whole stack is to apply magnetic field along the layers . the magnetic field promotes strong inductive interaction between the neighboring junctions . large magnetic field generates a dense josephson vortex ( jv ) lattice homogeneously filling all junctions . in the case of large - size system , interaction between the static jv arrays in neighboring junctions leads to formation of the triangular lattice corresponding to the @xmath7 phase shift between the phases in the neighboring junctions . the fiske resonances excited by the moving jv lattice have been observed experimentally @xcite . due to its triangular ground state , the jv lattice typically excites the antiphase modes . only a very weak outside radiation at double josephson frequency is expected in this case.@xcite to achieve a powerful emission it would be desirable to prepare aligned rectangular arrangement of jvs . such configuration is expected at certain conditions in small - size mesas due to interaction with edges . at small lattice velocities transition from triangular to rectangular configuration with increasing magnetic field has been observed experimentally as a crossover from @xmath8- to @xmath9-periodic magnetic oscillations of the flux - flow voltage @xcite . the transitions between _ static _ configurations have been studied theoretically in ref . . the possibility to have the aligned configuration at large velocities is an open issue . recent large - scale numerical simulations @xcite suggest that excitation of the resonance in - phase mode and interaction via the radiation electromagnetic field may promote alignment of jvs . a possibility to use a mesa with small lateral size and a very large number of junctions ( about 10@xmath10 ) in zero magnetic field as a source of terahertz radiation has been proposed in ref . . in such a design oscillations in different junctions are synchronized by the external electromagnetic radiation field generated by the oscillations themselves . small lateral size increases the strength of interaction due to the radiation field and allows to avoid excessive heating . such a source is frequency - tunable with the maximum power conversion efficiency about 30% . the obvious technological challenge of this design is a requirement to fabricate a mesa with such a large number of almost identical junctions . in this paper we explore a different way to excite resonance mode and synchronize oscillations in a junction stack . we propose to use a stack with strongly modulated josephson critical current ( jcc ) with modulation identical in all junctions . such a modulation dramatically enhances coupling between the josephson oscillations and the in - phase resonance modes . for a single junction such a mechanism of excitation of the fiske resonances has been considered in ref . . the frequencies of in - phase resonance modes are set by the lateral size of the mesa , @xmath11 , @xmath12 where @xmath13 is the c - axis dielectric constant . in particular , assuming @xmath14 , the resonance at @xmath15 thz for the fundamental mode , @xmath16 , takes place for a mesa with width 43 @xmath17 m . in the resonance a powerful almost standing wave is excited by josephson oscillations which synchronizes oscillations in the whole stack . such synchronization function of the cavity mode does not exist in a single junction.@xcite in such a design the frequency tunability in a single device is sacrificed in favor of larger power and better efficiency . we consider several specific cases of modulation , both symmetric and asymmetric , which correspond to excitation of the mode with the wavelength equal to @xmath11 and @xmath18 . for simplicity , we assume that the mesa size along the field direction , @xmath19 , is larger than the wavelength of outcoming electromagnetic wave , @xmath20 ( 0.3 mm for 1 thz ) . the calculation can be straightforwardly generalized to the opposite case @xmath21 . we calculate the i - v dependences , radiated electromagnetic power , and power conversion efficiency for such systems . recently , the resonant electromagnetic emission from the mesas fabricated out of underdoped bscco crystals has been detected experimentally @xcite . the resonance frequencies vary from 0.4 to 0.85 thz , and they increase roughly inversely proportional to the mesa widths . the origin of the observed resonances is most probably due to the mechanism described in this paper , even though no jcc modulation has been introduced deliberately . suppression of superconductivity near the edges during the fabrication process occurs to be sufficient for excitation of the resonances . one can expect that deliberately introduced jcc modulation will significantly enhance the amplitude of the resonance and radiation power . the paper is organized as follows . in section [ sec - gen ] we present the equation and boundary conditions for the oscillating phase when it is identical in all junctions . derivation of the boundary condition for this case is summarized in appendix [ app - bound ] . we also present the radiation power in terms of the oscillating phase and power conversion efficiency . in section [ sec - enbal ] we derive the energy balance relations in the dynamic state . in section [ sec - nearres ] , using these relations , we analyze the behavior near the resonances and derive approximate results for resonant enhancements of the current , radiated power , and power conversion efficiency . appendix [ app - rad ] presents derivation of the radiation losses for the resonance mode in a thin rectangular mesa . in section [ sec - cases ] we consider several special cases of modulation for which the problem allows for exact analytical solutions , see fig . [ fig - modulschem ] . we perform a detailed analysis of transport and radiation properties for these cases . we consider a stack of intrinsic josephson junctions ( mesa ) located at @xmath22 , with modulated jcc @xmath23 , where @xmath24 is the jcc density at the reference point at which @xmath25 . when all junctions oscillate in - phase , the @xmath0-axis homogeneous phase obeys the following reduced equation @xmath26 in which the unit of length is the c - axis london penetration depth , @xmath27 , and the unit of time is the inverse plasma frequency , @xmath28 . both @xmath27 and @xmath29 are also defined at the reference point at which @xmath25 . we will use these reduced units throughout the paper , converting to real units only in some important final results . the reduced damping parameter , @xmath30 , is related to the quasiparticle tunneling conductivity , @xmath31 . we will neglect inhomogeneity in the dissipation parameter @xmath32 , because dissipation plays a minor role in the following consideration . in the resistive state the phase is given by @xmath33 .\label{phaseresst}\ ] ] where @xmath34 is the reduced josephson frequency . we will use the linear approximation for the oscillating phase @xmath35 valid for @xmath36 . as @xmath37 $ ] , the amplitude of the oscillating phase , @xmath38 , obeys the following equation @xmath39 the boundary conditions follow from the relation between the oscillating electric and magnetic fields in outside medium and the josephson relations between the oscillating phase and these fields @xcite . in general , the boundary conditions for the @xmath0-axis homogeneous oscillating phase describing radiation can be written as @xmath40 or , in fourier representation , @xmath41 here the kernel @xmath42 depends on electromagnetic properties of the outside media . these boundary conditions assume that there is only outcoming waves at both boundaries . this means that we neglect reflected waves , coming back to the stack and mixture of radiation coming from the opposite sides . such a mixture can be suppressed , if the mesa is bounded by large metallic contacts on both sides acting like screens . derivation of the boundary condition in such a situation is presented in appendix [ app - bound ] and gives the following result for @xmath43 @xmath44 .\label{betan}\ ] ] the radiation losses are determined by the real part of @xmath45 . its imaginary part only slightly displaces the resonance frequencies . in the following , we will neglect the imaginary part in the analytical estimates . using typical values @xmath46 and @xmath47m@xcite , we estimate @xmath48 , indicating that @xmath49 is typically very small . the oscillating phase determines transport and radiation properties of the mesa . without interference , the total radiation loss , @xmath50 , is a sum of radiation powers coming from the left and right , sides , @xmath51 . the left - side power , @xmath52 , is given by poynting vector at the boundary , which is determined by the oscillating electric and magnetic fields at this side . these fields , in turn , may be related to the boundary value of the phase , @xmath53 . in the case of the boundary with free space , in real units , @xmath54 can be presented as @xcite , @xmath55 correspondingly , the power radiated from the right side , @xmath56 , is obtained from this formula by replacement @xmath57 . for @xmath58 thz and @xmath46 , we obtain an estimate for the prefactor , @xmath59 this estimate provides the upper limit for possible radiation power in the case of strong resonance @xmath60 . we can also obtain a useful general expression for the power conversion efficiency , @xmath61 , the fraction of the total power fed to the junction which is converted to radiation . the total power fed into the stack can be represented as @xmath62 where @xmath63 is the quasiparticle current @xmath64 in units of @xmath24 and @xmath65 is the reduced jcc density , @xmath66 . \label{joscurrgen}\ ] ] combining eqs . ( [ poylay ] ) and ( [ totalpowergen ] ) , we derive @xmath67}{2l}\frac{|\theta_{\omega}(0)|^{2 } + |\theta_{\omega}(l)|^{2}}{\nu_{c}\tilde{\omega}+i_{j}}. \label{powerfrac}\ ] ] in the following sections , we will first consider the behavior near the resonance frequencies and then we will present the most interesting special cases of modulation allowing for exact solutions . we consider first the energy - balance relations . the reduced energy in units of @xmath68 accumulated in the phase oscillations inside the stack is given by @xmath69 . \label{oscenergy}\ ] ] therefore , the energy - change rate is given by @xmath70 \!+\!\left [ \frac{\partial\theta}{\partial\tau}\frac { \partial\theta}{\partial x}\right ] _ { l}\!-\left [ \frac{\partial\theta } { \partial\tau}\frac{\partial\theta}{\partial x}\right ] _ { 0}.\label{enrate1}\ ] ] as follows from the eq . ( [ homphaseeq ] ) , the first term can be transformed to @xmath71 \\ & = -\nu_{c}\int_{0}^{l}dx\left ( \frac{\partial\theta}{\partial\tau}\right ) ^{2}-\int_{0}^{l}dx\frac{\partial\theta}{\partial\tau}g(x)\sin\tilde{\omega } \tau.\end{aligned}\ ] ] here , the first term accounts for the quasiparticle damping , while the second term gives the driving force from the josephson oscillations leading to pumping of energy from a dc source into the electromagnetic oscillations inside the stack . the last two terms in eq . ( [ enrate1 ] ) account for the radiation losses at the boundaries . for the general boundary conditions ( [ boundcondgen ] ) these terms can be transformed as @xmath72 _ { l}-\left [ \frac{\partial\theta}{\partial\tau}\frac { \partial\theta}{\partial x}\right ] _ { 0}=-\left [ \frac{\partial\theta } { \partial\tau}\hat{\beta}\frac{\partial\theta}{\partial\tau}\right ] _ { l}-\left [ \frac{\partial\theta}{\partial\tau}\hat{\beta}\frac { \partial\theta}{\partial\tau}\right ] _ { 0}\ ] ] where we introduce a notation for the operator @xmath73 . therefore , the total energy - change rate can be written as @xmath74 _ { l}-\left [ \frac{\partial\theta}{\partial\tau } \hat{\beta}\frac{\partial\theta}{\partial\tau}\right ] _ { 0}.\label{enrate2}\end{aligned}\ ] ] for a steady state , the energy has to remain constant , meaning that the energy supplied by the josephson oscillations has to be exactly compensated by the quasiparticle and radiation losses . in this section , we obtain approximate results for the current and radiation in the vicinity of the resonance frequency @xmath75 , where the phase can be approximated as the corresponding cavity mode @xmath76 we neglect the small influence of the radiation on the shape of the resonance mode . in this case , the energy in the mode ( [ oscenergy ] ) and energy - change rate ( [ enrate2 ] ) can be approximated as @xmath77 , \label{oscenermode}\\ \frac{\partial\mathcal{e}}{\partial \tau } & \approx\!-\frac{l}{2}\left [ \nu_{c}\left ( \frac { \partial\psi}{\partial\tau}\right ) ^{2}\!+\frac{\partial\psi}{\partial\tau } g_{m}\sin\tilde{\omega}\tau+\frac{4}{l } \frac{\partial\psi } { \partial\tau } \hat{\beta}\frac{\partial\psi}{\partial\tau } \right ] , \label{enratemode}\end{aligned}\ ] ] where @xmath78 is the coupling parameter . therefore , equation for the mode amplitude is given by @xmath79 using complex representation @xmath80 $ ] , we obtain a solution @xmath81 where @xmath82 is the parameter of the radiation damping ( the last formula is written in real units ) . one can see that both the quasiparticle dissipation and radiation contribute to the resonance damping.@xcite the cavity quality factor is given by @xmath83 . optimal power conversion is achieved when the main contribution to damping is coming from the radiation , @xmath84 . comparing the damping channels using eq . ( [ betan ] ) , we obtain that this is achieved for a sufficiently large number of junctions in the stack @xmath85 taking @xmath86 nm , we also rewrite this formula in the practically convenient form as @xmath87l[\mu \mathrm{m}]/f[\mathrm{thz}].\ ] ] for typical values @xmath88 ( @xmath89cm)@xmath90 , @xmath91 m , and @xmath92 thz , we estimate @xmath93 . in the regime of dominating radiation losses , the cavity quality factor is simply given by @xmath94 . the solution ( [ modesolutions ] ) allows us to obtain the average jcc @xmath95 \cos(m\pi x / l ) \right\ } \right\rangle \nonumber\\ & = \frac{1}{4}\frac{g_{m}^{2}\left ( \nu_{c}+\nu_{r}\right ) \tilde{\omega}}{\left [ \tilde{\omega}^{2}-\tilde{\omega}_{m}^{2}\right ] ^{2}+\left ( \nu_{c}+\nu_{r}\right ) ^{2}\tilde{\omega}^{2}}. \label{joscurrentmode}\end{aligned}\ ] ] this gives the maximum current enhancement in the resonance @xmath96 a similar result has been derived in ref . for the case of a single junction without radiation losses . comparing this result with the reduced quasiparticle current , @xmath97 , we see that the resonance feature in i - v characteristic is pronounced if @xmath98 . in the case of dominating radiation losses we can rewrite this condition in a more transparent form , @xmath99 . for @xmath100 , @xmath101 , @xmath102 m , and @xmath103 m corresponding to @xmath16 , we obtain that the resonance feature in the i - v dependence becomes strong when the coupling parameter exceeds @xmath104 . in the case of strong resonance , the total current @xmath105 nonmonotonically depends on the josephson frequency @xmath106 ( voltage ) . in this case , only the increasing part @xmath107 is stable . the total radiated power from both sides is given by @xmath108\tilde{\omega}^{2}|\psi_{\omega } & = \frac{2p_{\mathrm{sc}}\beta_{\omega}\tilde{\omega}^{2}g_{m}^{2}}{\left [ \tilde{\omega}^{2}-\tilde{\omega}_{m}^{2}\right ] ^{2}+\left ( \nu_{c } + \nu_{r}\right ) ^{2}\tilde{\omega}^{2}}.\label{radpowermode}\end{aligned}\ ] ] here , the scale of @xmath50 is given by @xmath109 , where @xmath110 is the electric field corresponding to the plasma frequency . for the maximum radiated power in the resonance we obtain @xmath111 . in the regime of dominating radiation losses , @xmath84 , using eq . ( [ betan ] ) and @xmath112 , we obtain a very simple and universal estimate for maximum total radiated power ( in real units ) @xmath113 an important observation is that _ for a tall stack in resonance , the radiated power does not depend on _ @xmath114 due to compensation between the factor @xmath115 in front of the radiated power ( [ poylay ] ) and the amplitude of phase oscillations in the resonance , which , due to the increasing radiation losses , drops as @xmath116 , see also ref .. this compensation only exists in the regime when the damping of the resonance is caused by the radiation which is realized under the condition ( [ condition ] ) . the power conversion efficiency is given by @xmath117 in resonance , it can be represented in a quite transparent form as a product of two factors @xmath118 where the first factor represents the relative current increase in the resonance , @xmath119 , and the second factor is the relative contribution of the radiation to the resonance damping . we can see that , remarkably , in resonance the conversion efficiency can approach @xmath120 provided ( i ) the resonance feature is pronounced in i - v dependence , @xmath121 and ( ii ) the losses are dominated by the radiation , @xmath122 . both conditions are quite realistic . remember that simple and transparent results for the typical number of junctions ( [ condition ] ) , current ( [ joscurrmax ] ) , radiation power ( [ radpowgenreson ] ) , and conversion efficiency ( [ conveffres ] ) are valid only in the case @xmath123 and without mixing of radiation coming from the opposite sides . these results can be generalized for other cases . radiation losses of the resonance mode in the short rectangular mesa can be approximately calculated similarly to radiation out of a rectangular capacitor.@xcite these calculations are summarized in appendix [ app - rad ] , and the results for the radiation damping parameter for different cases are presented in fig . [ fig - mesacases ] . in the case of long mesas , @xmath124 , the radiation damping parameters for different geometries differ only by numerical factors of order unity . in the opposite limit , @xmath125 , @xmath126 acquires an additional small factor @xmath127 . in this section we consider several special cases of modulation for which the problem allows for exact analytical solution and full analysis of transport and radiation properties . we will consider three cases : linear modulation , parabolic modulations , and steplike suppression of the critical current . practical ways to prepare such modulations are suggested in the discussion section [ sec - disc ] . and the same parameters as in the previous figure . one can see that the difference is rather small and amounts to a slightly different asymmetry of the peaks . , width=307 ] , for @xmath128 and the same parameters as in fig . [ fig - linmodulplots ] . note that while the radiation power increases with @xmath114 , the resonant feature in current becomes less pronounced due to the increasing radiation damping @xmath129 . the inset in the upper plot shows the @xmath114 dependence of the maximum radiation power , and the dashed line in this plot indicates the large-@xmath114 limit obtained from eq . ( [ radpowgenreson]).,width=288 ] in this section , we consider a mesa , with linearly modulated jcc , @xmath130 , see fig . [ fig - modulschem]a . here , we introduce the new coordinate , @xmath131 , which is symmetrical with respect to the mesa , @xmath132 . this means that the jcc at the left side is larger by the factor @xmath133 than the jcc at the right side . such a modulation couples homogeneous josephson oscillations with the odd fiske modes ( [ resmodes ] ) , @xmath134 , including the fundamental mode , @xmath16 , and the coupling parameter ( [ modecouplingconst ] ) to this mode is connected with the modulation parameter as @xmath135 solution of eq . ( [ oscpheq ] ) with the boundary conditions ( [ boundcond ] ) in the case of linear modulation can be found exactly . splitting the solution into the symmetric and antisymmetric parts , @xmath136 , we derive @xmath137 } , \label{linantiphase}\\ \theta_{\omega}^{(s)}(\bar{x } ) & = \frac{i}{\tilde{\omega}^{2}+i\nu_{c } \tilde{\omega}}\nonumber\\ & + \frac{\zeta\cos(p_{\omega}\bar{x})}{\left ( \tilde{\omega}^{2}+i\nu _ { c}\tilde{\omega}\right ) \left ( p_{\omega}\sin(\chi)+i\zeta\cos ( \chi)\right ) } \label{linsymphase}\\ \text{with } p_{\omega}^{2 } & \equiv\tilde{\omega}^{2}+i\nu_{c}\tilde{\omega } \text { and } \chi\equiv p_{\omega}l/2.\nonumber\end{aligned}\ ] ] only the antisymmetric phase is coupled to the resonance mode . in particular , for the boundary phases , we have @xmath138 } { p_{\omega } \left [ p_{\omega}\cos ( \chi)\!-\!i\zeta\sin(\chi)\right ] } , \label{linboundanti}\\ \theta_{\omega}^{(s)}\left ( \pm\frac{l}{2}\right ) & = \frac{i \sin(\chi ) } { p_{\omega } \left [ p_{\omega}\sin(\chi)\!+\!i\zeta\cos(\chi)\right ] } .\label{linboundsym}\end{aligned}\ ] ] the radiated power is determined by the boundary phases by eq . ( [ poylay ] ) , where we have to replace @xmath53 with @xmath139 . near the resonance @xmath140 , using @xmath141 and @xmath142 , we obtain @xmath143 } , \ ] ] where @xmath144 is defined in eq . ( [ raddamppar ] ) . using the coupling parameter ( [ linmodcouplcnst ] ) , we can see that this result is consistent with the general formula ( [ modesolutions ] ) near the resonance . the maximum antisymmetric phase in the resonance can be estimated as @xmath145 . it exceeds the nonresonant symmetric part approximately by the factor @xmath146 . in resonance , using eqs.([radpowgenreson ] ) and ( [ linmodcouplcnst ] ) , we obtain for the radiation power from one side in real units @xmath147 it exceeds nonresonant emission by the factor @xmath148 . in particular , for @xmath58 thz , @xmath149 a/@xmath150 , and @xmath151 , we estimate @xmath152 0.05 w / cm . in the average jcc ( [ joscurrgen ] ) , the contributions from symmetric and antisymmetric parts split , @xmath153 . direct calculation gives @xmath154 \left ( 2i / l+\zeta\right ) /\chi}{\left ( \tilde{\omega } ^{2}+i\nu_{c}\tilde{\omega}\right ) \left [ p_{\omega}\cos(\chi)-i\zeta \sin(\chi)\right ] } \right ] \right\ } , \label{linjoscurranti}\\ i_{j , s } & = \frac{1}{2}\operatorname{re}\left [ \frac{i}{\tilde{\omega } ^{2}+i\nu_{c}\tilde{\omega}}\left ( 1\!-\!\frac{2i\zeta\sin(\chi)/l } { p_{\omega}\sin(\chi)\!+\!i\zeta\cos(\chi)}\right ) \right ] . \label{linjoscurrsym}\end{aligned}\ ] ] from the general formula for @xmath155 near the resonance we obtain a much simpler result @xmath156 and the maximum current enhancement in the resonance is given by @xmath157 these results are also consistent with the corresponding general formulas ( [ joscurrentmode ] ) and ( [ joscurrmax ] ) if we use the coupling parameter ( [ linmodcouplcnst ] ) . for comparison , the symmetric part of the jcc at the resonance frequency can be estimated as @xmath158 as expected , the resonant enhancement of the current exceeds the nonresonant radiation correction by the same factor @xmath159 as for the radiation power . for illustration , we present the behavior near the resonance for mesas with two modulation parameters , @xmath128 and @xmath151 . as a unit of the radiation power , we selected the quantity @xmath160 which is independent on the sizes @xmath11 and @xmath161 . this choice of unit is suggested by the result ( [ radpowgenreson ] ) . for @xmath162 @xmath17 m @xmath163 w / cm . figure [ fig - linmodulplots ] shows the josephson - frequency dependences of ( i ) the current density @xmath164 ( in units of the jcc density in the center ) , ( ii ) radiated power @xmath165 ( in units of @xmath166 ) , ( iii ) the power conversion efficiency , @xmath167 , and ( iv ) the amplitude of oscillating phase at the boundary . for comparison , the case of homogeneous mesa ( @xmath168 ) is also shown . in calculation we used the following parameters : @xmath169 , @xmath170 , and @xmath171=0.00035\tilde{\omega}^{2}$ ] ( corresponding to @xmath172 , @xmath173 m , and @xmath174 [ @xmath175 cm]@xmath90 ) . we can see that the modulation leads to the appearance of a strong resonance feature in the i - v dependence . note that only i - v regions with positive differential resistivity are stable . current enhancement in the resonance is mainly caused by the generation of the powerful electromagnetic wave and it is accompanied by a huge enhancement of outside radiation . the maximum radiation power for used parameters for the case @xmath151 corresponds to @xmath176w / cm and it exceeds the nonresonant radiation from the homogeneous mesa by more than 3 orders of magnitude . it is important to note that the power conversion efficiency is also strongly enhanced in the resonance , reaching 20% for @xmath151 . the plot of @xmath177 shows that for selected parameters it remains smaller than one in the resonance and , therefore , the linear approximation used in calculations is not violated . in spite of the asymmetry of the jcc , the powers radiated to the opposite sides of the mesa in resonance are approximately the same , because the radiation is mostly promoted by the resonance mode which has identical amplitudes of the oscillating electric field at the opposite sides this is illustrated in fig.[fig - powerlinmodasym ] , where these powers are plotted for @xmath128 . one can see that the peaks have slightly different asymmetries originating from the symmetric phase . figure [ fig - linmodndep ] illustrates the evolution of the radiation power and resonance feature in the i - v dependence with increasing number of junctions in the stack , @xmath114 , for @xmath128 and the same parameters as in fig.[fig - linmodulplots ] . the number of junctions above which the radiation losses dominate ( [ condition ] ) can be estimated for used parameters as @xmath178 . we can see that the current and radiation have opposite tendencies : while the radiation power increases with @xmath114 , the resonant feature in current becomes less pronounced due to the increasing radiation damping . . [ fig - symplots ] in this section we consider a symmetric modulation . for simplicity , we assume a simple parabolic profile of the jcc density , @xmath179 , see fig . [ fig - modulschem]b , where , again , @xmath131 and @xmath180 is the ratio of jccs at the edge and in the center . the cases @xmath181 and @xmath182 correspond to current suppression and enhancement at the edges respectively . such a modulation will lead to excitation of only even frequency modes ( [ resmodes ] ) , @xmath183 . in the following , we will focus on the lowest even mode with @xmath184 . to have this resonance at @xmath185 thz assuming @xmath186 , the mesa size has to be rather large , @xmath187 @xmath17 m . the coupling parameter ( [ modecouplingconst ] ) to this mode in our case is given by @xmath188 the oscillating phase in the case of parabolic modulation also can be found exactly . from symmetry , the solution of eq . ( [ oscpheq ] ) with boundary conditions ( [ boundcond ] ) must be an even function of @xmath189 , and it has the following form @xmath190 where the first two terms give a particular solution of the inhomogeneous equation and the last term is the solution of the homogeneous equations . from the boundary conditions ( [ boundcond ] ) , we obtain @xmath191 \left ( \tilde{\omega}^{2}+i\nu_{c } \tilde{\omega}\right ) } , \label{constsol}\ ] ] where @xmath192 and @xmath193 are defined after eq . ( [ linsymphase ] ) . the oscillating phase given by eqs . ( [ oscphasesol ] ) and ( [ constsol ] ) determines all other observable properties . the powers radiated from both sides ( [ poylay ] ) are determined by the boundary phase @xmath194 , which we obtain from eqs . ( [ oscphasesol ] ) and ( [ constsol ] ) , @xmath195 \cos(\chi)}{\left ( p_{\omega}\sin(\chi)+i\zeta\cos(\chi)\right ) \left ( \tilde { \omega}^{2}+i\nu_{c}\tilde{\omega}\right ) } .\end{aligned}\ ] ] in the resonance , @xmath196 , using @xmath197 , we obtain for the power radiated from one side in real units @xmath198 this result is consistent with the general formula ( [ radpowgenreson ] ) with the coupling parameter ( [ parabcouplconst ] ) . for @xmath58 thz , @xmath199a/@xmath150 and @xmath200 , we obtain an estimate for the radiated power in the resonance , @xmath201w / cm . note that @xmath24 in eq.([poyntres ] ) is the jcc density in the center while the the jcc density at the edge , @xmath202 , is given by @xmath203 . in the case of @xmath182 vanishing of superconductivity in the middle , which corresponds to the limits @xmath204 and @xmath205 , does not lead to vanishing of radiation power because @xmath206 and the radiation is determined by the jcc density at the edge . to find the voltage - current characteristic , we calculate the average reduced jcc ( [ joscurrgen ] ) . using oscillating phase ( [ oscphasesol ] ) , we obtain @xmath207 \nonumber \\ & + \operatorname{re}\left\ { \frac{c}{2}\left ( \frac{\sin \chi } { \chi } \!-\!2r\left [ \left ( 1\!-\!\frac{2}{\chi ^{2}}\right ) \frac{\sin \chi } { 2\chi } + \frac{\cos \chi } { \chi ^{2}}\right ] \right ) \right\}. \label{joscurr}\end{aligned}\ ] ] at the resonance frequency , @xmath196 , we estimate @xmath208 . figure [ fig - symplots ] shows the representative josephson - frequency dependences of the current density @xmath164 , radiated power @xmath165 to one side , the power conversion efficiency , @xmath167 , and the amplitude of oscillating phase at the boundary for negative modulation parameters , @xmath209 and @xmath210 , corresponding to the case of stronger superconductivity at the edges . in calculation , we used the following parameters : @xmath211 , @xmath212 , and @xmath171=0.003\tilde { \omega}^{2}$ ] ( corresponding to @xmath172 and @xmath213 m in the center ) . overall , the behavior is very similar to the case of linear modulation shown in fig . [ fig - linmodulplots ] with minor quantitative differences . we also see that the modulation leads to the appearance of a strong resonance feature in the i - v dependence accompanied by a huge enhancement of the outside radiation and power conversion efficiency . . all units and parameters are the same as in fig . [ fig - linmodulplots ] ] . [ fig - stepmodwdep ] in this section , we consider the case when there is a region with suppressed jcc on one side , see fig . [ fig - modulschem]c @xmath214 the coupling parameter ( [ modecouplingconst ] ) to the fundamental mode in this case is given by @xmath215 the solution of equation for the oscillating phase ( [ oscpheq ] ) can be built in the piecewise form , @xmath216{l}% i/(\tilde{\omega}^{2}\!+\!i\nu_{c}\tilde{\omega})+a_{+}\exp(ip_{\omega}x)\\ \ + a_{-}\exp(-ip_{\omega}x),\text { for } 0<x < l - w\\ \\ i(1-r)/(\tilde{\omega}^{2}\!+\!i\nu_{c}\tilde{\omega})+\left ( a_{+}\!-\!c_{+}\right ) \exp(ip_{\omega}x)\\ \ + \left ( a_{-}\!-\!c_{-}\right ) \exp(-ip_{\omega}x),\text { for } l\!-\!w\!<\!x\!<\!l \end{array } \right.\ ] ] with @xmath217 . matching @xmath38 and @xmath218 at @xmath219 , we obtain @xmath220\ ] ] using this result , from the boundary conditions ( [ boundcond ] ) we find the coefficients @xmath221 , @xmath222 \right\ } -irp_{\omega}\left ( p_{\omega}\mp\zeta\right ) \sin\eta}{2\left ( \tilde{\omega}^{2}+i\nu_{c}\tilde{\omega}\right ) \left\ { \left [ p_{\omega}^{2}+\zeta^{2}\right ] \sin\bar{\chi}+2ip_{\omega}\zeta\cos \bar{\chi}\right\ } } , \ ] ] where @xmath223 and @xmath224 . this gives for the boundary phases which determine the outside radiation @xmath225 -i\zeta\sin \bar{\chi}\right\ } -rp_{\omega}\left\ { \zeta\left ( 1-\cos\eta\right ) + ip_{\omega}\sin\eta\right\ } } { \left ( \tilde{\omega}^{2}+i\nu_{c } \tilde{\omega}\right ) \left [ ( p_{\omega}^{2}+\zeta^{2})\sin\bar{\chi}+2ip_{\omega } \zeta\cos\bar{\chi}\right ] } , \label{stepph0}\\ \theta_{\omega}(l)\ ! & = \!\frac{i\left\ { 1-r(1-\cos\eta)\right\ } } { \tilde{\omega}^{2}+i\nu_{c}\tilde{\omega}}\nonumber\\ & + \!\frac{\zeta\!\left [ p_{\omega}\!\left ( 1+\cos\bar{\chi}\right ) \!-\!i\zeta\sin\bar{\chi}\right ] \!-\!r\left [ \zeta\left ( 1\!-\!\cos\eta\right ) \!+\!ip_{\omega}\sin\left ( \eta\right ) \right ] \left [ p_{\omega}\cos \bar{\chi}\!-\!i\zeta\sin\bar{\chi}\right ] } { \left ( \tilde{\omega}^{2}+i\nu_{c } \tilde{\omega}\right ) \left\ { \left [ p_{\omega}^{2}+\zeta^{2}\right ] \sin\bar{\chi}+2ip_{\omega}\zeta\cos\bar{\chi}\right\ } } .\label{stepphl}\end{aligned}\ ] ] near the resonance , @xmath226 , the coefficients @xmath227 and @xmath228 can be strongly simplified @xmath229 /\left ( 2\omega^{2}\right ) } { 2(\tilde{\omega}-\pi / l)-i(\nu_{c}+\nu_{r})}.\ ] ] in the resonance @xmath230 /(4\tilde{\omega}^{2}\beta_{\omega})$ ] giving @xmath231 } { 2\tilde{\omega}\beta_{\omega}}.\ ] ] at @xmath232 the resonance is strong , @xmath233 , if @xmath234 or , in real units , @xmath235 the total radiated power in resonance can be estimated as @xmath236 which is consistent with the general formula ( [ radpowgenreson ] ) . the reduced jcc flowing through the stack ( [ joscurrgen ] ) can be computed as @xmath237 \!\nonumber\\ & + \operatorname{re}\left [ \!\frac{\mathcal{n}/\bar{\chi}}{\left ( \tilde{\omega}^{2}\!+\!i\nu_{c}\tilde{\omega}\right ) \left\ { \left [ p_{\omega } ^{2}\!+\!\zeta^{2}\right ] \sin\bar{\chi}\!+\!2ip_{\omega}\zeta\cos\bar{\chi}\right\ } } \right ] \label{stepjoscurr}\end{aligned}\ ] ] with @xmath238 \right\ } \\ & + ir^{2}\sin^{2}\frac{\eta}{2}\left ( p_{\omega}\cos\frac{\eta}{2 } -i\zeta\sin\frac{\eta}{2}\right ) \left ( p_{\omega}\cos\left [ \bar{\chi } -\frac{\eta}{2}\right ] + i\zeta\sin\left [ \bar{\chi}-\frac{\eta}{2}\right ] \right)\end{aligned}\ ] ] near the resonance , we estimate @xmath239 which gives @xmath240 the maximum current enhancement in the resonance is given by @xmath241 in agreement with eqs . ( [ joscurrmax ] ) and ( [ stepcouplconst ] ) . general cumbersome formulas ( [ stepph0 ] ) , ( [ stepphl ] ) , and ( [ stepjoscurr ] ) can be significantly simplified if we assume the conditions @xmath30 , @xmath242 and @xmath232 valid in most practical situations . in this case these equations can be represented in approximate , simpler form , @xmath243 and @xmath244 -2\tilde{\omega } rw\cos\left ( \bar{\chi}-\frac{\eta}{2}\right ) \right ] } { 2l\tilde{\omega } ^{3}\left [ \sin\bar{\chi}-i\left ( \nu_{c}l/2 + 2\beta_{\omega}\right ) \cos \bar{\chi}\right ] } \right\}\end{aligned}\ ] ] with @xmath245 and@xmath246 . figure [ fig - stepmodwdep ] illustrates the evolution of the radiated power and resonance feature in the i - v dependence with increasing width of the suppressed region . the jcc density in the suppressed region is assumed to be half of its value in the rest part , @xmath247 . for used parameters , the maximum radiation power in this plot is around , @xmath248w / cm . let us discuss now practical ways to prepare mesas with lateral modulation of the critical current density . mesas with linear jcc modulation can be fabricated in a crystal with inhomogeneous doping . one possible way to prepare such inhomogeneity is to utilize the sensitivity of doping in bscco to the oxygen concentration . due to strong temperature dependence of the oxygen diffusivity @xcite , in principle , the oxygen concentration profile in the crystal can be prepared by short - time annealing by carefully selecting the annealing temperature and time . in a similar way , mesas with parabolic - like profiles can be prepared by short - time annealing of mesas themselves already after fabrication . another way to prepare modulation in a controlled way is to use radiation with high - energy electrons , protons , or heavy ions . if part of the mesa is protected by a mask , this radiation will produce a mesa with steplike suppression of the critical current . the major technical challenge is to prepare a mesa with significant modulation of the josephson coupling _ identical in all junctions_. variation of parameters in different junctions , which may be caused by composition variations , inhomogeneous heating , and different junction areas would strongly reduce the optimal performance . the quantitative analysis of the radiation properties of mesas with such parameter variations in different layers will be done elsewhere . as the optimal mesa size is rather large , another major technical problem is sample heating due to quasiparticle damping . the self - heating in the bscco mesas has been investigated by many experimental groups @xcite . the major focus of these studies was the influence of heating on the gap feature in i - v characteristics , which is located at voltages 30 - 60 mv / junction . even though our voltage range is significantly lower , @xmath12 mv / junction , the heating is still expected to be significant due to the required large lateral size of the mesa . for example , for @xmath249 @xmath250[@xmath175 cm ] , @xmath46 , and @xmath251 m , 10 mw of power will be dissipated inside the mesa . this heat has to be removed from the mesa faces . therefore , efficient heat removal is crucial for operation of the device . recent experimental observations of the resonant emission using underdoped bscco @xcite demonstrate that the heating effects can be manageable even in large - size mesas with lateral sizes of several hundred micrometers in the voltage range corresponding to the josephson frequencies around 1 thz . the designs with improved thermal management may include , for example , fabrication of underdoped mesas on the top of overdoped crystal , using massive gold contacts on the top and bottom of the mesa , and placing an insulator with high thermal conductivity , such as sapphire , in good thermal contact at the side of the mesa . from these considerations , mesas with asymmetric modulation look more preferable than ones with symmetric modulation , because they need a smaller lateral size for the lowest resonance mode . in the case of symmetric modulation , the design with suppression of the jcc in the middle , @xmath182 , looks more practical for better thermal management . in fact , the material in the middle can even be made insulating , because this part is needed only to form almost standing wave at the working frequency . to excite resonance mode , it is sufficient to have superconducting regions only at the edges . in conclusion , we demonstrated that a stack of the intrinsic josephson junctions with modulated josephson coupling represents a very powerful and efficient source of electromagnetic radiation at the resonance frequency set by its lateral size . selecting this size , the generation frequency can be tuned to the terahertz range . power levels up to several milliwatts look plausible in such structures . aek would like to acknowledge very useful discussions and joint work on practical implementation of the device discussed in this manuscript with u.welp , k. gray , l. ozyuzer , and c. kurter . in argonne this work was supported by the department of energy under contract no . de - ac02 - 06ch11357 . in the los alamos national laboratory this work was carried out under the auspices of the national nuclear security administration of the department of energy under contract no . de - ac-06na25396 . in this appendix , we consider the boundary conditions for the oscillating phase at the edges and the radiation power for a stack of intrinsic josephson junctions . we will limit ourselves to the case when the oscillating phase is identical in all junctions . a more general case will be considered elsewhere . the oscillating phase @xmath38 defined by eq . ( [ phaseresst ] ) is connected with the electric and magnetic fields by the josephson relations@xmath252 therefore , the boundary conditions for the oscillating phase at the edges are determined by the relation between the fields @xmath253 and @xmath254 in the outside media , which we assume to be monochromatic with time dependences @xmath255 . outside dielectric media at @xmath256 is characterized by the dielectric constant @xmath257 , and we assume only outgoing wave in this space . the fourier components of fields with @xmath258 propagate in the media , while the field components with @xmath259 decay . in particular , for @xmath260 at @xmath261 , we have @xmath262 \label{ez - kz}\ ] ] with @xmath263 other field components , @xmath264 and @xmath254 , are also expressed via @xmath265 . first , @xmath266 is obtained from eq.([ez - kz ] ) and the maxwell equation @xmath267 , and then @xmath268 is obtained using the maxwell equation @xmath269 leading to the following result @xmath270.\ ] ] this gives the relation between the fields at the boundary @xmath271 , @xmath272 the condition at @xmath273 has opposite sign , @xmath274 . note again that the term @xmath275 for @xmath276 originates from outgoing electromagnetic wave ( radiation ) , while the term @xmath275 for @xmath259 is due to the wave decaying at distance @xmath277 from the crystal boundary . the latter term does not carry energy out of the junctions . in particular , for @xmath278 , we have @xmath279 , leading to the simple boundary condition for the homogeneous oscillating phase in the limit @xmath280 , @xmath281 for @xmath282 . the relation ( [ byez - kz ] ) can also be rewritten in the frequency - space representation as @xmath283 , \nonumber\end{aligned}\ ] ] where @xmath284 and @xmath285 are the bessel functions . the same approach can be used in the realistic case of a crystal small along the @xmath286 axis , @xmath287 , if we know the radiated electric field at the planes @xmath288 outside of the crystal , at @xmath289 . if we put well conducting screens there , we can approximate @xmath290 at @xmath289 . in this case for the homogeneous @xmath291-independent electric field , we obtain for the average magnetic field at the edge@xmath292 e_{z}(l,\omega),\ ] ] with @xmath293 , where @xmath294 is the euler constant . we can see that for a small - size mesa , the magnetic field at the boundary is reduced by the factor @xmath295 in comparison to the infinite-@xmath161 case . this gives the following boundary condition for the oscillating phase@xmath296 \theta_{\omega},\\ ] ] for @xmath282 . this corresponds to the boundary conditions ( [ boundcond ] ) and ( [ betan ] ) in reduced coordinates and @xmath297 used in the paper . therefore , for short crystals , the boundary condition can not be written in the form of an instantaneous relation in between the space and time derivatives of the phase . this significantly complicates their numerical implementation . screens also completely isolate semispaces @xmath261 and @xmath298 and eliminate interference of radiation coming from the opposite edges . the radiation power from a short rectangular mesa can be found approximately . for such a mesa radiation influences weakly the shape of the resonance mode . in such a situation , the radiation is mostly determined by the distribution of the oscillating electric field at the mesa edge , which , in turn , is determined by the shape of the internal mode . finding the radiation occurs to be a somewhat easier problem than finding general boundary conditions for the oscillating phase . an approximate expression for the radiated electric field far away from the crystal and , thus , the radiation power can be calculated using the huyghens principle , as it is developed in the theory of antennas , see , e.g. , ref . . this approach has been applied to resonance modes in rectangular capacitors in ref . . such a consideration clearly shows the role of screens and crystal geometry in the formation of the radiation . the huyghens principle in the formulation of schelkunoff ( equivalence principle ) states that we can find fields outside of real sources ( currents and charges ) if we know equivalent sources placed on some boundary surface surrounding real sources . in particular , the equivalent magnetic current , @xmath299 , is related to the tangential components of the electric field @xmath300 on the surface as @xmath301 where @xmath302 is the normal vector on the boundary surface . in the following , we consider the crystal inside the volume @xmath303 , @xmath304 , and @xmath305 bounded by thin metallic contacts on the top and bottom , see fig . [ fig - radrectgeom ] . the contacts are highly conductive , and there we can neglect the tangential electric field . at the crystal edges @xmath306 we can neglect the magnetic field when the radiation power is small , as in the case of radiation from a capacitor with a small distance @xmath161 between electrodes , @xmath307 , see ref . . in this case , we need to account only for the electric field at the crystal edges , which produces the magnetic equivalent currents ( [ magcurr ] ) . they are related to the oscillating phases @xmath38 at these edges according to eq . ( [ e - jos ] ) . we consider modes inside the crystal which are uniform along the @xmath286-axis ( synchronized josephson oscillations in all ijjs ) . in this approximation , the electric field @xmath253 inside the crystal is described by standing waves characterized by the indices @xmath308 and @xmath291 : @xmath309 faraway radiated electric field in terms of @xmath310 is given by the expression @xmath311 where the integral is taken over the perimeter of the crystal , and the coordinate system as well as definitions of the unit vectors @xmath312 and @xmath313 are given in fig . [ fig - radrectgeom ] . integration over contour @xmath314 gives the following result@xcite @xmath315 where @xmath316 and @xmath317 the interference factors @xmath318 and @xmath319 describe the contribution of waves coming to the faraway point @xmath320 from opposite sides of the crystal @xmath321 and @xmath322 along the @xmath323 axis as well as from opposite sides @xmath324 and @xmath325 , respectively . their role in the formation of the radiation becomes clear if we will consider different modes . from the radiated electric field , we can compute the total radiated power as@xmath326 .\label{radpower}%\ ] ] @xmath328 \left [ 1-\exp\left ( -ik_{\eta}l_{y}\right ) \right]\ ] ] corresponding to @xmath329 \left [ 1-\exp\left ( -ik_{\omega}l_{y}\sin\theta\sin\phi\right ) \right ] } { \sin\theta\cos\phi\sin\phi},\ ] ] and @xmath330 . the radiated power ( [ radpower ] ) is given by @xmath331 \left [ 1-\cos\left ( k_{\omega}l_{y}\sin\theta\sin\phi\right ) \right ] } { \sin\theta\cos^{2}\phi\sin^{2}\phi}%\ ] ] for small - size crystal @xmath332 with almost uniform josephson oscillations , we obtain @xmath333 this is the result for dipole radiation because all sizes of the crystal are small in comparison with the wavelength of the radiated field . for the crystal with size @xmath19 bigger than the radiation wavelength , @xmath334 , the result is quite different : @xmath335 . \label{radhommode}%\ ] ] now the waves coming from opposite sides of the crystal along the @xmath336 axis do not interfere with each other and radiation power becomes proportional to @xmath19 . for @xmath337 , we obtain the power proportional to @xmath338 due to destructive interference of the waves coming from opposite sides of the crystal along the @xmath323 axis . if we put highly conductive metallic screens separating the spaces @xmath339 and @xmath298 so that the edge @xmath273 radiates only into @xmath298 half - space , while that at @xmath340 radiates only into @xmath339 half - space ( see fig . 2 ) , the interference will be eliminated . such screens also double the radiation coming from one side . this can be demonstrated in the simplest way using image technique@xcite : radiation from the real electric currents induced at the screens is equivalent to radiation from the image magnetic current placed next to the original magnetic current . this leads to doubling of the effective magnetic current , @xmath341 , and quadruples the radiated power density . as the radiation now is limited only by half - space , the total radiated power doubles . therefore , in the presence of screens , the factor @xmath342 $ ] in eq . ( [ radhommode ] ) has to be replaced by the factor @xmath343 . this means that the screens strongly enhance the radiation induced by the homogeneous mode in the case @xmath337 . such a design with screens for a crystal thin along the @xmath323 axis was proposed in ref . . this design gives the possibility of frequency tuning . in addition , heating is reduced due to small @xmath344 . however , the crystal should have a large number of layers to synchronize oscillations in all junctions and work in the super - radiation regime . @xmath346 \left [ 1+\exp\left ( -ik_{\eta}l_{y}\right ) \right]\ ] ] the components of the faraway electric field are given by the expressions @xmath347 \left [ 1-\exp\left ( -ia_{y}% \sin\theta\sin\phi\right ) \right ] , \\ \frac{e_{\phi}}{e_{z}(1,0 ) } & = -\frac{l_{z}\exp(ik_{\omega}r)}{4\pi r}% \frac{(\pi / a_{x})^{2}}{\sin^{2}\theta\cos^{2}\phi-(\pi / a_{x})^{2}}\frac { \cos\theta}{\sin\theta}\left [ 1+\exp\left ( -ia_{x}\sin\theta\cos \phi\right ) \right ] \left [ 1-\exp\left ( -ia_{y}\sin\theta\sin\phi\right ) \right]\end{aligned}\ ] ] with @xmath348 and @xmath349 . the radiated power ( [ radpower ] ) can be represented as @xmath350 with@xmath351 \left [ 1-\cos\left ( a_{y}\sin\theta\sin\phi\right ) \right ] } { \sin \theta\sin^{2}\phi}\nonumber\\ & \times\frac{\left ( \sin^{2}\theta-(\pi / a_{x})^{2}\right ) ^{2}\cos^{2}% \phi+(\pi / a_{x})^{4}\cos^{2}\theta\sin^{2}\phi}{\left ( \sin^{2}\theta\cos ^{2}\phi-(\pi / a_{x})^{2}\right ) ^{2}}.\label{i10}%\end{aligned}\ ] ] in the regime @xmath334 this gives the following result @xmath352.\label{radfundmode}%\ ] ] now we have a constructive interference of waves coming from opposite sides of the crystal along the @xmath323 axis because electric field on these sides has opposite signs generating the same - sign magnetic fields . for such mode , screens do not influence much the radiation in the limit @xmath337 . the reduced parameter of radiation damping @xmath144 introduced in eqs . ( [ modesolutions ] ) and ( [ raddamppar ] ) is related to the radiation power as @xmath353 the above results can also be straightforwardly generalized to the case when a stack is bounded by large - size ground plate at the bottom , @xmath354 . this is the case , for example , for the mesa fabricated on the top of bulk crystal . if we treat the ground plate as an ideal conductor , its influence can again be taken into account by the image technique@xcite . this just leads to the doubling of the effective magnetic current , @xmath355 , and to the doubling of the total radiated power @xmath356 . p. barbara , a. b. cawthorne , s. v. shitov , and c. j. lobb , phys . 82 * , 1963 ( 1999 ) ; b. vasili , s. v. shitov , c. j. lobb , and p. barbara , appl . phys . letters , * 78 * 1137 ( 2001 ) ; b. vasili , p. barbara , s. v. shitov , and c. j. lobb , phys . rev . b * 65 * , 180503(r ) ( 2002 ) . y. matsuda , m.b . gaifullin , k. kumagai , k. kadowaki , and t. mochiku , phys . rev . lett . * 75 * , 4512 ( 1995 ) ; y. matsuda , m.b . gaifullin , k. kumagai , m. kosugi , and k. hirata , phys . * 78 * , 1972 ( 1997 ) . throughout the paper , in estimates we use parameters typical for moderately underdoped bscco at low temperatures . usage of the underdoped bscco in the discussed design is preferable for reducing heating due to smaller quasiparticle conductivity . note that , in general , the radiation damping parameter @xmath126 depends not only on stack geometry but also on the excited mode . only in the case considered here ( @xmath357 and no interference between radiation from the opposite edges ) @xmath126 is the same for all @xmath308 . v. m. krasnov , a. yurgens , d. winkler , and p. delsing , j. appl . phys . , * 89 * , 5578 ( 2001 ) ; j. c. fenton and c. e. gough , j. appl . phys . , * 94 * , 4665 ( 2003 ) ; v. n. zavaritsky , phys . rev . b * 72 * , 094503 ( 2005 ) ; v. m. krasnov , m. sandberg , and i. zogaj , phys . lett . , * 94 * , 077003 ( 2005 ) ; h. b. wang , t. hatano , t. yamashita , p. h. wu , and p. muller , appl . lett . , * 86 * , 023504 ( 2005 ) ; myung - ho bae , jae - hyun choi , and hu - jong lee , appl . lett . , * 86 * , 232502 ( 2005);x . b. zhu , y. f. wei , s. p. zhao , g. h. chen , h. f. yang , a. z. jin , and c. z. gu , phys . rev . b , * 73 * , 224501 ( 2006 ) .

we present a new model to compute the effects of dust on the integrated spectral properties of galaxies , based on an idealized prescription of the main features of the interstellar medium ( ism ) . the model includes the ionization of hii regions in the interiors of the dense clouds in which stars form and the influence of the finite lifetime of these clouds on the absorption of radiation . we compute the production of emission lines and the absorption of continuum radiation in the hii regions and the subsequent transfer of line and continuum radiation in the surrounding hi regions and the ambient ism . this enables us to interpret simultaneously all the observations of a homogeneous sample of nearby ultraviolet - selected starburst galaxies , including the ratio of far - infrared to ultraviolet luminosities , the ratio of h to h luminosities , the h equivalent width , and the ultraviolet spectral slope . we show that the finite lifetime of stellar birth clouds is a key ingredient to resolve an apparent discrepancy between the attenuation of line and continuum photons in starburst galaxies . in addition , we find that an effective absorption curve proportional to reproduces the observed relation between the ratio of far - infrared to ultraviolet luminosities and the ultraviolet spectral slope . we interpret this relation most simply as a sequence in the overall dust content of the galaxies . the shallow wavelength dependence of the effective absorption curve is compatible with the steepness of known extinction curves if the dust has a patchy distribution . in particular , we find that a random distribution of discrete clouds with optical depths similar to those in the milky way provides a consistent interpretation of all the observations . a noteworthy outcome of our detailed analysis is that the observed mean relations for starburst galaxies can be closely approximated by the following simple recipe : use an effective absorption curve proportional to to attenuate the line and continuum radiation from each stellar generation , and lower the normalization of the curve typically by a factor of 3 afteryr to account for the dispersal of the birth clouds . this recipe or our full model for absorption can be incorporated easily into any population synthesis model .

astro-ph0003128

to interpret the observed spectral properties of galaxies , we require models for both the production of stellar radiation and its transfer through the interstellar medium ( ism ) . currently , the accuracy of population synthesis models contrasts with the rudimentary way in which the absorption of starlight by dust is often treated . in many applications , dust is either ignored or assumed to be distributed in a uniform screen in front of the stars . the resulting uncertainties in the absorption of the ultraviolet radiation in galaxies can be as much as an order of magnitude . this problem has become especially acute in studies of galaxies at high redshifts . nearby starburst galaxies may be suitable analogs of high - redshift galaxies and provide important clues to interpreting their spectral properties . the observations of nearby ultraviolet - selected starburst galaxies are numerous , including the ratio of far - infrared to ultraviolet luminosities , the ratio of h@xmath0 to h@xmath1 luminosities , the h@xmath0 equivalent width , and the ultraviolet spectral slope . in particular , there is a remarkably tight correlation between far - infrared luminosity and ultraviolet spectral slope ( meurer et al . 1995 ; meurer , heckman , & calzetti 1999 ) . this wealth of observations can potentially help us quantify the effects of dust on various spectral properties of galaxies . there have been several analyses of the spatial distribution and optical properties of the dust in nearby starburst galaxies based on various subsets of the observations ( e.g. , fanelli , oconnell , & thuan 1988 ; calzetti , kinney , & storchi - bergmann 1994 , 1996 ; puxley & brand 1994 ; meurer et al . 1995 ; gordon , calzetti , & witt 1997 ) . a generic result of these studies is that if the dust is distributed in a uniform foreground screen , it must have an unusually grey extinction curve . otherwise , the distribution must be patchy . however , these analyses also raise questions of how to account self - consistently for all the observations . for example , the absorption inferred from the h@xmath0/h@xmath1 ratio in starburst galaxies is typically twice as high as that inferred from the ultraviolet spectral slope . one interpretation of this result is that the ionized gas and ultraviolet - bright stars have different spatial distributions ( calzetti 1997 ) . issues such as this highlight the need for a simple yet versatile model to interpret simultaneously a wide range of phenomena related to the absorption of starlight by dust in galaxies . the purpose of this paper is to present such a model . we begin with the conventional view that young stars ionize hii regions in the interiors of the dense clouds in which they are born . line photons produced in the hii regions and the non - ionizing continuum photons from young stars are absorbed in the same way by dust in the outer hi envelopes of the birth clouds and the ambient ism . the birth clouds , however , have finite lifetimes . thus , non - ionizing ultraviolet and optical photons from stars that live longer than the birth clouds are absorbed only by the ambient ism . this allows the ultraviolet continuum to be less attenuated than the emission lines . our model builds on several previous studies . for example , silva et al . ( 1998 ) considered the effects of finite lifetimes of stellar birth clouds on the continuum but not the line emission from galaxies . here , we treat the transfer of radiation ( especially scattering ) in an approximate way , which would preclude a detailed description of surface brightnesses but should be appropriate for angle - averaged quantities such as luminosities . our model succeeds in accounting quantitatively for all the available observations of a homogeneous sample of nearby starburst galaxies . we present our model in 2 , where we express the effective absorption curve describing the global transmission of radiation in terms of the different components of the ism . in 3 , we compare our model with observations and identify the specific influence of each parameter on the different integrated spectral properties of galaxies . one outcome of our detailed analysis is a remarkably simple recipe for absorption , which provides a good approximation to the observed mean relations and is easy to incorporate into any population synthesis model . in 4 , we explore how the spatial distribution of dust can be constrained by the observations . our conclusions are summarized in 5 . we first introduce some notation and nomenclature that will help specify our model . the luminosity per unit wavelength @xmath5 emerging at the time @xmath6 from a galaxy illuminated by an internal stellar population can be expressed generally as @xmath7 here @xmath8 is the star formation rate at the time @xmath9 , @xmath10 is the luminosity emitted per unit wavelength and per unit mass by a stellar generation of age @xmath11 , and @xmath12 is the `` transmission function , '' defined as the fraction of the radiation produced at the wavelength @xmath13 at the time @xmath6 by a generation of stars of age @xmath11 that escapes from the galaxy . thus , @xmath12 must be regarded as the average transmission along rays emanating in all directions from all stars of age @xmath11 within the galaxy . since absorption and scattering by dust will cause the radiation to emerge anisotropically to some degree , @xmath14 can only be measured by a hypothetical set of detectors completely surrounding the galaxy . in practice , this means that equation ( [ basic ] ) gives the mean of @xmath15 , where @xmath16 is the distance and @xmath17 is the observed flux , for an ensemble of randomly oriented but otherwise similar galaxies . for some purposes , it is convenient to reexpress @xmath5 as the product of a mean transmission function weighted by the luminosity of each stellar generation @xmath18 and the total unattenuated stellar luminosity : @xmath19 @xmath20 we often reexpress @xmath18 in terms of an `` effective absorption '' optical depth @xmath21 indicated by a caret and defined by @xmath22\ , . \label{taueff}\ ] ] this nomenclature emphasizes that the luminosity of a galaxy is diminished purely by the absorption of photons . defined by equation ( [ taueff ] ) has also been referred to by several other names in the literature , including the optical depth of `` apparent extinction , '' `` effective extinction , '' `` attenuation , '' `` obscuration , '' and `` absorption . '' ] however , since scattering may also affect the path lengths of the photons through the ism before they are absorbed , @xmath21 will differ in general from the true absorption curve determined by the optical properties of the dust grains . the fraction of stellar radiation not transmitted by the ism , @xmath23 , is absorbed by dust . using equation ( [ basic2 ] ) and integrating over all wavelengths , we obtain the total luminosity absorbed and reradiated by dust @xmath24\ , \int_0^t\,dt'\,\psi(t - t')\,s_\lambda(t')\ , . \label{ldust}\ ] ] we now specify our model for computing the above properties of galaxies . this amounts to specifying the transmission function @xmath12 . our goal is to parameterize only those features of the ism that are essential to understanding the main physical effects of dust on the integrated spectral properties of galaxies . we adopt the simple but fairly conventional view illustrated in figure 1 . young stars ionize hii regions in the inner parts of the dense ( mostly molecular ) clouds in which they are born . line photons produced in the hii regions and the non - ionizing continuum photons from young stars propagate through the outer hi envelopes of these `` birth clouds '' and then through the `` ambient ism '' before they escape from the galaxy . both the hii and hi regions of the birth clouds may contain dust , and in the ambient ism , the dust may be distributed in a smooth or patchy way . the birth clouds , however , have finite lifetimes . thus , photons emitted by longer - lived stars propagate only through the ambient ism where they are treated in the same way as photons emerging from the birth clouds before they escape from the galaxy . to incorporate these basic features into our model , we make the following simplifying assumptions . we assume that stars form in clusters with the galaxy - wide stellar initial mass function ( imf ) . this should be an excellent approximation for the purpose of computing the integrated spectral properties of a galaxy , although it could break down in clusters or regions with small numbers of stars . in our model , all star clusters are embedded in their birth clouds for some time and then disrupt them or migrate away from them into the ambient ism of the galaxy . we assume for simplicity that the effective absorption in the birth clouds depends only on the stellar age @xmath11 and is constant in the ambient ism . hence , we can write the total transmission function in the form @xmath25 where @xmath26 and @xmath27 are the transmission functions of the birth clouds and the ambient ism , respectively . in our model , the birth clouds , assumed to be identical , consist of an inner hii region ionized by young stars and bounded by an outer hi region . for simplicity , we assume that all stars remain within their birth clouds with a constant transmission function for a time @xmath28 and are then introduced immediately into the ambient ism ; thus @xmath29 where @xmath30 and @xmath31 are the transmission functions of the hii and hi regions , respectively . by writing the transmission functions as products in equations ( [ tsingle ] ) and ( [ tbc ] ) , we have assumed that the propagation of photons in the hii and hi regions and the ambient ism are independent of each other . this prescription is exact for forward scattering and should be a good approximation for the purpose of computing luminosities even with some backward scattering . an alternative but equivalent way of expressing the transmission functions of the hii and hi regions and the ambient ism is in terms of the corresponding effective absorption optical depths , @xmath32 , @xmath33 , and @xmath34 . for non - ionizing photons ( with @xmath35 ) , we thus write @xmath36 @xmath37 @xmath38 for some purposes , it is also convenient to define the total effective absorption optical depth of dust in the birth clouds and the fraction of this contributed by the internal hii regions @xmath39 @xmath40 we note that @xmath41 may or may not coincide with the ratio of the hii to total gas column densities of the birth clouds , depending on whether or not the dust - to - gas ratios in the hii regions are the same as those in the hi regions . finally , we assume that the emission lines are produced only in the hii regions of the birth clouds . the justification for this is that the lifetimes of the birth clouds are generally greater than the lifetimes of the stars producing most of the ionizing photons , i.e. about @xmath42yr . in this case , line photons are absorbed in the hi regions and the ambient ism in the same way as non - ionizing continuum photons . the emergent luminosity of a line of wavelength @xmath43 is thus @xmath44 where @xmath45 is the luminosity of the line produced in the hii regions by a stellar generation of age @xmath11 , and @xmath46 and @xmath47 are given by equations ( [ thi ] ) and ( [ tism ] ) . the luminosities of h - recombination lines are proportional to the ionization rate , @xmath48 where @xmath49 is the fraction of lyman continuum photons absorbed by the gas rather than the dust in the hii regions . we adopt the formula @xmath50\rbrace}}\ , . \label{absion}\ ] ] this equation , which is equivalent to equation ( 5 - 29 ) and table 5.4 of spitzer ( 1978 ) , was derived by petrosian , silk , & field ( 1972 ) and shown by them to provide a good approximation to the results of more detailed numerical computations . we are especially interested in the emergent luminosities of the h@xmath0 and h@xmath1 balmer lines of hydrogen , @xmath51 and @xmath52 . we assume that , given @xmath53 , the production rates of all h - recombination photons are the same as for the dust - free case b recombination . this is validated by detailed photoionization models of hii regions , including the effects of internal dust ( hummer & storey 1992 ; ferland 1996 ; bottorff et al . 1998 ) . for electronic densities @xmath54@xmath55 and temperatures @xmath56k , we then have ( osterbrock 1989 ) @xmath57 with @xmath58 and @xmath59 . it is straightforward to extend this procedure to compute the luminosity of any other h - recombination line ( e.g. , p@xmath1 , b@xmath60 ) . we assume that all the power radiated in the form of l@xmath0 photons , @xmath61 , is eventually absorbed by dust as a consequence of resonant scattering ( @xmath62 ) . we have checked that our conclusions are not affected even if a significant fraction of l@xmath0 photons escapes from the galaxy . for completeness , we compute the power from ionizing radiation that is neither absorbed by dust nor emerges in the h@xmath0 and h@xmath1 lines , @xmath63 and distribute this uniformly in wavelength between 3000 and 6000 ( a range that includes most of the relevant emission lines ) . this is only a few percent of all the stellar radiation and has a negligible influence on our results . we now consider how the transmission function of the ambient ism depends on the spatial distribution and optical properties of the dust . a particularly enlightening way to express @xmath27 is in terms of the probability density for the absorption of photons emitted in all directions by all stars within a galaxy . we define @xmath64 as the optical depth of absorption along the path of a photon , from its emission by a star to its escape from the galaxy , including the influence of scattering on the path . from now on , we refer to @xmath64 simply as the `` optical depth '' of the dust . is sometimes referred to as an effective optical depth and denoted by @xmath65 ( rybicki & lightman 1979 ) . in this paper , we use the term `` effective optical depth '' to refer to the geometry - dependent optical depth @xmath66 defined by equation ( [ taueff ] ) . ] we further define @xmath67 as the probability that the optical depth lies between @xmath64 and @xmath68 . integrating over rays emanating in all directions from all stars within the galaxy , we can then write the transmission function of the ambient ism as @xmath69 if the spatial distribution of the dust is specified , this determines @xmath70 , and @xmath27 then follows from equation ( [ tlamism ] ) . for some purposes , however , it is more informative to specify @xmath70 itself rather than any of the spatial distributions of the dust that could give rise to it when evaluating @xmath27 . the optical depth @xmath64 of the dust depends on a combination of absorption and scattering . if scattering were only in the forward direction , @xmath64 would equal the true absorption optical depth . in the case of isotropic scattering , @xmath64 is given by the geometrical mean of the true absorption and extinction ( absorption plus scattering ) optical depths , i.e. , @xmath71^{1/2 } = \tau_\lambda^a ( 1- \omega_\lambda)^{-1/2}\ , , \label{tscat}\ ] ] where @xmath72 is the single - scattering albedo ( rybicki & lightman 1979 ) . ) is exact in the limit of large extinction optical depths but overestimates @xmath64 in the opposite limit . ] although this formula was derived for an infinite homogeneous medium , it also provides an excellent approximation for more complicated geometries ( see , e.g. , fig . 5 of silva et al . 1998 , which compares this approximation with a monte carlo calculation by witt , thronson , & capuano 1992 ) . in reality , both observations and models suggest that interstellar dust in the milky way has pronounced forward scattering at ultraviolet and optical wavelengths ( e.g. , draine & lee 1984 ; kim , martin , & hendry 1994 ; calzetti et al . 1995 ) . thus , we expect the truth to lie between the idealized formulae @xmath73 and @xmath74 . in our model , the details of scattering and absorption are subsumed in our approximation of the optical depth of the dust as a power law in wavelength , @xmath75 ( 4 ) . this approach would be questionable if we wished to compute directional quantities such as surface brightnesses . however , since we compute only angle - averaged quantities ( i.e. , luminosities ) , our simple treatment of scattering should be an adequate approximation . the reason for this is that , since many of the photons scattered out of a line of sight will be replaced by other photons scattered into it , inaccuracies in the treatment of scattering will partially ( but not totally ) cancel out when the emission is averaged over all directions . we now consider three specific , idealized distributions of dust in the ambient ism : a uniform foreground screen , a mixed slab , and discrete clouds . these are fairly standard representations , which may be regarded as illustrative of a wide range of possibilities . other models could be constructed for more realistic and hence more complicated distributions of dust , for example by multiplying the transmission functions derived from different @xmath70 s . it should be recalled in any case that the total transmission function in our model is the product of that for the birth clouds and that for the ambient ism , as specified by equation ( [ tsingle ] ) . _ foreground screen_@xmath76 . the simplest model is that of a uniform foreground screen , which absorbs the light from all stars by the same amount . while unrealistic , this model has been adopted in many previous studies , and we include it here for reference . we call @xmath77 the optical depth of the screen . we then have @xmath78 @xmath79 where @xmath80 is the dirac delta function . _ mixed slab_@xmath76 . a potentially more realistic model is that of a slab filled with a uniform mixture of sources and absorbers ( i.e. , stars and dust ) . this can account for differences in the absorption affecting stars near the midplane or near the surface of a galactic disk . we call @xmath81 the full optical depth through the slab in the direction normal to the surface . by considering sources located at all points within the slab and rays emanating at all angles from the normal , we obtain , after some manipulation , @xmath82 @xmath83\ , , \label{tslab}\ ] ] where @xmath84 is the exponential integral of the first order . _ discrete clouds_@xmath76 . both the foreground screen and mixed slab models are based on smooth representations of the ambient ism . in real galaxies , however , the ism is organized into clouds and other structures . the photons emitted in different directions by different stars can encounter any number of these clouds on their way out of the galaxy . thus , @xmath70 may be better approximated by a poisson distribution ( spitzer 1978 ; natta & panagia 1984 ) . we call @xmath85 the optical depth per cloud , and @xmath86 the mean number of clouds encountered along different lines of sight . the mean optical depth is thus @xmath87 . in this case , we have ) corresponds to a poisson distribution of identical , face - on screens . for spherical clouds , the factor @xmath88 would have to be replaced by a function with a finite width to account for the different optical depths seen by photons with different impact parameters . we ignore this subtlety here . ] @xmath89 @xmath90\ , . \label{tpois}\ ] ] in the limits @xmath91 and @xmath92 , we obtain the familiar expressions @xmath93 and @xmath94 . this model should provide an adequate description of the absorption along individual random lines of sight , as required by equation ( [ tlamism ] ) , but it neglects possible correlations between the absorption along multiple nearby lines of sight , which may be important in other applications . in principle , we could compute models with arbitrary star formation rates . for simplicity , in all the following applications , we take the star formation rate to be constant and include the age @xmath6 of the stellar population as a parameter . this should be regarded as the effective age of the most recent burst of star formation , which can range from a few times @xmath4yr for a galaxy in an active starburst to several times @xmath95yr for a more quiescent galaxy . with the assumption @xmath96 , equations ( [ ttrans ] ) , ( [ taueff ] ) , and ( [ tsingle])([taubc ] ) allow us to reexpress the effective absorption optical depth at the time @xmath6 in terms of the effective absorption optical depths in the birth clouds and the ism as @xmath97 with @xmath98 , & for $ t > t_{\rm bc}$.\cr}\cr & & \label{deltataueff}\end{aligned}\ ] ] we compute @xmath10 using the latest version of the bruzual & charlot ( 1993 ) population synthesis code . our conclusions would not be affected by the use of other standard codes . unless otherwise indicated , we adopt solar metallicity and a salpeter imf truncated at 0.1 and 100@xmath99 . we now have a completely specified model for the absorption of starlight by dust in galaxies . the main adjustable parameters describing the birth clouds are : the lifetime of the clouds , @xmath100 ; the wavelength dependence and normalization of @xmath101 ; and the fraction of @xmath102 contributed by dust in hii regions , @xmath41 . for the ambient ism , the parameters depend on whether we adopt a foreground screen model ( @xmath77 ) , a mixed slab model ( @xmath81 ) , or a discrete cloud model ( @xmath85 and @xmath86 ) . the other main parameter is the effective age of the starburst , @xmath6 . we compute the continuum luminosity per unit wavelength emerging from a galaxy using equations ( [ basic ] ) and ( [ tsingle])([tism ] ) . the emergent h@xmath0 and h@xmath1 line luminosities , @xmath103 and @xmath104 , are obtained from equations ( [ thi ] ) , ( [ tism ] ) , and ( [ emline])([hlums ] ) . the ratio of the h@xmath0 line and continuum luminosities is then the emission equivalent width of the emergent h@xmath0 line , @xmath105 . finally , we compute the far - infrared luminosity of the galaxy , @xmath106 , using equations ( [ ttrans ] ) and ( [ ldust ] ) . in the remainder of this paper , we take two approaches to compare our model with observations . as equations ( [ tism ] ) and ( [ tlamism ] ) indicate , the optical properties and spatial distribution of the dust affect observable quantities only through the effective absorption curve @xmath34 . different combinations of @xmath64 and @xmath70 , however , could lead to the same wavelength dependence of @xmath107 . thus , it is of primary interest to study how the wavelength dependence of @xmath108 is constrained by observations . this is our goal in 3 below . then , in 4 , we explore whether there are examples of optical properties and spatial distributions of the dust that satisfy these constraints , recognizing that there is no unique solution . in this section , we use observations of nearby starburst galaxies to constrain the wavelength dependence of the effective absorption curve @xmath109 defined by equation ( [ taueff ] ) without specifying the spatial distribution and optical properties of the dust . we also identify the specific influence of each parameter in the model on different integrated spectral properties of galaxies and investigate the physical origin of the observed relationships between these properties . we assume for simplicity that the effective absorption curves of the birth clouds and the ambient ism have the same power - law form , @xmath110 @xmath111 where @xmath112 is a parameter to be determined , and @xmath113 and @xmath114 are the normalizing coefficients at 5500 . to calibrate our model , we appeal to a homogeneous set of observations of nearby ultraviolet - selected starburst galaxies , for which the ultraviolet , far - infrared , h@xmath0 , and h@xmath1 fluxes , h@xmath0 equivalent widths , and ultraviolet spectral slopes are available . the sample , compiled by meurer et al . ( 1999 ) , includes 57 non - seyfert galaxies from the _ international ultraviolet explorer _ ( _ iue _ ) atlas of kinney et al . ( 1993 ) classified into one of the starburst categories ( i.e. , starburst nucleus , starburst ring , blue compact dwarf galaxy , or blue compact galaxy ) . this sample has been further restricted by meurer et al . ( 1999 ) to galaxies with optical angular diameters less than 4 , for which most of the ( usually concentrated ) ultraviolet emission could be observed within the _ iue _ aperture . the galaxies span a wide range of morphological types , from sab to irr , and a wide range of absolute @xmath115-band magnitudes , @xmath116 ( we adopt a hubble constant @xmath117 km@xmath76s@xmath118mpc@xmath118 ) . from the _ iue _ spectra of all galaxies in the sample , meurer et al . ( 1999 ) compute the restframe ultraviolet flux at 1600 , @xmath119 , and the slope @xmath1 of the ultraviolet continuum . the quantity @xmath119 is defined as @xmath120 , where @xmath121 is the mean flux density over a square passband with a central wavelength of 1600 and a width of 350 . the slope @xmath1 is determined from a power - law fit of the form @xmath122 to the observed ultraviolet spectrum in ten continuum bands that avoid strong stellar and interstellar absorption features ( in particular , the 2175 dust feature ) in the range @xmath123 , as defined by calzetti et al . these ultraviolet properties were derived for galaxies in the sample after removing foreground extinction by the milky way . observations with the _ infrared astronomical satellite _ ( _ iras _ ) are available for 47 galaxies in the sample . to estimate the total far - infrared flux @xmath124 from the observed _ iras _ flux densities , @xmath125 and @xmath126 , we follow the prescription of meurer et al . they first calculate the quantity @xmath127\,\,$]ergs@xmath76@xmath128s@xmath118 defined by helou et al . ( 1988 ) and then compute @xmath124 ( @xmath129 in their notation ) from @xmath130 and @xmath131 using the bolometric correction appropriate for dust with a single temperature and an emissivity proportional to frequency @xmath132 . this prescription agrees reasonably well with observations from near - infrared to submillimeter wavelengths for a small sample of galaxies ( see fig . 2 of meurer et al . the corrections from @xmath130 to @xmath124 are relatively small , with a mean @xmath133 and a standard deviation of 0.1 . finally , for 31 galaxies , the ratio of h@xmath0 to h@xmath1 fluxes and the h@xmath0 equivalent width are available from storchi - bergmann , kinney , & challis ( 1995 ) and mcquade , calzetti , & kinney ( 1995 ) , with corrections for stellar absorption by calzetti et al . ( 1994 ) . neglecting possible anisotropies , we equate flux ratios at all wavelengths to the corresponding luminosity ratios . figures 2 , 3 , and 4 show @xmath134 , @xmath135 , and @xmath136 , respectively , as a function of @xmath1 for this sample . the typical scatter at fixed @xmath1 amounts to a factor of about 2 in @xmath137 and @xmath138 , but nearly a factor of 10 in @xmath139 . thus , both @xmath134 and @xmath140 , to a first approximation , are functions of @xmath1 , while there is little or no correlation between @xmath139 and @xmath1 . we note , however , that the scatter even in the first two relations appears to be real because it is larger than the typical measurement errors ( which are shown in the upper left panels of figs . we now use these observations to constrain the parameters in our model . our goal is to identify models that account for the typical properties , scatter , and trends seen in the sample , rather than to interpret in detail the emission from individual galaxies . such an attempt might be of limited validity because the predictions of our model , unlike the observations , are averaged over angles . and @xmath141 in figure 2 could be a galaxy for which some of the ultraviolet emission was missed with _ iue _ ( see the discussion in 3.1 in meurer et al . ] we have checked , however , that the properties of every galaxy in the sample could be reproduced with at least one combination of parameters . the influence of each parameter on observable quantities can best be explored by keeping all other parameters fixed at `` standard '' values . after some experimentation , we adopted : @xmath142 while these values are not the results of a rigorous optimization procedure , they do enable the standard model to match roughly the observed typical ( i.e. , median ) properties of the starburst sample . we note that @xmath143 is typical of the observed optical depths to o - type and supergiant stars in the milky way ( humphreys 1978 ) . the dense cores of molecular clouds have higher optical depths , but even short - lived stars spend only the first 10%@xmath14420% of their lifetimes in these regions ( e.g. , mezger & smith 1977 ) . also , @xmath145yr is typical of the timescale for the dispersal of giant molecular clouds in the milky way ( blitz & shu 1980 ) . we now compare the predictions of our model with the observations described above . each panel in figures 24 shows the effect of increasing and decreasing one parameter with respect to its standard value with the others held fixed . we can summarize the role of each parameter as follows . _ effective starburst age_@xmath76 . the effect of increasing @xmath6 is most noticeable on the h@xmath0 equivalent width , which drops from almost 300 at @xmath146yr to less than 100 at @xmath147yr because of the rising contribution by old stars to the continuum near the line ( fig . increasing @xmath6 also makes the ultraviolet spectral slope larger because old stars have intrinsically red spectra . however , this does not affect the h@xmath148h@xmath1 ratio , since the emission lines are produced mainly by stars younger than about @xmath149yr ( fig . 3_a _ ) . _ lifetime of the birth clouds_@xmath76 . increasing @xmath28 makes both @xmath1 and the ratio of far - infrared to ultraviolet luminosities larger because the radiation from young blue stars is heavily absorbed for a longer time ( fig . changes in @xmath28 do not affect @xmath150 because we have assumed that the birth clouds last longer than @xmath149yr , the lifetime of stars producing most of the ionizing photons ( 2 ) . however , since more of the continuum luminosity near h@xmath0 is absorbed for larger @xmath28 , the h@xmath0 equivalent width also increases ( fig . 4_b _ ) . _ effective absorption optical depth in the birth clouds_@xmath76 . increasing @xmath113 above unity has the remarkable property of making @xmath150 larger at almost fixed @xmath1 and @xmath151 ( figs . 2_c _ and 3_c _ ) . the reason for this is that , while @xmath150 traces the absorption in the hi envelopes of the birth clouds , the ultraviolet radiation from embedded stars is so attenuated for @xmath152 that @xmath153 saturates , and @xmath1 and @xmath154 are dominated by stars in the ambient ism . this has fundamental observational implications ( see below ) . figure 4_c _ also shows that the h@xmath0 equivalent width is very sensitive to changes in @xmath113 , which affect @xmath155 . _ effective absorption optical depth in the ambient ism_@xmath76 . since @xmath114 controls the absorption of stars that have emerged from their birth clouds , it has a major influence on the ultraviolet spectra and hence on @xmath1 and @xmath151 ( fig . changes in @xmath114 also affect the h@xmath148h@xmath1 ratio , although with a strong dependence on @xmath1 ( fig . as expected , @xmath114 has no effect on @xmath139 . _ fraction of dust in the hii regions_@xmath76 . increasing @xmath41 causes more ionizing photons to be absorbed by dust and hence reduces the h@xmath0 equivalent width considerably ( fig . dust in hii regions has a negligible effect on the h@xmath148h@xmath1 ratio . however , since we have kept @xmath113 fixed , increasing @xmath41 reduces the absorption in the hi envelopes of the birth clouds and hence the h@xmath148h@xmath1 ratio in figure 3_e_. changes in @xmath41 have no effect on @xmath1 and @xmath154 and a negligible effect on @xmath153 . _ exponent of the effective absorption curve_@xmath76 . increasing @xmath112 steepens the effective absorption curve . this increases the ratio of far - infrared to ultraviolet luminosities , the h@xmath148h@xmath1 ratio , and the ultraviolet spectral slope and reduces slightly the h@xmath0 equivalent width ( figs . 2_f_4_f _ ) . we elaborate below on the reason for adopting @xmath156 as a standard value . _ other parameters_@xmath76 . we have also computed models with different imfs and metallicities ( not shown ) . changes in these parameters have no influence on @xmath150 and affect @xmath1 and @xmath134 at the level of less than 20% . the influence on the h@xmath0 equivalent width is slightly stronger . for example , adopting a scalo imf instead of a salpeter imf would reduce @xmath139 by about 50% for the standard model because of the smaller proportion of massive stars . in contrast , adopting a metallicity 20% of the solar value would increase @xmath139 by about 20% because metal - poor stars are hotter than solar - metallicity stars . we also tested the influence of a possible leakage of some ionizing photons from hii regions into the ambient ism by allowing 50% of the line photons produced by stars in the standard model to escape without absorption in the hi envelopes of the birth clouds ( oey & kennicutt 1997 , and references therein ) . while the h@xmath148h@xmath1 ratio was found to be only 12% lower in this case than in the standard model , the h@xmath0 equivalent width increased by 60% . it is also interesting to examine the shape of the total effective absorption curve defined by equation ( [ taueff ] ) and its dependence on the age of the stellar population . in figure 5 , we show @xmath21 at different ages for our standard model ( computed from eqs . [ [ newtaueff ] ] and [ [ deltataueff ] ] ) . initially , for @xmath157 , @xmath158 is a power law of index @xmath156 . at later times , the effective absorption curve steepens markedly at wavelengths @xmath159 ( @xmath160 ) . it is more similar to the initial power law in the wavelength range where @xmath1 is determined ( @xmath161 ) , but steeper again at wavelengths @xmath162 ( @xmath163 ) . the reason for this is that short - lived , massive stars , which produce most of the luminosity at @xmath164 , are embedded in the birth clouds , where their radiation is heavily absorbed . their contribution is substantial and nearly constant across the wavelength range @xmath165 but less important at longer wavelengths . the steepening of @xmath158 at @xmath166 is also enhanced by the appearance in the ambient ism of long - lived stars with a strong 4000 break in their spectra , which contribute more light redward than blueward of the break . the absorption in the birth clouds , therefore , has a major influence on the shape of the effective absorption curve . it is worth pausing here to emphasize the important implications of the finite lifetimes of the birth clouds . as mentioned in 1 , emission lines appear to be more attenuated than the continuum radiation in starburst galaxies . this led calzetti ( 1997 ) to suggest that ultraviolet - bright stars must be somehow physically detached from the gas they ionize . in our model , the lines produced in hii regions and the non - ionizing continuum radiation from young stars are attenuated in the same way by dust in the outer hi envelopes of the birth clouds and the ambient ism . however , since the birth clouds have finite lifetimes , the non - ionizing continuum radiation from stars that live longer than the birth clouds is attenuated only by the ambient ism . the ultraviolet and optical continuum radiation , therefore , appears to be less attenuated than the lines . this can be seen clearly in figures 3_c _ and 4_c _ from the rise in the h@xmath167h@xmath1 ratio accompanying the drop in the h@xmath0 equivalent width at nearly constant ultraviolet spectral slope as the effective absorption optical depth of the birth clouds increases above unity . thus , the finite lifetime of the birth clouds is a key ingredient to resolve the apparent discrepancy between the attenuation of line and continuum photons in starburst galaxies . since each parameter in our model has a specific influence on the various integrated spectral properties of starburst galaxies , we can gain insight into the physical origin of the relations defined by these observed quantities . an inspection of figures 24 leads to the following conclusions . ( 1 ) the wide observed range of ultraviolet spectral slopes can be most naturally understood as a spread in the effective absorption optical depth in the ambient ism , although variations in the lifetimes of the birth clouds could also play some role . it is worth mentioning that the steep ultraviolet spectra of the galaxies with largest @xmath1 can not originate from old stars because these galaxies also have large @xmath139 . ( 2 ) as noted above , absorption in the birth clouds is required to explain the large h@xmath148h@xmath1 ratios observed in galaxies with intermediate ultraviolet spectral slopes . ( 3 ) the fact that @xmath134 , @xmath168 , and @xmath1 are all very low at one end of the observed relations and all very high at the other end further suggests that variations in the effective optical depths of the birth clouds and the ambient ism may be related ( figs . 2 and 3 ) . to explore the effects of varying simultaneously @xmath113 and @xmath169 , we assume that they are related by @xmath170 and fix all other parameters at the standard values given by equation ( [ standard ] ) . the results are shown in figure 6 . evidently , the models reproduce well the observed relations between @xmath171 , @xmath172 , and @xmath1 in figures 6_a _ and 6_b_. they also fall roughly in the middle of the observed ranges of @xmath173 and @xmath1 in figure 6_c_. an important result of our analysis is that the shape of the relation between @xmath151 and @xmath1 actually _ requires _ that the exponent of the effective absorption curve @xmath174 be @xmath175 . in figure 6 , we also show models with @xmath176 as above , but for @xmath177 and 1.0 instead of 0.7 . these models miss the observed relation , the sequence being too steep for @xmath177 and too shallow for @xmath178 . this observational constraint on @xmath112 applies essentially to @xmath179 , since @xmath151 and @xmath1 are almost independent of @xmath180 for @xmath181 . for @xmath156 , therefore , the relations defined by @xmath182 , @xmath183 , and @xmath1 can be understood most simply as a sequence in the overall dust content of galaxies . our assumption of @xmath184 should be taken merely as illustrative . with @xmath156 , the observations can be matched by a variety of models that range from @xmath185 at one end of the relations ( with @xmath186 ) to @xmath187 and @xmath188 at the other end ( with @xmath189 ) . for reference , the fraction of the total stellar radiation absorbed by dust in the models of figure 6 ranges from 0% to about 90% along this sequence and is roughly 70% for our standard model . non - ionizing continuum photons account for about 90% of the heating in all models except for those with very low dust content , in which the relative contribution by l@xmath0 photons is more important ( over 20% for @xmath190 ) . as figures 24 indicate , the intrinsic scatter about the relations at fixed @xmath1 can arise from variations in the effective age of the starburst , the parameters controlling the birth clouds , and the fraction of dust in the hii regions ( and possibly the imf and metallicity of the stars ) . = 10.7 cm our models are also consistent with observations of near - infrared h - recombination lines in starburst galaxies . the fluxes of the paschen @xmath1 ( 1.28 @xmath191 m ) and brackett @xmath60 ( 2.17 @xmath191 m ) lines of 11 galaxies in the meurer et al . ( 1999 ) sample are available from calzetti et al . we compute the @xmath192 and @xmath193 luminosities , @xmath194 and @xmath195 , using the same procedure as described in 2 for @xmath196 and @xmath197 . figure 7 shows the @xmath198 and @xmath199 ratios as a function of the @xmath200 ratio for the observations and for the same models as in figure 6 . sequences of models are straight lines in these log - log diagrams , reflecting the assumed power - law form of the effective absorption curves in the birth clouds and the ambient ism . figure 7 shows that @xmath156 is consistent with , although not tightly constrained by , the observed strengths of optical and near - infrared recombinations lines in nearby starburst galaxies . as noted above , the exponent @xmath112 of the effective absorption curve is much better constrained by the relation between the ratio of far - infrared to ultraviolet luminosities and the ultraviolet spectral slope . it is worth noting that the observed mean relations can be reproduced by the following simple recipe : use an effective absorption curve proportional to @xmath3 to attenuate the non - ionizing continuum radiation and emission lines ( as predicted by dust - free case b recombination ) from each stellar generation , and lower the normalization of the curve typically by a factor of 3 after @xmath4yr to account for the dispersal of the birth clouds . this recipe approximates the results of our full calculations shown in figures 6 and 7 but differs from them in that it omits the absorption of ionizing photons by dust in hii regions ( i.e. , it assumes @xmath201 ) . as mentioned earlier , this has no effect on the ultraviolet spectrum and a negligible effect on the far - infrared luminosity ( @xmath202% underestimate ) and the @xmath203 ratio ( from @xmath202% to about 7% overestimate along the observed relations ) . the effect is more important for the h@xmath0 equivalent width ( from 3% to over 70% overestimate ) . of course , the simple recipe above does not allow one to reproduce the scatter in the observed relations , which requires variations in several parameters of our model . however , if only the mean properties are considered , the simple recipe should suffice for many purposes . both our complete model and the simple recipe derived from it can be incorporated easily into any population synthesis model . = 10.7 cm in an attempt to understand the physical basis for the observed relation between @xmath151 and @xmath1 , we plot other properties of the galaxies against @xmath1 in figure 8 : the oxygen abundance of hii regions o / h and the absolute @xmath115 magnitude @xmath204 . the oxygen abundances of 31 galaxies of the meurer et al . ( 1999 ) sample , measured in the same aperture as @xmath1 , are available from calzetti et al . ( 1994 ) , while the absolute @xmath115 magnitudes of all the galaxies are available from kinney et al . the spearman rank correlation coefficients are @xmath205 for the relation between o / h and @xmath1 and @xmath206 for that between @xmath204 and @xmath1 . for the different numbers of galaxies in each case , these correlations are significant at the @xmath207 and @xmath208 levels . thus , they are probably both statistically significant , especially the one between o / h and @xmath1 ( which was noted by calzetti et al . 1994 from a slightly different sample ) , but both have very large scatter . this suggests that neither o / h nor @xmath204 are entirely responsible for the relations between @xmath209 , @xmath210 , and @xmath1 . the ultimate physical explanation for these relations therefore requires further investigation . for reference , we have also computed models using the calzetti et al . ( 1994 ) effective absorption curve . this was derived from the analysis of _ iue _ spectra of a sample of 33 starburst galaxies , nearly all of which are included in the meurer et al . ( 1999 ) sample . calzetti et al . ( 1994 ) condensed all the spectra in their sample into six `` templates '' corresponding to different values of the h@xmath148h@xmath1 ratio . they then derived an effective absorption curve by dividing the five spectral templates with the highest @xmath168 by the template with the lowest @xmath211 ( assuming that the latter represented the dust - free case ) . the result is strictly an effective absorption curve in the sense defined by equation ( [ taueff ] ) , although it is often referred to by other names ( see footnote 2 ) . figure 5 shows that the calzetti et al . ( 1994 ) effective absorption curve is greyer than those derived here . in figure 6 , we show the results obtained when adopting the calzetti et al . ( 1994 ) curve for @xmath34 and assuming @xmath212 for consistency . , we use the formula provided by calzetti ( 1999 ) . to compute @xmath34 at wavelengths @xmath213 , we extrapolate the curve linearly in @xmath214 . this has a negligible influence on the results because typically less than 25% of @xmath153 is produced by the absorption blueward of 1200 . in fact , adopting @xmath215 for @xmath216 gives similar results . ] as anticipated , the models predict a steeper relation between @xmath217 and @xmath1 than is observed . the greyness of the calzetti et al . ( 1994 ) curve probably results from a bias inherent in the method used to derive it . as figure 6_b _ shows , galaxies with the lowest observed h@xmath148h@xmath1 ratios extend over a fairly wide range of ultraviolet spectral slopes . thus , the template spectrum with the lowest @xmath210 , built by calzetti et al . ( 1994 ) from eight spectra with @xmath218 , may be relatively but not perfectly dust free , as is assumed by the method . up to now , we have compared our model with observations only in terms of the effective absorption curve , i.e. , without specifying the spatial distribution and optical properties of the dust . in this section , we study the influence of different spatial distributions of dust on the integrated spectral properties of galaxies . as shown in 3 , the effective absorption curve favored by observations of nearby starburst galaxies is significantly shallower than most known extinction curves . this is a natural consequence of any spatial distribution of dust in which the least extincted stars contribute most to the emergent light at short wavelengths . such an effect has been quantified for many distributions of dust , both patchy and smooth ( e.g. , natta & panagia 1984 ; caplan & deharveng 1986 ; bruzual , magris , & calvet 1988 ; disney , davies , & phillipps 1989 ; witt et al . 1992 ; calzetti et al . 1994 , 1996 ; puxley & brand 1994 ; di bartolomeo , barbaro , & perinotto 1995 ; bianchi , ferrara , & giovanardi 1996 ; witt & gordon 1996 ; gordon et al . 1997 ) . these studies , however , did not include the finite lifetimes of stellar birth clouds . thus , they can not account for the different attenuation of line and continuum photons in starburst galaxies . we now investigate the influence of the spatial distribution of the dust on the effective absorption curve in the context of our model . for simplicity , we assume that the optical depth of the dust has a power - law form @xmath219 where the choice of @xmath220 is discussed below . this is consistent with recent evidence that the optical depth of the dust in nearby starburst galaxies does not exhibit a strong feature near 2175 ( gordon et al . we further assume for simplicity that @xmath180 has the same wavelength dependence as @xmath64 . for comparison , figure 9 shows @xmath64 plotted against @xmath13 for milky way - type , lmc ( 30 doradus)-type , and smc - type dust and for the two cases of forward - only and isotropic scattering ( eq . [ [ tscat ] ] ) . these optical properties are based on the standard draine & lee ( 1984 ) grain model but with the proportions of graphite and silicates adjusted so as to fit the observed mean extinction curves of the three galaxies ( pei 1992 ) . the inclusion of scattering makes @xmath64 greyer because the albedo is larger at optical than at ultraviolet wavelengths , the effect increasing from milky way- to lmc- to smc - type dust . power - law fits to these curves over the wavelength range @xmath221 yield values of @xmath220 between @xmath222 and @xmath223 , with @xmath224 representing a middle value ( fig . the ambient ism of starburst galaxies can not be idealized as a foreground screen if the dust within them is similar to that in the milky way , the lmc , or the smc because the effective absorption curve would then take the form @xmath225 with @xmath226 . as shown in 3 , this would lead to a relation between @xmath217 and @xmath1 shallower than observed . we now examine the other two models of the ambient ism specified in 2 . again , these should be regarded only as convenient approximations to more realistic distributions of dust . in the following , we adopt the same effective starburst age , @xmath6 , lifetime of birth clouds , @xmath28 , and fraction of dust in the ionized gas , @xmath41 , as in our standard model in 3 ( eq . [ [ standard ] ] ) . we first consider the mixed slab model for the ambient ism . in this case , @xmath34 is related to the optical depth through the slab in the direction normal to the surface , @xmath227 , by equations ( [ tism ] ) and ( [ tslab ] ) . by analogy with our standard model in 3 , we adopt @xmath228 for the optical depth of the birth clouds and take @xmath229 . figure 10 shows the total effective absorption curve at @xmath230yr with @xmath224 . a comparison with figure 5 shows that this is very similar to the total effective absorption curve of the standard model with @xmath156 in 3 . to understand the origin of this result , we plotted @xmath34 as a function of @xmath81 and noticed that the simple formula @xmath231 approximates @xmath34 with an accuracy or 5% or better over the whole range @xmath232 ( at low optical depths , @xmath34 is closer to @xmath81 ) . this weak dependence of @xmath34 on @xmath233 arises from the fact that , for large @xmath81 , sources located at high optical depths from the surface of the slab contribute very little to the emergent light . in the ultraviolet , therefore , where @xmath81 is greater than unity , the wavelength dependence of @xmath108 is closer to @xmath234 than to @xmath235 in the model of figure 10 . hence , for reasonable assumptions about the optical properties of the dust , a mixed slab model for the ambient ism can produce an effective absorption curve similar to that required by the observed relation between @xmath151 and @xmath1 . mixed slab models , however , can not accommodate the full range of observations . in figure 11 , we show the analog of figure 6 for models with @xmath236 and @xmath224 . at low optical depths , i.e. for @xmath237 , the models reproduce well the observed relation between @xmath151 and @xmath1 , as noted above . at larger optical depths , however , the models deviate markedly from the observations . this upturn , which persists over the range @xmath238 , is entirely a consequence of absorption in the ambient ism because , for large optical depths of the birth clouds , i.e. for @xmath239 , @xmath1 and @xmath151 are controlled by @xmath34 ( figs . 2_c _ and 2_d _ ) . the reason for this behavior can be traced back to the fact that @xmath108 scales only as @xmath240 for @xmath241 . as @xmath242 increases in figure 11 , @xmath243 approaches a power law of index @xmath244 over a range that expands toward progressively longer wavelengths ( since @xmath245 ) . once the power - law shape extends over the entire ultraviolet wavelength range , increasing @xmath242 no longer affects @xmath1 but continues to alter @xmath151 , @xmath246 , and @xmath136 . this is an intrinsic limitation of the mixed slab model . for large @xmath242 , @xmath1 can not reach the very large values observed because sources at high optical depths from the surface of the slab contribute very little to the emergent ultraviolet radiation . we now consider the discrete cloud model for the ambient ism . in this case , @xmath34 is related to the optical depth per cloud , @xmath247 , and the mean number of clouds along a line of sight , @xmath86 , by the expression @xmath248 $ ] ( eqs . [ [ tism ] ] and [ [ tpois ] ] ) . thus , @xmath86 does not affect the wavelength dependence of @xmath34 . to illustrate the effective absorption curves of the discrete cloud models , we fix @xmath228 and experiment with different optical depths of the clouds , @xmath249 , 0.3 , and 0.5 . by analogy with our standard model in 3 , we adjust @xmath86 so that the mean optical depth along a line of sight is @xmath250 in each case . figure 12 shows the total effective absorption curves at @xmath251yr for these models with @xmath224 . the models exhibit the characteristic signatures of the absorption in the birth clouds ( fig . 5 ) . as @xmath252 increases , @xmath109 becomes greyer because the wavelength below which the clouds absorb all incident radiation , i.e. where @xmath253 , becomes progressively larger . figure 12 indicates that the model with @xmath254 has the ultraviolet slope closest to that of the standard model in 3 . for reasonable assumptions about the optical properties of the dust , therefore , discrete cloud models also lead to effective absorption curves similar to that required by the observed relation between @xmath151 and @xmath1 . the advantage of discrete cloud models over mixed slab models is that they can account for the full range of observations . in figure 13 , we show the analog of figure 6 for models with @xmath255 , @xmath86 a continuous variable , @xmath249 , 0.3 , and 0.5 , and @xmath224 . the models with @xmath254 agree with all the observations . in particular , they match remarkably well the relations between @xmath134 , @xmath246 , and @xmath1 over the entire range observed . models with @xmath256 and 0.1 lie above and below those with @xmath254 , as expected from the different effective absorption curves in figure 12 . the mean number of clouds in the models with @xmath257 ranges from @xmath258 at one end of the relations ( with @xmath259 ) to @xmath260 at the other end ( with @xmath261 ) . these values of @xmath86 are small enough that the distribution of dust can be regarded as patchy over most of the range . we also find that the relations between @xmath262 , @xmath263 , and @xmath1 are very stable to @xmath220 over the range @xmath238 for @xmath264 , but curiously less stable for @xmath265 and 0.5 . the above values of @xmath252 would be somewhat smaller if we had adopted an optical depth @xmath64 with a strong feature near 2175 . our results demonstrate that , for reasonable assumptions about the wavelength dependence of the dust optical depth @xmath64 , discrete cloud models can account for all the observed integrated spectral properties of nearby starburst galaxies . poisson distributions of clouds have traditionally been used to interpret variations in the observed reddening of stars at the same distance within the milky way . these observations can be well represented by a model with two types of clouds , one with @xmath266 and another , less abundant one with @xmath267 ( spitzer 1978 ) . this is similar to our model , which has a population of optically thin clouds in the ambient ism and a population of optically thick birth clouds . it is worth emphasizing that in any model with discrete clouds , the optical depth has a large variance along different lines of sight within a given galaxy , depending on the number of clouds encountered . in particular , some lines of sight can be quite opaque , even for @xmath264 and @xmath268 . for starburst galaxies with the largest values of @xmath151 , @xmath269 is indicated , and most lines of sight are opaque . thus , there is no contradiction between the relatively small values of @xmath252 found here and the very dusty appearance of some starburst galaxies . we have developed a simple model to compute the effects of dust on the integrated spectral properties of galaxies , based on an idealized prescription of the main features of the ism . our model includes the ionization of hii regions in the interiors of the dense clouds in which stars form . emission lines from hii regions and the non - ionizing continuum from young stars are attenuated in the same way by dust in the outer hi envelopes of the birth clouds and the ambient ism . however , since the model also includes the finite lifetimes of the birth clouds , the non - ionizing continuum radiation from stars that live longer than the birth clouds is attenuated only by the ambient ism . we show that this can fully resolve the apparent discrepancy between the attenuation of line and continuum photons in starburst galaxies . this enables us , in turn , to interpret in a consistent way all the observations of a homogeneous sample of nearby ultraviolet - selected starburst galaxies , including the ratio of far - infrared to ultraviolet luminosities ( @xmath151 ) , the ratio of h@xmath0 to h@xmath1 luminosities ( @xmath150 ) , the h@xmath0 equivalent width ( @xmath139 ) , and the ultraviolet spectral slope ( @xmath1 ) . the different parameters in our model , including the effective age of the starburst , the lifetime and effective optical depth of the stellar birth clouds , the effective optical depth in the ambient ism , and the fraction of dust in the ionized gas , each have a specific influence on the integrated spectral properties @xmath137 , @xmath270 , @xmath136 , and @xmath1 . this provides new insights into the origin of the mean relations defined by the data and the scatter about these relations . in particular , the relation between the ratio of far - infrared to ultraviolet luminosities and the ultraviolet spectral slope in starburst galaxies reflects the wavelength dependence of the effective absorption in the ambient ism . we find that a power law of the form @xmath271 accounts remarkably well for all the observations . the relation between @xmath137 and @xmath1 can then be interpreted as a sequence in the overall dust content of the galaxies . interestingly , this relation is accompanied by much weaker trends of the oxygen abundance and the optical luminosity with the ultraviolet spectral slope . the fact that our model reproduces the observed spectral properties of nearby starburst galaxies relatively easily leads us to suspect that , with suitable adjustment of the parameters , it could also reproduce those of more quiescent ( but still star - forming ) galaxies . the effective absorption curve required by the observations is much greyer than would be produced by a foreground screen of dust like that in the milky way , the lmc , or the smc . we have explored whether this could be accounted for by a much steeper wavelength dependence of the optical depth , i.e. @xmath75 with @xmath272 , combined with a more realistic spatial distribution of the dust . we find that a mixed slab model for the ambient ism can produce the required effective absorption curve for low dust content but can not explain the observations of starburst galaxies with very reddened ultraviolet spectra . in contrast , we show that a random distribution of discrete clouds provides a consistent interpretation of all the observed integrated spectral properties of starburst galaxies . while these results were anticipated in some previous studies , we have shown here for the first time how to reconcile them with the large h@xmath148h@xmath1 ratios and other observations . we also find that the optical depths of the clouds favored by our analysis are similar to those inferred from the statistics of stellar reddening in the milky way . the model we have developed for computing the absorption of starlight by dust in galaxies can be combined easily , by design , with any population synthesis model . the observed mean relations for starburst galaxies can also be reproduced by the following simple recipe : use an effective absorption curve proportional to @xmath3 to attenuate the line and continuum radiation from each stellar generation , and lower the normalization of the curve typically by a factor of 3 after @xmath4yr to account for the dispersal of the birth clouds . this recipe accounts at least as well as the one by calzetti et al . ( 1994 , and as modified by calzetti 1997 , 1999 ) for the effects of dust on the non - ionizing continuum radiation . in addition , it fully resolves the apparent discrepancy between the attenuation of line and continuum photons in starburst galaxies . we believe , therefore , that our model and the recipe derived from it provide simple yet versatile tools to interpret the integrated spectral properties of starburst and possibly other types of galaxies . in future work , we plan to apply them to the growing body of observations of high - redshift galaxies . we thank p. boiss , a. ferrara , and n. panagia for valuable discussions . s.c . appreciates the hospitality of the stsci , and s.m.f . that of the iap , during the course of several visits . this research was supported in part by the national science foundation through grant no . phy94 - 07194 to the institute for theoretical physics . bianchi , s. , ferrara , a. , & giovanardi , c. 1996 , apj , 465 , 127 blitz , l. , & shu , f. h. 1980 , apj , 238 , 148 bottorff , m. , lamothe , j. , momjian , e. , verner , e. , vinkovi , & ferland , g. 1998 , pasp , 110 , 1040 bruzual , a. g. , & charlot , s. 1993 , apj , 405 , 538 bruzual , a. g. , magris , g. , & calvet , n. 1988 , apj , 333 , 673 calzetti , d. 1997 , aj , 113 , 162 . 1999 , private communication calzetti , d. , bohlin , r. c. , gordon , k. d. , witt , a. n. , & bianchi , l. 1995 , apj , 446 , l97 calzetti , d. , kinney , a. l. , & storchi - bergmann , t. 1994 , apj , 429 , 582 . 1996 , apj , 458 , 132 caplan , j. , & deharveng , l. 1986 , a&a , 155 , 297 di bartolomeo , a. , barbaro , g. , & perinotto , m. 1995 , mnras , 277 , 1279 disney , m. , davies , j. , & phillipps , s. 1989 , mnras , 239 , 939 draine , b. t. , & lee , h. m. 1984 , apj , 285 , 89 fanelli , m. n. , oconnell , r. w. , & thuan , t. x. 1988 , apj , 334 , 665 ferland , g. j. 1996 , in hazy , a brief introduction to cloudy , univ . of kentucky dept . of physics & astronomy int . gordon , k. d. , calzetti , d. , & witt , a. n. 1997 , apj , 487 , 625 helou , g. , khan , i. r. , malek , l. , & boehmer , l. 1988 , apjs , 68 , 151 hummer , d. , & storey , p. 1992 , mnras , 254 , 277 humphreys , r. m. 1978 , apjs , 38 , 309 kim , s .- h . , martin , p. g. , & hendry , p. d. 1994 , apj , 422 , 164 kinney , a. l. , bohlin , r. c. , calzetti , d. , panagia , n. , & wyse , r. f. g. 1993 , apjs , 86 , 5 mcquade , k. , calzetti , d. , & kinney , a. l. 1995 , apjs , 97 , 331 meurer , g. r. , heckman , t. m. , & calzetti , d. 1999 , apj , 521 , 64 meurer , g. r. , heckman , t. m. , leitherer , c. , kinney , a. , robert , c. , & garnett , d. r. 1995 , aj , 110 , 2665 mezger , p. g. , & smith , l. f. 1977 , in iau symp . 75 , star formation , ed . t. de jong & a. maeder ( dordrecht : reidel ) , 133 natta , a. , & panagia , n. 1984 , apj , 287 , 228 oey , m. s. , & kennicutt , r. c. 1997 , mnras , 291 , 827 osterbrock , d. e. 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( mill valley : university science books ) pei , y. c. 1992 , apj , 395 , 130 petrosian , v. , silk , j. , & field , g. b. 1972 , apj , 117 , l69 puxley , p. j. , & brand , p. w. j. l. 1994 , mnras , 266 , 431 rybicki , g. b. , & lightman , a. p. 1979 , radiative processes in astrophysics ( new york : wiley ) , 36 silva , l. , granato , g. l. , bressan , a. , & danese , l. 1998 , apj , 509 , 103 spitzer , l. 1978 , physical processes in the interstellar medium ( new york : wiley ) , 154 storchi - bergmann , t. , kinney , a. l. , & challis , p. 1995 , apjs , 98 , 103 witt , a. n. , & gordon , k. d. 1996 , apj , 463 , 681 witt , a. n. , thronson , h. a. , & capuano , j. m. 1992 , apj , 393 , 611

we report the first planet discovery from the two wheeled _ kepler _ ( k2 ) mission : hip116454b . the host star hip116454 is a bright ( , ) k1dwarf with high proper motion , and a parallax based distance of pc . based on high - resolution optical spectroscopy , we find that the host star is metal - poor with [ fe / h ] , and has a radius r and mass m . the star was observed by the _ kepler _ spacecraft during its two - wheeled concept engineering test in february 2014 . during the 9 days of observations , k2 observed a single transit event . using a new k2 photometric analysis technique we are able to correct small telescope drifts and recover the observed transit at high confidence , corresponding to a planetary radius of r . radial velocity observations with the harps - n spectrograph reveal a 1.33 m planet in a 9.1 day orbit , consistent with the transit depth , duration , and ephemeris . follow up photometric measurements from the most satellite confirm the transit observed in the k2 photometry and provide a refined ephemeris , making hip116454b amenable for future follow up observations of this latest addition to the growing population of transiting super - earths around nearby , bright stars .

1412.5674

after four years of nearly continuous photometric monitoring and thousands of planet discoveries ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the primary _ kepler _ mission came to an end in in may 2013 with the failure of the second of four reaction wheels used to stabilize the spacecraft . without at least three functioning reaction wheels , the spacecraft is unable to achieve the fine pointing necessary for high photometric precision on the original target field . however , an extended mission called k2 was enabled by pointing along the ecliptic plane and balancing the spacecraft against solar radiation pressure to mitigate the instability caused by the failed reaction wheels . the recently extended k2 mission enables renewed opportunities for transit science on a new set of bright target stars , albeit with somewhat reduced photometric precision compared to the original _ kepler _ mission @xcite . searching for transiting exoplanets around bright , nearby stars is important because measuring the precise masses and radii of transiting planets allows for characterization of their interior structures and atmospheres @xcite . this is particularly desirable for planets with masses intermediate to those of the earth and uranus , commonly referred to as super earths , because no such planet exists in our solar system @xcite . however , while the radii of _ kepler _ planets are often measured to high precision @xcite , their masses are generally unknown because the host stars are faint ( @xmath10 ) and exposure times needed for radial velocity ( rv ) measurements are prohibitive for all but the brightest _ kepler _ planet candidates ( e.g. * ? ? ? * ; * ? ? ? preparations for the extended two wheeled _ kepler _ mission included a 9day test of the new observing mode in february of 2014 . after the data were released to the public , ( * ? ? ? * hereafter vj14 ) presented a photometric reduction technique that accounts for the motion of the spacecraft , improves photometric precision of raw k2 data by a factor of 25 , and enables photometric precision comparable to that of the original _ kepler _ mission . while the data collected during the engineering test were intended primarily as a test of the new spacecraft operating mode , inspection of light curves produced with this technique nonetheless revealed a single transit event in engineering data taken of hip116454 . in this paper , we provide an analysis of that light curve along with archival and follow up spectroscopy , archival and adaptive optics imaging , radial velocity measurements from the harps - n spectrograph , and photometric observations from the wide angle search for planets ( wasp ) survey and the microvariability and oscillations of stars ( most ) space telescope . these measurements allow us to verify and characterize the first planet discovered by the two - wheeled _ kepler _ mission , a new transiting super - earth orbiting the bright , nearby , high - proper - motion k dwarf hip116454 . hip116454 and nearly 2000 other stars were observed by the _ kepler _ spacecraft from 4 february 2014 until 12 february 2014 during the _ kepler _ two - wheel concept engineering test . after the first 2.5 days of the test , _ kepler _ underwent a large , intentional adjustment to its pointing to move its target stars to the center of their apertures , where they stayed for the last 6.5 days of the test . we downloaded the full engineering test dataset from the mikulski archive for space telescopes ( mast ) , and reduced the _ kepler _ target pixel files as described in vj14 . in brief , we extracted raw aperture photometry and image centroid positions from the _ kepler _ target pixel files . raw k2 photometry is dominated by jagged events corresponding to the motion of the spacecraft , as _ kepler _ s pointing drifts due to solar radiation pressure and is periodically corrected with thrusters . for the last 6.5 days after the large pointing tweak , we removed the systematics due to the motion of the spacecraft by correlating the measured flux with the image centroid positions measured from photometry . we essentially produced a `` self flat field '' ( sff ) similar to those produced by , for instance , @xcite , for analysis of _ spitzer _ photometry . we fit a piecewise linear function to the measured dependence of flux on centroid position , with outlier exclusion to preserve transit events , and removed the dependence on centroid position from the raw light curve . similarly to vj14 , we excluded datapoints taken while _ kepler _ s thrusters were firing from our reduced light curves because these data were typically outliers from the corrected light curves . for hip116454 , the median absolute deviation ( mad ) of the 30 minute long - cadence datapoints improved from @xmath11 parts per million ( ppm ) for the raw light curve , to @xmath12 50 ppm for the sff light curve . visual inspection of light curves from the @xmath12 2000 targets observed during the engineering test revealed a one millimagnitude ( mmag ) deep candidate transit in photometry of hip116454 , designated epic60021410 by the _ kepler _ team . raw and corrected k2 photometry for hip116454 are shown in figure [ sff ] . we fit a @xcite model to the transit and measured a total duration of approximately 2.25 hours , and a planet to star radius ratio of approximately 0.03 . unfortunately , the data point during transit ingress happened during a thruster firing event , and was excluded by our pipeline . this particular data point does not appear to be anomalous , but we choose to exclude it to minimize risk of contaminating the transit with an outlier . slow photometric variability , presumably due to starspot modulation , is evident in the k2 light curve at the sub - percent level . we also performed a similar sff correction to the data taken in the 2.5 days of data before the large pointing tweak . even though the resulting data quality is somewhat worse , we are able to confidently exclude any other events of a similar depth during that time . because k2 only observed one transit event , we were not able to measure a precise orbital period for the planet candidate . nonetheless , we were able to put rough constraints on the orbital period from our knowledge of the transit duration and estimates of the stellar properties . the nine - day time baseline of the k2 observations allowed us to constrain the period of the candidate transiting planet to be greater than 5 days . to put a rough upper bound on the allowed planet period , we compared the transit duration of the candidate transit around hip116454 to the distribution of transit durations from the ensemble of _ kepler _ planet candidates ( retrieved from the nasa exoplanet archive ; * ? ? ? * ) . we found that of the 413 _ kepler _ planet candidates with transit durations between 2 and 2.5 hours , 93% had orbital periods shorter than 20 days . because transit duration is a function of the mean stellar density , we repeated this calculation while restricting the sample to the 64 planet candidates with transit durations between 2 and 2.5 hours and host star effective temperatures within 200 k of hip116454 . we find that similarly , 94% of these candidates had orbital periods shorter than 20 days . lrcl @xmath13 ( j2000 ) & 233549.28 & ... & _ hipparcos _ + @xmath14 ( j2000 ) & + 002643.86 & ... & _ hipparcos _ + @xmath15 ( mas yr@xmath16 ) & -238.0 & 1.7 & _ hipparcos _ + @xmath17 ( mas yr@xmath16 ) & -185.9 & 0.9 & _ hipparcos _ + @xmath18 ( mas ) & 18.1 & 1.72 & _ hipparcos _ + @xmath19 & 11.08 & 0.01 & _ tycho _ + @xmath20 & 10.190 & 0.009 & _ tycho _ + @xmath21 & 9.71 & 0.03 & tass + @xmath22 & 9.25 & 0.03 & tass + @xmath23 & 14.786 & 0.02 & sdss + @xmath24 & 10.837 & 0.02 & sdss + @xmath25 & 9.908 & 0.02 & sdss + @xmath26 & 9.680 & 0.02 & sdss + @xmath27 & 8.60 & 0.02 & 2mass + @xmath28 & 8.14 & 0.03 & 2mass + @xmath29 & 8.03 & 0.02 & 2mass we used a combination of modern and archival imaging to limit potential false positive scenarios for the transit event . hip116454 was observed in the national geographic society palomar observatory sky survey ( poss - i ) on 28 november 1951 . hip116454 has a proper motion of 303 mas yr@xmath16 @xcite , and therefore has moved nearly 20 arcseconds with respect to background sources since being imaged in poss - i . inspection of the poss - i image reveals no background objects within the k2 aperture used in our photometric reduction . we show the poss - i blue image overlaid with the k2 photometric aperture in figure [ imaging]a . the poss - i survey has a limiting magnitude of 21.1 in the blue bandpass @xcite , 10 magnitudes fainter than hip116454 . the depth of the detected transit is 0.1% , so if a background eclipsing binary were responsible the depth would correspond to a total eclipse of a star 7.5 magnitudes fainter . a low proper motion background star such as that would have readily been detected in poss - i imaging . we conclude that our aperture is free of background objects whose eclipses could masquerade as planet transits . the poss - i imaging also reveals a companion star about 8 arcseconds to the southwest of hip116454 . the companion is not fully resolved in poss - i because the photographic plate was saturated by the bright primary star , but an asymmetry in the stellar image is visible . hip116454 was also observed during the second palomar observatory sky survey ( poss - ii ) on 31 august 1992 . improvements in photographic plate technology over the previous forty years allowed the companion star to be resolved . the companion shares a common proper motion with the primary at a projected distance of @xmath11 au , so we conclude the two stars are a gravitationally bound visual binary system . hip116454 was observed during the sloan digital sky survey ( sdss ) , and the secondary star was detected @xcite . the secondary star falls on a diffraction spike caused by the much brighter primary star , but the sdss pipeline flagged its photometry as `` acceptable . '' the sdss photometry indicates that the secondary star is 67 magnitudes fainter than the primary depending on the filter , so because the two stars are gravitationally associated , the secondary must be intrinsically much fainter than the k dwarf primary . this implies that the companion must either be a late m dwarf or a white dwarf . the sdss colors are relatively flat , indicating a hot star . to quantify this , we fit the _ ugri _ sdss colors to a blackbody model , excluding _ z _ due to its low throughput and assuming photometric errors of 5% . we included no corrections for extinction due to the proximity of the target and our ability to accurately predict broadband photometry using stellar models in section [ stellarproperties ] . we find that the data are best described by an object radiating at a temperature of @xmath30 k. we used the stefan boltzmann law combined with the _ hipparcos _ parallax and derived temperature to estimate a radius of @xmath31 , which is consistent with our white dwarf hypothesis . using a simple analytic white dwarf cooling law @xcite , we estimate a cooling age of the white dwarf of @xmath32 gyr . the formal uncertainty on the cooling age is 0.2 gyr , but this neglects uncertainties due to the unknown composition of the white dwarf and inaccuracies in the simple model . the true uncertainty on this quantity is likely on order of a factor of two @xcite . the cooling age of the white dwarf is a lower limit on the age of the system , and the total age of the system is the sum of the main sequence lifetime of the progenitor and the white dwarf s cooling age . the secondary star is close enough to the primary that it is blended in the k2 image and is bright enough that if it were a totally eclipsing binary , it could cause the transit event we observed . this situation is unlikely because the duration and minimum period of the event are generally inconsistent with an object eclipsing a white dwarf . with the baseline of k2 data we can exclude orbital periods shorter than 5 days . while 5day period companions eclipsing main sequence stars are common , and have relatively high transit probabilities , the probability of a transit or eclipse goes as @xmath33 at a given stellar mass and orbital period . furthermore , in order for an earth sized object eclipsing a white dwarf to have an eclipse duration of two hours , the orbital period would have to be roughly 600 years in the case of a circular orbit and impact parameter @xmath34 . even with a highly elliptical orbit transiting at apastron , which is _ a priori _ unlikely , the orbital period would be of order centuries , and the semimajor axis would be roughly 50 au . the probability of an orbit such as that eclipsing the white dwarf is @xmath35 , where @xmath36 is the semimajor axis and @xmath37 is the radius of the occulting body . in the worst case scenario of a non - luminous jupiter sized object occulting the white dwarf , the orbital period would have to be on order 3 years and have a semimajor axis of roughly 1.5 au , correpsonding to a transit probability of @xmath38 . we conclude that the transit event we observed was far more likely caused by a short period planet orbiting the primary star than a long period object eclipsing the secondary . we also obtained high angular resolution imaging of the primary star to rule out any very close associated companions . we observed hip116454 with the robo - ao laser adaptive optics and imaging system on the 60inch telescope at palomar observatory @xcite . we obtained seven images with robo - ao between 15 june 2014 and 11 july 2014 in three different bandpasses : sloan @xmath26band , sloan @xmath39band , and a long pass filter with a cutoff at 600 nm ( lp600 ) which more closely approximates the _ kepler _ bandpass . each observation consisted of a series of images read out from the detector at a maximum rate of 8.6 hz , for a total integration time of 90 seconds . the frames were co added in post - processing using a shift and add technique with hip116454 as the tip tilt guide star . the quality of the robo - ao images varied between the observations , but none of the images showed evidence for companions within 3 magnitudes of the primary outside of 0.2 arcseconds . some but not all of the images , however , showed an elongation that could be consistent with a bright close binary companion at a separation of 0.15 arcseconds at the < 5@xmath40 level , similar to koi 1962 in @xcite . to investigate this possibility further , we obtained higher resolution adaptive optics images on 2 august 2014 using the keck ii natural guide star adaptive optics ( ngsao ) system with the nirc2 narrow detector at keck observatory . we obtained unsaturated frames of hip116454 in @xmath27 , @xmath28 , and @xmath29-band filters to search for close companions near the diffraction limit ( @xmath4140 mas in @xmath28-band ) . we also acquired deeper , saturated images in @xmath28 ( 70 sec total ) and @xmath29-bands ( 200 sec total ) with the primary positioned in the lower right quadrant of the array and rotated so the white dwarf companion falls in the field of view . we calibrated and processed the data as described in @xcite . we corrected the data for optical aberrations using the distortion solution from b. cameron ( 2007 , priv . communication ) and north - aligned the images using the detector orientation measured by @xcite . we found no evidence for the companion suggested by some of the robo - ao data . our 7-@xmath40 @xmath28-band limiting contrasts are \{3.0 , 5.9 , 6.8 , 9.2 , 10.8 , 12.7 } mag at separations of \{0@xmath421 , 0@xmath423 , 0@xmath425 , 1@xmath420 , 2@xmath420 , 5@xmath420}. we are able to exclude roughly equal brightness companions to an angular separation of 0.04 arcseconds ( projected distance of 2.2 au ) . hip116454 was observed 9 times for the carney - latham proper motion survey @xcite with the cfa digital speedometer spectrograph over the course of 9.1 years from 1982 until 1991 . the digital speedometer measured radial velocities to a precision of approximately 0.3 km s@xmath16 , and detected no significant radial velocity variations or trends in the velocities of hip116454 . when corrected into an absolute radial velocity frame , the digital speedometer measurements indicate an absolute radial velocity of @xmath43 km s@xmath16 and when combined with proper motion , a space velocity of @xmath44 km s@xmath16 . this somewhat unusual space velocity corresponds to an elliptical orbit in the plane of the galaxy , indicating that hip116454 originated far from the stellar neighborhood . a detailed analysis of hip116454 s elemental abundances could reveal patterns that are dissimilar to stars in the solar neighborhood . we obtained three observations of hip116454 in june of 2014 with the tillinghast reflector echelle spectrograph ( tres ) on the 1.5 m tillinghast reflector at the fred l. whipple observatory . the spectra were taken with a resolving power of @xmath45 with a signal to noise ratio ( snr ) of approximately 50 per resolution element . when corrected into an absolute radial velocity frame , the tres spectra indicate an absolute radial velocity for hip116454 of @xmath46 km s@xmath16 . when combined with the absolute radial velocities from the digital speedometer , there is no evidence for radial velocity variation of greater than 100 m s@xmath16 over the course of 30 years . the three individual radial velocities from the tres spectra revealed no variability at the level of 20 m s@xmath16 over the course of eight days . we also find no evidence for a second set of stellar lines in the cross correlation function used to measure the radial velocities , which rules out many possible close companions or background stars . when the adaptive optics constraints are combined with a lack of radial velocity variability of more than 100 m / s over 30 years and the lack of a second set of spectral lines in the cross correlation function , we can effectively exclude any close stellar companions to hip116454 . we obtained 44 spectra of hip116454 on 33 different nights between july and october of 2014 with the harps - n spectrograph @xcite on the 3.57 m telescopio nazionale galileo ( tng ) on la palma island , spain to measure precise radial velocities and determine the orbit and mass of the transiting planet . each harps - n spectrum was taken with a resolving power of @xmath47 , and each measurement consisted of a 15 minute exposure , yielding a snr of 50100 per resolution element at 550 nm , depending on weather conditions . the corresponding ( formal ) radial velocity precision ranged from 0.90 m s@xmath16 to 2.35 m s@xmath16 . radial velocities were extracted by calculating the weighted cross correlation function of the spectrum with a binary mask @xcite . in some cases , we took one 15 minute exposure per night , and in other cases , we took two 15 minute exposures back - to - back . in the latter case we measured the two consecutive radial velocities individually and report the average value . the harps - n radial velocity measurements are listed in table [ rvs ] , and are shown in figure [ rvplot ] . a periodic radial velocity variation with a period of about nine days and a semiamplitude of about 4 m s@xmath16 is evident in the rv time series . we checked that we identified the correct periodicity by calculating a lomb - scargle periodogram @xcite , shown in figure [ periodogram ] . we found a strong peak at a period of 9.1 days and an aliased peak of similar strength with a period close to 1 day , corresponding to the daily sampling alias of the 9.1 day signal ( e.g. * ? ? ? we estimated the false alarm probability of the rv detection by scrambling the rv data and recalculating the periodogram numerous times , and counting which fraction of the scrambled periodograms have periods with higher power than the unscrambled periodogram . we found that the false alarm probability of the 9.1 day periodicity is significantly less than @xmath48 . in addition to the 9.1 day signal , we also found evidence for a weaker 45 day periodic rv variation.to help decide whether to include the second periodicity in our rv modeling , we fit the harps - n radial velocities with both a one - planet and a two - planet keplerian model . the one planet model was a keplerian function parameterized by : @xmath49 , time of transit , @xmath50,@xmath51 , @xmath52 , where @xmath53 is the planet s orbital period , @xmath54 is the orbital eccentricity , and @xmath55 is the argument of periastron . we also fit for a radial velocity zero - point and a stellar jitter term , for a total of 7 free parameters . we fit each of these parameters with an unbounded uniform prior except for @xmath51 and @xmath52 , which had uniform priors over the interval between -1 and 1 . the two planet model was the sum of two keplerian functions , each of which was parameterized by : @xmath49 , time of transit , @xmath56,@xmath51 , and @xmath52 . once again , we also fit for a radial velocity zero point and stellar jitter term , for a total of 12 free parameters . we fit each of these parameters with an unbounded uniform prior except for @xmath51 and @xmath52 , which had uniform priors over the interval between -1 and 1 , and for @xmath57 , the period of the outer planet , which we constrained to be between the period of the inner planet and 1000 days . we performed the fits using _ emcee _ @xcite , a markov chain monte carlo ( mcmc ) algorithm with an affine invariant ensemble sampler . we note that upon exploring various different periods for the outer planet , our mcmc analysis found the 45 day period to be optimal . we calculated the bayesian information criterion ( bic , * ? ? ? * ) to estimate the relative likelihoods of the two models . while the bic does not provide a definitive or exact comparison of the fully marginalized likelihoods of the models , it allows us to roughly estimate the relative likelihoods . upon calculating the bic , we estimate that the two - planet model is favored over the one - planet model with confidence @xmath58 . from here on , we therefore model the radial velocity observations as the sum of two keplerian functions . for both periods , we find an amplitude consistent with that of a transiting super - earth . the nine - day periodicity in the rvs is consistent with the orbital period we estimated from the duration of the k2 transit event . we `` predicted '' the time of transit for the nine day period planet during the k2 observations , and found that the harps - n measurements alone constrain the expected time of transit to better than one day , and we find that the k2 transit event is consistent with the transit ephemeris predicted by only the harps - n rvs at the 68.3% ( 1-@xmath40 ) level . we show the k2 light curve with the transit window derived from only the harps - n rvs in figure [ transitwindow ] . thus , the 9.1 day periodicity is consistent with being caused by a transiting planet . the more tenuous 45 day periodic variation , on the other hand , may be due to to an outer planet , but it may also be caused by stellar variability . the harps - n spectra include regions used to calculate activity indicators such as the mount wilson s@xmath59 index and the r@xmath60 index ( e.g. * ? ? ? we calculated the s@xmath59 index for each harps - n spectrum and found a mean value of 0.275 @xmath61 0.0034 , and an associated @xmath62 . there were no obvious correlations between the s@xmath59 index and either the measured rv or the residuals to either a one or two planet keplerian fit . lrc 2012.7150 & -6.51 & 1.13 + 2013.7062 & -3.15 & 1.37 + 2014.7001 & -4.44 & 2.35 + 2015.6955 & -2.62 & 1.12 + 2016.6307 & 1.01 & 1.88 + 2017.7029 & 4.31 & 1.24 + 2018.6971 & 1.03 & 1.10 + 2019.6985 & 0.33 & 1.83 + 2020.7000 & -3.17 & 1.31 + 2025.6645 & -0.57 & 1.22 + 2026.6780 & -0.05 & 1.39 + 2027.6258 & 2.33 & 1.00 + 2028.6266 & -0.57 & 1.08 + 2030.7261 & -0.48 & 1.19 + 2031.7186 & -3.69 & 0.90 + 2032.7231 & -2.82 & 1.05 + 2033.7197 & -1.53 & 1.87 + 2041.6353 & -1.32 & 1.28 + 2043.6442 & -1.72 & 2.30 + 2044.6658 & 7.28 & 1.32 + 2045.7016 & 5.02 & 1.21 + 2050.7205 & -5.16 & 1.41 + 2051.7258 & -3.64 & 0.97 + 2052.7229 & -1.40 & 1.11 + 2053.7166 & 2.53 & 1.60 + 2054.7304 & 4.29 & 0.92 + 2055.6129 & 1.02 & 1.34 + 2056.5885 & 0.20 & 1.43 + 2057.6126 & -3.09 & 0.87 + 2086.5397 & -1.97 & 1.07 + 2090.5445 & 6.01 & 0.76 + 2091.5401 & 4.58 & 0.97 + 2098.5448 & 2.10 & 1.48 + hip116454 was observed by the superwasp - n instrument on la palma , spain and the superwasp - s instrument at the south african astronomical observatory @xcite . the observations spanned three observing seasons , from 2008 to 2010 , and consisted of roughly 15,000 data points with a typical precision of 0.6% . the wasp observations are consistent with a typical level of stellar variability at the sub - percent level . the wasp data rule out deep transits but are not of high enough quality to detect the 0.1% transit depth observed by k2 . in section [ prot ] , we use the wasp light curve in combination with light curves from k2 and most to attempt to derive hip116454 s rotation period . after detecting the k2 transit , we obtained follow up photometric observations with the microvariability and oscillations of stars ( most , * ? ? ? * ) space telescope , during august and september of 2014 . most observed hip116454 during three nearly continuous time spans : 13 days from 3 august 2014 to 16 august 2014 , 18 days from 21 august 2014 to 9 september 2014 , and 3.5 days from 15 september 2014 to 18 september 2014 . during the first segment of the most data , observations of hip116454 were interleaved with observations of other stars during the satellite s orbit around earth , but for the second and third segments most observed hip116454 continuously . during the first and third segments , the exposure time for individual datapoints was 1 second , and during the second segment , the exposure time was 2 seconds . we processed the most data using aperture photometry techniques as described in @xcite . background scattered light ( modulated by the 101 minute orbital period of most ) and interpixel sensitivity variations were removed by subtracting polynomial fits to the correlations between the stellar flux , the estimated background flux , and the centroid coordinates of the star . at each stage , outlying data points were excluded by either sigma clipping or hard cuts . the resulting precision of the most light curve was approximately 0.2% per 2second exposure . when we search the most light curve at the predicted times of transits from a simultaneous analysis of the k2 and harps - n data , we detect a weak signal with the same depth , duration , and ephemeris as the k2 transit . the most data refine our estimate of the transiting planet period to a precision of roughly 30 seconds . we take this detection as confirmation that the 9.1 day period detected in radial velocities is in fact caused by the transiting planet . from here on , we refer to the 9.1 day period planet as hip116454b . we measured the spectroscopic properties of the host star using the stellar parameter classification ( spc ) method @xcite on the spectra from tres and harps - n . analysis of spectra from both instruments showed consistent results for the stellar parameters . we adopt the results from the harps - n spectra , due to their higher spectral resolution and snr . the spc analysis yields an effective temperature of @xmath63 k , a metallicity [ m / h ] @xmath64 , and a surface gravity @xmath65 . we did not detect significant rotational broadening even with the high resolution harps - n spectra . the upper limit on the projected rotational velocity is roughly @xmath66 km s@xmath16 . we used several different approaches to estimate the stellar mass and radius of hip116454 . first , we used the spc parameters , in particular the metallicity , surface gravity , and effective temperature , to interpolate onto the yonsei yale stellar evolution model grids @xcite using a monte carlo approach . the resulting stellar parameters were @xmath67 , and @xmath68 . hip116454 was observed by _ hipparcos _ and has a measured parallax , allowing us to interpolate model grids using a separate luminosity indicator . we used the online padova model interpolator , which uses the technique of @xcite to interpolate a measured effective temperature , metallicity , @xmath20band magnitude , and parallax onto parsec isochrones , yielding estimates of stellar mass , radius , surface gravity , and @xmath69 color . when the spc parameters , _ hipparcos _ parallax , and the v magnitude from the _ tycho _ catalog @xcite are provided as input , the models output @xmath70 , and @xmath71 , along with @xmath72 dex and @xmath73 mag . the model predicted surface gravity is consistent with the spectroscopically measured surface gravity and more precise due to the additional constraint from parallax . the model output @xmath69 color is discrepant with the measured _ tycho _ @xmath69 at the 1.5@xmath40 level , but is still within 0.04 magnitudes of the measured _ tycho _ @xmath74 . this discrepancy is small enough ( 3% ) that it could be due to differences in the filters used by _ tycho _ and the filter transmission assumed by the padova models . we adopt the outputs from the padova model interpolator as our stellar parameters due to their more precise constraints and ability to predict @xmath75 and @xmath69 . we attempted to measure the rotation period of hip116454 using photometric measurements from wasp , most , and k2 to see if stellar activity might contribute to the possible 45 day periodicity in the rv measurements . a constraint or measurement of the rotation period consistent with the possible 45 day periodicity in radial velocities could affect our interpretation of the signal . we only used photometric measurements for our analysis given the relatively short time coverage and sparseness of the spectroscopic observations . we first started by analyzing the wasp data only , because its time baseline far exceeded that of the k2 and most data . we binned the wasp data into nightly datapoints , calculated a lomb - scargle periodogram and fourier transformed the resulting power spectrum to obtain an autocorrelation function . we performed this analysis on each season of wasp data individually . in the first season ( 2008 ) of wasp data , we found a moderately strong peak in both the autocorrelation function and the lomb - scargle periodogram at a period of about 16 days . we evaluated the significance of this peak by scrambling the binned data , recalculating the lomb - scargle periodograms , and counting the number of times the maximum resulting power was greater than the power in the 16 day peak . we found a false alarm probability of 2% for the peak in the first season . we did not find any convincing signals in the second ( 2009 ) or third ( 2010 ) observing seasons . a possible explanation for the inconsistency between observing seasons is that hip116454 experienced different levels of starspot activity , but it is also possible that the 16 day period detected in the first season is spurious . we concluded that the wasp data showed a candidate rotation period at 16 days , but the relatively high false alarm probability and lack of consistency between observing seasons precluded a confident detection . after our analysis of the wasp data yielded suggestive yet ambiguous results , we attempted to measure the rotation period of hip116454 by fitting all of the photometric data with a gaussian process noise model . stochastic yet time correlated processes such as the variability produced by rotation in stellar light curves can be modeled as a gaussian process by parametrizing the covariance matrix describing the time correlation between data points with a kernel function and inferring kernel hyperparameters from the data . we use a quasi periodic kernel function for the specific problem of measuring a rotation period from a light curve . this in principle is better suited to inferring the rotation period of a star than a periodogram analysis because the variability produced by active surface regions on a rotating star is typically neither sinusoidal nor strictly periodic . the gaussian process analysis also allows us to simultaneously model multiple datasets , and to take advantage of datasets ( like k2 and most ) with relatively short time coverage . we conducted our analysis using _ @xcite , a gaussian process library that employs hierarchical off - diagonal low rank systems ( hodlr ) , a fast matrix inversion method @xcite . we used the following kernel function in our analysis : @xmath76\exp\left[\frac{-\sin^2\left(\frac{\pi(x_i - x_j)}{p}\right)}{g_q^2}\right],\ ] ] where @xmath77 is the amplitude of correlation , @xmath78 is the timescale of exponential decay , @xmath79 is a scaling factor for the exponentiated sinusoidal term and @xmath53 is the rotation period . an additional hyperparameter , @xmath80 was used to account for additional white noise in the time series , where @xmath80 is added to the individual measurement uncertainties in quadrature . we modeled the three continuous periods of most data , the three seasons of wasp data , and the k2 photometry simultaneously , with @xmath77 , @xmath78 , @xmath79 and @xmath53 constrained to be the same across all seven data sets and with @xmath80 allowed to take a different value for each . emcee _ to explore the posterior distributions of the hyperparameters . the resulting posterior distribution was not well constrained , with significant power at essentially all rotation periods greater than about 8 days and less than about 50 days . there were a few periods which seemed to be preferred to some extent in the posterior distribution a strong peak at 12 days , and weaker peaks at 16 , 20 , and 32 days . we conclude that with our data , we can not conclusively identify a rotation period for hip116454 , which leaves us unable to rule out stellar activity as the cause of the 45 day signal in the radial velocities . while we do not find any strong evidence in the photometry that the rotation period is close to 45 days , we can not conclusively rule out a 45 day rotation period . more photometric ( or spectroscopic ) observations will be important to determining hip116454 s rotation period . lcccc _ orbital parameters _ & & & + orbital period , @xmath53 [ days ] & 9.1205 & @xmath61&@xmath81 & a + radius ratio , @xmath82 & 0.0311 & @xmath61&@xmath83 & a + transit depth , @xmath84 & 0.000967 & @xmath61&@xmath85 & a + scaled semimajor axis , @xmath86 & 27.22 & @xmath61&@xmath87 & a + orbital inclination , @xmath26 [ deg ] & 88.43 & @xmath61&@xmath88 & a + transit impact parameter , @xmath89 & 0.65 & @xmath61&@xmath90 & a + eccentricity & 0.205 & @xmath61 & @xmath91 & a + argument of periastron @xmath55 [ degrees ] & -59.1 & @xmath61&@xmath92 & a + velocity semiamplitude @xmath93 [ m s@xmath16 ] & 4.41&@xmath95&@xmath96 & a + time of transit @xmath97 [ bjd ] & 2456907.89 & @xmath61 & 0.03 & a + & & + _ stellar parameters _ & & & + @xmath98 [ @xmath99 & 0.775 & @xmath61&@xmath100 & b , d + @xmath101 [ @xmath102 & 0.716 & @xmath61&@xmath103 & b , d + @xmath104 [ @xmath105 & 2.11 & @xmath61&@xmath106 & b , d + @xmath107 [ cgs ] & 4.590 & @xmath61&@xmath108 & b + [ m / h ] & @xmath109 & @xmath61&@xmath110 & b + distance [ pc ] & 55.2 & @xmath61&@xmath111 & d + @xmath112 [ k ] & 5089 & @xmath61&@xmath113 & b + & & + _ planet parameters _ & & & + @xmath114 [ m@xmath8 ] & 11.82 & @xmath61&@xmath116 & b , c , d + @xmath117 [ r@xmath8 ] & 2.53 & @xmath61&@xmath118 & b , c , d + mean planet density , @xmath119 [ g @xmath120 & 4.17 & @xmath61&@xmath121 & b , c , d + @xmath122 [ cgs ] & 3.26 & @xmath61&@xmath110 & b , c , d + equilibrium temperature @xmath123 [ k ] & 690 & @xmath61&@xmath124 & b , c , d , e we conducted an analysis of the k2 light curve , the harps - n radial velocity observations , and the most light curve and the wasp light curve to determine orbital and planetary properties . we first re - processed the k2 data using a different method from that described in vj14 to minimize the possibility of any bias due to using the in transit points in the flat field . we re - derived the sff correction by fitting the k2 light curve using an mcmc algorithm with an affine invariant ensemble sampler ( adapted for idl from the algorithm of * ? ? ? * ; * ? ? ? we fit the transit light curve with a @xcite model , as implemented by @xcite , with quadratic limb darkening coefficients held at the values given by @xcite . we modeled the stellar out of transit variations with a cubic spline between 10 nodes spaced evenly in time , the heights of which were free parameters . similarly , we modeled the sff correction as a cubic spline with 15 nodes spaced evenly in `` arclength , '' a one dimensional metric of position on the detector as defined in vj14 . upon finding the best - fit parameters for the sff correction , we applied the correction to the raw k2 data to obtain a debiased light curve . after rederiving the correction to the k2 light curve , we simultaneously fit a transit light curve to the k2 light curve , the harps - n radial velocities , and the most and wasp photometry using _ emcee_. we modeled the radial velocity variations with a two planet keplerian model ( fitting the 9.1 day period and the 45 day period simultaneously ) , and modeled the transits of the 9.1 day planet with a @xcite model . for the k2 light curve , we accounted for the 29.4 minute long cadence exposure time by oversampling the model light curve by a factor of 13 and binning . we allowed limb darkening coefficients ( parametrized as suggested by * ? ? ? * ) to float . we used a white noise model for the radial velocity observations with a stellar jitter term added in quadrature to the harps - n formal measurement uncertainties . for the light curves , we used a gaussian process noise model , using the same kernel described in equation [ kernel ] . we used an informative prior on the stellar @xmath75 from section [ stellarproperties ] in our fits , which we converted to stellar density to help break the degeneracy between the scaled semimajor axis and impact parameter . using this prior let us constrain the impact parameter despite having only one k2 long cadence datapoint during transit ingress and egress . in total , the model had 28 free parameters . we used _ emcee _ to sample the likelihood space with 500 `` walkers '' , each of which we evolved through 1500 steps . we recorded the last 300 of of these steps as samples of our posterior distribution , giving a total of 150000 mcmc links . we calculated correlation lengths for all 28 parameters which ranged from 5.6 to 19.0 , corresponding to between 8000 and 27000 independent samples per parameter . we assessed the convergence of the mcmc chains using the spectral density diagnostic test of @xcite and found that the means of the two sequences are consistent for 21/28 parameters ( 75% ) at the 1@xmath40 level , for 27/28 ( 96% ) at the 2@xmath40 level , and 28/28 at the 3@xmath40 level . these fractions are consistent with draws from a normal distribution , which is the expected behavior for the mcmc chains having converged . in table [ tab : bigtable ] , we report the best fitting planet and orbit parameters and their uncertainties for hip116454b by calculating the median link for each parameter and 68% confidence intervals of all links for each parameter , respectively . we summarize the priors used in the fits and the full model outputs in table [ tab : combinedtable ] , including nuisance parameters like the noise model outputs . we find that our data are best described by the presence of a planet with @xmath125 = 2.53 @xmath61 0.18 m@xmath8 , @xmath126 = 11.82 @xmath61 1.33 m@xmath8 , in a 9.1205 day orbit . while we find some evidence for an outer planet in the radial velocity measurements , we can not conclusively claim its existence based on the data presently at our disposal . figure [ massradius ] shows hip116454b on a mass - radius diagram with other known transiting , sub size exoplanets with measured masses and radii . we overlaid the plot with model composition contours from @xcite . we first note that hip116454b has a mass and radius consistent with either a low density solid planet , or a planet with a dense core and an extended gaseous envelope . the relatively low equilibrium temperature of the planet ( @xmath127 k , assuming zero albedo and perfect heat redistribution ) makes it unlikely that any gaseous envelope the planet started with would have evaporated over its lifetime . in terms of mass and radius , this planet is similar to kepler-68b ( 8.3 m@xmath8 , 2.31 r@xmath8 , * ? ? ? * ) and hd97658b ( 7.9 m@xmath8 , 2.3 r@xmath8 * ? ? ? * ; * ? ? ? * ) , but is likely slightly larger . hip116454b and hd97658b ( @xmath128 k ) have similar effective temperatures while , kepler-68b ( @xmath129 k under the same assumptions ) is somewhat hotter . like these planets , hip116454b has a density intermediate to that of rocky planets and that of ice giant planets . on the mass - radius diagram , hip116454b lies close to the 75% h@xmath130o-25% mgsio@xmath131 curve for solid planets . it could be either a low density solid planet with a large fraction of h@xmath130o or other volatiles ( which have similar equations of state to h@xmath130o ) , or it could have a dense core with a thick gaseous layer . we made inferences about the structure and composition of hip116454b if it indeed has a dense core and thick gaseous envelope using analytic power law fits to the results of @xcite . assuming the thick gaseous envelope is composed of hydrogen and helium in solar abundances , an equilibrium temperature calculated with perfect heat redistribution and an albedo of 0 , and a stellar age of about 2 gyr , the models predict that hip116454b has a hydrogen / helium envelope making up about 0.5% of the planet s mass . the model suggests hip116454b to have a 1.8 r@xmath8 core with virtually all of the planet s mass , surrounded by a gaseous envelope with thickness 0.35 r@xmath8 , and a radiative upper atmosphere also with thickness 0.35 r@xmath8 . using different assumptions to calculate the equilibrium temperature like imperfect heat distribution and a nonzero albedo ( for instance , the value of * ? ? ? * ) , and different assumptions about the envelope s composition and age does not change the calculated thickness and mass of the gaseous envelope by more than a factor of two . we note that this envelope fraction is consistent with the population of _ kepler _ super - earth / sub - neptune sized planets studied by @xcite , who found the envelope fraction of these candidates to be distributed around 1% with a scatter of 0.5 dex . we also explored the composition of hip116454b assuming it is solid and has little in the way of a gaseous envelope using an online tool , based on the model grids of @xcite . we investigated a three component model , with layers of h@xmath130o , mgsio@xmath131 and fe . in this case , hip116454b must have a significant fraction of either h@xmath130o or other volatiles in an ice form like methane or ammonia . the composition in this case would be more similar to the ice giants in the solar system than the rocky planets like earth . we find that the pressure at the core of the planet can range from 1400 gpa for an iron free planet to 2800 gpa for a silicate free planet . assuming a ratio of iron to silicates similar to that of the earth and other solar system bodies , we find that the core pressure of hip116454b is about 2400 gpa . under this assumption , hip116454b would consist of 8% fe,17% mgsio@xmath131 and 75% h@xmath130o by mass . using the ratio of iron to silicates in solar system bodies is usually a relatively good assumption because this ratio is largely determined by element synthesis cosmochemistry , which does not vary greatly on the scale of 50 parsecs . however , hip116454 s unusual space motion indicates that it might have formed elsewhere in the galaxy , so this assumption might not hold . more detailed spectral analysis , in particular measuring elemental abundances for mg and si compared to fe in the parent star could put additional constraints on the composition of hip116454b assuming it is solid . if solid , hip116454b would be one of the `` h@xmath130o rich '' planets described in @xcite , for which it is possible to make inferences about the phase of the planet s h@xmath130o layer , given knowledge of the star s age . various evidence points to hip116454 having an age of approximately 2 gyr . using relations from @xcite , the @xmath132 level indicates an age of 2.7 gyr and the rotation indicates an age of 1.1 gyr , if the rotation period is indeed close to 16 days . the white dwarf s cooling age , however , sets a lower limit of approximately 1.3 gyr . future observations , like a mass measurement of the white dwarf to estimate its progenitor s mass ( and therefore age on the main sequence ) , could constrain the age further . if hip116454 s age is indeed about 2 gyr and the planet lacks a gaseous envelope , then it is likely to have water in plasma phases near its water - silicate boundary ( the bottom of the h@xmath130o layer ) , while if it is slightly older ( @xmath41 3 gyr or more ) , or has a faster cooling rate , it could have superionic phases of water . hip116454b is a promising transiting super - earth for follow up observations due to the brightness of its star , especially in the near infrared . we used the exoplanet orbit database @xcite to compare hip116454b to other transiting sub - neptune sized planets orbiting bright stars . we found that among stars hosting transiting sub - neptunes with @xmath133 , only kepler 37 , 55 cnc , and hd 97658 have brighter @xmath134band magnitudes . hip116454 is particularly well suited for additional follow up photometric and radial velocity observations both to measure the mass of the planet to higher precision and to search for more planets in the system . hip116454 is chromospherically inactive and has low levels of stellar radial velocity jitter ( 0.45 @xmath61 0.29 m s@xmath16 ) . this combined with its brightness makes it an efficient radial velocity target . moreover , the brightness of the host star makes hip116454 ideal for follow up with the upcoming cheops mission @xcite . hip116454b could be important in the era of the james webb space telescope ( jwst ) to probe the transition between ice giants and rocky planets . in the solar system , there are no planets with radii between 13 r@xmath8 while population studies with _ kepler _ data have shown these planets to be nearly ubiquitous @xcite . atmospheric studies with transit transmission spectroscopy can help determine whether these planets are in fact solid or have a gaseous envelope , and give a better understanding on how these planets form and why they are so common in the galaxy . also of interest is the fact that hip116454b is very similar to hd97658b , in terms of its orbital characteristics ( both are in @xmath41 10 day low - eccentricity orbits ) , mass and radius ( within 10% in radius , and within 25% in mass ) , and stellar hosts ( both orbit k dwarfs ) . comparative studies of these two super earths will be valuable for understanding the diversity and possible origins of close in super earths around sun like stars . this being said , despite hip116454 s brightness , the relatively shallow transit depth will make it a somewhat less efficient target than super - earths orbiting smaller stars ( for instance , gj 1214 b , * ? ? ? * ) . hip116454b has demonstrated the potential of k2 to increase the number of bright transiting planets amenable to radial velocity follow up . despite its degraded pointing precision , it is possible to calibrate and correct k2 data to the point where super - earths can be detected to high significance with only one transit . despite the increased expense of bright stars in terms of kepler target pixels required for the aperture , k2 data is of high enough precision to produce many transiting exoplanets around bright stars . many k2 fields , including the engineering test field , are located such that observatories in both hemispheres can view the stars , a significant difference between k2 and the original _ kepler _ mission . even though all of our follow up observations for hip116454b took place at northern observatories , the star s equatorial location enables follow up from southern facilities like the original harps instrument at la silla observatory @xcite and the planet finding spectrograph at las campanas observatory @xcite just as easily as with northern facilities with instruments like harps - n or the high resolution echelle spectrograph at keck observatory @xcite . many _ kepler _ planet candidates were confirmed in part thanks to precise measurements of the _ kepler _ image centroid as the planet transited the expected motion of the image centroid could be calculated based on the brightness and position of other stars near the aperture , and deviations from that prediction could signal the presence of a false positive planet candidate . such an analysis will be substantially more difficult for k2 data , because the unstable pointing leads to large movements of the image centroid . in this work , we were able to exclude the possibility of background objects creating false transit signals taking advantage of the star s high proper motion and archival imaging . this will be more difficult for more distant stars . however , the focus of the k2 mission on nearby late k and m dwarfs , which typically have high proper motions , could make this technique of background star exclusion more widely applicable than it was for the original _ kepler _ mission . we thank ball aerospace and the _ kepler_/k2 team for their brilliant and tireless efforts to make the k2 mission a possibility and a success . without their work , this result would not have been possible . we thank sarah ballard and kevin apps for helpful conversations . we acknowledge many helpful comments from our anonymous reviewer , as well as from eric feigelson , our scientific editor . some of the data presented in this paper were obtained from the mikulski archive for space telescopes ( mast ) . stsci is operated by the association of universities for research in astronomy , inc . , under nasa contract nas526555 . support for mast for non hst data is provided by the nasa office of space science via grant nnx13ac07 g and by other grants and contracts.this paper includes data collected by the _ kepler _ mission . funding for the _ kepler _ mission is provided by the nasa science mission directorate . this research has made use of nasa s astrophysics data system , the simbad database and vizier catalog access tool , operated at cds , strasbourg , france , the exoplanet orbit database and the exoplanet data explorer at http://www.exoplanets.org , pyastronomy , the repository and documentation for which can be found at https://github.com/sczesla/pyastronomy , and the nasa exoplanet archive , which is operated by the california institute of technology , under contract with the national aeronautics and space administration under the exoplanet exploration program . b.t.m are supported by the national science foundation graduate research fellowship , grants no . dge 1144152 and dge 1144469 , respectively . j.a.j is supported by generous grants from the david and lucile packard and alfred p. sloan foundations . c.b . acknowledges support from the alfred p. sloan foundation . pf acknowledges support by fundao para a cincia e a tecnologia ( fct ) through investigador fct contracts of reference if/01037/2013 and poph / fse ( ec ) by feder funding through the program `` programa operacional de factores de competitividade - compete '' . www was supported by the austrian science fund ( fwf p22691-n16 ) . the research leading to these results has received funding from the european union seventh framework programme ( fp7/2007 - 2013 ) under grant agreement n. 313014 ( etaearth ) . this publication was made possible through the support of a grant from the john templeton foundation . the opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the john templeton foundation . this work is based on observations made with the italian telescopio nazionale galileo ( tng ) operated on the island of la palma by the fundacin galileo galilei of the inaf ( istituto nazionale di astrofisica ) at the spanish observatorio del roque de los muchachos of the instituto de astrofisica de canarias . the harps - n project was funded by the prodex program of the swiss space office ( sso ) , the harvard university origin of life initiative ( huoli ) , the scottish universities physics alliance ( supa ) , the university of geneva , the smithsonian astrophysical observatory ( sao ) , and the italian national astrophysical institute ( inaf ) , university of st . andrews , queens university belfast and university of edinburgh . the robo - ao system is supported by collaborating partner institutions , the california institute of technology and the inter - university centre for astronomy and astrophysics , and by the national science foundation under grant nos . ast-0906060 , ast-0960343 , and ast-1207891 , by the mount cuba astronomical foundation , by a gift from samuel oschin . some of the data presented herein were obtained at the w.m . keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california and the national aeronautics and space administration . the observatory was made possible by the generous financial support of the w.m . keck foundation . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of maunakea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . wasp - south is hosted by the saao and superwasp by the isaac newton group and the instituto de astrofsica de canarias ; we gratefully acknowledge their ongoing support and assistance . funding for wasp comes from consortium universities and from the uk s science and technology facilities council ( stfc ) . the digitized sky surveys were produced at the space telescope science institute under u.s . government grant nag w-2166 . the images of these surveys are based on photographic data obtained using the oschin schmidt telescope on palomar mountain and the u.k . schmidt telescope . the plates were processed into the present compressed digital form with the permission of these institutions . the national geographic society palomar observatory sky atlas ( poss - i ) was made by the california institute of technology with grants from the national geographic society . the second palomar observatory sky survey ( poss - ii ) was made by the california institute of technology with funds from the national science foundation , the national geographic society , the sloan foundation , the samuel oschin foundation , and the eastman kodak corporation . the oschin schmidt telescope is operated by the california institute of technology and palomar observatory . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . , r. , lovis , c. , pepe , f. , et al . 2012 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 8446 , society of photo - optical instrumentation engineers ( spie ) conference series , j. d. , shectman , s. a. , butler , r. p. , et al . 2010 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 7735 , society of photo - optical instrumentation engineers ( spie ) conference series , s. s. , allen , s. l. , bigelow , b. c. , et al . 1994 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 2198 , instrumentation in astronomy viii , ed . d. l. crawford & e. r. craine , 362 lcccc @xmath135 & @xmath136 & 0.244 & -0.052 & + 0.049 + @xmath137 & @xmath136 & -0.395 & -0.091 & + 0.109 + @xmath138 , [ bjd ] & @xmath139 & 2456907.895 & -0.037 & + 0.015 + @xmath140 & @xmath139 & 2.210472 & -0.000176 & + 0.000079 + @xmath141 & @xmath139 & -3.286 & -0.111 & + 0.089 + @xmath142 & @xmath136 & -0.0296 & -0.0038 & + 0.0023 + @xmath143 & @xmath136 & 0.19 & -0.26 & + 0.22 + @xmath144 & @xmath136 & -0.581 & -0.122 & + 0.186 + @xmath145 , [ bjd ] & @xmath139 & 2456930.8 & -5.4 & + 5.7 + @xmath146 & @xmath139 & 3.838 & -0.082 & + 0.093 + @xmath147 & @xmath139 & -2.22 & -0.29 & + 0.28 + @xmath148 & @xmath139 & -3.443 & -0.039 & + 0.042 + @xmath149 & @xmath150 & 0.37 & -0.24 & + 0.33 + @xmath151 & @xmath152 & 0.41 & -0.39 & + 0.29 + rv zero point [ ms@xmath153 & @xmath139 & -0.42 & -0.34 & + 0.33 + jitter [ ms@xmath153 & @xmath139 & 0.45 & -0.28 & + 0.32 + @xmath75 & @xmath154 & 4.591 & -0.020 & + 0.022 + @xmath155 & @xmath156 & -13.05 & -0.33 & + 0.35 + @xmath157 & @xmath158 & 1.33 & -0.93 & + 1.33 + @xmath159 & @xmath160 & 2.7 & -1.6 & + 1.4 + @xmath161 & @xmath162 & 3.31 & -0.78 & + 0.53 + @xmath163 & @xmath164 & -4.1 & -1.8 & + 1.8 + @xmath165 & @xmath164 & -4.2 & -1.8 & + 1.8 + @xmath166 & @xmath164 & -4.3 & -1.7 & + 1.9 + @xmath167 & @xmath168 & -9.284 & -0.044 & + 0.048 + @xmath169 & @xmath170 & -6.685 & -0.059 & + 0.064 + @xmath171 & @xmath170 & -7.227 & -0.051 & + 0.051 + @xmath172 & @xmath170 & -6.710 & -0.122 & + 0.149 +

for markovian dynamics of field fluctuations we present here an extended strong collision approximation , thereby putting our previous strong collision approach ( phys . rev . lett . 83 ( 1999 ) 4215 ) into a systematic framework . our new approach provides expressions for the free induction and spin echo magnetization decays that may be solved analytically or at least numerically . it is tested for the generic cases of dephasing due to an anderson - weiss process and due to restricted diffusion in a linear field gradient . * * ii 2

cond-mat0204325

the understanding of spin dephasing is of paramount interest in all fields of nuclear magnetic resonance ( nmr ) sciences . in nmr spectroscopy it determines the line shape , in nmr imaging it is besides longitudinal relaxation the major mechanism determining the contrast and contains morphological as well as functional information . the processes contributing to spin dephasing are related to the spin environment . in biological tissues , for example , spin dephasing may result from dipole - dipole interaction of water proton spins with paramagnetic ions like fe@xmath0 . another cause is diffusion within inhomogeneous magnetic fields generated by native or contrast agent induced susceptibility differences that are related to tissue composition and/or cellular and sub cellular compartments . in magnetic resonance imaging spin dephasing in external gradient fields is exploited to get information about diffusion within biological systems . these diffusion sensitive imaging techniques are applied to study tissue anisotropy and restrictions of diffusion that are given by membranes of cells and sub cellular structures . essential for dephasing of spins are the field fluctuations that induce the phase modulations . it is important to note that in biological tissues the relevant processes cover almost the whole range of time scales . for example , the dynamics of interactions of water proton spins with paramagnetic macromolecules as ferritin is so fast that it can be considered to be within the motional narrowing regime . on the other hand , dephasing of spins in magnetic field gradients around larger vessels is almost coherent . i.e. it is in the static dephasing regime @xcite . hence , for biological applications it is important to obtain results from theory that are valid over the whole motion regime . however , in most cases this is not possible analytically . therefore , most efforts have focused on limiting cases . the _ motional narrowing limit _ is well investigated and a number of analytical results were obtained for it@xcite . the characteristic of this limit is that the mean phase shift induced by a field realization is much smaller than one , i.e. @xmath1 , where the correlation time @xmath2 gives the mean duration of some field realization , and @xmath3 is the variance of the inhomogeneous field . the relaxation time is then obtained as @xmath4 . in the other limiting case , i.e. the _ static dephasing regime _ ( @xmath5 ) , yablonski@xcite derived analytical expressions for coherent dephasing of spins in inhomogeneous fields around magnetic centers like cylinders or spheres . kiselev @xcite extended yablonski s static dephasing approach by considering diffusion of spins within local linear gradients . however , this approach requires that the diffusion length @xmath6 during dephasing is within the linear approximation of the inhomogeneous fields @xmath7 . note that expansion around the limiting cases by perturbation approaches leads to divergences in the respective other limits . therefore , the intermediate motion regime , i.e. almost everything between the static dephasing and motional narrowing limit , was in most situations accessible by simulations only @xcite . recently , we used a strong collision ( sc ) approach to characterize spin dephasing in a particular situation : an inhomogeneous field around regularly arranged parallel cylinders filled with a paramagnetic substance @xcite , a model reflecting the capillary network of the cardiac muscle . the results agreed well with simulations @xcite over the whole dynamic range , and with experimental data @xcite . the basic idea behind the sc approach is to replace the original generator of the markov process by a simpler one , the sc generator , that conserves particular features of the original process . in particular , by an appropriate choice of its parameter the sc process reproduces the correlation time of the field fluctuations induced by the original markov process . there are several advantages of the sc approximation . first , it is is correct both in the motional narrowing and the static dephasing limits ; thereby also the error in the intermediate regime is reduced considerably , when compared to perturbation approaches . and second , it provides a simple expression for the magnetization decay which may be solved analytically or at least numerically . however , the drawback of the sc approach was up to now that is not part of a systematic approximation to or an expansion of the original generator . therefore , it was unclear how results could be improved beyond the sc approximation . the aim of this paper is to extend the sc approach and provide a framework for a systematic approximation . in the next section we will present a formal description of spin dephasing that will be the basis for our analytical analysis . in sect . iii we will introduce the extended strong collision ( esc ) approximation proper and show how it is used to describe free induction and spin echo decay . in sect . iv we will apply it to two generic cases : spin dephasing induced by an anderson - weiss process @xcite and by restricted diffusion in a linear field gradient . we will close the paper with a summary and a discussion of our results . we assume that dephasing of transversely polarized nuclear spins exposed to an external field is induced by randomly fluctuating magnetic perturbation fields with frequency @xmath8 , where @xmath9 is a discrete or continuous variable . the transition dynamics between two distinct states @xmath9 and @xmath10 is that of a stationary continuous time markov process described by rates @xmath11 for the transition @xmath12 . the matrix @xmath13 as the generator of the markov process conserves the probability to find a spin within one state , i.e. @xmath14 . the eigenvalues @xmath6 of @xmath15 fulfill the condition @xmath16 where @xmath17 corresponds to the equilibrium probability distribution . to simplify the notation we denote the normalized left and right eigenvectors of @xmath15 as @xmath18 and @xmath19 , respectively , with @xmath20 . the time evolution between @xmath21 and @xmath22 of the transverse magnetization of spins in the state @xmath10 ( in polar notation @xmath23 ) results from the linear superposition of the transition and the precession dynamics , i.e. @xmath24 . the precession within the external field was omitted since it only induces a constant offset of the frequency which may be gauged to zero . with the diagonal frequency matrix @xmath25 one obtains for the magnetization @xmath26 @xmath27 which is a generalization of the bloch torrey equation @xcite originally formulated for diffusing spins , i.e.@xmath28 . in most cases it is reasonable to assume that the initial magnetization @xmath29 is proportional to the equilibrium probability distribution @xmath30 , e.g. when free diffusion is considered this would imply a homogeneous transverse magnetization . equation ( [ gbt ] ) then provides the time evolution of the transverse magnetization ( free induction decay ) as @xmath31\;|0\rangle\ ; , \label{fid1}\ ] ] where the initial magnetization was normalized to one . the overall magnetization is then determined as @xmath32\;|0\rangle\label{magn}\;.\end{aligned}\ ] ] the free induction decay as determined by eqs . ( [ fid1 ] ) and ( [ magn ] ) results from coherent and incoherent spin dephasing . the incoherent contribution is determined from spin echo experiments . in - plane polarized spins are rotated by a @xmath33 ( @xmath34-pulse ) after a time @xmath35 . this pulse transforms the original magnetization @xmath36 to its complex adjoint @xmath37\;|0\rangle$ ] . this procedure cancels the coherent spin dephasing after the time @xmath21 ( echo time ) , i.e. the decay of magnetization at @xmath21 is solely due to incoherent spin dephasing . the time course of the magnetization after the pulse , i.e. for times @xmath38 , is determined by @xmath39\exp[(\r-{\rm i}\o)t/2]|0\rangle \;,\ ] ] i.e. the overall spin echo magnetization at the echo time @xmath21 is @xmath40\;\exp[(\r-{\rm i}\o)\;t/2]\;|0\rangle\cr & = & \langle m(t/2)|m^*(t/2)\rangle\label{se2}\;.\end{aligned}\ ] ] equation ( [ se2 ] ) relates the overall spin echo magnetization to the magnetization of the free induction decay . the analytical determination of the free induction decay according to eqs . ( [ magn ] ) is restricted to very few cases , e.g. free diffusion in a linear gradient or stochastic fluctuations between two precession frequencies . the idea of the strong collision approach and its extension is to replace the generator of the markov process @xmath15 by a more simple generator @xmath41 that conserves specific features of the original dynamics . in many cases the stochastic fluctuations of the perturbation fields occur on a much shorter time scale than spin - dephasing , i.e. the correlation time @xmath2 of field fluctuations is much shorter than the relaxation time of the magnetization . for ergodic markov processes one can estimate that after a few @xmath2 a spin has visited almost all relevant states with the equilibrium probability . on the other hand , there is only little change of the magnetization during this time interval . therefore , spin dephasing in this situation can be described equivalently by a process in which the transition rate between two states @xmath42 is independent of the initial state . consequently , the transition rate for @xmath42 is proportional to the equilibrium probability of the final state , @xmath43 . such a dynamics is referred to as _ strong collision dynamics_. the generator @xmath41 of this process has the form @xmath44 where @xmath45 is the projection operator onto the eigenspace generated by the the equilibrium eigenvector of @xmath15 , and @xmath46 is the identity operator . the factor @xmath47 has to be determined self consistently . since the starting point of the strong collision approximation is the observation that in many cases of interest the correlation of the stochastic field fluctuations appear on a shorter time scale than that of changes of the magnetization , only the long time behavior of the field fluctuations is of importance . this long time behavior is characterized by the correlation time of the two - point autocorrelation function of the field fluctuations ( see also the appendix ) , @xmath48 which is defined as @xmath49\o|0\rangle}{\langle 0|{\bf\omega}^2|0\rangle-\langle 0|{\bf\omega}|0\rangle^2}\label{2pointcf}\end{aligned}\ ] ] stochastic field fluctuations determined by the strong collision ( sc ) process should have the same correlation time as the original process , leading to the self - consistency condition @xmath50 it is easy to determine that the correlation time for the strong collision approximation is @xmath51 , see the appendix eq . ( [ sctau ] ) , leading to @xmath52 the extension of the strong collision approximation is based on a comparison with the spectral expansion of the original operator , @xmath53 where @xmath54 are the ordered eigenvalues of @xmath55 and @xmath56 is the projection operator onto the eigenspace corresponding to @xmath57 . since the time evolution operator is @xmath58 , it is clear that the low order eigenvalues determine the long - time behavior , while higher orders dominate shorter and shorter time scales . a comparison with a rewriting of eq . ( [ strcollgen ] ) , @xmath59 ( note that @xmath60 ) , shows that in the strong collision approximation just the lowest order term of ( [ spectralexpansion ] ) is taken into account explicitly , while the contribution of the higher eigenvalues is approximated by the self - consistently determined parameter @xmath47 . a natural extension , therefore , would be to take into account more low order eigenvalues explicitly , thereby increasing the accuracy of the description of the long - time behavior : @xmath61 with @xmath62 . a stochastic process generated by an operator @xmath63 in eq . ( [ gstrocollgen ] ) will be referred to as simplified extended strong collision ( esc@xmath64 ) approximation of order @xmath65 . as before , the contribution of the higher eigenvalues is approximated by the parameter @xmath47 , which is determined again self - consistently from condition ( [ ctsc ] ) . here it leads to @xmath66 with @xmath67 and @xmath68 . note that for @xmath69 this equation becomes ( [ lambda ] ) again . however , there are several problems involved with an approximation based on ( [ gstrocollgen ] ) and ( [ lambda ] ) . practically , an exact determination of the low order eigenvalues and eigenvectors is possible only in special cases . therefore , one has to deal with the problem that the eigenvalues and eigenvectors are known either only approximately or not at all . furthermore , even with eigenvalues and eigenfunctions known , it turns out that the esc approximation may be not applicable at all in certain situations : if the autocorrelation function of the field fluctuations is determined fully by the eigenfunctions included in @xmath70 , eq . ( [ lambda ] ) is undetermined . in that case additional self - consistency requirements would be necessary for a better description of the intermediate time regime . nevertheless , the above approach can be readily adapted to these situations . equation ( [ gstrocollgen ] ) can be viewed as an optimized reduced description of the relaxation in various subspaces of the original operator @xmath15 . such a optimized description should also be possible for subspaces that are not eigenvectors of @xmath15 . we can , therefore , set @xmath71 however , now the rates @xmath72 , @xmath73 are not eigenvalues anymore , but have to be determined by additional self - consistency requirements , see below . moreover , the @xmath74 are not projectors onto the eigenspace of a particular eigenvalue , but onto the spaces defined by arbitrarily chosen mutually orthogonal functions @xmath75 , @xmath73 , with @xmath76 and @xmath77 ; i.e. the projectors have the form @xmath78 and @xmath79 . naturally , one would try to choose the functions @xmath75 close to the eigenfunctions @xmath80 , although it is not required for the extension to work . another natural function space for example is based on polynomials in the frequency operator @xmath81 , i.e. @xmath82 where @xmath83 is some polynomial of degree @xmath9 , the coefficients of which are chosen in such a way that the orthonormal relations are fulfilled . in the following we will refer to this base of functions as the @xmath84-base . in analogy to eq . ( [ gstrocollgen ] ) a stochastic process generated by an operator @xmath85 in eq . ( [ gstrocollgen2 ] ) will be referred to as extended strong collision ( esc@xmath64 ) approximation of order @xmath65 . it is evident that the esc@xmath86 approximation refers to the strong collision approximation . we mentioned already that the rates @xmath72 , @xmath73 , in eq . ( [ gstrocollgen2 ] ) have to be determined now by additional self - consistency requirements . as it was with the sc approximation , the aim of the esc@xmath64 approximation is to approximate more closely the correlation of field fluctuations . this is achieved by considering also higher order correlation functions @xmath87 following the same arguments as for the strong collision approximation , the long time behavior of the @xmath88 is of interest . in the same way as for the strong collision approximation this should be characterized by some first order statistical moment which is obtained by integration of the correlation function over @xmath89 . however , direct usage of @xmath88 is hampered by its non - vanishing asymptotic behavior : it is easily seen that from the relation @xmath90 follows @xmath91 which does not necessarily vanish . in the strong collision approximation we avoided this problem by considering the operator @xmath92 $ ] instead of @xmath93 in eq . ( [ 2pointcf ] ) , i.e. we considered only the relaxational part of the stochastic process . when we perform the same replacement in eq . ( [ hocf ] ) we obtain modified correlation functions @xmath94 that we will call _ quasi cumulants _ ( s. appendix ) . they vanish asymptotically for all @xmath95 . we now require that the generalized correlation times derived from these quasi cumulants , @xmath96 are equal for the exact process and for the extended strong collision description . the relaxation rates are , therefore , determined by @xmath97 which replace the single self - consistency condition ( [ ctsc ] ) . note that in many systems the correlation functions @xmath94 vanish for odd values of @xmath98 due to symmetry . therefore , we require the equivalence of relaxation times in eq . ( [ ctsc2 ] ) for even values of @xmath98 only . otherwise one has to determine the correlation times of the first @xmath99 non - vanishing correlation functions . it is important to emphasize some properties of the esc approximation . first of all , it usually does not reduce to the esc approximation when eigenfunctions are used for the projection operator ; i.e. the @xmath100 do not take on the numerical values of the corresponding eigenvalues , although they usually do approximate them . in the light of the problems with the esc approximations mentioned above , it will turn out that this is an advantage : the self - consistent determination of the relaxation parameters @xmath47 , and @xmath100 according to eq . ( [ ctsc2 ] ) is more balanced than when only eq . ( [ ctsc ] ) is used , and gives rise to an improved approximation . moreover , the self - consistency conditions ( [ ctsc2 ] ) imply that both processes , the esc process and the original markov process have the same motional narrowing expansion of the transverse relaxation , as it is shown in the appendix , eq . ( [ mnarrowing2 ] ) . in this section we will exploit the simple structure of the generator @xmath101 to determine the time course of magnetization . we will consider both : the free induction decay , i.e. the superposition of coherent and incoherent spin dephasing , and the spin echo decay , i.e. pure incoherent spin dephasing . in the extended strong collision approximation the generator @xmath15 of the free induction decay in the generalized bloch torrey equation ( [ gbt ] ) is replaced by the generator @xmath101 of eq . ( [ gstrocollgen2 ] ) . instead of solving the propagator @xmath102 it is more convenient to solve its laplace transform @xmath103 which may be expanded as @xmath104 where @xmath105 is the laplace transform in the static dephasing limit ( @xmath106 ) , and the operator @xmath107 is defined as @xmath108 where we set @xmath109 . we will now confine the operators in eq . ( [ laplace1 ] ) onto the subspace defined by the projection operator @xmath110 . using the abbreviation @xmath111 for denoting any operator @xmath112 confined to that subspace , we obtain @xmath113 where we exploited the fact that @xmath114 and the idempotency of projection operators , i.e. @xmath115 . equation ( [ laplace2 ] ) is of fundamental importance . it demonstrates that the relation ( [ laplace1 ] ) between the esc - approximation and the static dephasing is also valid in the subspace @xmath116 $ ] . this simplifies determination of spin relaxation considerably since one only has to determine the @xmath117 matrices of the static dephasing limit @xmath118 and @xmath107 , i.e. @xmath119 the laplace transform of the overall magnetization decay @xmath120}(s)$ ] in the extended strong collision approximation has the form @xmath121}(s)=\langle 0|\;\hat{\u}^\pi(s)\;|0\rangle\ ] ] for the special case of the strong collision approximation , esc@xmath86 , eqs . ( [ laplace3 ] ) and ( [ propagatoresc ] ) result in @xmath122}(s)={{\hat{m}_{sd } ( s+\lambda)}\over{1-\hat{m}_{sd}(s+\lambda)\ ; \lambda}}\;. \label{lsc}\ ] ] with @xmath123 as the laplace transform of the overall magnetization in the static dephasing regime . the time evolution @xmath124 can be obtained from eqs ( [ laplace3 ] ) and ( [ propagatoresc ] ) either by numerical inverse laplace transform or using the generalized moment approach @xcite which allows a multi - exponential approximation . the spin - echo decay is obtained by inserting of the generator @xmath101 into eq . ( [ se2 ] ) , i.e. @xmath125}(t)&=&-\lambda\ ; m_{se,[n]}(t)+\langle m(t/2)|{\bf\lambda}|m^*(t/2)\rangle\cr&=&-\lambda\ ; m_{se,[n]}(t)+\langle 0|{\bf u}(t/2){\bf\lambda}{\bf u}^*(t/2)|0\rangle\cr & = & -\lambda\ ; m_{se,[n]}(t)+\langle 0|{\bf u}^\pi(t/2){\bf\lambda}{\bf u}^{*\;\pi}(t/2)|0\rangle\cr % % & = & -\lambda\ ; m_{se,[n]}(t)+\lambda\;\left|m_{[n ] } % % ( t/2)\right|^2\cr & & + % % \sum_{i=1}^n(\lambda-\lambda_i)\langle 0\left| { \bf % % u}(t/2)|i\rangle \right|^2 % % % \langle i|{\bfu}^*(t/2)|0\rangle\cr \label{esc4}\end{aligned}\ ] ] i.e. the spin - echo decay is expressed as a function of the spin - echo amplitude @xmath126 , and the projection of the free induction decay onto the subspace @xmath127 . this projection of the free induction decay is obtained from eq . ( [ laplace3])by inverse laplace transform , i.e. @xmath128 . the solution of eq . ( [ esc4 ] ) is @xmath129}(t)=e^{-\lambda t}\left[1 + 2\int_{0}^{t/2}d\xi e^{2\lambda\xi}\langle 0|{\bf u}^\pi(\xi){\bf\lambda}{\bf u}^{*\pi}(\xi)|0\rangle\right ] % \left[\lambda \;\left|m_{[n ] } ( \xi/2)\right|^2 + \right.\right.\cr % & & \left.\left . \sum_{i=1}^n(\lambda-\lambda_i ) \left|\langle % 0|{\bf u}(\xi/2)|i\rangle\right|^2 % % % \langlei|{\bf u}^*(\xi/2)|0\rangle \right]\right\}\cr\ ] ] the free induction and the spin echo decay are usually described by the time constants @xmath130 and @xmath131 . however , there is no unique definition of these parameters . one definition of the relaxation times is @xmath132 for the esc decay one has to replace @xmath133 by @xmath134}$ ] and @xmath126 by @xmath135}$ ] . this definition implies a dependence of relaxation times on @xmath21 , except for single exponential decay . another definition of relaxation times is based on the assumption that these constants provide the best single exponential approximation of magnetization decays , i.e. @xmath136 . according to the mean relaxation time approximation the relaxation times are then the first long time moments of the decays @xcite , i.e. @xmath137 for a single exponential the mean relaxation time definition and the definitions ( [ t2se ] ) give the same results . according to definition ( [ t2 ] ) the relaxation times of the esc decays can be related to their laplace transforms as @xmath138}(0)\cr t_2&=&\hat{m}_{se,[n]}(0)\;,\label{t2esc}\end{aligned}\ ] ] the term @xmath120}(0)$ ] , which provides @xmath130 , is obtained from eq . ( [ propagatoresc ] ) . applying some rules of laplace transforms the term @xmath139}(0)$ ] giving @xmath131 is obtained from eq . ( [ esc4 ] ) as @xmath140 where @xmath141 is the mean relaxation time of the absolute squared overall free induction magnetization @xmath142}(t)|^2=|\langle 0|{\bf u}(t)^\pi|0\rangle|^2 $ ] , and for @xmath143 @xmath144 are transit times describing the transient occurrence of the non - equilibrium components of the free induction decay @xmath145 . equation ( [ esc6 ] ) relates @xmath131 which describes the incoherent , i.e. irreversible , component of spin dephasing to the stochastic field dynamics ( @xmath146 ) and time constants of the free induction decay ( @xmath147 ) , i.e. eq . ( [ esc6 ] ) is a dissipation - fluctuation - coherence relation . note that the eqs . ( [ theta0 ] ) to ( [ thetai ] ) directly relate the time constants @xmath147 to the laplace transform of the free induction decay @xmath148 given by the fundamental equation ( [ laplace2 ] ) . > from eq . ( [ esc6 ] ) one can derive asymptotic relations for very fast and slow stochastic field fluctuations . let @xmath149 be some scaling parameter of @xmath150 , i.e. @xmath151 , then eq . ( [ esc6 ] ) reads in the static dephasing limit ( @xmath152 ) @xmath153 where we exploited that @xmath154 approaches its finite static dephasing limit . for very fast fluctuations i.e. in the motional narrowing limit ( @xmath155 ) one exploits that @xmath156 , as a power expansion demonstrates , i.e. one obtains @xmath157 this implies that the spin echo relaxation time is almost identical with the relaxation time of the absolute squared magnetization of the free induction decay , or vice versa that the free induction decay is almost irreversible . the dissipation - fluctuation - coherence relation ( [ esc6 ] ) takes a very simple form in the strong collision approximation , when we assume that the overall magnetization decay is well approximated by a single exponential , i.e. @xmath158}(t)\approx e^{-t / t2^*}$ ] . since @xmath159 , see eq . ( [ lambda ] ) , eq . ( [ esc6 ] ) reads @xmath160 > from eqs . ( [ esc7 ] ) follows that in the motional narrowing limit @xmath161 holds whereas in the static dephasing limit of the strong collision approximation the relation @xmath162 holds . the anderson - weiss model @xcite is one of the rare approaches besides the esc approximation which describes spin dephasing over the whole dynamic range of stochastic field fluctuations . the approach is suitable , for example , when dephasing is induced by spin interaction with a great number of independently fluctuating perturbation fields in the spin environment . then analytical results are obtained for the free induction and the spin echo magnetization decay as @xmath163\label{mft2st}\\ m_{se}(t)&=&\exp\bigg[-4\int_0^{t/2}(t/2-\xi)\;c_2(\xi)\ ; d\xi \cr&&+\int_0^t(t-\xi)\ ; c_2(\xi)\ ; d\xi\bigg]\ ; , \label{mft}\end{aligned}\ ] ] where @xmath164 is the modified two point correlation function ( s. appendix ) . in this section we will first characterize the class of markovian processes which fulfills the conditions of the anderson - weiss model . this leads to a generalized bloch - torrey equation according to eq . ( [ gbt ] ) which is solved . finally we compare the anderson - weiss model with its esc@xmath86 and esc@xmath165 approximation . the anderson - weiss approach is based on a gaussian distribution of perturbation field frequencies @xmath166 . even more important is the _ additional _ assumption that the stochastic phase accumulation of a spin @xmath167 also exhibits a gaussian distribution . this latter condition implies that the greens function @xmath168 , i.e. the probability that a spin initially precessing with frequency @xmath8 precesses at @xmath21 with @xmath169 , is also a gaussian function in @xmath170 with the condition @xmath171 . this implies that only nearest neighbor transitions rates are non - vanishing . markovian processes in a continuous variable @xmath166 with this property are described equivalently by a fokker planck equation @xcite , i.e. the probability density @xmath172 satisfies @xmath173 where @xmath174 is a possibly @xmath166-dependent diffusion coefficient and @xmath175 is some driving force . since the equilibrium probability density is a gaussian function one obtains @xmath176 with @xmath177 . the generalized bloch - torrey equation ( [ gbt ] ) which determines the dynamics of magnetization as a superposition of precession and stochastic transitions then reads @xmath178m(\omega , t)\;. \label{fokkerbt}\ ] ] the derivation of the eqs . ( [ fokker ] ) , ( [ fokkerbt ] ) is of fundamental importance since it states that a markovian dynamics of a variable @xmath166 which satisfies the anderson - weiss conditions is equivalent to a diffusion process in this variable within a harmonic potential @xmath179 and vice versa . transformation of variables @xmath180 and @xmath181 simplifies eq . ( [ fokkerbt ] ) to @xmath182 m(\omega , t)\label{nfokkerbt}\ ] ] where we continue to denote also the transformed variables as @xmath166 and @xmath21 and @xmath183 is the transformed diffusion coefficient . in the following we will restrict consideration to the case of a constant diffusion coefficient @xmath184 . the left and right sided eigenfunctions of the transition operator @xmath185 are the hermite functions , i.e. @xmath186 with eigenvalues @xmath187 > from the definition of the hermite functions and the operator intertwining relation @xmath188=1 $ ] follow the recursive equations @xmath189 which also provide the normalization of eigenfunctions . the advantage of the markovian formulation of the anderson - weiss model is that it does not only provide global parameters but also local ones , e.g. the time course of the magnetization with frequency @xmath166 . straightforward application of the eqs . ( [ rekursiv ] ) and some operator algebra provides the solution of eq . ( [ nfokkerbt ] ) as @xmath190| 0\rangle\cr\cr & = & \exp[-\beta^{-1 } t+\beta^{-2}(1-e^{-\beta t})]\cr & & ( 2\pi)^{-1/2}\exp[-1/2(\omega-\i\;\beta^{-1}(e^{\beta t}-1 ) ) ] \label{magaw}\end{aligned}\ ] ] integration over @xmath166 just gives the free induction decay of the overall magnetization @xmath191 \label{maw}\ ] ] which is just equivalent to the result of eq . ( [ mft2st ] ) , since the 2-point correlation function of eq . ( [ nfokkerbt ] ) is @xmath192 ( s. appendix eq . ( [ corrf2ho ] ) ) . insertion of this 2-point correlation function into eq . ( [ mft ] ) provides the spin echo decay as @xmath193\;.\label{mseaw}\ ] ] relaxation times of the free induction and spin echo decay were determined according to eqs . ( [ t2 ] ) . the esc propagator is determined from the propagator in the static dephasing limit @xmath194 $ ] , and the @xmath195-matrix of eq . ( [ lambdamatrix ] ) both restricted either to the function space @xmath196 $ ] for the esc@xmath86 or @xmath197 $ ] for the esc@xmath165 approximation eq . ( [ laplace3 ] ) . the special structure of the transition rate operator of the anderson weiss model implies that the base of eigenfunctions , eqs . ( [ hermite ] ) , is identical with the @xmath84-base , eq . ( [ omegabase ] ) , i.e. @xmath198 . hence , we will set in the following @xmath199 [ [ esc_0-approximation ] ] esc@xmath86 approximation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the matrix element of the laplace transformed propagator in the static dephasing limit required for the esc@xmath86 approximation is @xmath200 where @xmath201 is the complementary error function . the coefficient @xmath47 which guarantees the self consistency condition eq . ( [ ctsc ] ) is determined form the eqs . ( [ sctau],[corrt1ho ] ) ( s. appendix ) as @xmath202 [ [ esc_1-approximation ] ] esc@xmath165 approximation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the matrix elements of the laplace transformed propagator in the static dephasing limit required for the esc@xmath165 approximation in the @xmath84-base are that of eq . ( [ fidhos ] ) and @xmath203\cr & = & n \;(-\i){\cal l}[\partial_t\langle 0|\exp(\i\o t)|0\rangle]\cr & = & n\ ; \i ( 1-s \langle 0|{\hat u}_0(s)|0\rangle)\ ; , \label{fidho1}\end{aligned}\ ] ] where the factor @xmath204 generally is some normalization factor with @xmath205 , i.e. in the case of the anderson weiss model it is simply @xmath206 . consequently , using some elementary rules of laplace transforms , one derives the other matrix elements as @xmath207 it has to be stressed that the equations ( [ fidho1 ] ) to ( [ fidho ] ) are generally valid for all esc@xmath165 approximations in the @xmath84-base . the coefficients @xmath208 guaranteeing the self consistency condition eq . ( [ ctsc2 ] ) are obtained from eqs . ( [ hot2],[corrt1ho ] ) and ( [ tau4esc1simple],[corrt3ho ] ) @xmath209 [ [ relaxation - in - the - esc_0-and - esc_1-approximation ] ] relaxation in the esc@xmath86 and esc@xmath165 approximation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the matrix @xmath210 and the coefficients @xmath47 and @xmath211 determine the laplace transformed esc propagator @xmath148 in eq . ( [ laplace3 ] ) , which itself is the base for all other calculations . it directly provides @xmath130 when defined as the first moment , eq . ( [ t2esc ] ) , of the free induction decay eq . ( [ propagatoresc ] ) . insertion of @xmath212 into eqs . ( [ theta0],[thetai ] ) provides according to eq . ( [ esc6 ] ) the relaxation time of the spin echo decay when defined as its first long time moment eq . ( [ t2esc ] ) . inverse laplace transformation of @xmath212 gives the esc propagator @xmath213 which itself allows determination of the spin echo decay eq . ( [ esc4solved ] ) . the relaxation time @xmath214 of the anderson - weiss process is well approximated by the esc@xmath86 and esc@xmath165 approximation over the whole dynamic range of stochastic field fluctuations ( fig . 1 ) . in the static dephasing regime all curves approach @xmath215 . the successive approximation of the spin echo relaxation by the esc approximation is seen from the magnetization decay curves ( fig . 2 ) and the curves showing the dependence of @xmath131 obtained by either definition ( eqs . ( [ t2se])-([t2 ] ) ) on the diffusion coefficient @xmath184 as figure 3 demonstrates . the latter curves all run parallel in the motional narrowing regime ( @xmath216 ) and exhibit a similar location of the maximum relaxation rate . towards the static dephasing regime ( @xmath217 ) the rate of the anderson - weiss process declines less than the rates of the esc processes . whereas dephasing of free diffusing spins in a linear gradient field can be treated analytically , only numerical solutions exists for the restricted diffusion case @xcite . on the one hand restricted diffusion in a linear gradient field provides a simple model to study principle features of spin dephasing by diffusion . on the other hand treatment of this problem is not only of academic interest as already mentioned in the introduction . we will approximate the free induction and spin echo decay of the global magnetization for the case of restricted diffusion by the strong collision approximation ( esc@xmath86 ) and its first extension ( esc@xmath165 ) . the esc@xmath165 approximation will be performed for both , in the @xmath84-polynomial base , i.e. @xmath218 and in the eigenfunction base , i.e. @xmath219 . we assume that the spins diffuse within an interval of size @xmath220 in a linear gradient field @xmath221 . reflecting boundary conditions at @xmath222 imply that @xmath223 . with @xmath224 as the diffusion coefficient and @xmath225 $ ] the bloch - torrey eq . ( [ gbt ] ) has the form @xmath226+{\rm i } g x ) m(x , t)$ ] , where the brackets @xmath227 $ ] denote that application of the operator @xmath228 is restricted to functions which fulfill the reflecting boundary conditions . transformation of variables @xmath229 and @xmath230 results in @xmath231+{\rm i}\;x)\ ; m(x , t)\label{btlg}\;,\ ] ] and vanishing derivatives at the edges of the unit interval @xmath232 with the diffusion coefficient @xmath233 . we continue to denote also the transformed variables as @xmath234 and @xmath21 to reduce the number of symbols . when the initial magnetization is proportional to the equilibrium probability , i.e. @xmath235 , the laplace transform @xmath236 of the local magnetization decay satisfies @xmath237+{\rm i}\;x)\hat{m}(x , s)=-1\;.\label{btlgl}\ ] ] the equation ( [ btlg ] ) was solved numerically . integration of the result over the unit interval provided the free induction decay of the overall magnetization , and application of eq . ( [ se2 ] ) on the result gave the spin echo decay . when the spin echo relaxation time was defined as the first statistical moment of the magnetization decay eq . ( [ t2 ] ) was applied . for determination of @xmath130 , defined as the first moment of the free induction decay , eq . ( [ btlgl ] ) was solved numerically , and integration @xmath238 gave @xmath239 . the determination of the esc@xmath86 and esc@xmath165 approximation is completely analogous to that for the anderson weiss model , except that the @xmath84-polynomial and the eigenfunction base are not identical . [ [ esc_0-approximation-1 ] ] esc@xmath86 approximation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the equilibrium function for the restricted diffusion within the unit interval is @xmath240 i.e. one obtains @xmath241 the self consistency condition for the strong collision approximation ( [ ctsc ] ) determines the parameter @xmath47 as ( s. appendix eqs . ( [ sctau ] ) ( [ tau1lg ] ) ) @xmath242 insertion of the results of eqs . ( [ lapfid1 ] ) and ( [ scc1 ] ) into eq . ( [ laplace3 ] ) determines the laplace transformed propagator in the esc@xmath86 approximation @xmath243 from which @xmath130 , @xmath131 and spin - echo decay curves are obtained . [ [ esc_1-approximation - in - the - omega - polynomial - base ] ] esc@xmath165 approximation in the @xmath84-polynomial base : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the lowest order function besides the equilibrium state in the @xmath84-polynomial base has the form @xmath244 the matrix element ( [ lapfid1 ] ) and the equations ( [ fidho1 ] ) to ( [ fidho ] ) then directly provide the static dephasing operator @xmath245 in the @xmath246 $ ] base . the parameters @xmath247 of the esc@xmath165 approximation are determined from the self consistency condition ( [ ctsc2 ] ) , i.e. with eqs . ( [ hot2 ] ) , ( [ tau1lg])and and eqs . ( [ tau4esc1simple]),([tau3lg ] ) one obtains @xmath248 [ [ development - in - the - eigenfunction - space ] ] development in the eigenfunction space : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the normalized non equilibrium eigenfunctions of the restricted diffusion operator are @xmath249 since @xmath250 $ ] is a symmetric operator left and right sided eigenfunctions are identical . with @xmath219 and @xmath251 one obtains @xmath252\big|_{\xi =- z^*}^{\xi = z}\cr \langle 1|\hat\u_0(s)|1\rangle&=&-2\arctan(2s)-\cr & & \quad \big[\i\;\cosh(2\pi s)\ci(\xi)+ \cr & & \quad\quad\quad \sinh(2\pi s)\si(\xi)\big]\big|_{\xi=-2z^*}^{\xi=2z}\cr & & \label{lapfid}\end{aligned}\ ] ] where @xmath253 and @xmath254 denote the integral cosine and integral sinus function respectively . the parameters @xmath255 in the eigenfunction base are determined similarly as in the @xmath84-base ( s. appendix ) and one obtains @xmath256 [ [ relaxation - in - the - esc_0-and - esc_1-approximation-1 ] ] relaxation in the esc@xmath86 and esc@xmath165 approximation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + figure 4 demonstrates the first long time moment of the free induction decay , that of the strong collision approximation , and its first extension for both bases as a function of the diffusion coefficient @xmath184 . all curves show the same asymptotic behavior in the static dephasing ( @xmath257 ) and in the motional narrowing limit ( @xmath258 ) . furthermore the better approximation by the esc@xmath165 curves compared to the esc@xmath86 curve in the intermediate motion regime is evident . there is no significant difference between the esc@xmath165 approximation in the eigenfunction and in the @xmath84-polynomial base . the spin echo magnetization decay is shown in fig . 5 . especially in the long time behavior near the static dephasing regime , the esc@xmath165 curves either in the eigenfunction space or in the @xmath84-space demonstrate a better approximation than the esc@xmath86 curve . this is also reflected by the dependence of spin echo relaxation rate @xmath259 on the diffusion coefficient ( figs . 6,7 ) . when defined by the echo time ( eq . ( [ t2se ] ) ) the esc@xmath86 and esc@xmath165 curves run parallel with the curve obtained for restricted the diffusion dynamics for short echo times . for longer echo times and decreasing diffusion coefficients the esc@xmath165 curve provides a better approximation . again as for the free induction decay there is no significant difference between esc@xmath165 approximations in the eigenfunction and that in the @xmath84-base . analytical results on transverse spin relaxation due to stochastic phase modulation exist mainly for limiting cases , like the motional narrowing and the static regime . perturbation approaches are only valid close to their respective limits , and they diverge as one tries to extend them towards the opposite motion regime . particularly the intermediate motion regime can not be described reliably by such a treatment . we choose a different approach . our aim was to approximate the dynamics , assumed to be markovian , by a more simple one that conserves specific features of the original . the starting point was the strong collision approximation@xcite that assumes the transition probability between two states being independent from the initial state , an approximation that holds when spin dephasing occurs on a time scale significantly longer than the stochastic phase modulations . hence , all states perpendicular to the equilibrium state relax with the same exponential factor that is determined self - consistently by comparison with the field fluctuations . note that the motional narrowing limit as well as the static dephasing regime are described correctly by this approximation . consequently , the error in the intermediate motion regime is already less than it would be by perturbation approaches of a comparable low order . nevertheless , there is still room for improvement . also , one would like to have higher order approximations that can be used to check the quality of low order descriptions . a systematic extension of the strong collision ansatz is to include the relaxation of states of an appropriate finite function base explicitly . we require that correlation times of original and approximate dynamics are identical to a certain order . this self consistency condition assures that both dynamics have the same motional narrowing expansion of spin dephasing . as it was already in the strong collision ansatz , spin dephasing is asymptotically identical for both dynamics in the limit of the static motion regime . the finite function base of the esc@xmath64 approximation may be given by the first @xmath65 ordered eigenfunctions of the generator of the original phase modulations . obviously , then the esc generator directly reflects the dynamics of original generator up to a time scale corresponding to the @xmath65-th eigenvalue . for practical applications the esc approach within an eigenfunction space may be a safe way to approximate spin dephasing . however , when the determination of the eigenfunctions is tedious , the application of the @xmath84-base latexmath:[$[|0\rangle,\;|f_1\rangle\sim{\bf\omega}|0\rangle , appropriate , at least for the esc@xmath165 approximation . within the @xmath84-base the determination of the two- and four - point correlation times ( s. appendix ) and the propagator in the static motion regime , eq . ( [ fidho ] ) , is considerably simplified . the mechanism by which the esc@xmath165 approach in the @xmath84-base works becomes evident by the following consideration : terms of the motional narrowing expansion eq . ( [ narrowing1 ] ) may be interpreted as repetitive interactions of the spin system with the inhomogeneous field @xmath261 and intermediate evolution with the free propagator @xmath262 . in the motional narrowing limit , one obtains from eq . ( [ mnarrowing2 ] ) @xmath263 where the factor @xmath264 is due to the normalization of @xmath265 , @xmath266 . equation ( [ f1v ] ) implies that in the motional narrowing limit the long time behavior of spin dephasing solely depends on the free propagator related relaxation of the state @xmath265 , i.e. this state remains the only relevant one . hence , it is obvious that in the intermediate motion regime an esc@xmath165 approximation including the state @xmath265 in its generator is superior to the esc@xmath86 approximation . within the function base the propagator of spin dephasing is directly related to the propagator of spin dephasing in the absence of stochastic phase modulations . this specific feature of the esc dynamics tremendously facilitates the actual determination of spin dephasing for the following reasons : ( @xmath9 ) in many cases the propagator in the static motion regime ( which is an average phase factor ) may be determined analytically or at least numerically ; ( @xmath267 ) the determination of the propagator from that in the static motion regime is self contained within the base , i.e. it is obtained from a combination of finite dimensional matrices . the two lowest order esc approximations were applied to two generic models : spin dephasing in the anderson weiss model , i.e. gaussian frequency distribution and gaussian transition dynamics , and dephasing by restricted diffusion in a linear frequency gradient . the reason for this choice was that besides their generic character these models allow either an analytical ( anderson weiss ) or , at least , a simple numerical treatment ( linear gradient ) of magnetization decay . these features are helpful to prove the esc approach . for the anderson weiss model we determined the corresponding markov generator of the phase modulations , which to our knowledge was done here for the first time . for both generic models the subsequent improvement by esc@xmath64 approximations of dephasing parameters and magnetization decays could be demonstrated . one of our next aims will be the application of the esc approach to more realistic scenarios . in closing , we would like to emphasize that the esc approach is actually not limited to spin dephasing only . it can be applied , in principle , in any situation where the time behavior of complicated observables of stochastic processes is of interest . in each case , however , an appropriate function base has to be chosen , corresponding to the @xmath84-base for spin dephasing . we thank p. grassberger for a critical reading of the manuscript . this work was supported by the deutsche forschungsgesellschaft : sonderforschungsbereich 355 `` pathophysiologie der herzinsuffizienz '' , sfb 237 `` unordnung und groe fluktuationen '' , and graduiertenkolleg `` nmr '' ha 1232/8 - 1 . the general @xmath65-point autocorrelation function of stochastically fluctuating fields @xmath169 is defined as @xmath268 where @xmath269 is the probability to find at @xmath270 the frequency @xmath271 , at @xmath272 the value @xmath273 , ... and at @xmath274 the frequency @xmath275.when the stochastic dynamics is determined by a markov process , this probability can be factored into transition probabilities between sequential states @xmath276 after the interval @xmath277 and the initial ( @xmath270 ) probability distribution , i.e. @xmath278 the transition probabilities after the interval @xmath279 are the matrix elements of evolution operator @xmath280 . since the dynamics is assumed to be stationary the initial probability @xmath281 is the equilibrium state probability distribution , i.e. we can rewrite eq . ( [ npointcorrfunction ] ) @xmath282 where @xmath283 is the diagonal frequency matrix . a modification of the correlation functions occurs if one exchanges the evolution operator @xmath284 with the operator @xmath285 , where @xmath45 is the projection operator onto the equilibrium state space . this modified evolution operator describes the relaxation of observables minus their equilibrium state values . the modified autocorrelation functions will be denoted as quasi cumulants and they are then defined as @xmath286\o .... \cr .... [ & \exp&(\r t_1)-{\bf \pi}_0]\o\;|0\rangle\;.\label{corrf3}\end{aligned}\ ] ] the laplace transform of the correlation function in eq . ( [ corrf3 ] ) has the form @xmath287&\o .... \cr\cr .... & \left[{1\over{s_1-\r}}-{1\over{s_1}}\;{\bf \pi}_0\right]&\o\;|0\rangle\;.\label{lapcorrf3}\end{aligned}\ ] ] this laplace transform allows the determination of temporal moments of the normalized autocorrelation function @xmath288 as the generalized correlation times @xmath289 the motional narrowing expansion is a perturbation approach to determine the overall magnetization @xmath124 or its laplace transform @xmath290 in terms of powers of the fluctuating fields @xmath81 . it is based on the assumption , that the stochastic fluctuations are more rapid than the precession frequencies of the perturbation fields ( motional narrowing limit ) . we will present a general relation between the relaxation of the magnetization and the correlation of the field fluctuations that contains the motional narrowing limit as a limit case . the first step is to expand the laplace transform of the overall magnetization , eq . ( [ magn ] ) , in @xmath81 , i.e. @xmath291 where @xmath292 are the laplace transformed n - point correlation functions of eq.([npointcorrfunction2 ] ) . to avoid singularities at @xmath293 it is better to consider @xmath294 . when we set @xmath295 one obtains @xmath296 rearrangement of terms of equal order in @xmath81 provides @xmath297 where the coefficients @xmath298 have the form @xmath299 a comparison of this sum with the modified correlation functions @xmath300 , eqs . ( [ corrf3 ] ) and ( [ lapcorrf3 ] ) , shows that @xmath301 i.e. one obtains @xmath302 equation ( [ mnarrowing ] ) expands the relaxation of the magnetization in terms of correlation functions to an arbitrary order . the long time behavior of @xmath124 is determined by the laplace transform in the range of small @xmath303 , i.e. in this range the relation @xmath304 is valid . the series in eq . ( [ mnarrowing2 ] ) contains terms of magnitude @xmath305 , where @xmath6 denotes non vanishing eigenvalues of @xmath15 . the latter determine the fluctuation frequency . in the motional narrowing limit these fluctuations are much higher than the precessing frequencies @xmath306 , i.e. after normalization of the average perturbation field @xmath307 to zero , i.e. @xmath308 , @xmath124 is given a single exponential decay with the well known result for the transverse relaxation rate as @xmath309 in the strong collision ( esc@xmath86 ) approximation the quasi cumulants take a very simple form . insertion of the generator @xmath310 into eq . ( [ lapcorrf3 ] ) results in @xmath311 i.e. the quasi cumulant is a product of single exponential functions @xmath312 and the generalized correlation times , eq . ( [ gcorrtimes ] ) , are all identical namely @xmath313 we restrict ourselves here to the @xmath314 approximation and determine the correlation functions and generalized relaxation times for the @xmath84-polynomial base only in order to show the principle . extensions to higher order approximations and to other function bases are straightforward , although they may be more tedious to calculate . for @xmath314 the generator of the stochastic field fluctuations has the form @xmath315 . we will determine only the laplace transforms of the 2- and 4-point correlation functions since the 3-point correlation functions vanishes in the models we consider . according to eq . ( [ lapcorrf3 ] ) the determination of the correlation functions requires of the operator @xmath316 assuming that the average frequency vanishes , i.e. @xmath317 , which can always be achieved by normalization . , @xmath318 . hence , the projector @xmath319 takes the form @xmath320 for the laplace transformed two point correlation function one obtains then @xmath321 i.e. the two point correlation function exhibits a single exponential decay with relaxation rate @xmath322 the 4-point correlation function is @xmath323 the four point correlation time is then determined as @xmath324^{1/3 } \cr\cr & = & \bigg[\frac{1}{\lambda_1 ^ 2}\bigg(\frac{1}{\lambda_1}- \frac{1}{\lambda}\bigg)\frac{\langle 0|{\bf\omega}^3|0\rangle^2}{\langle 0|{\bf\omega}^2|0\rangle ( \langle 0|{\bf\omega}^4|0\rangle-\langle 0|{\bf\omega}^2|0\rangle^2)}\cr\cr & & + \frac{1}{\lambda_1 ^ 2\lambda}\bigg]^{1/3}\cr\label{tau4esc1}\end{aligned}\ ] ] in the case of the anderson weiss model and for the restricted diffusion linear gradient one has @xmath325 , i.e. eq . ( [ tau4esc1 ] ) simplifies to @xmath326{\frac{1}{\lambda_1 ^ 2\lambda } } \label{tau4esc1simple}\ ] ] in this final appendix we will determine the generalized relaxation times of stochastic field fluctuations up to the 4-th order for the generic models we discuss in the main text . when field fluctuations result from diffusion in a harmonic potential according to eq . ( [ fokker ] ) and spin dephasing is described by eq . ( [ nfokkerbt ] ) , the corresponding laplace transformed 2-point correlation function , eq . ( [ lapcorrf3 ] ) , is @xmath327 where we applied the operator properties of @xmath328 according to eqs . ( [ rekursiv ] ) . this result shows that the two point correlation function exhibits a single exponential decay . since @xmath329 one obtains @xmath330 since @xmath331 vanishes , the next relevant correlation function is @xmath332 . similarly , one obtains for its laplace transform @xmath333 and since @xmath334 the corresponding correlation time is @xmath335{\frac{1}{2}}\;\beta^{-1 } \label{corrt3ho}\ ] ] in this section we will determine the correlation times @xmath336 of eq . ( [ gcorrtimes ] ) for @xmath337 for 1-d restricted diffusion of spins in a unit box in which they are affected by a linear gradient field . in dimensionless parameters one obtains for the generator @xmath338 $ ] where @xmath184 is the dimensionless diffusion coefficient . the brackets denote that the application of the operator @xmath228 is restricted to functions with vanishing derivative at @xmath339 ( reflecting boundary conditions ) . the frequency operator is @xmath340 and the equilibrium state eigenfunction is @xmath341 . the determination of correlation times @xmath342 requires the calculation of the laplace transformed quasi cumulants @xmath343 , eq . ( [ gcorrtimes ] ) . the definition of this quasi cumulants according to eq . ( [ lapcorrf3 ] ) shows the need of recurrent determination of terms of the form @xmath344 g\;,\label{he1}\ ] ] with some function @xmath345 , i.e. @xmath346 fulfills @xmath347 i.e. with @xmath348 $ ] eq . ( [ he2 ] ) becomes a second order differential equation . the application of the inverse second order differential operator leaves in general 2 integration constants . one may be determined from the reflective symmetric boundary conditions ; however , a further condition is needed to get the second constant . a spectral decomposition @xmath15 in eq . ( [ he1 ] ) shows @xmath349 , i.e. we obtain as a further condition @xmath350 we define @xmath351 which obviously fulfills the reflecting boundary condition at @xmath339 and eq . ( [ vanishf ] ) . hence , the two point laplace transformed quasi cumulant is @xmath352 and with @xmath353 one obtains @xmath354 iterative application of eq . ( [ he1 ] ) defines @xmath355 which , according to eq . ( [ he2 ] ) , fulfills @xmath356 f_2(x)&=&(1-\pib)\ ; xf_1(x)\cr & = & xf_1(x)-\int_{-1/2}^{1/2}dx\;xf_1(x)\cr & = & xf_1(x)+\frac{1}{120}\beta^{-1}\end{aligned}\ ] ] insertion of @xmath357 , eq . ( [ f1 ] ) , and considering the reflective boundary conditions and the condition ( [ vanishf ] ) yields @xmath358 similarly to the procedure above we could determine a function @xmath359 but instead we use a different approach which exploits the symmetry of eigenfunctions of the operator @xmath360 $ ] . the laplace transformed 4-point quasi cumulant is @xmath361}^{=f_1(x)}\ ; x \cr & & \lim_{s_2\to 0}\left[-{1\over{s_2-\r}}+{1\over{s_2}}\;\bf\pib\right]\cr & & x \lim_{s_1\to 0}\left[-{1\over{s_1-\r}}+{1\over{s_1}}\ ; \bf\pib\right]x|0\rangle\cr\cr & = & -\;\int_{-1/2}^{1/2}dx\;f_1(x)xf_2(x)\cr & = & \beta^{-3}\frac{89}{79833600}\;,\end{aligned}\ ] ] and with @xmath362 we finally have @xmath363{\frac{89}{443520}}\;\beta^{-1}\ ] ]

we study self - diffusion within a simple hopping model for glassy materials . ( the model is bouchaud s model of glasses [ j .- p . bouchaud , _ j. physique i _ * 2 * 1705 ( 1992 ) ] , as extended to describe rheological properties [ p. sollich , f. lequeux , p. hbraud and m.e . cates , _ phys . rev . lett . _ * 78 * 2020 ( 1997 ) ] . ) we investigate the breakdown , near the glass transition , of the ( generalized ) stokes - einstein relation between self - diffusion of a tracer particle and the ( frequency - dependent ) viscosity of the system as a whole . this stems from the presence of a broad distribution of relaxation times of which different moments control diffusion and rheology . we also investigate the effect of flow ( oscillatory shear ) on self - diffusion and show that this causes a finite diffusivity in the temperature regime below the glass transition ( where this was previously zero ) . at higher temperatures the diffusivity is enhanced by a power law frequency dependence that also characterises the rheological response . the relevance of these findings to soft glassy materials ( foams , emulsions etc . ) as well as to conventional glass - forming liquids is discussed .

cond-mat9901267

the dynamics of systems close to a glass transition remains a central problem in statistical physics @xcite . because the glass transition is a many - body phenomenon , models for it invariably involve some approximation and it is important to disentangle the physical phenomena from the approximation scheme used . in mode - coupling theories , for example , the dynamics are dominated by a small number of collective degrees of freedom @xcite . this is appealing , but ignores activated processes , and obscures the fact that glassy materials are dynamically heterogeneous with a variety of local environments @xcite . another approach is to use disordered hopping models , which treat instead single - particle degrees of freedom coupled to a random environment , and this is the route followed here . such models , though unrealistic in some respects , are likely to give a better account of those properties , such as self - diffusion , which are not dominated by relaxation of collective modes . an important breakthrough in the approach based on hopping models was that of bouchaud @xcite , who showed that an ensemble of _ non - interacting _ particles , moving by thermally activated hopping at temperature @xmath0 in an uncorrelated fashion through a sequence of traps , can show a glass transition . this occurs if and only if the density of states ( the prior probability distribution of trap depths ) @xmath1 has an exponential tail . though grossly over - simplified , this allows many properties of the model to be calculated exactly . the exponential tail to the prior distribution @xmath1 , which leads to a power - law spectrum of relaxation times and is vital to the appearance of a glass transition , is supported by evidence from theory on spin glasses @xcite and experiments on low - temperature fluids @xcite . such a power - law relaxation spectrum was the point of departure for a related model of glasses @xcite , and similar features appear in models of dispersive transport in disordered semiconductors @xcite . the original model of bouchaud has no explicit spatial coordinates . nevertheless , the extension to self - diffusion is unambiguous ( if each hop corresponds to a spatial displacement with a well - defined second moment ) and indeed monthus and bouchaud @xcite gave an expression for the distribution of displacements on a hypercubic lattice . for flow properties , more choices are possible in the way spatial coordinates are treated , but a minimal extension of the model was offered by sollich et al . this model , which we call the gr ( glassy rheology ) model allows both linear and nonlinear rheological properties to be calculated . details of it are recalled in section [ gr ] below . the model introduced in ref . @xcite was in fact proposed to describe a class of soft glassy materials " ( argued to include foams , dense emulsions , etc . ) , in which context it was found necessary to replace the thermodynamic temperature @xmath0 in bouchaud s model by an effective ( noise ) temperature @xmath2 . this replacement converts the gr model studied below into the _ soft _ glassy rheology model of refs . @xcite , and can be made throughout our calculations . the glass transition in the gr model shows interesting features which may need further explanation for readers whose background lies in conventional glasses . specifically , in bouchaud s model the glass transition @xmath3 is identified as the temperature below which the system shows weak ergodicity breaking " : its boltzmann distribution is not normalizable . this means the system will evolve into deeper and deeper traps as time goes by , and will never attain a steady state although at no finite time are there infinite barriers partitioning phase space . when flow - related degrees of freedom are included in the model ( see section [ gr ] below ) , one finds , as expected , that the system has a finite elastic modulus ( at zero frequency ) for temperatures @xmath4 . less obviously , the _ viscosity _ of the material diverges , not at @xmath5 but at @xmath6 . between these two temperatures , the static modulus is zero but the viscosity infinite ; the material is what is known in rheological language as a power - law fluid " . for applications of the model to soft glassy materials this is a very attractive feature , since such viscoelastic behaviour is frequently observed in these systems @xcite . in a model of conventional glasses , on the other hand , it might be considered undesirable . however , with the introduction of a high energy cutoff in the prior distribution @xmath1 , the phenomenology of the gr model can be adapted to model that of conventional glasses . we discuss the conceptual features of this modification at the end of section [ concl ] ; for the actual calculations in the present paper we restrict ourselves to the simpler case without cutoff . below we calculate the statistics ( additional to the results of ref . @xcite ) of self - diffusion in the gr model , and explore quantitatively the breakdown , near the glass transition , of the ( generalized ) stokes - einstein relation [ ( g)ser ] between self - diffusion of a representative particle in the fluid and the ( frequency - dependent ) viscosity of the system as a whole . ( this breakdown does not , of course , imply that gser also fails for _ macroscopic _ probe particles which see the surrounding material as a continuum . ) we also investigate the effect of shear flow on self - diffusion and show that this causes a finite diffusivity in the low - temperature regime where this was previously zero . in the next section we review the status of the gser in both conventional and soft glassy systems . the gr model is described in section [ gr ] and used in section [ test ] to study and discuss the breakdown of the gser . in section [ analysis ] , we further elucidate the physics of particle transport in glassy systems , using both analysis and simulation of the model . in section [ shear ] we calculate the effects of shear on the rate of self - diffusion , finding the diffusion constant in the presence of continuous and periodic shear strains of various rates and amplitudes . in section [ concl ] we conclude with a discussion of our results in the context of both conventional super - cooled / glass - forming liquids and soft glassy materials . the stokes - einstein relation ( ser ) between coefficients of diffusion and viscosity ( and its generalization to frequency - dependent quantities ) is of great importance to our understanding of transport in fluids . this is especially true since the development of new techniques to measure viscoelastic response by light scattering from ( or tracking of ) probe particles @xcite . yet the relation s range of validity has been called into question by experiments on super - cooled and glass - forming liquids @xcite . to find the linear storage and loss moduli , @xmath7 and @xmath8 , of a viscoelastic fluid , a rheometer is normally used to apply an oscillatory stress to the fluid and measure the resultant strain ( or _ vice versa _ ) as a function of ( angular ) frequency @xmath9 . a technique has recently been developed whereby the complex viscoelastic modulus @xmath10 ( @xmath11 ) may be found is the fourier transform of the time derivative of the stress relaxation modulus @xmath12 , which relates stress on the fluid , @xmath13 , to strain rate @xmath14 . ] , even when the available sample of fluid is too small to fill a rheometer . the technique involves introducing one or more neutrally buoyant probe spheres of radius @xmath15 into the fluid and measuring their brownian motion . for a newtonian fluid , which is characterized by a single , frequency - independent viscosity @xmath16 , that viscosity is given by the familiar stokes - einstein formula @xmath17 . here @xmath18 is boltzmann s constant and @xmath0 temperature . as the fluid is newtonian , the probe executes an unbiased random walk , and its mean - square displacement as a function of time is @xmath19 , which defines the diffusion constant @xmath20 appearing in the stokes - einstein formula . in this purely viscous fluid , @xmath10 is purely imaginary , given by @xmath21 . as the derivative of the stress relaxation modulus is a well - behaved function of time , its laplace transform @xmath22 is simply related to its fourier transform @xmath23 . combining the above expressions , we see that probe spheres _ in a newtonian fluid _ respect the relation @xmath24 where @xmath25 is the laplace transform of the mean - square displacement of a probe . equation [ gser ] says that the modulus of a viscoelastic fluid is the ratio of a driving force ( in this case induced by @xmath26 rather than by a rheometer ) to a response ( mean - square distance moved by the sphere ) . although derived above only for the newtonian case , eq . [ gser ] is the generalized stokes - einstein relation ( gser ) , and when written in this form it applies under relatively general conditions to _ all linearly viscoelastic fluids _ , not just newtonian ones @xcite . the basic stokes - einstein relation ( ser ) is recovered from eq . [ gser ] in the zero frequency limit , since viscosity is _ defined _ by @xmath27 . in all cases , the fluid around the probe is treated as a featureless continuum , and therefore the result is only valid when the sphere radius greatly exceeds the microscopic length scale ( _ e.g_. particle size ) of the fluid . ( furthermore , probe particles must be sufficiently dilute to ensure independent stokesian flow fields around each . ) if , as is sometimes the case , the gser is applied to probe particles which are not large compared to the microscopic scale , then it ceases to be a rigorous result . as an approximation , it is then close to the spirit of effective medium theory ( wherein a representative particle is viewed as embedded in a continuum and its contribution to the continuum properties then calculated self - consistently ) . with macroscopic probe particles , the gser has been fruitfully employed @xcite to measure the rheological properties of viscoelastic fluids from light scattering or optical tracking observations of the suspended probes . in ref.@xcite , mason and weitz applied this method successfully to various systems , using colloidal spheres as probes . remarkably , in one case the medium analysed was itself a suspension of the colloidal particles , of volume fraction @xmath28 , which lies close to the glass transition concentration . the scattering method used was diffusion - wave spectroscopy ( dws ) @xcite whose direct interpretation in terms of self - diffusion is somewhat ambiguous ; but in any case , both criteria for the rigorous validity of the gser were violated ( no separation of length scales , and probe particles not dilute ) . yet the diffusion data for this system , analysed as though dws was probing self - diffusion and transformed into putative rheological data via the gser , compare remarkably well with frequency - dependent rheological data measured directly . among various explanations that are possible for this , is the idea that the gser is in fact more widely valid than generally assumed @xcite . on the other hand , in conventional glass - forming fluids , the ( zero - frequency ) tracer diffusion and viscosity have been measured as a function of temperature , by experiment @xcite and simulation @xcite . while the ser held ( within a factor of order unity ) for probes just a few times larger than the fluid particles @xcite , for equal - size probes the translational diffusion was enhanced by orders of magnitude with respect to the ser prediction as the glass transition was approached . in view of the unexpected success of gser for colloidal glassy fluids , alongside the failure of ser for conventional glasses , a more detailed examination of the connection between diffusion and rheology seems worthwhile , and we pursue this in what follows . we use a simple hopping model to emulate properties of a fluid close to its glass transition . the model , which will be referred to as the ` glassy rheology ' ( gr ) model , was defined in ref . @xcite , where its response to a macroscopic shear was calculated . in the present study we shall investigate the relation between rheology and particle transport in the gr model . the gr model is based on bouchaud s model for glassy dynamics @xcite , with the addition of shear strain degrees of freedom which that model lacks . particles in the bouchaud model are thermally activated at temperature @xmath29 from traps of energetic depth @xmath30 in a time @xmath31 which has an exponential distribution with rate @xmath32 , where @xmath33 is some microscopic rate constant . having escaped , a particle selects its next trap _ at random _ ( _ i.e . _ there are no spatial correlations ) from the _ prior distribution _ of trap depths @xmath1 . we briefly recall the resulting behaviour of bouchaud s model . for high temperatures @xmath0 , the occupancy of traps of depth @xmath30 evolves towards the boltzmann distribution @xmath34 . as @xmath0 is lowered , this distribution may cease to be normalizable , leading to a glass transition at @xmath35 . for @xmath36 , no equilibrium state exists , and the system shows `` weak ergodicity breaking '' and various ageing phenomena . a finite value of @xmath37 implies an exponential tail in the density of states , @xmath38 . the gr model ascribes some internal features to the traps . this is done through a minimalist description in which the tensorial aspects of the problem are ignored and the shear rate and strain are treated as though they were scalar quantities @xcite . the trap s potential is taken to be quadratic in a local strain ( or relative displacement ) coordinate @xmath39 , so that the particle s energy is @xmath40 , with @xmath41 a constant . this applies up to a local yield value @xmath42 , whose energy @xmath43 is the depth of the trap . upon reaching this threshold , the system locally rearranges to a new configuration , thus relaxing the local strain @xmath42 . the yield energy of the new potential well is drawn from the prior distribution @xmath1 , and its strain @xmath39 set to zero . in the absence of an externally imposed shear rate @xmath44 , all @xmath39 values remain zero , and the gr model reduces to the bouchaud model , with hopping between traps caused by thermal activation alone . with shear imposed , changes in all local strains are assumed to follow the global strain rate , @xmath45 . ( this assumption is mean - field in character . ) with the particles thus dragged up their quadratic energy wells by global shear , the barrier to thermal activation is reduced , and the local yield rate becomes @xmath46 $ ] . if @xmath47 is the probability of finding an occupied trap of given @xmath39 and @xmath30 , the above dynamics imply @xcite @xmath48 the first term on the r.h.s . arises from local straining by the macroscopic deformation . the second describes the loss of occupied traps from the distribution by thermally activated yielding . the last term corresponds to creation of new , unstrained configurations , whose trap depth is drawn from the prior @xmath1 , at a rate equal to the total yield rate in the system , @xmath49\right\rangle_p$ ] . note that , as originally conceived in the context of soft glassy materials @xcite each occupied trap represented a small region of the medium , large enough for a local shear strain @xmath39 to be defined , but small enough for this to be approximately uniform within the region . however , the model can readily be interpreted as describing the motion of individual particles in a conventional super - cooled / glass - forming fluid , by associating the local strain @xmath39 with , say , the relative displacement of a particle from the centre of a trap formed by its neighbours . we have not yet defined the stress in the gr model . due to stochastic yielding events , the local stress , which we take to be @xmath50 , is inhomogeneous ( it follows the local strain @xmath39 , rather than the macroscopic strain ) . since we have stated that the strain rate @xmath51 is everywhere equal to the global strain rate @xmath44 , these local stresses must combine according to a straightforward average so that @xmath52 . as discussed in section [ breakdown ] , this corresponds to a fully parallel mechanical circuit ; this is the only combination consistent with affine shear , and may well be approached in many physical systems . the dynamics described ( eq . [ eom ] ) are further motivated in refs.@xcite , where the gr model s constitutive relation between stress and strain rate is calculated . we now briefly summarize the linear rheology that the model predicts . we use non - dimensional units for time and energy by setting @xmath53 ; we also re - scale our strain variables ( @xmath54 ) and stress @xmath55 so that @xmath56 . the complex dynamic shear modulus @xmath57 describes the stress response to small shear strain perturbations around the equilibrium state . as such , it is well defined ( _ i.e . _ time - independent ) only above the glass transition , @xmath58 . expanding eq . ( [ eom ] ) to first order in the amplitude @xmath59 of an oscillatory strain @xmath60 , we find @xmath61 where the average is taken with respect to occupied traps , and @xmath62 is the mean ( arrhenius ) residence time for a trap of depth @xmath30 . ( we distinguish this from the actual residence time @xmath31 in such a trap , which is a random variable . ) equation [ grresult ] corresponds to a distribution of maxwell modes whose spectrum of relaxation times is given by the equilibrium distribution @xmath63 . given that @xmath1 has an exponential tail ( required for @xmath64 to be finite ) , the distribution of residence times thus exhibits power - law behaviour for large @xmath31 : @xmath65 . this leads to power laws , which may be apparent in conventional glasses , maxwellian behaviour re - appears at low @xmath9 . see section [ concl ] . ] for @xmath10 in the low frequency range : @xmath66 for @xmath67 the system is maxwell - like at low frequencies , whereas for @xmath68 there is an anomalous power law in the elastic modulus . most interesting is the regime @xmath69 , where @xmath70 and @xmath71 have constant ratio ; both vary as @xmath72 . moreover , the frequency exponent approaches zero as @xmath73 , resulting in essentially constant values of @xmath71 and @xmath70 . note , however , that the ratio @xmath74 becomes small as the glass transition is approached . this increasing dominance of the elastic response @xmath70 prefigures the onset of a yield stress for @xmath75 @xcite . however , for @xmath75 the linear viscoelastic moduli show slow time evolution and ageing effects @xcite in accordance with the weak ergodicity breaking discussed in section [ introd ] . accordingly , in the following section , where the gser for this model is discussed , we consider only @xmath76 . however , that does include the regime @xmath77 which would lie below the glass transition temperature " were this to be defined operationally by the divergence of the viscosity . as @xmath64 is approached from above , the equilibration time following a quick quench diverges . hence , even for @xmath78 , non - equilibrium situations are also of interest in glass - forming materials , though we shall consider only equilibrium situations in the following section . in order to study self - diffusion @xcite , we must clarify further the spatial interpretation of our gr model . we shall associate each relaxation event with a step on an unbiased random walk . we draw each step from a gaussian of unit variance in each spatial direction ( thus defining a length unit for the model ) . this gives a meaning to the square displacement , @xmath79 . after each step , the randomly walking particle is temporarily trapped within a potential well . within each well , we assume rapid ` thermalization ' of the particle , which therefore samples a gaussian distribution of displacements at each site on the walk . in fact , we assume this thermalization to be instantaneous , thus discarding information about high frequency intra - well dynamics . hence , we model the time regime of @xmath80-relaxation , but not of @xmath81-relaxation . we have the complex linear viscoelastic modulus for the gr model , quoted in the section above . it is now our task to compare this with the prediction of the gser , applied to a probe particle which is representative of the fluid itself . the result of eq . [ gser ] , whose validity is in doubt for a small probe of this kind , will be denoted @xmath82 . calculating @xmath82 from eq . [ gser ] requires a knowledge of @xmath25 in the equilibrium fluid . the brackets @xmath83 denote an average over all random walks of the particles , in terms of choice of path and of trap depths drawn from the prior distribution @xmath1 . in the absence of intra - well structure , the treatment of continuous - time random walks is discussed in refs.@xcite . the mean - square displacement can be found from the correlation function @xmath84 , defined by @xmath85 . this is calculable @xcite via @xmath86 where @xmath87 is the probability of having performed exactly @xmath88 hops at time @xmath89 , and @xmath90 is the fourier transform of the positional probability distribution after an @xmath88-step random walk . as described in section [ gr ] , the positional distribution is the convolution of a standard random walk and a gaussian of thermalization " ( whose variance is @xmath91 , where @xmath92 is the local spring constant , and @xmath93 the dimension of space ) . so its fourier transform @xmath90 is the _ product _ of the usual function @xmath94 and another gaussian . the quantity @xmath87 depends only on the distribution of residence times . note that the time between hops along a random walk is drawn from the distribution @xmath95 , which is _ not _ the distribution of residence times of occupied states in the system , @xmath96 . the former distribution gives the inter - hop times _ available _ to a random walker upon selecting its next trap . the latter gives the residence times of all the states in the system which are _ presently occupied_. thus @xmath96 is related to @xmath95 by a weight factor @xmath31 , since the likelihood of finding a given trap occupied is proportional to the time for which it is occupied : @xmath97 accordingly , averages of some stochastic quantity @xmath98 with respect to the two distributions are related by @xmath99 in terms of a typical random walk , @xmath100 is the average with respect to the steps , while @xmath101 is the average with respect to time . note that , in calculating the probability @xmath87 of having performed @xmath88 hops at time @xmath89 , the time spent in each trap is drawn from @xmath102 _ except for the first trap _ @xcite , which is selected from @xmath96 . this reflects the fact that the walker is not introduced to the system at time zero , but is selected at random from the traps already occupied . this careful choice of the first trap is important to the diffusive behaviour at early times @xcite and , as we shall see , its influence persists for increasingly long times as the glass transition is approached . from the definition of @xmath103 it follows that , in terms of its temporal laplace transform @xmath104 , @xmath105 where @xmath106 denotes a derivative in @xmath107-space . from this prescription , we find @xmath108 here , @xmath109 is the laplace transform of @xmath110 . the distribution of available escape times @xmath95 is given by averaging the exponential distribution for a single trap over the prior energy distribution @xmath1 thus : @xmath111.\ ] ] so its laplace transform is @xmath112 with @xmath113 the mean residence time for trap depth @xmath30 . the expression for @xmath114 is the same , with the average over hops replaced by an average over occupied traps . so , from eq.[averages ] , @xmath115 substituting this expression into eq . [ r2s ] cancels the @xmath116-dependence . so we see that the second moment of the distribution of displacements , @xmath117 is independent of any properties of the distribution of trapping times except for its mean . if we use this expression for the self - diffusion of a small probe to calculate the modulus in eq . [ gser ] ( and substitute @xmath118 to find the complex viscoelastic modulus in the fourier domain ) , we find and particle radius @xmath119 to simplify the constant of proportionality in the gser ( eq . [ gser ] ) . ] that the gser , applied to a small probe , predicts a complex modulus @xmath120 this is _ exactly _ maxwellian behaviour at all temperatures for which an equilibrium distribution exists ( _ i.e . _ for all ) . clearly , this expression is completely at odds with the actual rheology of the gr model as described in section [ gr ] . the generalized stokes - einstein relation , when applied to a small probe , has failed in two distinct ways which we now discuss . the first is a ( fairly trivial ) discrepancy in the temperature dependence , which is inherent in any hopping model . to study the spurious @xmath0-dependence of the characteristic time in eq . [ wrong ] , let us take the inverse laplace transform of eq . thus we find the nature of the probe s diffusive motion as a function of time : @xmath121 the time - independent term @xmath122 is the variance of the local ` gaussian of thermalization ' in which the probe finds itself at each step of its walk , and respects equipartition of energy amongst the elastic degrees of freedom . the second term says that diffusion is exactly linear on all time scales is equivalent to the infinite - dimensional limit of the exact result for a random trap model on a hyper - cubic lattice obtained by schroeder @xcite and kehr et al . @xcite . ] , with a diffusion constant @xmath123 the gser implies that the mean - square speed ( also given by equipartition , @xmath124 ) sets the rate of translation along the random walk . indeed this is why the diffusion constant is proportional to temperature in dilute gases , where particles are not caged , so their directions rapidly become decorrelated . like any activated hopping process @xcite with a temperature - independent microscopic attempt rate ( _ e.g . _ the eyring model of dense fluids @xcite ) , the gr model lacks such a factor @xmath0 in the diffusion constant ( eq . this is the source of the rogue @xmath0-dependence of @xmath125 . as the gr model is designed to be applied close to the glass transition , factors of @xmath0 are of order unity . this source of deviation from the gser is therefore not of great importance to most experiments , which usually only test it up to factors of order unity @xcite . however , the gser is violated by much larger factors than this arising from the second source of error in eq . [ wrong ] . this is the misplacement " of the thermal averaging brackets , when compared with the correct expression in eq . [ grresult ] . in a maxwellian fluid , the distribution of lifetimes is narrow , since there is a single characteristic relaxation time . thermal averaging then commutes with the other algebraic operations , so the incorrect placement of the thermal brackets becomes irrelevant in that case . however , the gser result breaks down for broad distributions such as exist both in many real glass formers , and in the gr model . to illustrate this , let us compare the diffusion constant ( eq . [ d ] ) with that predicted by the stokes - einstein relation ( zero - frequency limit of the gser ) , given the gr model s actual viscosity . from eq . [ grresult ] with @xmath27 , the ser gives the diffusion constant as @xmath126 . the first moment of the distribution of times for which traps are occupied @xmath127 diverges in the gr model for temperatures below @xmath128 , whereas @xmath129 , which appears in eqs . [ d ] and [ wrong ] , remains finite . thus @xmath20 is _ enhanced _ with respect to the ser value as the viscous divergence is approached , as is widely observed in experiment and simulation @xcite . we note , however , that the gr model is not sufficiently elaborate to account for the power law ( @xmath130 ) which has been observed to replace the ser in some supercooled fluids @xcite . the rheology of the gr model is equivalent to a set of over - damped harmonic oscillators ( _ e.g_. masses on springs in dash - pots ) connected in _ parallel_. each over - damped element has the characteristics of a maxwellian fluid , with a different time constant . even if only a vanishingly small fraction of the population of maxwell models becomes infinitely viscous ( _ i.e . _ their time constants @xmath131 diverge ) , it will be apparent in the response of the system , whose stress is the _ sum _ of stresses of the population ) , will be different in many real systems , becoming a more elaborate network of series and parallel connections . ] . contrast this with the gser for a small probe , eq . [ wrong ] . it claims to derive the moduli from a knowledge of only the second moment , @xmath132 , of the distribution of displacements . this is ( qualitatively ) more like summing _ compliances _ of the population . this discussion shows that the second moment alone , @xmath132 , does not contain sufficient information about the distribution of displacements ( or , equivalently , about the full distributions @xmath110 ) to find the rheological behaviour of the gr model . specifically , this moment would be unaffected by a vanishingly small subset of the population being stuck for infinite time at @xmath133 ; yet this subset may dominate the rheology , _ e.g_. by causing @xmath127 to diverge . as we have mentioned , the gr model shows some intriguing dynamical features . the viscosity diverges at a temperature ( @xmath128 ) well above the glass transition temperature ( @xmath134 defined as the temperature below which the system has no equilibrium steady state ) . the diffusion constant @xmath20 vanishes at @xmath64 , and yet not all particles are static below this temperature . in this section we gain a fuller picture of the model s dynamics , by investigating its hopping statistics analytically and by simulation . let us begin by comparing the two distributions @xmath95 and @xmath96 , for the residence times of particles in a trap . the first is taken over the prior distribution ( of _ a priori _ available states ) ; the second over those states which are actually occupied at any given time . as stated in section [ test ] , these two distributions are so different that selecting just the first step on a random walk from the wrong distribution can have drastic effects on @xmath135 which persist up to rather late times . suppose that by mistake the _ initial _ residence times were drawn from the distribution @xmath95 ( which is the correct choice for all subsequent residence times ) , rather than from @xmath96 ( which is correct for the first one only @xcite ) . the mean square displacement @xmath136 over this hypothetical ensemble has laplace transform @xmath137 which , in contrast to eq . [ r2s2 ] , does not depend only on the mean of the distribution @xmath95 . the resultant @xmath136 is a non - linear function of time ( tending to eq . [ r2 t ] as @xmath138 @xcite ) which increases monotonically , even at @xmath139 ( where the true diffusion constant vanishes ) . the correct value of @xmath135 given in eq . [ r2 t ] is not found as just outlined , but instead by weighting each walk in the ensemble with the residence time of the trap it is initially in @xcite . this factor is required because the probability of a particle initially being found in a given trap is proportional to the residence time of that trap . ( since we are interested in a system which has already equilibrated , the particles have , in principle , had time to sample the distribution of traps , and are therefore more likely to be found in long - lived states than _ a priori_. ) it is somewhat unintuitive ( though necessary as a consequence of the system s steady state ) that this re - weighting of the ensemble should conspire to cancel all non - linearity from the time dependence of mean square displacements implicit in eq . [ r2prior ] and yield eq . [ r2 t ] instead . for the exponential prior @xmath140 , it can be deduced straightforwardly from eqs . [ laplace ] and [ averages ] that , in laplace time , @xmath141 from which it follows , showing temperature dependence explicitly , that @xmath142 thus , at equilibrium , the residence time distribution of occupied states coincides with the prior distribution of such times at a substantially lower temperature . to observe the process in action , we simulated the activated hopping of @xmath143 particles , using the exponential prior distribution of trap depths @xmath140 . for each particle , the time interval for successive hops was drawn from @xmath95 ( eq.[psioftau ] ) for all but the first hop ( as in @xcite ) . the algorithm used to simulate hopping in the system between times @xmath144 and @xmath89 is as follows . each of the @xmath143 particles is initially scheduled with a hopping time drawn from @xmath96 . if this is greater than @xmath89 , the particle is not visited again , and is recorded as performing zero hops during the simulation . the remaining particles are successively assigned further relaxation times is distributed according to @xmath95 if @xmath145 and @xmath146 are distributed uniformly on @xmath147 . ] drawn from @xmath95 and their numbers of hops incremented , until the next hop is scheduled for after @xmath89 . in this way the distribution @xmath87 of the number @xmath88 of relaxation events up to time @xmath89 was measured , as a function of temperature . the distribution of hops @xmath87 gives a clearer picture of dynamics in the system than the distribution of displacements , which is just its convolution with a random walk for which @xmath148 . the results appear in fig . [ p(n ) ] for a range of temperatures and times . we see that , at high temperature , all the particles are mobile and concentrated in a peak whose position increases linearly with time . as the temperature is lowered towards the glass transition ( @xmath149 ) , a second peak containing a significant fraction of the population appears at @xmath150 . of course , as this peak decays , it feeds the rest of the distribution . its lifetime increases as the transition is approached , yet the rest of the distribution , for which it is the source , remains concentrated in a mobile peak . we consider next the time dependence of @xmath151 , which is the probability that the time of the first hop exceeds @xmath89 : @xmath152 using eq . [ specialpsi ] and expressing @xmath95 in terms of an incomplete gamma function , @xmath153 ( from eq . [ psioftau ] with @xmath154 ) , we have @xmath155 so the population of stuck particles decays by an increasingly slow power law as the glass transition is approached . of course , this decay is fastest at early times , so this is when the maximum in the _ mobile _ peak is formed . the width of the mobile peak grows with the usual square - root relation to its mean so , at late times , the population is concentrated in two distinct and narrow regions . this confirms the result of the above simulation , that the diffusion process in the gr model leads , near the glass transition , to a strongly bimodal distribution of displacements . having hopped once , a particle is more likely to hop again in a given time , since relaxations in the mobile peak are drawn from the ( effectively higher - temperature eq . [ specialpsi ] ) hop distribution @xmath95 . nevertheless , the distribution of residence times throughout the whole system ( both peaks ) is @xmath96 . furthermore , the simulational data confirm that , although the mean of the mobile peak is a non - trivial function of time and temperature ( cf . eq.[r2prior ] ) , the two peaks together conspire to produce @xmath156 in agreement with eqs . [ r2 t ] and [ d ] . in ref . @xcite , such linearity in time was shown to hold in the steady state of any system of random walkers . nevertheless , in our highly bimodal distribution , it appears somewhat remarkable . that the diffusion coefficient " @xmath20 should actually comprise an average over two such distinct populations might call into question whether the dynamics of the model can properly be called diffusion at all : the mean square displacement may be linear in time , but is this diffusion in the ordinary sense ? to address the question , we calculate the _ variance _ in @xmath79 , that is @xmath157 , and find , for @xmath69 , @xmath158 the leading term is the standard value for simple diffusion . the way in which this limit is approached _ is _ temperature dependent , but such sub - dominant terms are unlikely to impinge on experimental measurements of the diffusion constant in the ergodic ( @xmath58 ) system . the deviation from a gaussian distribution of @xmath159 at finite times is often characterized by the ` non - gaussian parameter ' @xmath160 defined @xcite as @xmath161 in 3 dimensions . above the glass transition , this is expected @xcite to vanish in the infinite - time limit . for @xmath69 , we find @xmath162 whereas @xmath163 above @xmath128 . so the non - gaussian parameter in the gr ( and bouchaud s ) model does indeed vanish , but does so increasingly slowly as the glass transition is approached . the same qualitative results were observed by ( amongst others ) miyagawa et al . in md simulations @xcite . we conclude this section on unforced diffusion in the gr model by summarizing the important time - scales that are present in both diffusion and rheology . the diffusion is controlled by @xmath164 , equivalent to the _ mean rate _ , @xmath129 , which vanishes linearly at @xmath165 . mean time _ between yielding events in the system , @xmath127 controls the gr model s viscosity , and diverges at @xmath128 , although the mean hopping rate is finite here . the _ median _ of @xmath96 also remains finite below @xmath128 . from eq.[p(0 ) ] , it is the time at which half the population has relaxed ( when @xmath166 ) . from eq . [ decay ] this is @xmath167^{\frac{1}{t-1}}\ ] ] which has a rather strong divergence at the glass transition ( @xmath149 ) . meanwhile , the _ mode _ of the distributions @xmath95 and @xmath96 , _ i.e_. the most likely time between hops , is always zero . given these subtle statistical properties of the self - diffusion process , which arise from the power law distribution of relaxation times in the gr model , it is no surprise that the gser ( which from eq . [ d ] contains only @xmath168 ) fails for microscopic probes . so far we have explored only the _ linear _ response functions of the gr model , @xmath169 and @xmath170 , and their interrelation . these describe any system for which the applied shear stress is either zero or so small as not to disturb thermal equilibrium . let us now turn our attention to applied shear rates which are sufficiently large to influence self diffusion , and calculate the effect . of course , particles in brownian motion are also convected by the system s affine shear motion , but this effect ( in which we include taylor dispersion ) can be subtracted from the overall trajectory to reveal the perturbed diffusive contribution . ( in simple shear , no such subtraction is needed for the component of brownian motion transverse to the flow . ) the rheology of the gr model @xcite exhibits strong nonlinearities arising from shear - induced hops . these occur as a particle is strained within its well , and moves up the energy curve toward the yield value , making a thermal hop far more likely . clearly such shear - induced hops could have a radical effect on the diffusive properties too . we assume that , although biased by the affine motion , each such hop still entails a random displacement whose statistics is the same as when flow is absent . after subtraction of the affine motion , the diffusion constant is defined by @xmath171 in @xmath93 dimensions . for a random walk of @xmath88 hops , @xmath148 so , to calculate @xmath20 , we just require the number of hops executed by an average particle in a given long time . that is , @xmath172 since the time spent in each trap along the walk , which arises from the microscopic treatment , is completely equivalent to the time - weighted average @xmath173 which arises from the fokker - planck equation used in refs . [ averages ] . ] is drawn from @xmath95 , and each random walk becomes typical after a sufficiently long time . we now evaluate eq . [ diffusion ] in the presence of shear . given a time - dependent macroscopic shear @xmath174 , the yield rate for a trap which was entered at time @xmath175 was prescribed in section [ gr ] as @xmath176 ^ 2)\}\ ] ] since the trap has subsequently been sheared by an amount @xmath177 . at time @xmath89 , the probability of _ not _ having hopped from the trap of depth @xmath30 entered at @xmath175 ( which we call @xmath178 ) is therefore @xmath179 ^ 2 \right ) \right\}\ ] ] where @xmath180 and @xmath181 as before . integrating over the prior distribution of trap depths @xmath1 gives the overall survival probability , @xmath182 ^ 2 \right ) \right\}. \nonumber\end{aligned}\ ] ] the conditional probability density @xmath183 of hopping next at time @xmath89 , given that the present trap was entered at @xmath175 is then @xmath184 hence , the mean yield time , _ given that the present trap was entered at time @xmath175 _ is @xmath185 we next apply eqs . [ diffusion ] , [ f ] and [ timedependent ] to two particular forms of @xmath174 : steady shear , @xmath186 , and periodic shear , @xmath187 . with a time - independent shear rate , the system reaches a steady state . ( this is true for all temperatures , including those below @xmath3 : the presence of a finite shear rate destroys weak ergodicity breaking @xcite . ) therefore , the mean relaxation time becomes independent of the time of entry into a trap . thus eqs . [ diffusion ] and [ timedependent ] are related by @xmath188 for any @xmath175 we might choose . thus we obtain @xmath189 from the substitutions @xmath190 and @xmath191 . here , @xmath192 is a re - scaled shear rate . to proceed further we must approximate @xmath193 ( eq . [ j ] ) . let us define @xmath194 by @xmath195 . then for @xmath196 , @xmath197 grows rapidly with @xmath198 , while for @xmath199 it is small . so we can write @xmath200 this approximation is equivalent to saying that shear does not greatly modify the thermal activation rate while a particle is in the bottom of a quadratic well , but yielding is immediate once the particle has been sheared to the brink of its trap . the result is @xmath201 where @xmath202 is the inverse of @xmath203 in eq . we now approximate this inverse function by @xmath204 which is exact in the limits @xmath205 and @xmath206 . the constant @xmath207 solves @xmath208 and is the ( second , creates an upturn in the gradient of @xmath202 , generating artefacts . ] ) point of intersection of the two approximate functions . in fact , it can be shown that @xmath209 is always greater than @xmath193 in eq . [ steadyd ] which it approximates . so , the resulting approximation for @xmath20 is a strict _ lower bound_. general expressions for this are given in appendix [ steadyapp ] . two regimes of the scaled shear rate arise : @xmath210 and @xmath211 . using the exponential prior @xmath1 we find , in the limit of high shear rate , @xmath212 so , for fast shear , the diffusion rate is almost proportional to the shear rate . the imposed shear drags fluid elements rapidly through trap configurations , with very little time for thermal activation . more interesting behaviour occurs in the low - shear regime . with the exponential prior , the approximate ( lower bound ) @xmath20 is ( for @xmath213 see appendix [ steadyapp ] ) @xmath214^{-1 } \hspace{12pt } \mbox { for } \kappa < y_{0}.\ ] ] the function @xmath215 is given in fig . [ cfig ] . below the glass transition ( @xmath75 ) , the diffusion constant is zero in the absence of shear , and particle motion is sub - diffusive , with @xmath132 growing as some non - integer power of time @xcite . however , any finite steady shear rate ensures a steady rate of yielding events and a non - zero diffusion constant . in this regime , from eq.[bound ] , @xmath216 thus , @xmath20 is a finite , smooth function throughout all of the ( @xmath0,@xmath44 ) plane , except for the line segment @xmath217 , where it vanishes . for non - zero shear rate , there is no singularity at @xmath149 , where @xmath20 becomes @xmath218 for low scaled shear rate @xmath192 . a planar periodic shear strain , @xmath219 , is experimentally more easily realizable than a constant shear rate . if we define the phase @xmath220 to be @xmath221 _ modulo _ @xmath222 , then we can express the mean time @xmath223 spent in each trap on the walk ( defined in eq . [ timedependent ] ) as a function , @xmath224 , only of the _ phase _ at which the current trap was entered . this is done in appendix [ tau - and - p ] by averaging @xmath223 over all oscillatory cycles , for all times @xmath175 corresponding to phase @xmath220 . if @xmath225 is the probability that a trap was entered at phase @xmath220 of the cycle , then the mean residence time in eq . [ diffusion ] is @xmath226 let @xmath227 be the _ conditional _ probability density of a hop occurring at phase @xmath220 , _ given _ that the previous hop was at @xmath228 . then @xmath225 is the ( normalized ) solution to the integral equation @xmath229 the quantities @xmath227 and @xmath224 have simple relations to the distribution @xmath183 given in eq . [ conditional ] . in appendix [ tau - and - p ] they are given in terms of a single cycle of the imposed waveform . equation [ fredholm ] is a homogeneous fredholm equation of the second kind , for which the general solution is not known . however , the problem greatly simplifies in the limit of small amplitude oscillations , which we now treat . ( note that this perturbative treatment is only possible for @xmath58 , where an unperturbed steady state exists . ) in the absence of shear @xmath225 and @xmath224 must be independent of @xmath220 . in that case , we write @xmath230 and @xmath231 . if the oscillatory shear creates only small perturbations @xmath232 and @xmath233 , then eq . [ tau - periodic ] , to first order in small quantities , becomes @xmath234 we have used the fact that @xmath235 is normalized , so that @xmath236 . hence we need only expand eq . [ tauexpression ] to lowest order in strain amplitude and average it over a cycle to find @xmath168 and hence the diffusion constant . the result for an exponential prior @xmath140 and sinusoidal shear of amplitude @xmath59 is @xmath237 for @xmath213 , the integral is convergent as @xmath238 . in fact , for _ any _ waveform , the low - frequency limit for an exponential prior is @xmath239 where , in the sinusoidal case , @xmath240 $ ] . in the high - frequency limit , [ sinusoidal ] yields instead @xmath241 for larger amplitudes , we can use the results of section [ steadyshear ] . given a sufficiently low frequency ( @xmath242 ) , the shear rate at any given point on the cycle is almost constant . in that case , the steady - shear expressions for @xmath243 can be averaged over a sinusoidal cycle , yielding ( for @xmath242 ) @xmath244^{-1 } ; \hspace{5pt } \gamma_{0}^2 \omega^2 < 2t\ ] ] and , to within logarithmic factors , @xmath245 where @xmath246 , @xmath247 is given in fig . [ cfig ] and @xmath248 is euler s integral of the second kind . we now summarize the effects , calculated above , of low - frequency periodic shear on diffusion just above the glass transition . for the smallest amplitudes @xmath59 , we expect the change in the diffusion constant to be proportional to @xmath249 . at larger amplitude , this should cross over to @xmath250 , and at the largest amplitudes , @xmath20 is almost proportional to @xmath251 . we interpret these three regimes as follows . the smallest oscillations serve only to vibrate particles within the bottoms of their quadratic potential wells , thus effectively making all wells just a little shallower . larger amplitude strains cause yielding of even the deepest wells in the time taken for the affine shear to reach the yield point . however , the part of the population in the shallowest wells are thermally activated more quickly than this affine shear time - scale . the dividing line between those two parts of the population depends on the shear rate . at the highest shear rates , the global affine shear accounts for almost all yielding . we have studied quantitatively , within a simple hopping model of glassy dynamics , the relation between diffusion and rheological responses ( in the linear response regime ) and ( for the nonlinear regime ) the coupling between these two aspects of the dynamics . in the presence of a broad distribution of relaxation times , as the model possesses ( and as is generic in hopping models of the glass transition ) one can expect strong violations of the generalized stokes - einstein relation ( eq.[gser ] ) when applied @xcite to the self - diffusion of representative particles in the medium ( small probes ) . the relation holds only for probe particles large enough that their surroundings are properly viewed as a continuum @xcite . effective - medium theories , and also simple forms of mode - coupling theory @xcite ( in which a single mode of slow relaxation dominates ) can give misleading predictions under these conditions . for glassy systems , use of the gser will under - predict the diffusivity of particles and/or the rheological viscoelastic moduli . this is because the latter are dominated by the most immobile and the former is dominated by the most mobile particles . in fact , the nature of diffusion in the model is quite subtle ( see section [ analysis ] ) : close to ( but above ) the glass transition one has an apparently bimodal behaviour . starting from any initial equilibrium state , a fraction of particles remain stuck for a long time , but any particle that has hopped once remains mobile thereafter . it is peculiar that , despite this , the diffusive behaviour conspires to be relatively normal ( at least for low order moments of the displacement distribution ; moments of fractional negative order would presumably reveal a different story ) . an additional peculiarity of diffusion , which our model does not include , but which could also lead to enhanced diffusivity near the glass temperature , has been observed in a recent molecular dynamics simulation @xcite . in ref.@xcite the effective dimensionality of the most mobile random walks was observed to decrease with temperature . at low temperature , particles moved along string - like clusters ( also seen in ref . @xcite ) , and therefore covered greater distances than in an uncorrelated random walk in three dimensions . we have also applied our gr model to calculate the change in the self - diffusion constant when a material is sheared . after subtraction of affine motion ( including taylor dispersion ) @xcite , or equivalently restricting attention to diffusion perpendicular to the shear direction , one finds a strong effect of imposed flow on the mean jump rate @xcite and hence on the diffusion constant . the effect is particularly extreme below the glass transition , where an anomalous ( sub - diffusive ) behaviour is converted to a finite diffusivity which has , instead , a power law dependence on the steady shear rate . such effects should be accessible in scattering experiments ( with wavevector almost perpendicular to the flow direction ) on labelled small probes . in oscillatory shear , an enhancement in @xmath20 was predicted in section [ shear ] for systems above ( but near ) the glass temperature ; this could also be probed via scattering . the gr model as described in section [ gr ] was originally developed in ref . @xcite to reproduce the generic rheology of a class of soft glassy materials " . ( we comment further below on its relevance to conventional glasses . ) this class was argued to include , for example , foams , emulsions , pastes and slurries . experimentally , their linear viscoelastic behaviour is often characterized by a nearly constant ratio of the elastic and loss moduli @xmath7 , @xmath8 ( @xmath252 is usually about 0.1 ) with a frequency dependence that is either a weak power law ( clay slurries , paints , microgels ) or negligible ( tomato paste , dense emulsions , dense multi - layer vesicles ) @xcite . this behaviour persists down to the lowest experimentally accessible frequencies . sometimes a regime is seen at small @xmath9 where @xmath70 is constant and @xmath71 is decreasing ( which can be interpreted in terms of the model s behaviour below @xmath3 @xcite ) . as mentioned in sections [ introd ] and [ gr ] , there are two differences between the gr model used in this paper and its soft counterpart in @xcite . the first is that the _ soft _ gr model refers not to particles , as we have done , but to mesoscopic material elements " , large enough for a local elastic strain variable to be defined but small enough to have strong heterogeneity in local yield energies . ( for the case of a foam , say , an element " could correspond to a domain of several bubbles , and yield " to a local topological change . ) it is not necessarily clear what is meant by self - diffusion of such elements , so in the present paper we have retained a particle picture , although this is more natural for conventional glass - forming liquids than for soft glassy materials . the second difference is that in the _ soft _ gr model , @xmath0 is replaced by an effective noise temperature " @xmath253 . it was argued in @xcite that the resemblance to thermal activation is formal : the activated " yield processes are viewed as arising primarily by coupling to structural rearrangements elsewhere in the system . indeed , the elastic energies associated with local rearrangements in foams and the like are many orders of magnitude in excess of @xmath26 , so any interpretation of @xmath253 as a true temperature is somewhat unconvincing for these materials ( in contrast to conventional glasses ) . although this interpretation remains problematical , as discussed in @xcite , with it the soft gr model is able to reproduce many of the rheological properties of soft glassy materials . the results we have obtained for the gr model , concerning breakdown of the gser and the nature of shear - induced diffusion , equally apply ( with @xmath254 ) to the _ soft _ gr model . the breakdown of gser , arising from the fact that diffusion and rheology probe different aspects of the relaxation spectrum , is equally natural in this case ; the other contribution to its breakdown discussed in [ breakdown ] , arising from the temperature dependence of the local attempt frequency , even more so ( since @xmath253 is anyway not a true temperature ) . the results of section [ shear ] for shear - induced diffusion could also be quite interesting for soft glassy materials , in which it is easy to apply shear strains large enough to strongly perturb the intrinsic relaxation times . the length scales in these materials can be probed via light scattering ; index matching is often possible so that true tracer diffusion ( of a small subset of unmatched droplets or particles ) can be measured @xcite . for oscillatory flows , an important innovation is the echo technique @xcite in which the positions of a given scatter at identical points in the shear cycle are compared . results on dense emulsions @xcite suggest , in fact , that not only are shear - induced reorganizations easily detectable , but that these have strong temporal and spatial correlations regions that reorganize at a given point in the shear cycle will do so again in the next one . such correlations are not included in the ( soft ) gr model(s ) although they might be added in principle @xcite . ( it would require escape from a shallow trap to be preferentially into another shallow trap , though we do not advocate appending such ad - hoc correlations to this simplified model . ) it would be very interesting to know whether the same applies in conventional glasses ; preliminary work on colloidal suspensions ( which are traditionally thought of in these terms ) suggests not @xcite . finally , we return to the phenomenology of the gr model in relation to conventional glass - forming liquids . as explained in section [ gr]b , the gr model in its basic form predicts a viscosity divergence at @xmath128 , while the diffusion constant vanishes only at @xmath149 , where the system has a glass transition to a non - ergodic state . this existence of a temperature range with finite diffusivity but infinite viscosity appears to be at odds with the experimentally observed behaviour of conventional glasses . a common divergence of viscosity and inverse diffusion constant can however be incorporated into the gr model through a cutoff @xmath255 on the distribution of yield energies @xmath1 ; this modification of the model has in fact already been discussed in ref . it yields a viscosity which initially follows the original power - law divergence as @xmath128 is approached , but then crosses over to @xmath256 as @xmath0 is lowered further . similarly , the predicted @xmath257 would first seem to diverge as @xmath149 is approached from above , but actually remain finite there and eventually approach infinity at the same temperature as the viscosity ( @xmath258 ) . the shear moduli obey the power laws ( eq . [ powerlaws ] ) down to a cutoff frequency @xmath259 , but then cross over to low - frequency maxwell behaviour ( @xmath260 , @xmath261 ) . the above simple temperature dependences apply if , as we did throughout , we assume that the ( prior ) density of yield energies is temperature independent . this is of course an approximation ; odagaki , for example , suggested that the width of @xmath1 may in fact scale as the inverse of the amount of free volume @xmath262 in the system @xcite . in all temperature dependences , @xmath263 then replaces @xmath0 . as a consequence , if the free volume decreases smoothly to zero at a finite temperature @xmath264 , a vogel - fulcher - like divergence of @xmath16 and @xmath257 at @xmath264 ( rather than the above arrhenius behaviour ) would be predicted by the gr model with energy cutoff . one would then be inclined to locate the glass transition at that point ; if we revert to temperature independent @xmath1 , this corresponds to @xmath258 ( rather than @xmath149 , which is the appropriate choice in the absence of an energy cutoff ) . results in the range @xmath265 for the gr model with cutoff may therefore actually apply to supercooled liquids _ above _ the glass transition . these include a dynamic modulus @xmath266 which becomes _ more _ maxwellian in the low - frequency range as @xmath0 is lowered ( for @xmath75 , one has @xmath267 . and @xmath268 above the cutoff frequency @xmath269 @xcite , and hence maxwell behaviour with relaxation time @xmath270 for @xmath271 ) , in an intriguing correspondence with data taken by menon et al . @xcite . this work was funded by epsrc grant no . gr / k56025 ( rmle ) and a royal society dorothy hodgkin fellowship ( ps ) . we wish to thank p. n. pusey , f. lequeux , p. hbraud and j .- p . bouchaud for helpful discussions . substituting eqs . [ approxj ] and [ approxi ] into [ steadyd ] yields lower bounds on @xmath20 ( upper bounds on @xmath243 ) in two regimes : @xmath272 in each case , @xmath20 is approximately equal to the lower bound for @xmath213 . the probability of leaving a trap at time @xmath89 given that it was entered at @xmath175 ( at phase @xmath273 of the oscillatory cycle ) is @xmath183 . hence the mean time spent in the trap , given that it was entered at phase @xmath228 , is @xmath274 which , from eq . [ conditional ] , gives @xmath275 with @xmath276 defined in eq . [ f ] . for compact notation , we define @xmath277 and @xmath278 according to @xmath279 thus we find @xmath280 ^ 2 \right\ } \nonumber\end{aligned}\ ] ] since @xmath281 ^ 2 $ ] is a periodic function of @xmath220 , the integrand in eq . [ workings ] increases by a constant factor with each cycle @xmath282 . thus the r.h.s . is a geometric series of integrals over a single cycle , @xmath283 ^ 2 \right ) \right]^n \right\ } \ , \times \\ & & \int_{\phi_{0}}^{\phi_{0}+2\pi } \!\!\!\!\ ! d\phi \ , \exp \left\ { -u \int_{\phi_{0}}^{\phi } \!\ ! d\phi ' \ , \exp[a(\phi')-a(\phi_{0})]^2 \right\}.\end{aligned}\ ] ] finally , @xmath284 is expressed in terms of a single cycle as @xmath285 ^ 2 \right\ } } { 1 - \exp\left\ { -u\int_{0}^{2\pi } \!\ ! d\phi ' \ , \exp [ a(\phi')-a(\phi_{0 } ) ] ^2 \right\ } } \nonumber\end{aligned}\ ] ] where @xmath286 is the heaviside step function . similarly , @xmath287-\phi ) \;\ , \psi(t\,|\,\phi_{0}/\omega ) \ ; dt\ ] ] ( with @xmath288 the dirac delta function ) is also a geometric series , from which it follows @xmath289 ^ 2 \right\ } } { 1 - \exp\left\ { -u\int_{0}^{2\pi } \!\ ! d\phi ' \ , \exp [ a(\phi')-a(\phi_{0 } ) ] ^2 \right\ } } . \nonumber\end{aligned}\ ] ] w. kob , in _ experimental and theoretical approaches to supercooled liquids : advances and novel applications _ , edited by j. fourkas , d. kivelson , u. mohanty and k. nelson ( acs books , washington , 1997 ) ; w. gtze , in _ liquids , freezing and glass transition , les houches session li _ , edited by j .- p . hansen , d. levesque and j. zinn - justin , ( elsevier science pub . , 1989 ) , chap . 5 ; w. gtze and l. sjgren , z. phys . b condens.mat . * 65 * , 415 ( 1987 ) . k. schroeder , in _ proceedings of the international conference on fundamental aspects of radiation damage in metals , erda conf751006 , volume 1 _ , edited by m.t . robinson and f.w . young jr . ( 1975 ) , p. 525 . see also k.w . kehr and t. wichmann , materials sci . forum * 223 * , 151 ( 1996 ) . m. r. mackley , r. t. j. marshal , j. b. a. f. smeulders and f. d. zhao , chem . sci . * 49 * , 2551 ( 1994 ) ; r. j. ketz , r. k. prudhomme and w. w. graessley , rheol . acta * 27 * , 531 ( 1988 ) ; s. a. khan , c. a. schnepper and r. c. armstrong , j. rheol . * 32 * , 69 ( 1988 ) ; t. g. mason , j. bibette and d. a. weitz , phys . * 75 * , 2051 ( 1995 ) ; p. panizza _ et al . , _ langmuir * 12 * , 248 ( 1996 ) ; h. hoffmann and a. rauscher , coll . polymer sci . * 271 * , 390 ( 1993 ) . w. van megen , s. m. underwood and i snook , j. chem . phys . , 4065 ( 1986 ) ; w. van megen and s. m. underwood , j. chem . * 88 * , 7841 ( 1988 ) ; w. van megen and s. m. underwood , j. chem . * 91 * , 552 ( 1989 ) . for recent work on the glassy dynamics of hard sphere systems , see also van megen , transport theory and stat . * 24 * , 1017 ( 1995 ) and references therein .

using traditional morphological classifications of galaxies in 10 intermediate - redshift ( z 0.5 ) clusters observed with wfpc-2 on the hubble space telescope , we derive relations between morphology and local galaxy density similar to that found by dressler for low - redshift clusters . taken collectively , the `` morphology - density '' relationship , , for these more distant , presumably younger clusters is qualitatively similar to that found for the local sample , but a detailed comparison shows two substantial differences : ( 1 ) for the clusters in our sample , the relation is strong in centrally concentrated `` regular '' clusters , those with a strong correlation of radius and surface density , but nearly absent for clusters that are less concentrated and irregular , in contrast to the situation for low redshift clusters where a strong relation has been found for both . ( 2 ) in every cluster the fraction of elliptical galaxies is as large or larger than in low - redshift clusters , but the s0 fraction is 2 - 3 times smaller , with a proportional increase of the spiral fraction . straightforward , though probably not unique , interpretations of these observations are ( 1 ) morphological segregation proceeds hierarchically , affecting richer , denser groups of galaxies earlier , and ( 2 ) the formation of elliptical galaxies predates the formation of rich clusters , and occurs instead in the loose - group phase or even earlier , but s0 s are generated in large numbers only after cluster virialization .

astro-ph9707232

the observation that galaxy populations in clusters differ from those in the field goes back at least as far as hubble & humason ( 1931 ) . both spitzer & baade ( 1951 ) and gunn & gott ( 1972 ) suggested that dynamical processes within clusters might be responsible for transforming populations of spiral galaxies into s0 s . in a study of 10 rich clusters , oemler ( 1974 ) found that the cluster population was a function of both cluster structure and location within the cluster . regular , centrally concentrated clusters , whose appearance suggests that they are dynamically relaxed , have large populations of e s and s0 s , small numbers of spirals , and strong radial population gradients , with the ratio of early - to - late - type galaxies increasing towards the center . irregular , unrelaxed - looking clusters have populations that are much more spiral - rich and show no signs of radial gradients . oemler suggested , following gunn and gott , that the high s0/spiral ratio in relaxed clusters was due to dynamical processes within the clusters , but that the enhanced abundance of e s in the most centrally concentrated clusters , which are often dominated by cd galaxies , was due to an enhanced formation rate of ellipticals at early times in the densest environments that would later become the cores of rich clusters . dressler ( 1980a , hereafter d80 ) used data for @xmath06000 galaxies ( dressler 1980b , hereafter dcat80 ) to investigate correlations between morphological type , cluster properties , and spatial distribution in 55 low - redshift rich clusters . dressler correlated the spiral , s0 , and elliptical fractions with the _ local _ surface density of galaxies , defined in a rectangular area containing the 10 nearest neighbors . he found a smooth , monotonic relation , now commonly known as the _ morphology density relation _ , actually a correlation with surface density , which we will refer to here as the relation . dressler concluded that , to first order at least , the relationship was universal , that is , representative of every cluster in the sample , regardless of its global properties . ( indeed , other workers ( e.g. geller & beers 1982 ) have claimed that it extends beyond cluster environments into the general field population . ) of particular importance was dressler s finding that , while irregular clusters showed no radial segregation of cluster populations , as oemler had shown , they did display a relation as strong as that of regular , concentrated clusters . of course , the finite number of galaxies in each cluster prevents the testing of this hypothesis on a case - by - case basis , so that this conclusion was reached by combining together the data for regular , centrally - concentrated clusters and comparing them to those of irregular , clusters of low central concentration . both oemler s and dressler s portrayals of cluster populations are consistent with environment playing a critical role in determining present galaxy populations . indeed , for relaxed , centrally - concentrated clusters the two descriptions are equivalent , because there is a one - to - one correspondence between radial position and local density in such clusters . however , the difference for irregular clusters is significant . were irregular clusters to show little segregation by type , it would be reasonable to conclude that evolutionary processes which transform galaxies only work in the environment of a massive , relaxed system . on the other hand , finding local correlations between environment and populations even in small , cold subsystems suggests that much more local processes are at work . the purpose of this paper is to investigate these issues in clusters at higher redshift , @xmath1 , in order to trace the evolution of galaxy populations in clusters , specifically the processes that might alter galaxy types and/or be responsible for their spatial distribution within clusters . this paper is one of a series analyzing the data obtained by the `` morphs '' group using the wfpc-2 on the hubble space telescope . we have imaged 11 fields in 10 clusters of galaxies in the redshift range @xmath2 and cataloged the positions , photometric data , and morphological classifications for 1857 objects brighter than @xmath3 or @xmath4 , as described in smail et al . ( 1996a , hereafter , s97a ) . although the addition of further parameters derived from spectroscopy of these galaxies , such as cluster or field membership , internal dynamics , stellar population , and cluster dynamics , are likely to be revealing as to the evolutionary state of these populations , photometric / morphological information alone allows a simple and important comparison with the properties of present day clusters . this comparison offers clues as to how clusters of galaxies came to hold their atypical complements of galaxy types . the paper is organized as follows : in section 2 we briefly review the data used for this study . in section 3 we revisit the relation between morphology and surface density introduced in d80 . we include a short discussion of subsequent challenges to the original work , which is further explored in the appendix . section 4 gives our results for the higher - redshift sample , and in section 5 we suggest some possible conclusions about galaxy evolution in clusters that might be drawn from our results . as described in s97a , our data consist of images of 11 fields of 10 rich clusters , obtained with the wfpc-2 on the hubble space telescope . the exposures are typically 4 to 6 orbits each of @xmath02000 s duration . in a few cases fields were observed in more than one band , but for the purposes of this paper only f814w or f702w exposures are involved , since these alone were used for morphological classification . by inspecting the images in what is essentially the rest - frame b - band , we have tied our system to the revised hubble and de vaucouleurs schemes , which are traditionally based on images in the b - band this mitigates legitimate concerns about type - dependent _ k _ corrections . the morphological classifications used here follow the traditional method of visual inspection of images . as discussed in s97a , at least two , and often three or four of us assigned a morphological type for an image , with a high degree of repeatability and consistency . the clusters in our sample were not selected through any systematic criteria , but represent many of the best examples of well - studied clusters at intermediate redshift . as such , they are all rich ( populous ) clusters , though with a range in richness of a factor of three . they also represent a range in dynamical state , as evidenced by the regularity of the spatial distributions of their galaxies : the clusters range in appearance from highly concentrated and regular to very chaotic and dispersed . a further indication of their diversity of properties is the order - of - magnitude spread in their masses and x - ray fluxes ( smail et al . this is not to say that our small sample spans the distribution of richness and types that are characteristic of the abell catalog , for example , but not all of the clusters are analogs or richer versions of such unusual low - redshift clusters as abell 1656 ( the coma cluster ) or abell 2029 . the images cover the central 0.40.8 h@xmath5 mpc for a @xmath6 and @xmath7 km sec@xmath5 mpc@xmath5 cosmology , for which 1 arcsec is equivalent to 3.09 h@xmath5 kpc for our lowest redshift cluster and 3.76 h@xmath5 kpc in the most distant . for a resolution approaching 0.1 arcsecond , then , these images show detail at the level of @xmath0500 pc , equivalent to observing galaxies in the coma cluster with 1 `` seeing . while cruder than the resolution usually available for morphological classification of nearby galaxies , this is sufficient for the identification of basic morphological information the presence of disks and bulges , and large - scale spiral arms , dust lanes , and bars . in particular , this resolution is comparable to that obtained in the d80 study of morphology in low - redshift clusters , for which z = 0.04 and 1 '' seeing were the norm . object identification and measurement of photometric parameters were accomplished using a modified version of the sextractor image analysis package ( @xcite ) . because of the low sky background and long integration times , the exposures go very deep , to a 5@xmath8 limiting depth of @xmath9 or @xmath10 . however , after considerable testing , including intercomparisons of our several human classifiers , we concluded that reliable classification could not be done fainter than @xmath11 or @xmath12 with these images . in fact , as discussed below , for the purposes of the analysis described here , we reached this limit only for the highest redshift clusters . in s97a there is a detailed discussion of how the morphological classifications were assigned and what is their inferred accuracy . for the purposes of this work , only the objects which were given revised hubble types were used ; galaxies that appeared extremely chaotic or insufficiently detailed to fit into one of the categories a few percent in all were not included . this results in a small inconsistency with what has been done in studies of lower redshift clusters , for which galaxies were all assigned to one type or another , but we believe it is the safest thing to do for the classifications of distant samples . it is certainly not the case that all of this few percent of cases belong to one class , for example , the irregulars . probably half are cases of bulge - dominated , earlier types whose division into elliptical , s0 , and spiral can not be made because of too few pixels in an image . of the extremely chaotic systems , we believe that most of these should not be put into the irregular class , since they are much rarer at the present epoch and , as such , have no counterparts in the revised hubble scheme . ( the `` irregular '' class in the revised hubble system refers to a specific morphological type , and was not intended as a reposititory for galaxies that did not fit into the other classifications . ) following oemler ( 1974 ) , melnick and sargent ( 1977 ) found a strong gradient in morphological type with radial distance , what we will here refer to as a t - r relation , for a sample of clusters with strong thermal x - ray emission , in the sense of spiral fraction increasing outwards and elliptical and s0 fractions increasing inwards . as is typical of luminous x - ray clusters , these were centrally concentrated clusters with `` regular '' ( non - clumpy ) galaxy distributions . the dcat80 sample of 55 clusters with @xmath13 contained examples of this type , all of which showed strong radial population gradients . in addition , however , this sample included many irregular clusters with poorly defined centers and low concentration . as oemler had found , radial gradients for these were weak or even absent ; multiple centers often complicated the attempt to look for such trends . however , because such clusters often were found to contain several higher density clumps with increased fractions of e and s0 galaxies , dressler was led to look for a correlation of galaxy type with the local density , as discussed in appendix 1 . the basic interpretation of the relation was that the environmental density , possibly reaching back to quite early epochs , might have exercised substantial influence on the evolution of morphological types . at the time this might have been considered surprising , since it was commonly argued that galaxy mixing in the virializing collapse of a cluster would remove any correlation of a galaxy with its previous environment . however , subsequent n - body simulations confirmed that simple `` top hat '' analytical models of cluster virialization were misleading because the more realistic n - body models of structure growth showed that mixing was far less thorough . in fact , the simulations showed that galaxies found at the present epoch in high density regions were more than likely to have been in higher density regions throughout their lifetimes , even if there had been considerable mixing of the individual galaxies with those from somewhat lower density environments ( see , e.g. , evrard , silk , & szalay 1990 ) . the data from dcat80 have been reanalyzed for the purposes of this paper . following whitmore , gilmore , and jones ( 1993 ) , we have made a field correction that is dependent on morphological type , in addition to correcting the local surface density for field contamination . the percentages 18/23/59 adopted by these authors for field e / s0/s+i contamination have been adopted here as well . this produces a small but noticeable change in particular it lowers the spiral fraction at the lowest projected densities , since many of these are in fact projected field galaxies not associated with the cluster . we have also made finer divisions in binning the data , now 0.2 dex in surface density , to make full use of the large , statistical sample . this revised version of the d80 morphology - density relation for the entire 55 cluster sample is shown in histogram form in fig . 1 . also shown in fig . 1 , and in subsequent plots , is the number histogram of the membership in each bin . this is useful for estimating when the fractions in individual bins are very uncertain , i.e. , when fewer than 10 galaxies comprise a bin . however , here we mean to compare only overall trends and not individual bins , so we have chosen to omit error bars for each bin , as they greatly confuse the diagrams . ( we have , however , tabulated these values and their errors in tables as appendix 2 . ) the noise due to counting statistics can be judged by the scatter from one bin to the next of what would otherwise likely be monotonic relationships . subsequently , the relation was studied for poorer groups and the field ( bhavsar 1981 , de souza et al . 1983 , postman & geller 1984 , giovanelli , haynes , & chincarini 1986 ) with the general result that the trends found for rich clusters were found to continue to lower density environments , with lower precision due to the uncertainties of projection effects for low - density environments . however , salvador - sole et al . ( 1989 ) pointed out that if the relation were universal for surface densities , as d80 had suggested , then the relationship between morphology and three - dimensional space density could not be universal , implying that a further dependence on some more global property , one that compared whole clusters with one another , might also play a role . in fact , dressler had suggested that his sample showed some evidence for a global dependence on cluster concentration by noting a shift of the population levels for the clusters with the strongest x - ray emission . salvador - sole et al . pointed out that the sense of this effect , if real , was to imply an even greater dependence on concentration in the three - dimensional correlation . though this analysis might be questioned because of its reliance on a smooth power - law model for the distribution of galaxies in the cluster , both results seemed to point to some additional determinant of cluster population beyond the local density , perhaps connected with a global property of the cluster . the case for a dependence on cluster populations with a global cluster property was prosecuted even more vigorously by whitmore & gilmore ( 1991 ) and whitmore , gilmore , & jones , ( 1993 ) , who argued that the _ sole _ determinant of galaxy type within rich clusters is the radial distance from the cluster center ( a t - r relation ) to the exclusion of local density or even any other global property , such as x - ray luminosity , velocity dispersion , or central density . a thorough examination of this approach by whitmore and collaborators is beyond the scope of the present paper , but we include in this paper in appendix 1 a discussion of additional evidence that bears on this question of whether local density or radial distance from the cluster center is better correlated with morphological type . although we will show both and t - r relations for our distant clusters , our emphasis will be on the former . in particular , we divide our sample into centrally - concentrated , regular clusters and low - concentration , irregular clusters . a comparison of these two kinds of clusters in the dcat80 sample continues to provide the best evidence that the relation is a manifestation of a genuine physical process operating in clusters that relates environmental density to galaxy morphological type , either causally or through a common connection . we show such data in fig . 2a for 10 of the most centrally concentrated and fig . 2b for the lowest - concentration clusters . the global properties of these samples are very different : velocity dispersions and x - ray fluxes are much higher for the centrally concentrated clusters , and of course , cluster centers are well defined in these but not in the low concentration clusters . nevertheless , the relations for both kind of clusters look very similar , with a strong dependence of spiral and elliptical fraction with local density for both . as we show in the appendix , the increase in elliptical fraction with @xmath14 is , within the errors , the same in both , although there might be a significant excess of s0 s , and corresponding smaller spiral fraction , in the centrally - concentrated clusters compared to the low - concentration clusters . this may be indicative of an additional factor other than local density influencing cluster populations , but these differences are marginally significant at best . because the relation is evident for low - redshift clusters of all types , we use it as the basis for analyzing the relation for the intermediate - redshift clusters of our sample . while global influences may also play a role for low - redshift clusters , the evidence continues to support the conclusion that local density and morphological type are strongly connected . we now turn to the subject of how the or t - r relations are manifest in our sample of 10 intermediate - redshift clusters at @xmath1 . a list of these clusters , their positions , and redshifts are given in table 1 . we will assume that our morphological classifications are made comparably to the dcat80 study of 55 nearby clusters , that is , that the ability to discern salient features for classification are not appreciably diminished for these more distant clusters . with the possible exception of a difficulty in distinguishing face - on s0 from elliptical galaxies , a problem even for nearby galaxies , we believe that this assumption is justified , as discussed in s97a . ( we also discuss this issue further below . ) as in dcat80 , our classifications of morphological types in the intermediate - redshift clusters include cases designated e / s0 and s0/e . as explained in s97a , the order reflects the preference in the classification . therefore , as done in that study , we have added these types into e and s0 , respectively . as in d80 , the analysis of the relation for the dcat80 sample was done for galaxies down to a fixed apparent magnitude limit @xmath16 , which corresponds to @xmath17 for @xmath18 = 50 and z = 0.040 . we have chosen @xmath19 as the magnitude limit of the analysis of our clusters at intermediate redshift . using standard kron - cousins / johnson filter transmissions , the energy distribution of an sbc galaxy , and conversions from f702w to r and f814w to i given by holtzman et al . ( 1995 ) , we have derived the magnitude limits given in table 1 . also in table 1 we give the number of galaxies in each cluster brighter than the magnitude limit , and the fractional representation of the different morphological types for each field , corrected for field contamination . we obtained values of galactic extinction for each cluster from ned ; the maximum of these corresponds to 0.23 mag in the f702w band observation of cl0303 + 17 . these small corrections to the limiting magnitude have not been made , as also was the choice in d80 . they make no significant change to the fractional distributions of the populations . we also need to take note of the fact that our intermediate - redshift sample is studied over a smaller area ( by a linear factor of approximately 2.5 ) than that of the low - redshift sample of dcat80 . we have therefore reanalyzed the data for the 55 clusters for a restricted area a box that is 1094 arcsec on a side , or 1.2 mpc ( @xmath18 = 50 , adopted in d80 ) , for @xmath20 , and scales with redshift . for most of the low redshift clusters we have centered the box on the plate center , but have used the `` local density peak '' centers ( see appendix 1 ) for the centrally concentrated and low concentrated subsamples . the relation for the whole sample , and for the centrally concentrated and low concentration samples , are shown in fig . 3a - c , respectively . there is no obvious difference between these and figs . 1 & 2 , except that the lowest density range is unpopulated , as expected , and that the statistics for the two subsamples are substantially degraded , especially for the low concentration sample . we have applied a field correction , taken from the medium deep survey ( @xcite ) , of 11.2 gal / sq arcmin at @xmath4 . as we did for the 55 cluster sample , we have made differential corrections by type to the counts in the distant clusters . we adopt the percentages of the different types found in the medium deep survey 10% , 10% , and 80% for elliptical , s0 , and spiral galaxies , respectively ( s97a ) . these values are , coincidentally , the same as used by d80 , which have been replaced here in our reanalysis of the dcat80 sample with the whitmore , gilmore and jones s values of 18% , 23% , and 59% ) . there is no reason or expectation for using the same values for both low- and intermediate - redshift clusters , since the resolution of the images , and the composition of the field contamination are not necessarily the same . we now compare fig . 3a ( or , equivalently , fig . 1 ) to the relationship for the entire @xmath1 sample , fig . 4 , noting first that , while there is considerable overlap in density between the two samples , the density range encompassed by the more distant sample is shifted by half a dex to higher density . this undoubtedly reflects the fact that these clusters are systematically richer than the _ typical _ clusters of dcat80 , although there is considerable overlap with the projected density range covered by that sample . specifically , 13% of galaxies in the dcat80 sample lie at @xmath21 lower limit of the intermediate redshift sample , and 12% of this sample are found at @xmath22 , a higher projected density than found in any of the dcat80 clusters . because 75% of the galaxies in the two samples share the same range in projected density range ( somewhat greater if @xmath23 , we do not consider these two samples as composed of qualitatively different kinds of clusters . ( for example , cl1601 , cl0054 , and cl0303 , share the density range of the dcat80 sample . ) it would be desirable , of course , to include more clusters typical of the present epoch in future samples . before addressing the question of gradients in fig . 4 , we take note of the obvious differences between this and the nearby cluster sample . as is now well known , spirals are greatly overabundant in high density environments compared to present - epoch clusters , but , perhaps surprisingly , the difference seems to be made up entirely by a paucity of s0 galaxies rather than an underabundance of _ both _ s0 and e galaxies . in fact , e galaxies appear to be in even greater abundance than in the nearby sample ! in the overlapping density regimes of fig . 3a and fig . 4 , spirals are a factor of 2 overabundant , s0 s are a factor of 23 underabundant , and ellipticals are a factor of 1.5 overabundant . the e or s0 population is , of course , little affected by the uncertainty in the correction for field galaxies , since galaxies of this type make up a small ( @xmath24 ) fraction of the intermediate - redshift field population . in order to further investigate the possibility of misclassifications of e and s0 galaxies we have , as explained in s97a , compared the distribution in flattenings of s0 galaxies in our sample with that for the coma cluster , in an attempt to see if we have misclassified s0 galaxies as ellipticals , particularly for the face - on cases . the good agreement of these distributions indicates that this is not the case . our best estimate is that we have misclassified as ellipticals approximately @xmath25 of the total s0 population that is nearly face on . as discussed in s97a , this has been determined through comparison with the revised shapely ames galaxies ( as tabulated in sandage , freeman , and stokes 1970 ) , but we believe that in comparison with classifications of nearby clusters , such as dcat80 , the degree of misclassification is so similar as to make any systematic differences negligible . we demonstrate this in fig . 5 , where we compare the distribution of ellipticities for both e and s0 in the @xmath1 cluster sample with the e and s0 galaxies in the coma cluster ( andreon et al . 1996 ) and 11 clusters from dcat80 with @xmath26 . in both cases the distributions are in very good agreement , suggesting that any _ differential _ comparison of the e and s0 fractions in these clusters is justified . we thus conclude that the substantial difference seen in the s0/e fraction in distant clusters compared to nearby clusters is genuine for this sample ; of course , verification in a larger sample , to be certain that our sample is truly representative , would be highly desirable . the implications of this difference in e and s0 populations is discussed in section 5 . we now ask whether any trend of morphology with density is apparent for the distant sample . from fig . 4 it appears that a modest relation is present , but it is only for the bins of highest density over the last factor of 5 in surface density . over this range the spiral fraction plummets and the elliptical fraction rises sharply , but for the lower density zones , over which there is a very noticeable gradient in the nearby clusters , the relationships are basically flat . slicing the sample by redshift or richness does not change this result . however , when the sample is divided by the cluster concentration , which correlates well with degree of regularity , a very different picture emerges . we have used butcher & oemler s ( 1978 ) definition of concentration as @xmath27 , where the latter refer to the radii containing 60% and 20% of the cluster populations , respectively . these values are given in table 1 under the heading `` conc . '' 6 shows the relation for the four clusters of the @xmath1 sample with the highest central concentration 3c295 , cl0024 + 16 , cl0016 + 16 , and cl0054@xmath2827 . the relation for this subset is steep and well defined over the entire density range , as strong as the relation for the low redshift sample . as expected , the t - r relation ( fig . 7 ) , being more - or - less degenerate with for these cases , is also very strong for these clusters . though the gradients in fig . 6 are as strong as that found for low - redshift clusters , there are substantial differences , of course . most obvious is the prevalence of spiral galaxies even in very high density regions . and although the s0 galaxies show the same flat distribution in projected density , they are , as we have pointed out , far less abundant . finally , in comparison with fig . 3 , fig . 6 appears to show a 510% excess of elliptical galaxies at the same surface density , extending to much higher fractions , at higher density , than has been found for lower redshift clusters . this may again highlight the unusual richness of some of the high redshift clusters in this sample . the remaining clusters of lower concentration cl0303 + 17 , cl0412@xmath2865 , cl0939 + 47 , cl1447 + 23 , cl1601 + 42 have irregular , clumpy distributions of galaxies , just like their low redshift counterparts . we show in fig . 8 that , in contrast with the relatively strong gradients seen for the centrally concentrated clusters , there are no correlations at all for these 5 lowest - concentration , irregular clusters , except perhaps for a small _ anticorrelation _ of low statistical significance . ( the t - r relation is , of course , nonexistent as well . ) apparently , then , the morphology - density relation is well established at higher redshift for the clusters that appear well evolved dynamically , but there is no apparent segregation in clusters that have more chaotic distributions . of course , even the equivalent diagram for the centers of low - z clusters of low concentration ( fig . 3c ) suffers from small number statistics , so it is possible that this is an unrepresentative sample . we note however , that although the population gradients are not evident in the high - z , low - concentration clusters , the elliptical fraction is high and the s0 fraction is very low , as it is for the high - z , high - concentration clusters and the low - z sample . it is only the arrangement of these galaxies within the clusters , and not their relative proportions , that seems to differ from the centrally concentrated clusters . these , then , are our principal results from the small sample of intermediate redshift clusters we have studied . ( 1 ) over the range of density studied , which extends to somewhat higher densities , on average , than for the typical nearby cluster , elliptical galaxies are present in numbers comparable to what is seen today , while spirals are much more abundant and s0 galaxies much less abundant . ( 2 ) there is a modest relation , but only for the highest densities , when all the clusters are taken together . ( 3 ) when the sample is divided into high - concentration , regular clusters and low - concentration , irregular clusters , the former show strong gradients in morphology over the represented range in surface density , but the latter show no gradients at all . perhaps the most striking result of this study is that , whether the clusters appear dynamically `` mature '' or not , the incidence of elliptical galaxies is already very high . put another way , the elliptical fraction is high whether or not ellipticals are collected into dense , central regions . this suggests that elliptical galaxies predate , and are basically independent of , the virialization of a rich cluster . supporting evidence for the idea that elliptical galaxies in rich clusters are old can be found in the ellis et al . ( 1996 ) study of the @xmath29 colors of these same ellipticals . ellis et al . concluded , based on a scatter in rest frame @xmath30 color of 0.07 mag scarcely larger than the value of 0.035 mag found for ellipticals in the nearby coma cluster ( bower , lucey , & ellis 1992 ) that the stars in cluster elliptical galaxies formed early , @xmath31 ( or that , if they formed later , they formed within a very small time ) . this seems to argue against a `` late - merger '' origin for ellipticals , or at the least against dissipative mergers of star forming systems . in fact , as we will explore elsewhere , those systems in our sample for which a merger is likely to have `` recently '' occurred are still , for the most part , disk systems ; therefore , most are unlikely to be the ancestors of elliptical galaxies because there is no expectation that the disk will be destroyed without another significant encounter . we have further evidence of this in their distribution within the clusters : when we plot the fractional population of highly asymmetric or disturbed morphologies ( all morphological types with @xmath32 , as defined in s97a ) with local surface density ( fig . 9 ) , we see that these follow the s0 or spiral trend rather than the trend for ellipticals . that is , the fraction of asymmetric galaxies is decreasing , or perhaps flat , but certainly not rising , with increasing density . this is a result similar to that found by oemler , dressler , and butcher ( 1997 ) for the distribution of `` disturbed '' galaxies in four distant clusters . taken together , these three observations the prevalence of e galaxies , their small @xmath30 scatter , and the fact that most of the disturbed galaxies are disk systems and likely to remain so suggest that , at least for the environments of rich clusters , elliptical galaxies are not the result of mergers of star - forming , gas - rich systems after a redshift @xmath33 . this does not preclude the possibility of dissipationless mergers at @xmath34 , say , when these clusters might have been in the process of accreting small groups with lower velocity dispersion , but both the distribution and numbers of ellipticals we have found here , and their photometric and spectral properties , suggest that ellipticals in these regions are not produced by late mergers , or in any process that depended on the dynamical evolution of a rich cluster . instead , gaseous mergers or coherent collapse at high redshift , or growth of spheroids through dissipationless mergers continuing until later epochs , seems to be the history indicated for ellipticals . it is remarkable , we think , that the environments of proto - clusters of this richness were able to produce such a large population of ellipticals before the identities of the clusters themselves were well established . the situation for the s0 galaxies seems to be just the opposite . though the ones we find are , like the ellipticals , red and with little scatter in color , their numbers are so deficient as to suggest that many need to be added since @xmath1 , in order to reach the populations of present - epoch clusters . the source of these s0 s seems clear : the overabundance of spirals provides a reservoir of galaxies which may be stripped by ram pressure , tidally harassed ( moore et al . 1995 ) merged , or subjected to strong 2-body gravitational interactions , with the result of producing today s dormant disk galaxies in clusters . as we will describe in more detail in a later paper , our @xmath1 cluster sample includes a significant number of disturbed , distorted morphologies , often with spectroscopic evidence of strong episodes of recent in a few cases current star formation . these may be the result of mergers , strong interactions , accretions , harassment , or stripping we are still unable to tell which of these processes are responsible . but , as we have said , we do know from our morphological classifications that most of these are disk systems they do not seem destined to settle into luminous ellipticals galaxies when their jostling and bursts of star formation have ceased . though the exact mechanism(s ) may be yet unspecified , it seems that at least half of the s0 galaxies in today s clusters have been made by such processes since @xmath1 . some support for this model may also be found from the trend of the s0/e ratio with redshift , which we show in fig . omitting the two outer fields , which are properly excluded for this analysis , we see what appears to be a significant trend of increasing s0 abundance with decreasing redshift . even more provocative is the observation that this trend , in linear extrapolation , comes close to predicting the s0/e ratio of @xmath02 found in the d80 study of clusters at low redshift . although these data are of too low statistical weight to be conclusive , they suggest that we could be seeing the process of s0 production over the interval in cosmic time probed by these observations . on the other hand , according to ellis et al . ( 1996 ) , the spread in rest frame @xmath30 colors for s0 galaxies in these clusters is as small as it is for ellipticals , from which it was concluded that the stellar populations in ellipticals are very old . while we may with the present data simply postulate that these represent a population of old s0 s that formed with the ellipticals , our suggestion that most of the s0 s in these clusters have yet to form means that the color spread at some later time , at @xmath35 , for example , should be significantly larger , reflecting the more recent evolution of these systems from previously star forming galaxies . to make comparable morphological classifications in clusters at this redshift will require either a mosaic coverage with hst or larger - field ccd imaging with ground based images under conditions of superb seeing . at first glance it may not be surprising , then , to see a relation for the dynamically well - evolved clusters of our sample . one can imagine a large initial population of elliptical galaxies , formed at some earlier time , being diluted by infalling disk galaxies in increasing numbers working out from the cluster center . any mechanism for converting spiral galaxies to s0 s that is more efficient with increasing density would account qualitatively for the relation observed at @xmath1 and @xmath36 . in this connection it is interesting to note that in present - epoch low - concentration , irregular clusters ( fig . 3 ) ellipticals appear to outnumber s0 s in the highest density regimes , whereas the opposite situation seems to hold for the centrally concentrated clusters . again , the manufacture of s0 s by some density - dependent process seems to be a reasonable interpretation . what prevents this picture from being a tidy one , however , is the fact that the 5 low - concentration , more irregular clusters in our distant sample show no concentration of ellipticals towards the densest regions . instead , they possess large numbers of ellipticals spread willy - nilly over the area surveyed . does the gradient we see in the regular clusters mean that dilution alone by infalling galaxies has already been sufficient to set up an apparent concentration of ellipticals , or must there also be a mechanism for bringing ellipticals to what will become the dense cluster regions , and , as the whitmore & gilmore result suggests , even more strongly to one specific cluster center ? dynamical friction is the obvious candidate for concentrating the ellipticals as different subgroups merge , but would it be sufficiently effective ? by virtue of their luminosity function , which has the brightest characteristic magnitude of the morphological types ( sandage , binggeli , & tammann 1985 ) , and their high m / l ratios ( faber & gallagher 1976 ) , ellipticals are , on average , significantly more massive than most cluster galaxies , but it remains to be seen if the effect is sufficient to generate a substantial segregation . on the other hand , the lack of a radial gradient in the luminosity function in clusters of galaxies argues against this effect being an strong one . this discussion begs what must be the more puzzling question : why are the irregular clusters at @xmath1 different in terms of their relations from similar clusters today ? are any of these distant irregular clusters destined to remain so until the present epoch , in which case , mechanisms like dynamical friction and dilution through infall are certainly needed to generate trends that were not present at @xmath1 ? alternatively , are these distant irregular clusters the ancestors of some of today s regular clusters ? if so , the same processes must have occurred to create the relation , but there must be an _ additional _ explanation of why irregular clusters today , which are almost certainly younger dynamically than today s regular clusters , are amalgams of groups that already have , individually , well established segregations of galaxy types . a reasonable explanation for this behavior might be that morphological segregation has occurred hierarchically over time , first for the densest , most populous groups , and then working its way down to the smaller , less dense systems . in this picture , the groups that make up the irregular clusters at @xmath37 had not yet undergone significant morphological segregation , but by the present epoch , even these groups , from which the hercules cluster ( abell 2151 ) is now being composed , would exhibit such correlations . observations of smaller groups at intermediate redshift will be crucial for testing this notion . of course , testing these schematic ideas depends on understanding which mechanisms are responsible for the manufacture and distribution of s0 galaxies and , at an earlier epoch ( we believe ) , the ellipticals . our sample is still too small to say with certainty that these trends are universal ; a larger sample is essential to verify what we have found here . furthermore , our present data do not include sufficient area from the outer regions of intermediate - redshift clusters , a comparison with which might allow for a discrimination between the various candidates for s0 production . in forthcoming papers we hope to throw some light on this issue through the application of our large sample of spectroscopic data and what it tells us about the history of star formation in these distant cluster galaxies . it is clearly also the case that a push to higher redshift could be very revealing : it would seem probable , based on what we have found , that the few clusters that exist at @xmath34 will show little morphological segregation of any kind . also , by filling in the evolutionary sequence of clusters at @xmath35 with morphological types and colors , the progression of cluster evolution , particularly for the irregular clusters , might be better understood . we have revisited the relation between morphology and surface density and confirmed that it is a fundamental characteristic of all nearby clusters and groups . however , for a sample of distant clusters , @xmath38 , studied with hst , the relation is apparent only in high concentration , regular clusters that are presumably dynamically evolved . no correlation is found for less evolved , low concentration , irregular clusters . this suggests that the mechanisms that produce morphological segregation , either directly , by dynamical friction , for example , or through converting one type of galaxy into another , may work at different rates depending on the mass scale of the group or cluster . by the present epoch , even the smaller groups , presently coalescing to form today s irregular clusters , may have accomplished a morphological segregation that was absent only 1/3 of a hubble time ago . the large number of elliptical galaxies in these clusters , @xmath040% in both regular and irregular clusters , suggests that the creation of ellipticals predates cluster virialization . if mergers are responsible for making the ellipticals that now inhabit these rich clusters , they must have been dissipationless , in the `` group phase '' at @xmath34 , or much earlier , @xmath39 , if significant dissipation and star formation were involved . in contrast , the relative paucity of s0 s in the intermediate redshift clusters suggests that many of them have indeed been added since @xmath1 , by mechanisms that acted on the excessive numbers , compared to today s clusters , of spirals and irregulars . firstly , we wish to thank ray lucas at stsci for his enthusiastic , able help in the efficient acquisition of these hst observations . ad and ao acknowledge support from nasa through stsci grant 3857 . irs , rse and rms acknowledge support from the particle physics and astronomy research council . wjc acknowledges support from the australian department of industry , science and technology , the australian research council and sun microsystems . this work was supported in part by the formation and evolution of galaxies network set up by the european commission under contract erb fmrx - ct96 - 086 of its tmr program . the relation , the basis for our analysis in this paper , has been challenged as the primary correlation between galaxy type and some associated property of the galaxy clusters . our purpose here is review briefly the derivations of this relationship and to comment on these objections . a first step is to reexamine the raw data that led dressler to suspect that galaxy populations are in fact correlated with local density . in fig . 11 we show representative clusters from the samples , with the e and s0 galaxies highlighted . although only qualitative , one can observe even in these maps that the e and s0 galaxies are clumped together much more strongly than the cluster spirals , which are spread almost uniformly over the fields . this is true not only for reasonably regular clusters , such as abell 151 , but even for clusters without a well defined , single high density region , such as abell1631 , dc0003 - 50 , and dc0326 - 53 , all of which are shown in fig . although it is difficult to use relations to compare individual cases , we can verify the subjective impression of the greater clumping of e and s0 galaxies by a simpler statistic , the average nearest neighbor distances for the different types . we divided the sample into two types , e+s0 galaxies , and spirals . for the clusters just listed , the mean distance ( in h@xmath5 mpc ) from an e or s0 galaxy to its nearest - neighbor of the same type , compared to the mean distance of `` nearest - neighbor spiral galaxies '' are 0.202:0.344 , 0.394:624 , 0.365:0.459 , and 0.338:0.551 , respectively . in each case the nearest - neighbor distance between spirals is substantially larger . this result is consistent with the idea , unsubstantiated in 1980 but subsequently strongly supported by radial velocity studies , that subclustering is a real and important phenomenon is most clusters . while the change in galaxy populations with radial distance from the centers of the clusters can be seen in fig . 2 , the existence of subgroups with populations shifted towards early hubble types , even when removed from the centers , is also apparent . whitmore and collaborators claim that the t - r relation is more fundamental than the relation in understanding the distribution of galaxy types within a cluster . in order to evaluate this claim , it is important to remember that the and t - r relations are nearly degenerate for those clusters with a reasonably smooth , regular distribution of cluster galaxies . in such cases local surface density correlates very well with clustocentric radius . since the remaining variations are statistical and not representative of substantial subclustering , there is no reason to expect to be able to to distinguish between the two candidates for a fundamental relationship . thus , it is essential to compare these relationships for both centrally concentrated regular clusters _ and _ clumpy , irregular clusters to compare the efficacy of local density or distance from the cluster center as a driver of galaxy type . such a comparison was the basis of dressler s claim that the relation is universal , in the sense that it held irrespective of other global parameters . for example , dressler noted that the relations for the high - concentration and low concentration clusters were very similar , perhaps indistinguishable within sampling errors . in order to make quantitative comparisons between the relations of fig . 2 , we have fit the slopes and intercepts of the elliptical and spiral fractions as a function of @xmath14 . 2a and 2b respectively we find that the f(e ) = 0.061(0.069 ) + 0.122(0.026)@xmath14 and f(e ) = 0.094(0.041 ) + 0.090(0.039)@xmath14 ( with 1 @xmath8 errors given in parentheses ) , thus confirming the impression that these can not be distinguished within the errors . for the spirals , f(sp ) = 0.623(0.044 ) - 0.250(0.030)@xmath14 for fig . 2a and f(sp ) = 0.642(0.069 ) - 0.184(0.058)@xmath14 for fig . 2b , which indicates a marginally significant difference in the sense of a lower fraction of spirals , and a higher fraction of elliptical and s0 galaxies at a given density for the high - concentration sample . this difference might suggest a dependence on the cluster type , a global parameter , in addition to the relation , but the evidence is weak . while the relation holds to first order for both regular and irregular clusters , the t - r relation explored by whitmore and collaborators shows very different behavior in the two types of clusters . in fig . 12 we show the t - r relation for the same samples of fig . 2 , using center positions defined by the peak in local galaxy density over the field ( values taken from beers and tonry 1986 ) . as found by those studies , the t - r relation for the centrally concentrated clusters is strong . in contrast , the relationship for irregular clusters of low concentration is weak . in fact , it is arguably absent except for the points within 0.5 mpc , which result from the peaking up on a local density enhancement , in which domain one would expect the relation to be again apparent , which it is . our interpretation of fig . 12 is , then , that application of the t - r has weakened or even eliminated a true dependence further out in the cluster , by angular smoothing within the annular rings out what genuine variations do exist , leaving only the sharp gradients in the central region . whitmore and collaborators have used the flat - to - steep profile to argue that the t - r relation is the fundamental one , but our analysis here argues , as did dressler , that the dependence on distance from a single cluster center is only a second - order , though likely important supplement to the more fundamental correlation with local density . that fact that the relation holds for both types of clusters , and the t - r relation is substantially different for the two types , remains the best evidence that it local density is fundamental to the determination of galaxy type . a further challenge has been made by sanroma & salvador - sole ( 1990 ) . these authors used angular scrambling of the galaxy positions in dcat80 to argue that that , for the entire sample , the does not exist , but is instead a derivative of the t - r relation . again , separating centrally concentrated from low - concentration clusters is important . we have repeated this experiment for the samples of fig . 2 , scrambling around both `` density peak '' and `` median '' centers . the ranges in the distributions in density are largely maintained by the scrambling procedure , so a kolmogorov - smirnov test can be applied . we find that scrambling the centrally concentrated clusters produces no significant change in the relation : the fraction of e galaxies at a given projected density for the real data would match that of the scrambled data 38% of the time , and the spiral fraction would match 18% of the time . in stark contrast , the relation for the low - concentration clusters is in fact destroyed by the scrambling : all probabilities are @xmath40 , regardless of choice of galaxy center . it is our contention , then , that the relation remains important for understanding the distribution of morphological types in present - epoch clusters . this is not to say that there are no global dependences of cluster population , as whitmore and collaborators , sanroma & salvador - sole , oemler , and even dressler himself have suggested . in fact , in addition to the evidence mentioned above , both whitmore & gilmore s t - r relation , and the scrambling analysis of salvadore - sole , provide substantive evidence for global components . as can be seen in fig . 5 of whitmore & gilmore ( 1991 ) , the percentage of elliptical galaxies rises even higher , with a compensating _ drop _ in the s0 population , for those galaxies closest to the cluster center as defined by the presence of a `` d '' galaxy , compared to the bin of highest density of the relation . the same can perhaps be inferred from our experiments of scrambling the clusters of the high - concentration sample and comparing the relation to that of the real data . it appears that when all but the centrally located high - density peak are eliminated by the scrambling , the e fraction rises and the s0 fraction falls , as if the central location of these galaxies has an additional effect on the population . so , it appears that central location may indeed influence a galaxy s morphology , and be perhaps an important `` second - parameter '' to the relation . in table 2 we tabulate the values of the relations in the figures , as well as their errors . the errors stated in table 2 are 1 sigma errors derived from poisson statistics , assuming that the number of galaxies of a given type is independent of the number of galaxies of other types . it is also assumed that the number of background galaxies has a poisson distribution ( i.e. that the angular correlation function of galaxies is negligible at these magnitudes . ) lcccccccc & ra(j2000 ) dec(j2000 ) & @xmath41 & filter & m@xmath42 & ngal & % e : s0:sp & conc . a370#2 & 02:40:01.1 -01:36:45 & 0.37 & f814w & 21.98 & 71 & 28:20:52 & - cl1446 + 26 & 14:49:28.2 + 26:07:57 & 0.37 & f702w & 22.24 & 107 & 31:26:43 & 0.30 cl0024 + 16 & 00:26:35.6 + 17:09:43 & 0.39 & f814w & 22.11 & 170 & 37:24:39 & 0.53 cl0939 + 47 & 09:43:02.6 + 46:58:57 & 0.41 & f702w & 22.48 & 124 & 31:24:45 & 0.34 cl0939 + 47#2 & 09:43:02.5 + 46:56:07 & 0.41 & f814w & 22.23 & 72 & 35:07:58 & - cl0303 + 17 & 03:06:15.9 + 17:19:17 & 0.42 & f702w & 22.54 & 93 & 39:19:42 & 0.31 3c295 & 14:11:19.5 + 52:12:21 & 0.46 & f702w & 22.76 & 87 & 51:21:28 & 0.58 cl0412@xmath2865 & 04:12:51.7 -65:50:17 & 0.51 & f814w & 22.77 & 91 & 47:07:46 & 0.30 cl1601 + 42 & 16:03:10.6 + 42:45:35 & 0.54 & f702w & 23.20 & 100 & 33:12:55 & 0.34 cl0016 + 16 & 00:18:33.6 + 16:25:46 & 0.55 & f814w & 22.96 & 193 & 58:15:27 & 0.49 cl0054@xmath2827 & 00:56:54.6 -27:40:31 & 0.56 & f814w & 23.00 & 119 & 34:18:48 & 0.38 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & -0.05 & 0.000 0.000 & 0.460 0.148 & 0.540 0.148 0.15 & 0.110 0.039 & 0.289 0.053 & 0.601 0.059 0.35 & 0.090 0.021 & 0.320 0.032 & 0.589 0.034 0.55 & 0.119 0.016 & 0.346 0.023 & 0.535 0.025 0.75 & 0.133 0.014 & 0.348 0.020 & 0.519 0.021 0.95 & 0.177 0.014 & 0.391 0.018 & 0.432 0.019 1.15 & 0.182 0.014 & 0.425 0.018 & 0.392 0.018 1.35 & 0.196 0.015 & 0.443 0.019 & 0.361 0.018 1.55 & 0.225 0.018 & 0.473 0.021 & 0.302 0.020 1.75 & 0.287 0.024 & 0.454 0.026 & 0.259 0.023 1.95 & 0.376 0.033 & 0.448 0.034 & 0.176 0.026 2.15 & 0.398 0.050 & 0.521 0.051 & 0.081 0.028 2.35 & 0.360 0.096 & 0.561 0.100 & 0.079 0.055 2.55 & 0.500 0.134 & 0.429 0.132 & 0.071 0.069 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & -0.05 & 0.000 0.000 & 0.589 0.250 & 0.411 0.250 0.15 & 0.105 0.065 & 0.241 0.088 & 0.654 0.099 0.35 & 0.106 0.040 & 0.362 0.061 & 0.532 0.064 0.55 & 0.121 0.034 & 0.386 0.050 & 0.493 0.052 0.75 & 0.155 0.036 & 0.379 0.048 & 0.466 0.050 0.95 & 0.215 0.038 & 0.412 0.045 & 0.373 0.045 1.15 & 0.156 0.035 & 0.498 0.049 & 0.345 0.047 1.35 & 0.208 0.043 & 0.524 0.053 & 0.269 0.048 1.55 & 0.201 0.042 & 0.563 0.052 & 0.236 0.045 1.75 & 0.331 0.046 & 0.506 0.048 & 0.163 0.036 1.95 & 0.435 0.069 & 0.416 0.068 & 0.149 0.050 2.15 & 0.305 0.096 & 0.610 0.102 & 0.085 0.059 2.35 & 0.364 0.146 & 0.547 0.151 & 0.089 0.088 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & -0.05 & 0.149 0.242 & 0.595 0.328 & 0.256 0.316 0.15 & 0.177 0.112 & 0.234 0.135 & 0.589 0.153 0.35 & 0.113 0.066 & 0.192 0.079 & 0.695 0.093 0.55 & 0.153 0.049 & 0.318 0.062 & 0.529 0.067 0.75 & 0.169 0.038 & 0.298 0.047 & 0.533 0.051 0.95 & 0.181 0.036 & 0.364 0.046 & 0.454 0.048 1.15 & 0.180 0.031 & 0.457 0.040 & 0.363 0.039 1.35 & 0.189 0.034 & 0.459 0.043 & 0.351 0.042 1.55 & 0.252 0.055 & 0.393 0.062 & 0.355 0.061 1.75 & 0.318 0.108 & 0.369 0.112 & 0.313 0.108 1.95 & 0.481 0.101 & 0.280 0.090 & 0.238 0.086 2.15 & 0.501 0.251 & 0.499 0.251 & 0.000 0.000 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & 0.35 & 0.087 0.137 & 0.447 0.215 & 0.466 0.221 0.55 & 0.043 0.051 & 0.537 0.130 & 0.420 0.130 0.75 & 0.121 0.050 & 0.458 0.076 & 0.421 0.076 0.95 & 0.197 0.044 & 0.414 0.054 & 0.389 0.054 1.15 & 0.250 0.037 & 0.463 0.043 & 0.287 0.040 1.35 & 0.198 0.026 & 0.527 0.033 & 0.276 0.030 1.55 & 0.243 0.026 & 0.509 0.031 & 0.248 0.027 1.75 & 0.302 0.030 & 0.430 0.033 & 0.268 0.029 1.95 & 0.394 0.039 & 0.438 0.040 & 0.169 0.030 2.15 & 0.367 0.057 & 0.550 0.059 & 0.084 0.033 2.35 & 0.409 0.105 & 0.546 0.106 & 0.045 0.045 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & 0.55 & 0.000 0.000 & 0.524 0.298 & 0.476 0.298 0.75 & 0.096 0.100 & 0.375 0.170 & 0.529 0.177 0.95 & 0.099 0.103 & 0.517 0.171 & 0.384 0.169 1.15 & 0.308 0.132 & 0.542 0.143 & 0.149 0.105 1.35 & 0.174 0.051 & 0.577 0.068 & 0.249 0.060 1.55 & 0.202 0.044 & 0.546 0.055 & 0.252 0.049 1.75 & 0.331 0.048 & 0.492 0.051 & 0.177 0.039 1.95 & 0.432 0.070 & 0.433 0.070 & 0.135 0.049 2.15 & 0.263 0.101 & 0.685 0.107 & 0.052 0.052 2.35 & 0.364 0.146 & 0.547 0.151 & 0.089 0.088 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & 0.95 & 0.168 0.163 & 0.344 0.205 & 0.489 0.219 1.15 & 0.158 0.062 & 0.509 0.085 & 0.333 0.081 1.35 & 0.216 0.070 & 0.492 0.085 & 0.292 0.078 1.55 & 0.159 0.060 & 0.476 0.082 & 0.365 0.080 1.75 & 0.335 0.112 & 0.334 0.113 & 0.331 0.113 1.95 & 0.480 0.105 & 0.261 0.092 & 0.259 0.092 2.15 & 0.501 0.251 & 0.499 0.251 & 0.000 0.000 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & 1.35 & 0.437 0.159 & 0.137 0.090 & 0.426 0.186 1.55 & 0.355 0.100 & 0.238 0.087 & 0.407 0.126 1.75 & 0.334 0.054 & 0.138 0.039 & 0.528 0.061 1.95 & 0.318 0.036 & 0.184 0.030 & 0.498 0.041 2.15 & 0.370 0.033 & 0.182 0.026 & 0.448 0.036 2.35 & 0.478 0.041 & 0.216 0.033 & 0.305 0.041 2.55 & 0.485 0.055 & 0.184 0.041 & 0.331 0.055 2.75 & 0.770 0.084 & 0.230 0.084 & 0.000 0.000 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & 1.35 & 0.586 0.458 & 0.105 0.156 & 0.309 0.514 1.55 & 0.426 0.233 & 0.184 0.152 & 0.390 0.281 1.75 & 0.320 0.074 & 0.145 0.054 & 0.534 0.085 1.95 & 0.350 0.051 & 0.198 0.043 & 0.452 0.058 2.15 & 0.306 0.046 & 0.202 0.040 & 0.491 0.052 2.35 & 0.450 0.068 & 0.190 0.053 & 0.359 0.069 2.55 & 0.292 0.081 & 0.235 0.075 & 0.473 0.092 lcccccc @xmath14 & f(e ) @xmath8 & f(s0 ) @xmath8 & f(sp ) @xmath8 & 1.35 & 0.260 0.166 & 0.176 0.149 & 0.564 0.217 1.55 & 0.317 0.120 & 0.264 0.119 & 0.419 0.158 1.75 & 0.388 0.102 & 0.134 0.068 & 0.478 0.113 1.95 & 0.315 0.060 & 0.157 0.048 & 0.528 0.069 2.15 & 0.447 0.055 & 0.187 0.041 & 0.366 0.059 2.35 & 0.520 0.055 & 0.247 0.046 & 0.233 0.052 2.55 & 0.603 0.068 & 0.152 0.047 & 0.245 0.064 2.75 & 0.751 0.090 & 0.249 0.090 & 0.000 0.000 * figure 1 . * the ( morphology - density ) relation for the d80 55 cluster sample , reanalyzed , as discussed in the text . the histogram at the top shows the numbers of galaxies in each bin of surface density , for the total number in the upper right corner . * figure 2 . * the relation for 10 centrally concentrated , regular clusters of the d80 cluster sample ( above ) , a151 , a539 , a957 , a1656 , a1913 , a2040 , a2063 , dc0247 - 31 , dc0428 - 53 , dc1842 - 63 , and ( below ) , 8 low - concentration clusters a76 , a119 , a168 , a978 , a979 , a1644 , a2151 , and dc0003 - 50 . * figure 3 . * the relation for all 55 clusters of the d80 cluster sample , but limited to a smaller area of 1.2 mpc , comparable to that covered by the more distant hst sample . from top to bottom : all clusters ; centrally - concentrated clusters , low - concentration clusters . * figure 4 . * the relation for 10 clusters at intermediate redshift , @xmath2 . the histogram in the top box shows the number of galaxies in each bin , with the shaded portion indicating the number after correcting for field contamination . * figure 5 . * the distribution of ellipticities for e and s0 galaxies in the @xmath1 sample compared to that for the coma cluster ( andreon et al . 1996 ) and for 11 clusters @xmath26 of the dressler ( 1980 ) low redshift sample . the similarity of these distributions suggests that a fair comparison of the fraction of e and s0 galaxies can be made between @xmath36 and @xmath1 . * figure 9 . * the relation for 10 clusters at intermediate redshift , including those galaxies with substantial asymmetry or disturbance , irrespective of morphological type . these galaxies follow the spiral or s0 trend rather than the elliptical trend . * figure 10 . * the s0/e fraction for clusters in the sample as a function of redshift . the open circles are the outer fields in a370 and cl0939 + 47 , which are not used in the least - squares fit , shown with its 1-@xmath8 errors as the solid and dotted lines . the extrapolation of this linear relation to zero redshift approximately matches the value s0/e @xmath0 2 , shown by the solid box , found for 11 clusters @xmath26 of the d80 cluster sample . * figure 11 . * maps showing the positions of galaxies in four clusters of the d80 55 cluster sample , a151 , a1631 , dc0003 - 50 , and dc0326 - 53 . large solid dots are elliptical galaxies , dotted - circles are s0 s , and small dots are spirals . the enhanced clustering of e and s0 galaxies compared to spirals is evident . * figure 12 . * the t - r relation for the same clusters of fig . as expected , population gradients are strong for centrally concentrated clusters ( above ) , and weak or absent for low - concentration irregular clusters .

we derive a poisson random field model for population site polymorphisms differences within and between two species that share a relatively recent common ancestor . the model can be either equilibrium or time inhomogeneous . we first consider a random field of markov chains that describes the fate of a set of individual mutations . this field is approximated by a poisson random field from which we can make inferences about the amounts of mutation and selection that have occurred in the history of observed aligned dna sequences . and . .

1011.1724

a traditional goal of population genetics is to understand the relationship between darwinian selection and evolution . a particular problem is to estimate the distribution of the fitness of genes that become fixed in a natural population as a way of determining whether evolution is going `` uphill '' or `` downhill '' in that population ( both are possible ; @xcite ) . one approach is to use the numbers of site polymorphisms between and within a pair of closely related species at a genetic locus ( mcdonald and kreitman @xcite ; see also @xcite ) . sawyer and hartl @xcite developed a poisson random field ( prf ) model for these counts that can be used to estimate the amounts of selection that are occurring at individual loci ( @xcite ; see also @xcite ) . multilocus extensions of the model can be used for large numbers of different loci @xcite . the models assume a high level of recombination between nucleotides and are well suited for the analysis of polymorphism and divergence at multiple loci distributed across a genome @xcite . other authors have extended the basic prf model to more general biological settings and have used numerical simulations to study alternative models and the effects of deviations from model assumptions @xcite . williamson et al . @xcite used a time - dependent prf model to estimate the time since a hypothetical sudden change in human population size . the purpose of this paper is to provide a rigorous derivation of a time - inhomogeneous prf model by approximating discrete time markov chains by diffusion processes and , so that the model can be applied to data , to derive the corresponding sampling formulas for aligned dna sequences . first , a brief introduction to genetics and population genetics . the genetic material or dna of most living creatures is located in one or more chromosomes . a chromosome can be thought of as a long string of letters from the alphabet a , c , g , t , where each letter corresponds to a specific nucleotide . most higher plants and animals are _ diploid _ , which means that their dna is arranged in pairs of chromosomes . each of these paired chromosomes typically has double - stranded dna , amounting to four dna strands for each chromosome pair . the nucleotides on the two dna strands of each chromosome are mirrored as a way of providing redundant information . creatures with nonpaired chromosomes ( such as bacteria and viruses ) are called _ haploid_. a _ gene _ or _ genetic locus _ is a relatively - short segment of a chromosome that affects a particular trait or set of traits . _ random mating _ for a population of @xmath0 diploid individuals , along with mendel s laws for the offspring of two diploid individuals , can be statistically approximated by random sampling with replacement of @xmath1 haploid individuals to form the next generation . that is , random mating of @xmath0 diploid individuals is approximated by random mating of their @xmath1 haploid genomes . proteins are built from peptides , which are strings of amino acids . peptide strings are created by reading a sequence of _ codons _ , which are consecutive triples of nucleotides , from one or more subsegments of a gene . one amino acid is added to the peptide for each codon of dna . ( the amino acids and codons have no chemical similarity to one another . ) there are around twenty different amino acids as opposed to @xmath2 possible codons on one dna strand , so that the mapping of codons to amino acids can not be one to one . about half of amino acids are encoded by four different codons that vary in the third codon position . most of the remaining amino acids are encoded by two different codons . a mutation at one of the three sites in a codon position is called a _ silent _ mutation if the resulting new codon encodes the same amino acid . a site mutation that results in a different amino acid is called a _ replacement _ mutation . most mutations at first and second position sites in codons are replacement , while mutations at the third position can be either silent or replacement . the majority of replacement mutations severely damage the gene product and may be lethal to the host , but a substantial number have mild to moderate effects and can be advantageous . a dna site is _ polymorphic _ in a population if there is more than one nucleotide at that site . if there is only one nucleotide at that site in a population , the site is _ monomorphic _ and the population is _ fixed _ at that nucleotide . the _ fitness _ of a gene is defined as the expected relative number of surviving offspring ( genes ) in the next generation that are descendants of that gene , assuming constant lifetimes for individuals . suppose that we have @xmath3 haploid individuals that are of one of two types @xmath4 or @xmath5 at a particular site or gene , where @xmath5 is the mutant type and @xmath4 is the pre - existing or `` wild type . '' assume that @xmath5 has the fitness coefficient @xmath6 relative to the pre - existing @xmath4 with @xmath7 . if @xmath8 is scaled as @xmath9 for a large haploid population size @xmath3 , then @xmath10 is the _ selection coefficient _ of the mutant . the mutation is favorable if @xmath11 , detrimental if @xmath12 and neutral if @xmath13 . we first derive two results that lead to poisson random fields ( see below ) for the distribution of site polymorphisms in a limiting infinitely large random - mating population . the population results can not be directly applied to data , both because we can not sample the entire population and also because the results imply that there are an infinitely large number of site polymorphisms in each gene , most of which have very small population frequencies ( see below ) . however , we can use the population results to derive the distribution of sample statistics for aligned dna sequences arising from two related but not too distantly related species . the sample statistics will turn out to be independent poisson with means depending on population parameters for mutation and selection rates . these , in turn , lead to likelihood methods for the estimation of mutation , selection and divergence time parameters . the proofs of the population results are deferred to sections [ sec4][sec8 ] . mathematically , the key steps in the proofs of both of the main population results depend on trotter s theorem for the dual process of a diffusion process ( section [ sec6 ] ) . the sampling formulas are discussed in sections [ sec22 ] and [ sec23 ] with the details in section [ sec3].=1 the first result describes the limiting distribution of the population proportions of mutant nucleotides at polymorphic sites at a single genetic locus in a large population . under the assumptions of section [ sec1 ] , this distribution is a poisson random field ( prf ) @xcite on @xmath14 , where each point is a population frequency ratio at a different site ( see theorem [ theo21 ] below ) . the second result ( theorem [ theo22 ] ) describes the expected number of polymorphic sites that have become fixed in the population at the mutant nucleotide since time 0 . this random variable is also poisson and is independent of the prf derived in theorem [ theo21 ] . silent and replacement polymorphic sites are modeled separately and provide different information ( see below ) . we begin with a model for the frequency of a mutant nucleotide at a single site in a finite population of haploid size @xmath3 . assume that mutation is sufficiently rare at the site level that repeat mutations at the same site can be neglected . we use moran s second model @xcite for which , at each step , an individual is randomly chosen from the population with equal probabilities and replaced by a copy of a second individual ( perhaps the same as the first ) chosen with probability proportional to the fitness . then the number @xmath15 of individuals in the population that have the mutant nucleotide at a particular site is a markov chain on @xmath16 with transition function @xmath17 for @xmath18 with @xmath19 , where @xmath20 is the fitness of the mutant nucleotide @xmath5 with respect to @xmath4 . a generation corresponds to @xmath3 time steps . the states @xmath21 and @xmath3 are traps corresponding to the loss of all mutant ( @xmath5-type ) or original ( @xmath4-type ) nucleotides , respectively . we write @xmath22 instead of @xmath15 when we want to emphasize the size of the state space @xmath3 . we assume that , at time 0 , there are @xmath23 different sites that are population polymorphic with both mutant and nonmutant nucleotides . there are mutations at @xmath24 new sites at times @xmath25 the @xmath23 and @xmath24 are independent poisson with @xmath26 and typically @xmath27 ( @xmath28 ) . all mutations are assumed to occur at new sites . given @xmath23 and @xmath24 , the numbers of mutant nucleotides at these sites at times @xmath29 in the future ( for the initial polymorphic sites ) and at times @xmath30 ( for the new mutations ) are given by independent markov chains @xmath31 ( @xmath32 ) and @xmath33 ( @xmath34 ) , each of which has the transition function ( [ equ21 ] ) . by assumption , @xmath35 for @xmath28 and @xmath36 . the number of sites at which there are @xmath37 mutant nucleotides at time @xmath38 ( @xmath39 ) is @xmath40 where @xmath41 means cardinality , @xmath32 , @xmath42 and @xmath43 . thus @xmath44 for functions @xmath45 on @xmath46 $ ] . the expected value of @xmath47 in ( [ equ22 ] ) is @xmath48 where @xmath49 and @xmath50 is the @xmath38th matrix power of @xmath51 . the random variables @xmath47 are somewhat unintuitive in that each variable depends on information from many different polymorphic sites , rather than one , but turn out to have good statistical properties . we assume that the random variables @xmath24 and @xmath52 are independent and poisson distributed . then , by a slight extension of what is called bartlett s theorem @xcite , this implies that , for each fixed @xmath53 , the counts @xmath54 ( @xmath55 ) are independent poisson and thus define a poisson random field ( prf ) on @xmath56 . if the mean measures defined by ( [ equ24 ] ) for @xmath57 and @xmath58 converge weakly as @xmath59 on compact subsets of @xmath60 $ ] , then @xmath61 for @xmath62 converge weakly in the same sense to a prf on @xmath60 $ ] . the points of the limiting prf will be a countably - infinite set of diffusion processes on @xmath60 $ ] each fixed at a particular time @xmath63 . each of the markov chains @xmath64 can be approximated for large @xmath3 by a diffusion process @xmath65 on @xmath14 with time scaled as @xmath66 ( section [ sec4 ] ) . the diffusion process @xmath65 has traps at the endpoints @xmath67 , similarly to the markov chains . specifically , @xmath65 is the diffusion process on @xmath14 generated by the differential operator @xmath68 where @xmath69 for @xmath8 in ( [ equ21 ] ) . we write @xmath70 in the feller form @xmath71 where @xmath72 and @xmath73 for @xmath74 the function @xmath75 and measure @xmath76 are called the _ scale _ and _ speed measure _ of @xmath65 , respectively . the diffusion process @xmath65 has a smooth symmetric transition density @xmath77 with respect to @xmath76 such that @xmath78\\[-8pt ] & = & \int_0 ^ 1 p(t , x , y)f(y)m(dy ) \nonumber\end{aligned}\ ] ] for @xmath79 $ ] with @xmath80 . fix @xmath81 and assume that @xmath82 are scaled so that @xmath83 as mentioned earlier , each step in the moran model ( [ equ21 ] ) puts one individual at risk , so that each individual is put at risk on the average every @xmath3 time steps . thus @xmath84 and @xmath85 mean that selection at intensity @xmath10 is applied to each individual in the diffusion time scale @xmath63 . the relation @xmath86 implies that the rate of arrival of new mutations in the diffusion time scale is @xmath87 . however , the new mutant markov chains begin at @xmath35 which corresponds to states @xmath88 for the approximation diffusion . the limiting processes @xmath65 have a trap at @xmath89 , so that one might expect that only @xmath90 of the new mutant markov chains survive the first few generations . this would suggest that only of order @xmath91 of the new mutants would survive , so that the rate @xmath86 in ( [ equ29 ] ) is not paradoxical . the equilibrium distribution of the limiting prf of diffusion processes is @xmath92 ( @xcite ; see also corollary [ corr21 ] below ) . in particular , the equilibrium mean density has a @xmath93 singularity at @xmath89 in a large population but is bounded at @xmath94 . this means that , in the limit , there are an infinite number of sites that have polymorphic mutant alleles with population proportions @xmath95 . by properties of poisson random fields , @xmath96 thus the vast majority of the @xmath97 are too small for the mutant nucleotide to show up in finite samples . however , this does show that we need to allow for initial limiting prf mean measures that are not normalizable at zero . specifically , for the initial distribution of polymorphic sites , we assume that there exists a borel measure @xmath98 on @xmath14 such that @xmath99 and @xmath100 for all @xmath101 $ ] for @xmath102 in ( [ equ24 ] ) . since @xmath103 in ( [ equ27 ] ) , this is equivalent to @xmath104 our first result describes the limiting prf on @xmath14 which in turn describes the population distribution of polymorphic mutant sites . the proof is deferred to section [ sec7 ] . [ theo21 ] assume that @xmath105 satisfies ( [ equ29 ] ) and that @xmath54 defined in ( [ equ22 ] ) satisfies ( [ equ211 ] ) for @xmath102 . then for @xmath106 in ( [ equ28 ] ) @xmath107\\[-8pt ] & & \qquad= \int_0 ^ 1 q_tf(x ) \nu(dx ) + \theta\int_0 ^ 1 \frac{s(1)-s(x)}{s(1)-s(0 ) } \bigl(f(x ) - q_tf(x)\bigr ) m(dx ) \nonumber\end{aligned}\ ] ] for any @xmath79 $ ] with @xmath80 such that @xmath108 for @xmath109 extends to a continuous function on @xmath46 $ ] . the limiting prf mean density in ( [ equ213 ] ) is @xmath110 where @xmath111\\[-8pt ] & & { } -\theta\int_0 ^ 1 \frac{s(1)-s(x)}{s(1)-s(0 ) } p(t , x , y)m(dx ) . \nonumber\end{aligned}\ ] ] the first terms on the right in ( [ equ213 ] ) and ( [ equ214 ] ) are transient terms that are due to the initial ( or `` legacy '' ) polymorphisms at time @xmath112 . the remaining terms are due to new mutations that were introduced at times @xmath81 . if @xmath98 in ( [ equ211 ] ) and ( [ equ214 ] ) is the measure @xmath113 in ( [ equ210 ] ) , then the right - hand side of ( [ equ213 ] ) is identically @xmath114 , so that @xmath115 is an equilibrium measure . the following corollary shows that @xmath113 is an asymptotic measure as well ( section [ sec73 ] ) . [ corr21 ] let @xmath116 be the right - hand side of ( [ equ213 ] ) for any function @xmath45 satisfying the conditions of theorem [ theo21 ] . then @xmath117 theorem [ theo21 ] implicitly assumes that all new replacement mutants at a given genetic locus have the same selection coefficient @xmath10 . the model can be extended to allow within - locus random distributions of selective effects of new replacement mutants , for example , gaussian @xcite , exponential @xcite or gamma @xcite . the second main population result describes the limiting expected number of mutant sites that have become fixed in the population at the mutant nucleotide by time @xmath81 ( i.e. , such that the nonmutant or `` wild type '' nucleotide has been lost ) . the limit can be expressed in terms of the hitting times @xmath118 for the limiting diffusion process @xmath65 and for its dual process @xmath119 , but can also be expressed in terms of the transition density of @xmath65 . the proof is deferred to section [ sec8 ] . [ theo22 ] under the conditions of theorem [ theo21 ] , the asymptotic expected number of mutant sites that have become fixed in the population at the mutant nucleotide by time @xmath63 is @xmath120 where @xmath121 in ( [ equ216 ] ) refers to the diffusion process @xmath65 conditioned on @xmath122 , which has an entrance boundary at @xmath89 ( section [ sec6 ] ) . the right - hand side of ( [ equ216 ] ) can also be written ( section [ sec82 ] ) @xmath123\\[-8pt ] & & \hspace*{37.7pt}\qquad { } + \theta t -\theta\int_0^t\int_0 ^ 1 q(u,0+,y)s(y)^2m(dy ) \,du\biggr ) , \nonumber\end{aligned}\ ] ] where @xmath124 is the transition density of the dual - process diffusion ( section [ sec6 ] ) . theorems [ theo21 ] and [ theo22 ] give the distribution of polymorphic sites and fixations of mutant nucleotides in a large population but are not directly applicable to samples from a finite number of individuals . suppose that we have random samples from two species that are related but not extremely distantly related @xcite , for example , the _ drosophila _ species _ melanogaster _ and _ simulans _ @xcite or humans and chimpanzees @xcite . specifically , assume that we have a dna alignment of @xmath125 sequenced genes from the same genetic locus in two species , of which @xmath126 are randomly chosen from one species and @xmath127 from the second species . the sample can be viewed as an @xmath128 matrix , where @xmath129 is the number of nucleotide sites in the alignment and the matrix elements are chosen from the four letters a , c , g , t , which stand for the four nucleotides in dna . if mutation is sufficiently rare so that it never occurs more than once at the same site , then there will be at most two nucleotides at any one site in the population and hence at most two letters in any one column of the matrix . given the joint alignment , we say that a site is a _ fixed difference _ if it is monomorphic in the sample within each species but polymorphic in the two species ( i.e. , monomorphic within each species , but at different nucleotides ) . the site is _ polymorphic _ if it is polymorphic within either or both of the two species . let @xmath130 be the number of silent and replacement fixed differences in the joint alignment , and @xmath131 the number of ( within - species ) polymorphisms at silent and replacement sites . these counts can be arranged in the @xmath132 contingency table @xmath133},\ ] ] where the column headings @xmath134 refer to fixed differences and polymorphisms and the row headings @xmath135 to silent and replacement mutations , respectively . the table ( [ equ218 ] ) is called a mcdonald kreitman table @xcite and also a dprs table due to the row and column headings in ( [ equ218 ] ) . a statistically significant excess in the lower - left corner ( @xmath136 ) of the table suggests that one or both species has seen significant positive or directional selection at that gene since the two species diverged . in contrast , a statistically significant deficit of @xmath136 suggests that , instead , most new replacement mutations have been subject to strong negative selection . if the sites are assumed independent , one can apply traditional @xmath132 contingency table tests to infer an excess or deficit of replacement fixed differences @xcite . we say that a polymorphic site in the joint alignment is a _ legacy polymorphism _ if it is the descendent of a site that was polymorphic in the common ancestral species at the time of divergence . in contrast , a _ new polymorphism _ is a polymorphic site that was caused by a mutation since the time of divergence in one of the two daughter species . an important difference between the two types of polymorphisms is that legacy polymorphisms can lead to shared polymorphisms , which are sites that are polymorphic in both samples . in contrast , if mutations are sufficiently rare so that they never occur more than once at the same site , then new polymorphisms can cause sample polymorphisms in only one daughter species . another difference is that legacy polymorphisms begin at mutant population frequencies strictly between 0 and 1 , while new polymorphisms begin at mutant population frequency 0 ( or @xmath137 for the markov chain ) . thus legacy polymorphisms are more likely to have multiple copies of both the mutant and original nucleotides for relatively recent divergence times . an alternative to the @xmath132 table in ( [ equ218 ] ) , that may have more accuracy in estimating parameters for recently diverged populations , is @xmath138 } , \ ] ] where @xmath139 are numbers of sites in the joint sample that are polymorphic in only one sample and @xmath140 the numbers of sites that are polymorphic in both samples . thus @xmath141 and @xmath142 . a disadvantage of the dohrs table ( [ equ219 ] ) is that rare events in which multiple mutations occur at the same site may lead to a pair of new polymorphisms being misclassified as a legacy polymorphism . on the other hand , the more usual practice of counting sites that are polymorphic in both samples as two polymorphic sites may misclassify a single legacy polymorphism . ( see @xcite for other examples of higher - dimensional mcdonald kreitman tables . ) for simplicity , we assume that the effective population sizes @xmath143 , the mutation rates @xmath144 and the selection coefficients @xmath10 are the same in the two daughter populations . it then follows from theorem [ theo31 ] in section [ sec33 ] that , under the assumptions of theorem [ theo21 ] , the counts @xmath145 in ( [ equ218 ] ) and @xmath146 in ( [ equ219 ] ) are independent poisson with means depending on @xmath147 for replacement sites and @xmath148 at silent sites . if the effective population sizes for the two daughter species are different , the resulting formulas in section [ sec3 ] are easy to modify . let @xmath149 ( @xmath150 ) be the counts in ( for example ) ( [ equ218 ] ) in some order . let @xmath151 be the corresponding formulas in section [ sec33 ] for @xmath152 . then the likelihood of ( [ equ218 ] ) can be written @xmath153 or @xmath154 while in principle the parameters @xmath155 can be estimated by maximizing the log likelihood ( [ equ221 ] ) , normally only a few loci in a few species are sufficiently polymorphic to allow @xmath156 to be estimated from data from a single locus . more typically , likelihoods of the form ( [ equ220 ] ) are combined over many loci to form a single likelihood . the resulting expression can either be maximized to estimate the model parameters @xcite or else analyzed by bayesian methods @xcite . although the mutation rate per site per generation is normally assumed to be the same for sites in the same genetic locus , which would suggest @xmath157 , most replacement mutations are either lethal to their host or else are severely deleterious . this means that most replacement mutations are immediately lost on the diffusion time scale . this can show up in the model as a censored replacement mutation rate with @xmath158 and sometimes @xmath159 . in particular , there is no simple relation between @xmath160 and @xmath161 in ( [ equ220 ] ) . simulations have shown that methods based on ( [ equ220 ] ) for multilocus data are relatively robust to violations of basic model assumptions , such as lack of local independence of polymorphic sites @xcite . these results generalize sampling formulas in sawyer and hartl @xcite who made the approximation that the two species populations are individually at equilibrium . numerical simulations have shown that these lead to biased estimates of the population divergence time @xcite , particularly when the divergence time is small , but that estimates of the mutation and selection parameters are relatively unbiased . nonequilibrium results can also be applied to situations where a population has been subject in the past to an abrupt change in selection parameters or population size ( see , e.g. , @xcite ) . the sampling formulas in section [ sec3 ] for the poisson means in ( [ equ218 ] ) and ( [ equ219 ] ) are more complex than the equilibrium results @xcite , but likelihoods based on these means can be analyzed numerically as in the time homogeneous case . terms involving diffusion transition densities ( section [ sec3 ] ) can be estimated by the crank nicholson method @xcite . as in @xcite , gauss legendre quadrature can be used for integrals over a finite range . generalizations for random distributions of selection coefficients within loci have been handled by gauss hermite quadrature for within - locus normal variation @xcite and gauss laguerre quadrature for within - locus exponential variation @xcite . the behavior of the models described in this paper on simulated data , and applications to biological dna sequence data , will be discussed in future publications . since the two species in section [ sec22 ] are assumed to be relatively closely related on an evolutionary time scale , it is natural to assume that they have the same mutation and selection rates at each genetic locus and also the same average generation times . for simplicity , we also assume that the effective population sizes of the two daughter species are the same . this means that the scaled mutation rates @xmath162 and selection coefficients @xmath10 are the same at each genetic locus , and the scaled time @xmath63 since the divergence of the two species from their most recent common ancestral species is also the same [ see ( [ equ29 ] ) ] . as mentioned earlier , the formulas below can be easily modified if these assumptions are violated . as in section [ sec22 ] , assume that we have a dna alignment of @xmath125 sequenced genes from the same genetic locus in two species , of which @xmath126 are randomly chosen from one species and @xmath127 from the second species . each of the sample statistics @xmath145 in ( [ equ218 ] ) and @xmath146 in ( [ equ219 ] ) are counts of sites of two types , legacy and new polymorphisms , that lead to different formulas for the expected counts . it follows from section [ sec71 ] that the distribution of polymorphic site frequencies @xmath97 in a single daughter population at time @xmath81 that are derived from ancestral legacy polymorphic sites is a poisson random field with mean density @xmath163 where @xmath164 for silent mutations and @xmath165 for replacement mutations . it follows similarly from section [ sec8 ] that the number of population mutant fixations in a single daughter population by time @xmath81 that are derived from legacy polymorphic sites is poisson with mean @xmath166 assume that a particular legacy polymorphic site has population frequency @xmath167 ( @xmath168 ) for the mutant nucleotide at time @xmath112 . for a sample of size @xmath127 from a daughter population at time @xmath81 , let @xmath169 be the probability that this site is monomorphic in the sample at the wild - type ( nonmutant ) nucleotide , @xmath170 that it is polymorphic in the sample and @xmath171 the probability that it is monomorphic at the mutant nucleotide . then we have the following lemma . [ lemm31 ] for @xmath169 , @xmath170 and @xmath171 defined above , @xmath172 for @xmath173 as in section [ sec2 ] . the first terms in the formulas for @xmath174 and @xmath175 are due to population fixations . the remaining terms are due to sampling from polymorphic sites . note that @xmath176 in lemma [ lemm31 ] and that @xmath174 , @xmath177 and @xmath175 depend implicitly on @xmath178 through both @xmath179 and @xmath180 . proof of lemma [ lemm31 ] for a legacy polymorphism of initial mutant frequency @xmath167 , condition on its population frequency @xmath181 at time @xmath81 . given a sample of size @xmath125 from the two populations at time @xmath81 , let @xmath182 , @xmath183 , @xmath184 be the random numbers of legacy polymorphism sites that are fixed differences in the sample ( @xmath185 , polymorphic in only one of the two samples ( @xmath183 ) or polymorphic in both samples ( @xmath184 ) . then we have the following lemma . [ lemm32 ] the random variables @xmath186 defined above are independent poisson with means @xmath187 \nu(dx ) \\ & = & \int_0 ^ 1 \bigl(j(x , m ) + j(x , n ) - 2j(x , m)j(x , n)\bigr ) \nu(dx ) , \nonumber\\ e(l_3 ) & = & c_3(\beta ) = \int_0 ^ 1 j(x , m)j(x , n ) \nu(dx ) . \nonumber\end{aligned}\ ] ] consider the random mapping @xmath188 corresponding to random sampling at time @xmath81 for these three outcomes . then lemma [ lemm32 ] follows from lemma [ lemm31 ] and bartlett s theorem @xcite . in particular , it follows from lemma [ lemm32 ] that the expected number of polymorphic sites at time @xmath81 due to legacy polymorphisms is @xmath189 the third term in the integrand above corrects for double counting at shared polymorphisms . it follows from section [ sec72 ] that the contribution of new polymorphisms to site polymorphisms at time @xmath81 in a single population is a poisson random field with density @xmath190 it follows similarly from section [ sec8 ] that the number of mutant fixations in a single daughter population due to new polymorphisms is a poisson random variable with mean @xmath191 in particular , the limiting prf mean density for population site polymorphic frequencies in a single population is @xmath192 as in ( [ equ214 ] ) , and the limiting expected number of mutant fixed differences is @xmath193 as in ( [ equ216 ] ) . for a sample of size @xmath127 from a single population at time @xmath81 , let @xmath194 ( @xmath195 ) be the number of new polymorphic sites that have @xmath38 mutant nucleotides . then we have the following lemma . [ lemm33 ] the random variables @xmath194 defined above are independent poisson - distributed random variables with means @xmath196 for @xmath197 and @xmath198 in ( [ equ34 ] ) . consider the random mapping @xmath199 defined by binomial sampling with parameters @xmath181 and @xmath127 at each polymorphic site . the range of the mapping is a poisson random field by bartlett s theorem @xcite . thus the random variables @xmath194 for @xmath197 are independent poisson with the means in ( [ equ32 ] ) . note that bartlett s theorem applies here even though @xmath200 a.s . and @xmath201 . since the random variables @xmath194 in lemma [ lemm33 ] are independent , the expected number of polymorphic sites at time @xmath81 due to new polymorphisms is poisson with mean @xmath202 the expected number of sites due to new polymorphisms that are monomorphic in a sample of size @xmath127 at the mutant nucleotide is ( i ) the expected number of sites that have fixed in the population at the mutant nucleotide by time @xmath63 plus ( ii ) the expected number of sites that are not fixed in the population but are monomorphic at the mutant nucleotide in a sample , which is @xmath203 for @xmath204 in ( [ equ35 ] ) . we have now completed the proof of the following result . [ theo31 ] assume that the two species have the same effective population size @xmath143 and the same scaled parameter values @xmath205 and @xmath206 . then the counts @xmath146 in the table ( [ equ219 ] ) in section [ sec2 ] are independent and poisson distributed with means @xmath207\\[-9pt ] e(k_r ) & = & c_1(\beta_r ) + d_n(\beta_r , m ) + d_n(\beta_r , n ) , \nonumber\\ e(o_r ) & = & c_2(\beta_r ) + e_n(\beta_r , m ) + e_n(\beta_r , n ) , \nonumber \\ e(h_r ) & = & c_3(\beta_r ) \nonumber\end{aligned}\ ] ] for @xmath208 in ( [ equ33 ] ) , @xmath209 in ( [ equ37 ] ) and @xmath210 in ( [ equ38 ] ) . [ corr31 ] the counts @xmath145 in the @xmath132 table ( [ equ218 ] ) are independent poisson with means @xmath211 and @xmath212 in ( [ equ39 ] ) . this section describes the diffusion approximation @xcite of a single markov chain @xmath31 ( @xmath29 ) or @xmath33 ( @xmath213 ) in section [ sec2 ] , which are assumed to have the same transition function @xmath51 in ( [ equ21 ] ) . a major purpose of this section is to show that the transition density @xmath179 of the limiting diffusion process and its first partial derivative @xmath214 are smooth for @xmath81 and @xmath215 . ( the latter will be used in section [ sec82 ] . ) each of the markov chains @xmath216 will be approximated by a diffusion process @xmath65 for continuous @xmath63 scaled in terms of @xmath217 steps of the markov chain , so that @xmath66 . the limiting diffusion process @xmath65 is determined by the differential operator @xmath218 where @xmath219 for @xmath8 in ( [ equ21 ] ) . we write the operator @xmath70 in feller form @xmath220 where @xmath221 are called the _ scale _ and _ speed measure _ of @xmath70 , respectively @xcite . at silent sites , and in general if @xmath13 , the scale and speed measure are @xmath222 and @xmath223 . define @xmath224 where @xmath225 and @xmath226 . set @xmath227\dvtx f(0)=f(1)=0\},\ ] ] where @xmath228 $ ] is the class of continuous functions on @xmath229 . then , for any @xmath230}$ ] , @xmath231 is the unique solution @xmath232 such that @xmath233 for @xmath168 . ( for @xmath234 in ( [ equ42 ] ) , this does _ not _ imply @xmath235 $ ] , nor even that @xmath236 $ ] . ) since @xmath75 is increasing , @xmath237 by ( [ equ43 ] ) . hence by ( [ equ43 ] ) and ( [ equ42 ] ) @xmath238 and @xmath239 . thus @xmath240 and @xmath241 is a hilbert schmidt kernel on @xmath242 @xcite . this implies that there exists a complete orthonormal system of functions @xmath243 in @xmath242 such that @xmath244 where @xmath245 and @xmath246 . by the same arguments as after ( [ equ44 ] ) , @xmath247 and the integral equation in ( [ equ46 ] ) is equivalent to @xmath248 since @xmath249 , @xmath250 by ( [ equ41 ] ) and @xmath251 . in particular , we can assume @xmath252 . by cauchy s inequality in ( [ equ46 ] ) and by ( [ equ45 ] ) and ( [ equ43 ] ) @xmath253 since @xmath254 and @xmath255 , ( [ equ46 ] ) implies @xmath256 since @xmath257 is bounded and continuous by ( [ equ43 ] ) , it follows from ( [ equ46 ] ) and ( [ equ48 ] ) that @xmath258 extend to continuous functions on @xmath46 $ ] . if @xmath13 , the functions @xmath243 are polynomials related to jacobi polynomials @xcite . if @xmath259 , they are entire functions that are never polynomials . by mercer s theorem @xcite , @xmath260 converges absolutely and uniformly for @xmath261 . the series @xmath262 converges uniformly for @xmath261 and @xmath263 since @xmath264 . thus @xmath265 converges absolutely and uniformly for @xmath266 . in particular @xmath267 with uniform convergence for @xmath261 . it follows from ( [ equ410 ] ) and ( [ equ47 ] ) that @xmath268\\[-8pt ] ( \partial/\partial t)p(t , x , y ) & = & l_x p(t , x , y),\qquad t>0 , 0<x , y<1.\nonumber\end{aligned}\ ] ] choose @xmath269 with @xmath270 for @xmath271 and @xmath272 for some @xmath273 . let @xmath274 for @xmath275 . then by ( [ equ411 ] ) and ( [ equ412 ] ) @xmath276 where the series converges uniformly for @xmath229 and @xmath277 by ( [ equ411 ] ) . thus @xmath278)$ ] with @xmath279 it follows from maximum principles for parabolic partial differential equations @xcite that @xmath280 for @xmath281 . if @xmath282 , then @xmath283 uniformly for @xmath229 for a dense set of @xmath284 , and hence for all @xmath269 by standard arguments . it follows from the relations ( [ equ413])([equ415 ] ) that there exists a diffusion process @xmath65 with continuous sample paths with @xmath285 such that @xmath286 for @xmath79 $ ] with @xmath80 @xcite . the process @xmath65 satisfies @xmath287 up to the time that it is trapped at one of the endpoints @xmath67 , which happens eventually with probability one . the relation @xmath288 from ( [ equ410 ] ) and ( [ equ48 ] ) gives the rate at which @xmath65 is trapped at the endpoints . the exit point for @xmath65 is given by the scale function @xmath289 where @xmath290 @xcite . it follows from ( [ equ48])([equ415 ] ) that @xmath291 for all @xmath81 for @xmath292 in ( [ equ44 ] ) and that @xmath293 is a strongly continuous semigroup of linear operators on @xmath292 . the _ infinitesimal generator _ @xcite of a semigroup of linear operators @xmath294 on a banach space @xmath295 is the linear operator @xmath4 defined by @xmath296 on the linear subspace @xmath297 where @xmath298 is the norm in the banach space . if @xmath294 is strongly continuous , then @xmath299 is dense in @xmath295 and @xmath300 for some real @xmath175 . if @xmath301 as in ( [ equ417 ] ) , @xmath299 is the range of the resolvent operator @xmath302 on @xmath295 with @xmath303 this implies @xmath304 if @xmath305 @xcite . by ( [ equ412 ] ) @xmath306 is the green s operator defined by @xmath241 . a _ core _ of a strongly - continuous semigroup @xmath294 with @xmath301 is a subset @xmath307 such that @xmath308 is dense in @xmath295 . since @xmath309 is one one on @xmath295 , this is equivalent to specifying a dense subset @xmath310 and setting @xmath311 . let @xmath312 be the markov chain defined by the moran model ( [ equ21 ] ) . define @xmath313 , so that @xmath314 . it follows from standard arguments and ( [ equ21 ] ) that [ theo41 ] let @xmath315 be integers such that @xmath316 and such that @xmath317 for some @xmath167 , @xmath229 . then for any @xmath318 @xmath319 since the functions on the right - hand side of ( [ equ421 ] ) are continuous , convergence in ( [ equ421 ] ) is equivalent to uniform convergence in @xmath167 for @xmath320}$ ] . by taylor s theorem @xmath321\\[-8pt ] & & \qquad = l_xh(x ) = x(1-x)h''(x ) + \gamma x(1-x)h'(x ) \nonumber\end{aligned}\ ] ] uniformly for @xmath229 for any @xmath322 $ ] . then by trotter s theorem @xcite : [ theo42 ] for @xmath323 as above , @xmath324 $ ] , and the diffusion process @xmath65 in ( [ equ416 ] ) , @xmath325}\bigr ) \bigr ) = e_x(f(x_t ) ) = q_tf(x)\ ] ] uniformly for @xmath229 for any @xmath79 $ ] with @xmath80 . the convergence is also uniform for @xmath326 for any @xmath327 . we can not apply ( [ equ422 ] ) for @xmath322 $ ] directly for trotter s theorem , since in this case there exist @xmath328 with @xmath329 $ ] , let alone @xmath330 $ ] . however , it is sufficient to verify ( [ equ422 ] ) for all @xmath331 in a core for @xmath4 @xcite . if @xmath311 where @xmath332 is the set of all function @xmath333 such that @xmath270 for @xmath334 and @xmath335 for some @xmath336 , then @xmath337 is such a core . the result ( [ equ423 ] ) also holds for @xmath79 $ ] without the conditions @xmath80 with an appropriate modification of the definition of @xmath106 . see corollary [ corr51 ] in section [ sec5 ] below . thus , after suitable rescaling , the markov chains @xmath338 in section [ sec2 ] converge in distribution to diffusion processes @xmath339 with infinitesimal generator ( [ equ25 ] ) and scale and speed measure ( [ equ27 ] ) in section [ sec2 ] . let @xmath312 be the moran - model markov chain defined by ( [ equ21 ] ) . recall that @xmath84 as @xmath59 by ( [ equ29 ] ) . then we have the following lemma . [ lemm51 ] let @xmath340 be integers such that @xmath341 . then @xmath342 with the right - hand side replaced by @xmath343 if @xmath344 . by ( [ equ21 ] ) , @xmath345 for @xmath18 . since we can ignore `` wait states '' with @xmath346 , ( [ equ51 ] ) follows from the classical gambler s ruin problem ( see , e.g. , @xcite , pages 50 , 9294 , and moran @xcite ) . ( see lemma [ lemm82 ] for a second proof . ) as a consequence of lemma [ lemm51 ] : [ lemm52 ] let @xmath315 be integers such that @xmath316 and @xmath347 for some @xmath167 , @xmath229 . then @xmath348 for @xmath349 in ( [ equ418 ] ) and @xmath75 in ( [ equ42 ] ) . if @xmath350 $ ] , the convergence in ( [ equ52 ] ) is uniform in @xmath167 for @xmath229 . if @xmath351 and @xmath259 , it follows from ( [ equ51 ] ) that @xmath352 as @xmath59 . the proof is similar if @xmath13 . lemma [ lemm52 ] can be used to extend theorem [ theo42 ] to all @xmath79 $ ] : [ corr51 ] assume @xmath79 $ ] . let @xmath22 and set @xmath353 as in theorem [ theo41 ] . set @xmath320}$ ] . then @xmath354 } \bigr ) \bigr ) = e_x(f(x_t ) ) = \overline{q}_tf(x)\ ] ] uniformly for @xmath229 , where @xmath355 any @xmath79 $ ] can be written @xmath356 where @xmath357 , so that @xmath358 . since ( [ equ54 ] ) holds for @xmath359 by theorem [ theo42 ] , it only remains to prove ( [ equ54 ] ) for @xmath360 . however , @xmath361}\bigr)\bigr)\ ] ] for all @xmath81 . lemma [ lemm52 ] applied to both sides of ( [ equ56 ] ) implies ( [ equ54 ] ) for @xmath360 . a stronger result than lemma [ lemm52 ] is the `` local limit theorem . '' [ lemm53 ] let @xmath315 be integers with @xmath362 and set @xmath363 . then @xmath364 similarly , if @xmath365 and @xmath363 , then @xmath366 the most important cases of lemma [ lemm53 ] are when @xmath367 or @xmath368 . if @xmath369 + 1,1\}$ ] , then ( [ equ57 ] ) holds uniformly in @xmath167 for @xmath229 . similarly , ( [ equ58 ] ) holds uniformly in @xmath167 if @xmath350 $ ] . proof of lemma [ lemm53 ] the ratio in ( [ equ57 ] ) can be written @xmath370\\[-8pt ] & & \qquad= \biggl(\frac{\int_0^{x_n } e^{-yn\log(1+\sigma_n ) } \,dy } { \int_0^{x_n } e^{-y\gamma } \,dy}\biggr ) \biggl(\frac{\int_0 ^ 1 e^{-y\gamma}\,dy}{\int_0 ^ 1 e^{-yn\log(1+\sigma_n)}\,dy}\biggr ) , \nonumber\end{aligned}\ ] ] where the first line of ( [ equ59 ] ) holds for @xmath259 and the second line for all @xmath10 . since @xmath84 , then @xmath371 and the ratios converge uniformly in @xmath372 . thus ( [ equ57 ] ) follows from ( [ equ59 ] ) . similarly , @xmath373 with a corresponding relation for @xmath374 . then the ratio in ( [ equ58 ] ) can be written @xmath375 with a similar conclusion . define @xmath376 as in ( [ equ51 ] ) where , as before , @xmath377 means conditional on @xmath378 . for @xmath379 in ( [ equ21 ] ) and @xmath380 , define @xmath381 since @xmath382 for @xmath380 , @xmath383 defines a markov chain @xmath384 on @xmath385 that never attains @xmath386 and has @xmath3 as an absorbing boundary . similarly @xmath387 for all functions @xmath388 on @xmath389 . the chain @xmath390 [ or @xmath383 ] can be called an _ @xmath331-process _ of @xmath312 @xcite . we use the formula for @xmath374 in lemma [ lemm51 ] to find a diffusion approximation for @xmath384 . define @xmath392 , so that @xmath393 . the analog of theorem [ theo41 ] is : [ theo61 ] let @xmath315 be integers such that @xmath362 and @xmath317 where @xmath229 . then , for any @xmath318 , @xmath394 where @xmath395 if @xmath89 . if @xmath396 + 1,1\}$ ] , the convergence is uniform in @xmath167 . by ( [ equ21 ] ) and lemma [ lemm51 ] , writing @xmath397 for ease of notation , @xmath398\\ & & \qquad=\frac{j / n(1-j / n)n } { 1+\sigma_n j / n } \biggl [ \frac{\sigma_n + \sigma_n(1+\sigma_n)^{-j } } { 1-(1+\sigma_n)^{-j } } \biggr]\\ & & \qquad \to b(x ) = \gamma x(1-x)\frac{1+e^{-\gamma x}}{1-e^{-\gamma x}}.\end{aligned}\ ] ] similarly @xmath399\\ & & \qquad = \frac{j / n(1-j / n ) } { 1+\sigma_n j / n } \biggl [ \frac{2+\sigma_n - ( 2+\sigma_n)(1+\sigma_n)^{-j } } { 1-(1+\sigma_n)^{-j } } \biggr]\\ & & \qquad \to 2x(1-x)\end{aligned}\ ] ] and @xmath400 as in section [ sec4 ] , theorem [ theo61 ] implies , by taylor s theorem , that @xmath401\\[-8pt ] & & \qquad = \widetilde{l}_xh(x ) = x(1-x)h''(x ) + b(x ) h'(x ) \nonumber\end{aligned}\ ] ] uniformly for @xmath229 for any @xmath402 $ ] for @xmath403 defined in ( [ equ62 ] ) . as suggested by @xmath404 in ( [ equ61 ] ) , the operator @xmath405 in ( [ equ63 ] ) satisfies @xmath406 the operator @xmath407 can be written in feller form as @xmath408 for scale and speed measure @xmath409 for @xmath75 and @xmath76 in section [ sec4 ] . let @xmath119 be the diffusion process in @xmath14 generated by @xmath405 . since @xmath410 the boundary point 0 is an _ entrance boundary _ for @xmath411 or @xmath119 @xcite . since 0 is an entrance boundary , @xmath412 for @xmath109 where @xmath413 ( i.e. , 0 is inaccessible for @xmath414 ) , but @xmath415 exists for all @xmath79 $ ] with a probability distribution @xmath416 on @xmath60 $ ] for @xmath81 . thus @xmath417 for any @xmath81 and , given @xmath418 , @xmath119 leaves 0 immediately @xcite . the green function for ( [ equ64 ] ) and ( [ equ65 ] ) is @xmath419 by ( [ equ43 ] ) . since @xmath420 , the kernel @xmath421 is hilbert schmidt with respect to @xmath422 . the eigenfunction equation @xmath423 is the same as ( [ equ46 ] ) for @xmath424 , so that we can take @xmath425 with the same eigenvalues @xmath426 . define @xmath427\dvtx f(1)=0\}\ ] ] and @xmath428 for @xmath429 , by ( [ equ68 ] ) and ( [ equ48 ] ) , @xmath430 since @xmath431 by the discussion in section [ sec4 ] , the operators @xmath432 preserve @xmath433 . the principal result of this section is presented in section [ sec62 ] . let @xmath434 and @xmath383 be for the dual markov chain in ( [ equ61 ] ) and set @xmath435 . as in section [ sec4 ] , the relation ( [ equ63 ] ) defining @xmath436 with uniform convergence for @xmath402 $ ] does not cover all @xmath437 since @xmath438 contains functions @xmath329 $ ] . however , if @xmath332 is the set of all @xmath429 such that @xmath439 for @xmath334 and @xmath270 for @xmath335 for constants @xmath440 with @xmath336 , then ( [ equ63 ] ) holds for all @xmath331 in the core @xmath441 . then by trotter s theorem @xcite : [ theo62 ] let @xmath315 be integers such that @xmath362 and @xmath347 for some @xmath167 , @xmath229 . then @xmath442}\bigr ) \bigr ) = \widetilde{e}_x(f(\widetilde{x}_t ) ) = \widetilde { q}_tf(x)\ ] ] for any @xmath79 $ ] with @xmath443 . if @xmath444 + 1,1\}$ ] , the convergence is uniform in @xmath167 , and is also uniform in @xmath63 for @xmath445 for any @xmath327 . a weaker version of ( [ equ611 ] ) could be obtained from theorem [ theo42 ] directly : by ( [ equ61 ] ) @xmath446 , \nonumber\end{aligned}\ ] ] where @xmath362 and @xmath447 . now @xmath448 if @xmath429 , and @xmath449 uniformly for @xmath380 by lemma [ lemm53 ] . hence by theorem [ theo42 ] @xmath450}\bigr ) \bigr ) & = & \frac1{s(x)}q_t(sf)(x ) \\ & = & \frac1{s(x)}\int_0 ^ 1 p(t , x , y)s(y)f(y)m(dy)\\ & = & \widetilde{q}_tf(x)\end{aligned}\ ] ] uniformly for @xmath451 for any @xmath336 . however , this argument does not extend to uniform convergence for @xmath229 nor to @xmath89 . the purpose of this section is to prove theorem [ theo21 ] and corollary [ corr21 ] in section [ sec2 ] . by ( [ equ23 ] ) , the expected value of the left - hand side of ( [ equ213 ] ) is @xmath452 the expected value of the sum in ( [ equ71 ] ) that corresponds to legacy polymorphisms is @xmath453 by ( [ equ61 ] ) @xmath454 where @xmath455)f(x)$ ] . thus @xmath456\\[-8pt ] & = & \sum_{i=1}^{n-1 } \biggl(\frac{h_n(i)}{s(i / n ) } \biggr ) \widetilde{q}^n_k \biggl(\frac f{h_n } \biggr)(i ) s \biggl(\frac{i}n \biggr ) \omega^n_i . \nonumber\end{aligned}\ ] ] by lemma [ lemm53 ] , @xmath457 uniformly in @xmath167 for @xmath458 $ ] , and we can write @xmath459 for @xmath460 . by the assumptions of theorem [ theo21 ] , @xmath461 extends to a continuous function on @xmath228 $ ] with @xmath462 , so by theorem [ theo62 ] @xmath463 } g([nx ] ) = \widetilde { q}_tg(x)\ ] ] uniformly for @xmath229 and @xmath326 for any @xmath327 . by ( [ equ212 ] ) @xmath464 thus by ( [ equ73 ] ) and ( [ equ75 ] ) @xmath465\\[-8pt ] & = & \int_0 ^ 1 q_tf(y)\nu(dy).\nonumber\end{aligned}\ ] ] this completes the proof of theorem [ theo21 ] for the legacy terms in ( [ equ71 ] ) . the expected value of the double sum in ( [ equ71 ] ) , which corresponds to new mutations , is @xmath466 where @xmath467 . as in ( [ equ72 ] ) @xmath468\\[-8pt ] & = & ( n\mu_n ) ( nh_n(1 ) ) \int_0^{k_n / n^2 } \widetilde{q}^n_{[n^2u]}(f / h_n)(1 ) \,du , \nonumber\end{aligned}\ ] ] where @xmath469 + 1)$ ] . as @xmath59 , @xmath86 and @xmath470 by ( [ equ29 ] ) and , since @xmath103 , @xmath471 by lemma [ lemm53 ] . thus by ( [ equ79 ] ) and ( [ equ74 ] ) and ( [ equ75 ] ) @xmath472 where @xmath460 . by ( [ equ66 ] ) and ( [ equ69 ] ) , the green operator @xmath473 is bounded in the supremum norm , and @xmath474 since @xmath475 by ( [ equ610 ] ) and @xmath476 , we conclude @xmath477 and hence @xmath478 this is the second term on the right - hand side of ( [ equ213 ] ) and completes the proof of theorem [ theo21 ] . the first term @xmath479 in the last line in ( [ equ213 ] ) equals @xmath480 as in ( [ equ77 ] ) or ( [ equ610 ] ) , where @xmath481 is bounded and @xmath482 . thus the integral in ( [ equ711 ] ) is @xmath483 by ( [ equ69 ] ) and converges to zero as @xmath484 . by the same argument applied with @xmath485 in place of @xmath98 , @xmath486 as @xmath484 as well . this completes the proof of corollary [ corr21 ] . by bartlett s theorem @xcite , the number of processes @xmath487 and @xmath488 that have been trapped at state @xmath3 by time @xmath38 is poisson with mean @xmath489 the first term on the right - hand side of ( [ equ81 ] ) corresponds to legacy polymorphisms and the second to new polymorphisms . the proof of theorem [ theo22 ] for both terms depends on : proof of theorem [ theo22 ] given lemma [ lemm81 ] the first sum in ( [ equ81 ] ) can be written @xmath491 let @xmath492 be integers with @xmath362 and @xmath493 , and assume @xmath494 as in section [ sec2 ] . then @xmath495 by lemma [ lemm53 ] and ( [ equ82 ] ) , with uniform convergence for @xmath229 if @xmath496 + 1,1\}$ ] . thus by ( [ equ83 ] ) and ( [ equ212 ] ) @xmath497 the second term on the right - hand side of ( [ equ81 ] ) is @xmath498\\[-8pt ] & & \qquad= ( n\mu_n)(nh_n(1 ) ) \int_0^{k / n^2 } p_1\biggl(\frac1{n^2 } t^n_n \le\frac{[n^2u]}{n^2 } \bigm| t^n_n < t^n_0\biggr ) \,du \nonumber\\ & & \qquad\to \frac\theta{s(1 ) } \int_0^t p_0(t_1\le u \mid t_1<t_0 ) \,du \nonumber\end{aligned}\ ] ] by lemma [ lemm81 ] , since @xmath86 and @xmath471 as in ( [ equ79 ] ) . combining ( [ equ84 ] ) and ( [ equ85 ] ) completes the proof of theorem [ theo22 ] . proof of lemma [ lemm81 ] assume @xmath362 and @xmath347 for @xmath229 as before . then by theorem [ theo62 ] @xmath499}\bigr)\bigr ) = \widetilde{e}_x(f(x_t)),\ ] ] if @xmath79 $ ] , @xmath500 , and @xmath443 . since @xmath3 is a trap , this implies @xmath501 and @xmath502 by two applications of fatou s theorem . if we are able to prove @xmath503 then a standard compactness argument for weak convergence would imply equality in ( [ equ86 ] ) with @xmath504 replaced by @xmath505 , which would prove lemma [ lemm81 ] . hence it is sufficient to prove ( [ equ87 ] ) . consider an arbitrary birth - and - death markov chain @xmath506 on the state space @xmath507 with absorbing endpoints . as in the moran model in ( [ equ21 ] ) , assume that the transition function can be written @xmath508 where @xmath509 , @xmath510 , and @xmath511 for @xmath512 ( see , e.g. , @xcite , pages 50 , 9294 ) . then we have the following lemma . [ lemm82 ] let @xmath513 where @xmath514 are the matrix powers of @xmath515 in ( [ equ88 ] ) . then @xmath516\\[-8pt ] g_{ij } & = & a_n \frac{h_{i\wedge j}(1 - h_{i\vee j})}{\alpha_jq_j } \nonumber\end{aligned}\ ] ] for @xmath517 , @xmath518 and @xmath519 . here @xmath520 the probabilities @xmath521 satisfy the recurrence @xmath522 for @xmath18 . this implies @xmath523 and hence @xmath524 . by summation , @xmath525 . the conditions @xmath526 and @xmath527 imply @xmath528 and hence @xmath529 . the green matrix @xmath530 satisfies the recurrence @xmath531 and thus @xmath532 . the boundary conditions @xmath533 for @xmath534 and arguments similar to those for @xmath535 lead to the formula for @xmath530 . if @xmath540 for @xmath18 and @xmath541 , then @xmath542\\[-8pt ] & = & \sum_{j=1}^{n-1 } g_n(i , j),\nonumber\end{aligned}\ ] ] where @xmath543 . then by ( [ equ812 ] ) for the dual markov chain in ( [ equ61 ] ) and lemma [ lemm82 ] @xmath544 by ( [ equ21 ] ) and ( [ equ89 ] ) , since @xmath537 . by lemma [ lemm53 ] @xmath545 uniformly for @xmath517 . since @xmath9 by ( [ equ29 ] ) , the terms in the sum in ( [ equ813 ] ) are uniformly bounded . then by lemma [ lemm53 ] again , with @xmath493 and @xmath546 in ( [ equ813 ] ) , @xmath547 for @xmath241 in ( [ equ43 ] ) , @xmath76 in ( [ equ42 ] ) , @xmath548 in ( [ equ68 ] ) and @xmath422 in ( [ equ65 ] ) . this completes the proof of ( [ equ87 ] ) and hence of lemma [ lemm81 ] and theorem [ theo22 ] . for fixed @xmath81 @xmath549\\[-8pt ] & = & 1 - \int_0 ^ 1 \frac{p(t , x , y)}{s(x ) } s(y ) m(dy ) \nonumber\end{aligned}\ ] ] by ( [ equ68 ] ) . thus @xmath550 by ( [ equ43 ] ) , ( [ equ48 ] ) and ( [ equ46 ] ) @xmath551 and @xmath552 converges uniformly for @xmath263 and @xmath261 . thus by ( [ equ815 ] ) @xmath553

this paper deals with conservation laws on networks , represented by graphs . entropy - type conditions are considered to determine dynamics at nodes . since entropy dispersion is a local concept , we consider a network composed by a single node with incoming and outgoing arcs . we extend at the classical krukov entropy obtaining two conditions , denoted by ( e1 ) and ( e2 ) : the first requiring entropy condition for all krukov entropies , the second only for the value corresponding to sonic point . first we show that in case , no riemann solver can satisfy the strongest condition . then we characterize all the riemann solvers at satisfying the strongest condition ( e1 ) , in the case of nodes with at most two incoming and two outgoing arcs . finally we focus three different riemann solvers , introduced in previous papers . in particular , we show that the riemann solver introduced for data networks is the only one always satisfying ( e2 ) . _ key words : _ scalar conservation laws , traffic flow , riemann solver , networks , entropy conditions . _ ams subject classifications : _ 90b20 , 35l65 .

0905.4408

nonlinear hyperbolic conservation laws on networks have recently attracted a lot of interest in various fields : car traffic @xcite , gas dynamics @xcite , irrigation channels @xcite and supply chains @xcite . a network is modeled by a graph : a finite collection of arcs connected together by vertices . on each arc we consider a scalar conservation law . for instance one may think to the lighthill - whitham - richards model for car traffic @xcite . however , our results applies to the other application domains . it is easy to check that the dynamic at nodes is not uniquely determined by imposing the conservation of mass through vertices . then , to completely describe the network load evolution , the first step is to appropriately define the concept of solution at a vertex . + as in the classical theory of conservation laws , this problem is equivalent to giving the solution riemann problems ( now at vertices ) . more precisely , a riemann problem at a vertex is simply a cauchy problem with constant initial conditions in each arc of the vertex . the map , which associates the solution to each riemann problem at a vertex @xmath0 , is called a riemann solver at @xmath0 . similarly to the case of a real line , one has to resort to the concept of weak solutions in the sense of distributions and there are infinitely many riemann solvers producing weak solutions . first one uses entropy type conditions inside arcs as for the real line . then , in order to select a particular solution ( i.e. a riemann solver ) at the vertex , one has to impose some additional conditions . in @xcite , for example , the authors required some rules about the distribution of the fluxes in the arcs and a maximization condition ; see also @xcite . it is then natural to ask if entropy - like conditions can be imposed also at the vertex and not only inside arcs . in this paper , we focus on a single vertex @xmath0 , composed by @xmath1 incoming and @xmath2 outgoing arcs and we extend the krukov @xcite entropy - type conditions . more precisely , we propose two different entropy conditions for admissibility of solutions , called , respectively , ( e1 ) and ( e2 ) . the condition ( e1 ) is stronger than ( e2 ) , indeed the first asks for krukov entropy condition to be verified for all entropies , while the second asks only for the precise krukov entropy corresponding to sonic point . it is interesting to note that the entropy condition ( e1 ) imposes strong restrictions both on riemann solvers and on the geometry of the vertex . indeed , riemann solvers satisfying ( e1 ) can exist only in the case of vertices with the same number of incoming and outgoing arcs . we then test our conditions on riemann solvers considered in the literature . first we can prove that the riemann solver , introduced in @xcite for data networks , satisfies ( e2 ) and , in special situations , also ( e1 ) . + then we show that the riemann solvers defined in @xcite do not satisfy ( e2 ) . however , at least for the riemann solver in @xcite , the entropy condition and the maximization procedure agree on some particular set , over which the maximization is taken . roughly speaking the solver respects the entropy condition once traffic distribution is imposed . the paper is organized as follows . section [ se : def ] introduces the basic definitions of networks and of solutions . section [ se : riemann_problem ] deals with the solution to the riemann problem at the vertex @xmath0 . moreover , we introduce the entropy conditions ( e1 ) and ( e2 ) for riemann solvers at @xmath0 . in section [ se : rs_e1 ] , we determine which riemann solvers satisfy the entropy condition ( e1 ) . the paper ends with section [ se : examples ] , which considers the riemann solvers @xmath4 , @xmath5 and @xmath6 , introduced respectively in @xcite , and analyzes what entropy conditions these riemann solvers satisfy . consider a node @xmath0 with @xmath1 incoming arcs @xmath7 and @xmath2 outgoing arcs @xmath8 . we model each incoming arc @xmath9 ( @xmath10 ) of the node with the real interval @xmath11-\infty,0]$ ] and each outgoing arc @xmath12 ( @xmath13 ) of the node with the real interval @xmath14 . on each arc @xmath15 ( @xmath16 ) , the traffic evolution is given by @xmath17 where @xmath18 $ ] , is the _ density _ , @xmath19 is the _ average velocity _ and @xmath20 is the _ flux_. hence the network load is described by a finite collection of functions @xmath21 defined on @xmath22 . for simplicity , we put @xmath23 . on the flux @xmath24 we make the following assumption * @xmath25 \rightarrow { \mathbb r}$ ] is a piecewise smooth concave function satisfying 1 . @xmath26 ; 2 . there exists a unique @xmath270,1[$ ] such that @xmath24 is strictly increasing in @xmath28 and strictly decreasing in @xmath29\sigma,1]$ ] . [ deftau ] let @xmath30 \rightarrow [ 0,1]$ ] be the map such that : 1 . @xmath31 for every @xmath32 $ ] ; 2 . @xmath33 for every @xmath32\setminus\{\sigma\}$ ] . a function @xmath34 is an entropy - admissible solution to ( [ eq : lwr ] ) in the arc @xmath15 if , for every @xmath35 $ ] and every @xmath36 smooth , positive with compact support in @xmath290,+\infty[\times \left(i_l\setminus\{0\}\right)$ ] @xmath37 [ def : weak_solution ] a collection of functions @xmath34 , ( @xmath16 ) is a weak solution at @xmath0 if 1 . for every @xmath38 , the function @xmath21 is an entropy - admissible solution to ( [ eq : lwr ] ) in the arc @xmath15 ; 2 . for every @xmath38 and for a.e . @xmath39 , the function @xmath40 has a version with bounded total variation ; 3 . for a.e . @xmath39 , it holds @xmath41 where @xmath21 stands for the version with bounded total variation . given @xmath42 $ ] , a riemann problem at @xmath0 is a cauchy problem at @xmath0 with constant initial data on each arc , i.e. @xmath43 now , we give some definitions for later use . the first one is the definition of riemann solver , which is a map giving a solution to the riemann problem . [ def : riemann_solver ] a riemann solver @xmath44 is a function @xmath45^{n+m } & \longrightarrow & [ 0,1]^{n+m}\\ & ( \rho_{1,0},\ldots,\rho_{n+m,0 } ) & \longmapsto & ( \bar\rho_1,\ldots,\bar\rho_{n+m } ) \end{array}\ ] ] satisfying 1 . [ enum:1_def_rs ] @xmath46 ; 2 . [ enum:2_def_rs ] for every @xmath10 , the classical riemann problem @xmath47 is solved with waves with negative speed ; 3 . [ enum:3_def_rs ] for every @xmath13 , the classical riemann problem @xmath48 is solved with waves with positive speed . we introduce the concepts of equilibrium and consistency for riemann solvers . the fixed points of a riemann solver are called equilibria , while a riemann solver has the consistency condition when its image is contained in the equilibria . [ def : equilibrium ] we say that @xmath49 is an equilibrium for the riemann solver @xmath50 if @xmath51 [ def : consistency ] we say that a riemann solver @xmath50 satisfies the consistency condition if , for every @xmath52^{n+m}$ ] , then @xmath53 is an equilibrium for @xmath50 . we introduce now the concepts of entropy functions and admissible entropy conditions ( e1 ) and ( e2 ) for riemann solvers we are essentially extending the krukov entropy condition to the case of a node ; see @xcite . the function @xmath54^{n+m}\times[0,1]\to{\mathbb r}$ ] , defined by @xmath55 is called entropy - flux function . [ def : entropy_rs_e1 ] a riemann solver @xmath50 satisfies the entropy condition ( e1 ) if , for every initial condition @xmath49 and for every @xmath56 $ ] , we have @xmath57 where @xmath58 . if @xmath59 , then equation ( [ eq : entropy_rs_e1 ] ) becomes @xmath60 . + if @xmath61 , then equation ( [ eq : entropy_rs_e1 ] ) becomes @xmath62 . therefore the entropy condition ( e1 ) implies the conservation identity @xmath63 . [ def : entropy_rs_e2 ] a riemann solver @xmath50 satisfies the entropy condition ( e2 ) if , for every initial condition @xmath49 , we have @xmath64 where @xmath58 . the entropy condition ( [ eq : entropy_rs_e1 ] ) can be deduced in the following way . fix , for every @xmath65 , a smooth function @xmath66 with support contained in @xmath67 $ ] for some @xmath68 and assume that @xmath69 for every @xmath70 and @xmath71 . applying the divergence theorem to the inequality @xmath72 dx dt \ge 0,\ ] ] where @xmath73 is an equilibrium at @xmath0 , we deduce . obviously , these kinds of entropies are not justified by physical considerations . finally , let us introduce sets @xmath74 and @xmath75 , related to the points [ enum:2_def_rs ] and [ enum:3_def_rs ] of definition [ def : riemann_solver ] . 1 . for every @xmath10 define @xmath76 , & \textrm { if } 0\le\rho_{i,0}\le\sigma , \vspace{.2cm}\\ { } [ 0,f(\sigma ) ] , & \textrm { if } \sigma\le\rho_{i,0}\le1 , \end{array } \right.\ ] ] and @xmath77\tau(\rho_{i,0}),1 ] , & \textrm { if } 0\le\rho_{i,0}\le\sigma,\vspace{.2cm}\\ { } [ \sigma,1 ] , & \textrm { if } \sigma \le \rho_{i,0 } \le 1 . \end{array } \right.\ ] ] 2 . for every @xmath13 define @xmath78 , & \textrm { if } 0\le\rho_{j,0}\le\sigma , \vspace{.2cm}\\ { } [ 0,f(\rho_{j,0 } ) ] , & \textrm { if } \sigma\le\rho_{j,0}\le1 , \end{array } \right.\ ] ] and @xmath79 , & \textrm { if } 0\le\rho_{j,0}\le\sigma,\vspace{.2cm}\\ \{\rho_{j,0}\}\cup[0,\tau(\rho_{j,0 } ) [ , & \textrm { if } \sigma\le\rho_{j,0}\le1 . \end{array } \right.\ ] ] the following proposition links the previous sets with definition [ def : riemann_solver ] . [ prop : stati_ammissibili ] the following statements hold . 1 . for every @xmath10 , an element @xmath80 belongs to @xmath81 if and only if there exists @xmath82 $ ] such that @xmath83 and point [ enum:2_def_rs ] of definition [ def : riemann_solver ] is satisfied . 2 . for every @xmath13 , an element @xmath80 belongs to @xmath84 if and only if there exists @xmath85 $ ] such that @xmath86 and point [ enum:3_def_rs ] of definition [ def : riemann_solver ] is satisfied . the proof is trivial and hence omitted . the main result of this section is that , if @xmath3 , then every riemann solver @xmath50 at @xmath0 does not satisfy the entropy condition ( e1 ) . we first need the following result . [ prop : n_ne_m ] fix a node @xmath0 with @xmath1 incoming arcs and @xmath2 outgoing arcs and a riemann solver @xmath50 satisfying the entropy condition ( e1 ) . denote with @xmath87 the image through @xmath50 of the initial condition @xmath49 . 1 . if @xmath88 , then @xmath89 . 2 . if @xmath90 , then @xmath91 . consider first the case @xmath88 . suppose by contradiction that @xmath92 . define the set @xmath93 and fix @xmath94 . thus , the entropy inequality @xmath95 becomes , @xmath96\ge \sum_{j\in\{n+1,\ldots , n+m\}\setminus j}\left[f(\bar\rho_j)-f(k)\right ] + \sum_{j\in j}f(k).\ ] ] by point [ enum:1_def_rs ] of definition [ def : riemann_solver ] , we deduce that @xmath97 where @xmath98 denotes the cardinality of @xmath0 ; thus @xmath99 , which is a contradiction . consider now the situation @xmath90 . by contradiction we assume that @xmath100 . define the set @xmath101 and fix @xmath102 . thus , the entropy inequality @xmath95 becomes , @xmath103 -\sum_{i\in i}f(k)\ge \sum_{j = n+1}^{n+m}\left[f(k)-f(\bar\rho_j)\right].\ ] ] by point [ enum:1_def_rs ] of definition [ def : riemann_solver ] , we deduce that @xmath104 , which is a contradiction . [ thm : n_ne_m ] fix a node @xmath0 with @xmath1 incoming arcs and @xmath2 outgoing arcs and suppose that @xmath105 . every riemann solver @xmath50 at @xmath0 does not satisfy the entropy condition ( e1 ) . suppose , by contradiction , that there exists a riemann solver @xmath50 at @xmath0 satisfying the entropy condition ( e1 ) . assume @xmath88 and consider an initial condition @xmath49 satisfying @xmath106 for every @xmath10 . if @xmath107 , then , by proposition [ prop : n_ne_m ] , there exists @xmath108 such that @xmath109 , which is a contradiction since the wave @xmath110 has not negative speed . assume now @xmath90 and consider an initial condition @xmath49 satisfying @xmath111 for every @xmath13 . by proposition [ prop : n_ne_m ] , if @xmath107 , then there exists @xmath112 such that @xmath113 , which is a contradiction since the wave @xmath114 has not positive speed . in this section we determine which riemann solver satisfies the entropy condition ( e1 ) , in the sense of definition [ def : entropy_rs_e1 ] , for nodes with @xmath115 . in the case @xmath3 , theorem [ thm : n_ne_m ] implies that every riemann solver does not satisfy ( e1 ) . moreover if @xmath116 , then there exists exactly one riemann solver at @xmath0 satisfying ( e1 ) , while if @xmath117 , then there exist infinitely many riemann solvers satisfying ( e1 ) ; see sections [ sse : n = m=1 ] and [ sse : n = m=2 ] . we do not treat the case @xmath118 , for the huge number of different situations . in this subsection , we fix a node @xmath0 with one incoming and one outgoing arc . the following result holds . [ prop:1x1 ] a riemann solver @xmath50 at @xmath0 satisfies the entropy condition ( e1 ) if and only if , for every initial datum @xmath120 , the image @xmath121 satisfies either @xmath122 or @xmath123 consider first a riemann solver @xmath50 satisfying the entropy condition ( e1 ) . by [ enum:1_def_rs ] of definition [ def : riemann_solver ] , it is clear that @xmath124 . assume by contradiction that @xmath125 . since @xmath124 , we easily deduce that @xmath126 . putting @xmath127 in equation ( [ eq : entropy_rs_e1 ] ) we derive @xmath128 which is , by assumptions , equivalent to @xmath129 , and so we get a contradiction . consider now a riemann solver @xmath50 such that , for every initial datum @xmath120 , the image @xmath121 satisfies either ( [ eq : cond1_e1 ] ) or ( [ eq : cond2_e1 ] ) . it is trivial to prove that ( e1 ) holds . there exists a unique riemann solver @xmath50 at @xmath0 satisfying the entropy condition ( e1 ) . this riemann solver satisfies the consistency condition and coincides with the riemann solver introduced in @xcite for traffic or with the riemann solver introduced in @xcite . fix an initial datum @xmath120 . we show that there exists a unique @xmath130 , which is the image of an entropy admissible riemann solver . if @xmath131 , then we claim that @xmath132 . assume by contradiction that @xmath133 . in this case either @xmath134 or @xmath126 . by proposition [ prop:1x1 ] , the only possibility is @xmath134 . by proposition [ prop : stati_ammissibili ] , either @xmath135 or @xmath136 . in the first case @xmath137 , while in the second one @xmath138 . it is not possible . assume now that @xmath139 . we have some different possibilities . 1 . @xmath140 . by proposition [ prop : stati_ammissibili ] , we deduce that @xmath141 $ ] . moreover , by proposition [ prop:1x1 ] , we deduce that @xmath135 ; hence @xmath142 . this solution respects all the properties of definition [ def : riemann_solver ] and the entropy condition ( [ eq : entropy_rs_e1 ] ) . 2 . @xmath143 . by proposition [ prop : stati_ammissibili ] , we deduce that @xmath144 $ ] . moreover , by proposition [ prop:1x1 ] , we deduce that @xmath136 ; hence @xmath145 . this solution respects all the properties of definition [ def : riemann_solver ] and the entropy condition ( [ eq : entropy_rs_e1 ] ) . 3 . @xmath146 . by proposition [ prop : stati_ammissibili ] , we deduce that @xmath135 or @xmath147 and that @xmath136 or @xmath148 . + if @xmath149 , then , by proposition [ prop:1x1 ] , the only possibility is that @xmath135 and @xmath136 . + if @xmath150 , then , by proposition [ prop:1x1 ] , the only possibility is that @xmath151 . + finally , if @xmath152 , then , by proposition [ prop:1x1 ] , the only possibility is that @xmath132 . + in all the cases , the solution respects all the properties of definition [ def : riemann_solver ] and the entropy condition ( [ eq : entropy_rs_e1 ] ) . 4 . @xmath153 . by proposition [ prop : stati_ammissibili ] , we deduce that @xmath154 and @xmath155 . by proposition [ prop:1x1 ] , the only possibility is that @xmath156 . the solution respects all the properties of definition [ def : riemann_solver ] and the entropy condition ( [ eq : entropy_rs_e1 ] ) . the proof is completed . in @xcite , the authors described all the riemann solvers , with suitable properties , for nodes @xmath0 with @xmath116 . the unique riemann solver @xmath50 satisfying ( e1 ) corresponds to the riemann solver generated by the set @xmath157 and described in section 3.1 of @xcite . [ rmk : f1f2 ] one can try to generalize the entropy condition ( e1 ) , at least for nodes with @xmath116 , to the case of fluxes depending on the arcs . unfortunately this is not a trivial problem . consider indeed the following example . let @xmath158 \to { \mathbb r}$ ] , @xmath159 \to { \mathbb r}$ ] be two fluxes satisfying @xmath160 and assume that : 1 . @xmath161 is the flux in the arc @xmath162 ; 2 . @xmath163 is the flux in the arc @xmath164 ; 3 . @xmath165 is the point of maximum for both @xmath161 and @xmath163 ; 4 . @xmath166 for every @xmath1670,1[$ ] . choose @xmath168 such that @xmath169 and take @xmath170 $ ] ; see figure [ fig : remark_f1f2 ] . then , the entropy condition becomes @xmath171 which is equivalent to @xmath172 . the last inequality does not hold for @xmath173 and for all @xmath170 $ ] near to @xmath174 . .,width=226 ] consider a riemann solver @xmath50 for a node @xmath0 with two incoming and two outgoing arcs . in this subsection , we assume that @xmath175 denotes an equilibrium for @xmath50 . recall that the equilibrium must satisfy @xmath176 . by symmetry , we may assume also that ( h1 ) : : @xmath177 and @xmath178 . the results of this subsection are summarized in table [ tab : equilibrium ] . [ prop:0bad ] assume ( h1 ) and that every @xmath179 ( @xmath180 ) is a good datum . 1 . if @xmath50 satisfies the entropy condition ( e1 ) , then @xmath181 . 2 . if @xmath181 , then @xmath182 , for every @xmath56 $ ] . since all the data are good , then @xmath183 . if @xmath184 $ ] , then the entropy condition ( e1 ) becomes @xmath185 which is equivalent to @xmath186 . this implies that @xmath187 and so @xmath188 . if @xmath189 $ ] , then in the same way we deduce that @xmath190 . finally , if @xmath191 $ ] , then ( [ eq : entropy_rs_e1 ] ) , coupled with the previous results , becomes @xmath192 which is equivalent to @xmath193 . therefore @xmath194 and the conclusion follows . [ prop:1bad ] assume ( h1 ) and that the equilibrium @xmath175 for @xmath50 is composed by three good data and one bad datum . 1 . assume that the bad datum is in an incoming arc , say @xmath195 . + if @xmath50 satisfies ( e1 ) , then @xmath196 and both @xmath197 and @xmath198 belong to @xmath199 $ ] . + if @xmath196 and both @xmath197 and @xmath198 belong to @xmath199 $ ] , then @xmath200 for every @xmath56 $ ] . 2 . assume that the bad datum is in an outgoing arc , say @xmath201 . + if @xmath50 satisfies ( e1 ) , then @xmath202 and both @xmath203 and @xmath204 belong to @xmath205 $ ] . + if @xmath202 and both @xmath203 and @xmath204 belong to @xmath205 $ ] , then @xmath200 for every @xmath206 $ ] . first assume that the bad datum is in an incoming arc and the riemann solver satisfies the entropy condition ( e1 ) . without loss of generality , suppose that @xmath207 , @xmath208 and @xmath209 . we have three possibilities . ( a ) : : @xmath210 . ( b ) : : @xmath211 . ( c ) : : @xmath212 . consider the case * ( a)*. if @xmath213 $ ] , then the entropy condition ( e1 ) becomes @xmath214 equivalent to @xmath215 , which is true . + if @xmath216 $ ] , then the entropy condition ( e1 ) becomes @xmath217 equivalent to @xmath218 , which implies that @xmath196 . + if @xmath184 $ ] , then the entropy condition ( e1 ) reads @xmath219 equivalent to @xmath220 , which is true . consider the case * ( b)*. if @xmath221 $ ] , then the entropy condition ( e1 ) reads @xmath222 equivalent to @xmath223 . this implies that @xmath224 and so we are in the case * ( a)*. consider the case * ( c)*. if @xmath184 $ ] , then the entropy condition ( e1 ) becomes @xmath222 equivalent to @xmath223 . this implies that @xmath188 . + if @xmath191 $ ] , then the entropy condition ( e1 ) reads @xmath225 i.e. @xmath226 . this implies that @xmath227 and so we have a contradiction since , by case * ( a ) * , @xmath228 and so @xmath229 . the second statement in the case the bad datum is in an incoming arc easily follows . assume now that the bad datum is in an outgoing arc and that the riemann solver satisfies the entropy condition ( e1 ) . without loss of generality , suppose that @xmath230 , @xmath231 and @xmath232 . we have three possibilities . ( a ) : : @xmath233 . ( b ) : : @xmath234 . ( c ) : : @xmath235 . consider the case * ( a)*. if @xmath221 $ ] , then the entropy condition ( e1 ) becomes @xmath222 i.e. @xmath236 . this implies that @xmath202 . + if @xmath237 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath238 equivalent to @xmath239 , which is true . + if @xmath189 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath219 equivalent to @xmath240 , which is true . consider the case * ( b)*. if @xmath241 $ ] , then the entropy condition ( e1 ) reads @xmath242 which is equivalent to @xmath243 . thus we deduce that @xmath244 and so we are in the case * ( a)*. consider the case * ( c)*. if @xmath245 $ ] , then the entropy condition ( e1 ) reads @xmath246 equivalent to @xmath218 . this implies that @xmath190 . + if @xmath247 $ ] , then ( [ eq : entropy_rs_e1 ] ) reads @xmath222 i.e. @xmath223 and so @xmath248 . therefore @xmath181 , which is a contradiction . the second statement of the item 2 of the proposition easily follows . the proof is finished . [ prop:2bad ] assume ( h1 ) and that the equilibrium @xmath175 for @xmath50 is composed by two good and two bad data . 1 . assume that @xmath249 , i.e. the bad data are both in the incoming arcs . + if the riemann solver @xmath50 satisfies the entropy condition ( e1 ) , then @xmath250 . + if @xmath251 , then @xmath200 for every @xmath56 $ ] . 2 . assume that @xmath252 , i.e. the bad data are in the outgoing arcs . + if the riemann solver @xmath50 satisfies the entropy condition ( e1 ) , then @xmath253 . + if @xmath254 , then @xmath200 for every @xmath56 $ ] . 3 . assume that @xmath255 , i.e. the bad data are in the arcs @xmath162 and @xmath256 . + if the riemann solver @xmath50 satisfies the entropy condition ( e1 ) , then @xmath257 . + if @xmath257 , then @xmath200 for every @xmath56 $ ] . assume that @xmath249 and that the riemann solver satisfies the entropy condition ( e1 ) . since there are exactly two bad data , then @xmath258 . the conservation of mass at @xmath0 implies that we have the following possibilities . ( a ) : : @xmath250 . ( b ) : : @xmath253 . consider the case * ( a)*. if @xmath213 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 , which is true . + if @xmath184 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath219 which clearly holds . + if @xmath216 $ ] , then the entropy condition ( e1 ) reads @xmath217 equivalent to @xmath218 , which is true . consider the case * ( b)*. if @xmath259 $ ] , then the entropy condition ( e1 ) reads @xmath222 equivalent to @xmath223 . this implies that @xmath260 and consequently @xmath244 . the second statement of the item 1 of the proposition easily follows . assume now that @xmath252 and the riemann solver satisfies the entropy condition ( e1 ) . consequently @xmath261 . since @xmath176 , we have the following possibilities . ( a ) : : @xmath253 . ( b ) : : @xmath250 . consider the case * ( a)*. if @xmath259 $ ] , then the entropy condition ( e1 ) reads @xmath222 equivalent to @xmath223 , which is true . + if @xmath189 $ ] , then the entropy condition ( e1 ) reads @xmath219 which clearly holds . + if @xmath262 $ ] , then ( [ eq : entropy_rs_e1 ] ) reads @xmath238 equivalent to @xmath263 , which is true . consider the case * ( b)*. if @xmath213 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 . this implies @xmath260 and so @xmath244 . the second statement of the item 2 of the proposition easily follows . assume now @xmath255 , i.e. the bad data are in the arcs @xmath162 and @xmath256 , and that the riemann solver satisfies the entropy condition ( e1 ) . we have the following possibilities . ( a ) : : @xmath257 . ( b ) : : @xmath264 . consider the case * ( a)*. if @xmath213 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 , which is true . + if @xmath265 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath219 equivalent to @xmath266 , which is true . + if @xmath262 $ ] , then the entropy condition ( e1 ) becomes @xmath238 equivalent to @xmath263 , which is true . consider the case * ( b)*. if @xmath259 $ ] , then the entropy condition ( e1 ) reads @xmath222 equivalent to @xmath223 . this implies @xmath260 and so @xmath244 . the second statement of the item 3 of the proposition easily follows . the proof is finished . [ prop:3bad ] assume ( h1 ) and that the equilibrium @xmath175 for @xmath50 is composed by three bad data and one good datum . 1 . assume that @xmath208 , i.e. the good datum is in an incoming arc . + if the riemann solver satisfies the entropy condition ( e1 ) , then @xmath207 , @xmath252 , @xmath267 and @xmath268 . + if @xmath207 , @xmath252 , @xmath267 and @xmath268 , then @xmath200 for every @xmath56 $ ] . 2 . assume that @xmath269 , i.e. the good datum is in an outgoing arc . + if the riemann solver satisfies the entropy condition ( e1 ) , then @xmath249 , @xmath270 , @xmath271 and @xmath272 . if @xmath249 , @xmath270 , @xmath271 and @xmath272 , then @xmath200 for every @xmath56 $ ] . assume first that @xmath208 and that the riemann solver satisfies the entropy condition ( e1 ) . we easily deduce that @xmath273 . we have the following possibilities . ( a ) : : @xmath274 . ( b ) : : @xmath275 . ( c ) : : @xmath276 . consider the case * ( a)*. if @xmath189 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 . this implies that @xmath277 . + if @xmath278 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath279 equivalent to @xmath280 , which is true . + if @xmath184 $ ] , then the entropy condition ( e1 ) reads @xmath238 equivalent to @xmath263 , which is true . consider the case * ( b)*. if @xmath213 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 . this implies that @xmath266 . + if @xmath265 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath219 equivalent to @xmath266 . + if @xmath262 $ ] , then the entropy condition ( e1 ) reads @xmath238 equivalent to @xmath263 , which is true . consider the case * ( c)*. if @xmath216 $ ] , then the entropy condition ( e1 ) reads @xmath217 equivalent to @xmath218 . this implies that @xmath244 and so we are in the case * ( b)*. the second statement in the item 1 of the proposition easily follows . assume now that @xmath269 and that the riemann solver satisfies the entropy condition ( e1 ) . we easily deduce that @xmath281 . we have the following possibilities . ( a ) : : @xmath282 . ( b ) : : @xmath283 . ( c ) : : @xmath284 . consider the case * ( a)*. if @xmath189 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 , which is true . + if @xmath278 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath279 equivalent to @xmath285 , which is true . + if @xmath184 $ ] , then the entropy condition ( e1 ) reads @xmath238 equivalent to @xmath263 . this implies that @xmath286 . consider the case * ( b)*. if @xmath213 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 , which is true . + if @xmath265 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath219 equivalent to @xmath266 , which is true . + if @xmath262 $ ] , then the entropy condition ( e1 ) reads @xmath238 equivalent to @xmath263 . this implies that @xmath287 . consider the case * ( c)*. if @xmath259 $ ] , then the entropy condition ( e1 ) reads @xmath222 equivalent to @xmath223 . this implies that @xmath260 and so we are in the case * ( b)*. the second statement in the item 2 of the proposition easily follows . the proof is finished . [ prop:4bad ] assume ( h1 ) and that the equilibrium @xmath175 for @xmath50 is composed by four bad data . if the riemann solver satisfies the entropy condition ( e1 ) , then @xmath288 . moreover , if @xmath288 , then @xmath200 for every @xmath56 $ ] . it is sufficient to check the entropy condition ( e1 ) . if @xmath189 $ ] , then the entropy condition ( e1 ) reads @xmath214 equivalent to @xmath215 , which is true . + if @xmath278 $ ] , then ( [ eq : entropy_rs_e1 ] ) becomes @xmath279 equivalent to @xmath280 , which is true . if @xmath184 $ ] , then the entropy condition ( e1 ) reads @xmath238 equivalent to @xmath263 , which is true . this concludes the proof . .all the possible configurations for an equilibrium @xmath289 of a @xmath50 satisfying the entropy condition ( e1 ) . by symmetry , we assume that @xmath177 and @xmath178 , i.e. ( h1 ) holds . [ cols="^,<",options="header " , ] [ rmk : rs_e1 ] note that there exist riemann solvers satisfying the consistency condition and the entropy condition ( e1 ) . here we construct a riemann solver @xmath50 with such properties . + consider an initial condition @xmath290 . denote with @xmath291 the image of the initial condition through @xmath50 , i.e. @xmath292 if @xmath293 is the number of bad initial data , then we define @xmath50 according to the following possibilities . : : we put @xmath295 . by proposition [ prop:0bad ] , this provides an entropy admissible equilibrium . moreover @xmath296 @xmath297 . : : let @xmath298 be such that @xmath299 is a bad datum . we have two possibilities : @xmath300 or @xmath301 . + assume first @xmath302 . we put @xmath303 and @xmath304 for @xmath305 , @xmath306 . moreover we define @xmath307 and @xmath308 . + assume now @xmath309 . we put @xmath303 and @xmath304 for @xmath310 , @xmath306 . moreover we define @xmath311 and @xmath312 . + by proposition [ prop:1bad ] , these solutions provide entropy admissible equilibria . moreover @xmath296 @xmath313 . : : let @xmath314 , @xmath315 , be such that @xmath316 and @xmath317 are bad data . we have three different possibilities . + assume first that @xmath318 . in this case we put @xmath319 , @xmath320 , @xmath307 and @xmath308 . + assume now that @xmath321 . in this case we put @xmath319 , @xmath320 , @xmath311 and @xmath312 . + consider finally the last case . for simplicity suppose that @xmath322 and @xmath323 . we define @xmath319 , @xmath320 , @xmath324 and @xmath325 . by proposition [ prop:2bad ] , these solutions provide entropy admissible equilibria . moreover @xmath296 @xmath326 . : : let @xmath298 be such that @xmath299 is a good datum . we have two possibilities : @xmath300 or @xmath301 . + assume first @xmath300 ; say @xmath327 for simplicity . + if @xmath328 $ ] , then we put @xmath329 for every @xmath330 , @xmath306 and @xmath331 $ ] such that @xmath332 . + if @xmath333 and @xmath334 , then @xmath335 and @xmath336 . + if @xmath333 and @xmath337 , then @xmath338 and @xmath339 . + if @xmath340 , then @xmath341 and @xmath342 . + assume now @xmath301 ; say @xmath343 for simplicity . + if @xmath344 $ ] , then we put @xmath329 for every @xmath330 , @xmath306 and @xmath345 $ ] such that @xmath346 . + if @xmath347 and @xmath348 , then @xmath349 and @xmath350 . + if @xmath347 and @xmath351 , then @xmath336 and @xmath335 . + if @xmath352 , then @xmath353 and @xmath335 . + by propositions [ prop:2bad ] and [ prop:3bad ] , these solutions provide entropy admissible equilibria . moreover @xmath296 @xmath354 . : : we have some different cases . assume first that @xmath355 . we put @xmath356 , @xmath357 , @xmath358 and @xmath359 . + assume now that @xmath360 . for simplicity suppose that @xmath361 and @xmath334 . + if @xmath362 , then we put @xmath363 and @xmath364 . + if @xmath365 , then we put @xmath356 , @xmath357 , @xmath359 and @xmath366 $ ] such that @xmath367 . + assume finally that @xmath368 . for simplicity suppose that @xmath361 and @xmath334 . + if @xmath369 , then we put @xmath370 and @xmath371 . + if @xmath372 , then we put @xmath356 , @xmath358 , @xmath359 and @xmath373 $ ] such that @xmath374 . + by propositions [ prop:2bad ] , [ prop:3bad ] and [ prop:4bad ] , these solutions provide entropy admissible equilibria . moreover @xmath296 another example of riemann solver satisfying the entropy condition ( e1 ) for a node with two incoming and two outgoing arcs is a particular case of the riemann solver @xmath5 , defined in section [ ssec : rs2 ] ; see proposition [ prop : rs2_e1 ] . the riemann solver @xmath50 , constructed in remark [ rmk : rs_e1 ] , differs from the riemann solver @xmath5 . the key difference is that a permutation of initial data in incoming ( resp . outgoing ) arcs influences the solution in outgoing ( resp . incoming ) arcs in the case of @xmath50 , but not in the case of @xmath5 . + consider the following example . let @xmath375 be the flux . assume that @xmath376 and @xmath377 are two initial conditions . in both cases , we have only one bad datum and so , using the notation of remark [ rmk : rs_e1 ] , @xmath378 . hence we deduce @xmath379 and @xmath380 while @xmath381 see section [ ssec : rs2 ] . this section deals with some examples of riemann solvers , introduced in literature for describing car and data traffic . for each of them , we analyze the entropy conditions ( e1 ) and ( e2 ) . first we need some notation . let @xmath383 be the canonical basis of @xmath384 . for every @xmath385 , we denote @xmath386 . if @xmath387 , then we write , for every @xmath388 , @xmath389 and @xmath390 . let @xmath391 be the set of indices @xmath392 , @xmath393 , such that @xmath394 and for every @xmath395 define @xmath396 writing @xmath397 and following @xcite we define the set @xmath398 notice that , if @xmath399 , then @xmath400 . the matrices of @xmath401 will give rise of a unique solution to the riemann problem at @xmath0 . 1 . fix a matrix @xmath404 and consider the closed , convex and not empty set @xmath405 2 . find the point @xmath406 which maximizes the function @xmath407 and define @xmath408 . since @xmath409 , then @xmath410 is unique . 3 . for every @xmath10 , define @xmath411 either by @xmath412 if @xmath413 , or by the solution to @xmath414 such that @xmath415 . for every @xmath13 , define @xmath416 either by @xmath417 if @xmath418 , or by the solution to @xmath419 such that @xmath420 . finally , define @xmath421^{n+m}\to[0,1]^{n+m}$ ] by @xmath422 [ prop : no_e2_rs1 ] the riemann solver @xmath4 does not satisfy the entropy condition ( e2 ) in the sense of definition [ def : entropy_rs_e2 ] and , consequently , does not satisfy the entropy condition ( e1 ) in the sense of definition [ def : entropy_rs_e1 ] . consider a node with @xmath423 incoming and @xmath423 outgoing arcs , the flux function @xmath424 , a matrix @xmath425 and the initial conditions @xmath426 , @xmath427 , @xmath428 and @xmath429 . in this case the set @xmath430 in ( [ eq : omega ] ) is @xmath431\times\left[0,\frac7{16}\right ] : 0\le\frac{\gamma_1}3+\frac{\gamma_2}2\le\frac{15}{32},\ , 0\le\frac{2\gamma_1}3+\frac{\gamma_2}2\le1\right\}\,;\ ] ] see figure [ fig : omega_rs1_no_e2 ] . therefore we deduce that @xmath432 , @xmath433 , @xmath434 , @xmath435 , @xmath436 , @xmath437 , @xmath438 and @xmath439 . the entropy condition ( [ eq : entropy_rs_e2 ] ) in this case becomes @xmath440 which is equivalent to @xmath441 this concludes the proof . the maximization of the function @xmath442 over @xmath430 , which defines the riemann solver @xmath4 , is , however , in connection with the maximization of the entropy @xmath443 . in order to explain this fact , let us introduce some notations . + given @xmath430 in , define @xmath444 and the functional @xmath445 which is the restriction of @xmath443 on @xmath446 . note that the set @xmath447 consists in all the possible solutions at @xmath0 satisfying definition [ def : riemann_solver ] and the distribution rule , determined by the matrix @xmath404 . it is easy to see that there exists a one to one correspondence between @xmath430 and @xmath447 . + for every @xmath448 of cardinality @xmath293 , with @xmath449 , define @xmath450 and @xmath451 notice that @xmath452 and @xmath453 depend on the initial condition @xmath454 and on the matrix @xmath404 . there is a one to one correspondence between @xmath452 and @xmath453 , given by the one - to - one function @xmath455 moreover , if @xmath456 , then @xmath452 has , at most , topological dimension @xmath457 . + the following proposition holds . [ prop : entropy_restricted ] let @xmath448 be a set of cardinality @xmath293 , with @xmath449 and suppose that @xmath458 . the functional @xmath459 , restricted to @xmath453 , is given by @xmath460 + \sum_{l\in\mathcal h}\left[f(\sigma)-f(\rho_l)\right].\ ] ] 1 . @xmath462 and @xmath463 . in this case the term @xmath464 becomes @xmath465 . 2 . @xmath462 and @xmath466 . in this case the term @xmath464 becomes @xmath467 . 3 . @xmath468 and @xmath463 . in this case the term @xmath469 becomes @xmath465 . 4 . @xmath468 and @xmath466 . in this case the term @xmath469 becomes @xmath467 . let @xmath448 be a set of cardinality @xmath293 , with @xmath449 and suppose that @xmath458 . the problem of maximizing @xmath459 on the set @xmath453 is equivalent to the problem of maximizing the function @xmath442 , defined in ( [ eq : e ] ) , on the set @xmath452 . notice that , by proposition [ prop : entropy_restricted ] , the function @xmath459 on the set @xmath453 coincides with @xmath470 where @xmath471 is a constant , depending on @xmath472 and on the initial conditions . indeed , if @xmath463 , then @xmath21 is completely determined by the initial condition @xmath473 . more precisely , @xmath21 is equal to @xmath473 when @xmath473 is a bad datum , while @xmath21 is equal to @xmath474 in the other case . therefore , if @xmath475 , then we deduce that @xmath476 where @xmath477 and @xmath478 are constants . finally note that the function @xmath442 , restricted on @xmath452 , is given by @xmath479 this completes the proof . note that the set @xmath447 is , in general , disconnected , while the set @xmath430 is convex and so connected . the function @xmath459 , defined in ( [ eq : cal_g ] ) , i.e. the entropy function restricted on @xmath446 , is continuous , since it has not jumps in each connected component of @xmath447 . since there is a bijection between the sets @xmath430 and @xmath447 , then we can consider the entropy function on @xmath430 . more precisely , define the function @xmath480 satisfying @xmath481 for every @xmath482 , and consider the map @xmath483 . this map , in general , is discontinuous , since it can have jumps at every point @xmath484 with @xmath485 different subsets of @xmath486 of cardinalities less than or equal to @xmath487 . 1 . fix @xmath489 and define @xmath490 then the maximal possible through - flow at the crossing is @xmath491 2 . introduce the closed , convex and not empty sets @xmath492 3 . denote with @xmath410 the orthogonal projection on the convex set @xmath493 of the point @xmath494 and with @xmath495 the orthogonal projection on the convex set @xmath0 of the point @xmath496 . 4 . for every @xmath10 , define @xmath411 either by @xmath412 if @xmath413 , or by the solution to @xmath414 such that @xmath415 . for every @xmath13 , define @xmath416 either by @xmath417 if @xmath418 , or by the solution to @xmath419 such that @xmath420 . finally , define @xmath497^{n+m}\to[0,1]^{n+m}$ ] by @xmath498 assume @xmath500 and consider a node @xmath0 with @xmath1 incoming roads and @xmath2 outgoing roads . the riemann solver @xmath5 satisfies the entropy condition ( e2 ) in the sense of definition [ def : entropy_rs_e2 ] . : : in this situation , we deduce that @xmath503 for every @xmath482 . thus the entropy reads @xmath504 for every @xmath505 , the term @xmath506 can be minorized by @xmath507 and so @xmath508 @xmath509 . : : in this situation , we deduce that @xmath510 for every @xmath505 . thus the entropy reads @xmath511 for every @xmath482 , the term @xmath512 can be minorized by @xmath513 and so @xmath514 consider a node with @xmath423 incoming and @xmath423 outgoing arcs , the flux function @xmath424 , @xmath516 and the equilibrium configuration @xmath517 . in this case equation ( [ eq : entropy_rs_e1 ] ) becomes @xmath518 for every @xmath56 $ ] . if @xmath519 , then the previous inequality becomes @xmath520 which is clearly false . [ prop : rs2_e1 ] fix a node @xmath0 with two incoming and two outgoing arcs . if @xmath521 , then the riemann solver @xmath5 satisfies the entropy condition ( e1 ) , in the sense of definition [ def : entropy_rs_e1 ] . : : in this case we deduce that @xmath524 and so the entropy condition ( e1 ) is satisfied . @xmath525 . : : consider only the case @xmath526 , since the other case @xmath527 is completely symmetric . thus the bad datum is in an incoming arc and so we may assume that @xmath207 , @xmath208 and @xmath209 . since @xmath521 , then @xmath196 and @xmath528 . moreover , the fact that @xmath176 implies that @xmath529 by item 1 of proposition [ prop:1bad ] , the entropy condition ( e1 ) holds . @xmath530 . : : consider only the case @xmath526 , since the other case @xmath527 is completely symmetric . we have two possibilities : either the bad data are in the incoming arcs or one bad datum is in an incoming arc and the other bad datum is in an outgoing arc . + assume first that the bad data are in the incoming arcs . without loss of generality we may assume that @xmath531 and @xmath209 . since @xmath521 , then @xmath188 and so , the fact that @xmath176 implies that @xmath532 by item 1 of proposition [ prop:2bad ] , the entropy condition ( e1 ) is satisfied . + assume now that one bad datum is in an incoming arc and the other bad datum is in an outgoing arc . without loss of generality we may assume that @xmath255 and @xmath533 . since @xmath526 , then we deduce that @xmath196 . moreover @xmath521 implies that @xmath286 and so @xmath534 , since @xmath176 . therefore @xmath535 by item 3 of proposition [ prop:2bad ] , the entropy condition ( e1 ) is satisfied . @xmath536 . : : consider only the case @xmath526 , since the other case @xmath527 is completely symmetric . we have two possibilities : the good datum is in an incoming arc or in an outgoing arc . assume first that the good datum is in an incoming arc . without loss of generality , we may consider that @xmath537 and @xmath538 . since @xmath526 , then @xmath196 . moreover @xmath176 implies that @xmath539 . by item 1 of proposition [ prop:3bad ] , the entropy condition ( e1 ) is satisfied . + assume now that the good datum is in an outgoing arc . without loss of generality , suppose that @xmath540 , @xmath541 . since @xmath521 , then @xmath286 and so @xmath542 and @xmath271 since @xmath176 . by item 2 of proposition [ prop:3bad ] , the entropy condition ( e1 ) is satisfied . @xmath543 . : : in this case we have that @xmath544 . without loss of generality , suppose that @xmath545 and we conclude by proposition [ prop:4bad ] . in this subsection , we consider the riemann solver , introduced in @xcite for crossing nodes . consider a node @xmath0 with @xmath1 incoming and @xmath546 outgoing arcs and fix a positive coefficient @xmath547 , which is the maximum capacity of the node . the construction can be done in the following way . 1 . fix @xmath489 . for every @xmath10 , define @xmath548 then the maximal possible through - flow at @xmath0 is @xmath549 2 . introduce the closed , convex and not empty set @xmath550 \colon\sum_{i=1}^n\gamma_i=\min\left\{\gamma,\gamma_j\right\}\right\}.\ ] ] 3 . denote with @xmath410 the orthogonal projection on the convex set @xmath493 of the point @xmath551 and set @xmath552 . 4 . for every @xmath10 , define @xmath411 either by @xmath412 if @xmath413 , or by the solution to @xmath414 such that @xmath415 . for every @xmath13 , define @xmath416 either by @xmath417 if @xmath418 , or by the solution to @xmath419 such that @xmath420 . finally , define @xmath553^{n+m}\to[0,1]^{n+m}$ ] by @xmath554 consider a node @xmath0 with @xmath423 incoming arcs and @xmath423 outgoing ones , @xmath556 and @xmath557 . moreover , assume that @xmath558 . + we easily see that @xmath559 is an equilibrium for @xmath6 . thus we have @xmath560 consider a node @xmath0 with @xmath423 incoming arcs and @xmath423 outgoing ones , @xmath561 and @xmath562 . moreover , assume that @xmath558 . + we easily see that @xmath563 is an equilibrium for @xmath6 . thus we have @xmath564

gamma - ray bursts , which are among the most violent events in the universe , are one of the few viable candidates to produce ultrahigh energy cosmic rays . recently , observations have revealed that grbs generally originate from metal - poor , low - luminosity galaxies and do not directly trace cosmic star formation , as might have been assumed from their association with core - collapse supernovae . several implications follow from these findings . the redshift distribution of observed grbs is expected to peak at higher redshift ( compared to cosmic star formation ) , which is supported by the mean redshift of the swift grb sample , . if grbs are , in fact , the source of the observed uhecr , then cosmic - ray production would evolve with redshift in a stronger fashion than has been previously suggested . this necessarily leads , through the gzk process , to an enhancement in the flux of cosmogenic neutrinos , providing a near - term approach for testing the gamma - ray burst cosmic ray connection with ongoing and proposed uhe neutrino experiments .

astro-ph0610481

the origin of ultrahigh energy cosmic rays ( uhecr ) is one of the great remaining mysteries in astrophysics @xcite . the cosmic - ray spectrum has been measured to beyond @xmath1 ev , with a number of events with energy exceeding @xmath2 ev @xcite ; however , it is still debated how such highly energetic particles can be produced . it is now generally accepted that uhecr are of an extragalactic origin @xcite . however , above @xmath3 ev , the process of photopion production ( @xmath4 ) on the cosmic microwave background ( cmb ) is expected to lead to a significant diminution of the cosmic - ray spectrum , the well - known gzk effect @xcite . the relatively short attenuation length associated with the gzk process @xcite necessitates that the observed uhecr arise from local sources . the decay of charged pions produced in this process results in a flux of ultrahigh energy _ neutrinos @xcite . while the observed uhecr spectrum is somewhat insensitive to variations in cosmic source evolution @xcite , the cosmogenic neutrino flux can be greatly enhanced by strong evolution with redshift @xcite , as neutrinos can be produced in larger quantities due to the decreased photopion threshold ( since @xmath5 ) , and can themselves traverse cosmological distances without attenuation . few classes of astrophysical objects can possibly account for the observed cosmic - ray spectrum @xcite . active galactic nuclei ( agn ) have long been considered as possible uhecr sources @xcite . relatively recently , a potential connection between gamma - ray bursts ( grbs ) and uhecr has been explored @xcite . grbs , which are now generally accepted to be related to core - collapse supernovae @xcite , are violent events which release great amounts of energy in the form of gamma - rays ( @xmath6 erg ) @xcite . a number of models have been proposed to utilize their ultra - relativistic environment to accelerate protons to energies @xmath7 ev @xcite . it has also been noted that the source emissivity required to account for @xmath8 ev cosmic rays is comparable to that of gamma - ray bursts @xcite . , normalized to 1 at @xmath9 ) . from top - to - bottom , the metallicity - dependent grb rate density ( this work ) , the quasar ( qso ) evolution model used in refs . @xcite , and the fit to the cosmic star formation history ( sfh ) of ref . models similar to the latter two have been frequently used in cosmic - ray studies . [ fig : cr_evol],width=312 ] just as the core - collapse supernova rate density seems to follow the cosmic star formation history ( sfh ) @xcite , the same might be expected of the cosmological grb rate density @xcite . however , there is mounting evidence that grbs are _ not _ an unbiased tracer of the sfh @xcite . in particular , the host galaxies of grbs have a distinct tendency to be subluminous @xcite and metal - poor @xcite . this has been demonstrated for grb hosts both locally @xcite and at cosmological distances @xcite , which suggests that low metallicity is a key ingredient in the production of a gamma - ray burst @xcite . as we discuss in section [ ratesect ] , a rapidly rotating star , as required in the collapsar model @xcite , can retain much of its original mass and angular momentum if it is metal - poor @xcite . an anti - correlation with metallicity would imply that the cosmological grb rate peaks at a higher redshift @xcite , which now seems to be indicated by swift observations @xcite . simply put : the metallicity of the universe _ decreases _ with redshift , which implies a _ stronger _ evolution of the grb rate density than would be expected from the sfh alone . when grbs are considered as the source of uhecr ( with identical cosmic - ray production per burst ) , the change in the cosmological cosmic - ray emissivity is simply determined by the burst rate history . in the metallicity - biased grb model , this evolution is quite strong , as illustrated in fig . [ fig : cr_evol ] , even exceeding that of models used in past cosmic - ray studies , which have traced quasar ( qso ) luminosities @xcite or the sfh . we examine the effect of enhanced grb rate evolution on the expected flux of cosmogenic neutrinos , the measurement of which may provide the only way to break degeneracies between cosmic - ray models @xcite . in addition to naturally explaining the abundance of high - redshift bursts observed by swift , this strong evolution leads to a measurable neutrino signal , improving the near - term prospects for assessing this scenario with upcoming ultrahigh energy neutrino detectors @xcite . this result would still hold , in general , even if another mechanism is ultimately shown to account for the swift results through enhanced grb evolution . though we focus on the impact on cosmogenic neutrino production ( as dermer and holmes have also recently done for grb rates directly tracing several sfhs @xcite ) , the metallicity - biased evolution model would also have implications for predictions of diffuse neutrino fluxes directly produced in grbs ( e.g. , refs . @xcite ) and their prospects for detection @xcite . any treatment of cosmological cosmic - ray or neutrino production must account for source evolution . traditionally , the most commonly used cosmic - ray evolution models have tracked quasars @xcite or the sfh @xcite ( particularly in grb - related studies ) . observations indicate , however , that grbs do not faithfully trace the sfh @xcite . in fact , it now appears that the grb rate density is actually evolving more strongly than the standard sfh @xcite . a natural explanation may be found by considering the properties of grb host galaxies in the context of the single - star collapsar model @xcite . in this model , grb progenitors are rapidly - rotating wolf - rayet stars which undergo a core - collapse event that produces a black hole ( or possibly a rapidly - rotating neutron star ) @xcite . after collapse , this rapid rotation allows for the formation of highly relativistic jets which , when viewed on - axis , are observed as a burst @xcite . observationally , all known supernova counterparts of grbs are type ic , with the implication that the dying star lacked an outer hydrogen / helium envelope @xcite . the winds of wolf - rayet stars , which are typically the cause of the loss of this envelope , are known to increase in strength with stellar metallicity ( particularly iron ) @xcite . importantly , in this wind - induced loss process , angular momentum ( which is particularly important for forming jets @xcite ) is lost along with mass @xcite . this loss of angular momentum can be avoided if the progenitor has a very low metal content . in addition to having weaker winds , a rapidly - rotating , metal - poor massive star can avoid the production of an envelope altogether by completely mixing its interior , which results in the hydrogen being circulated into the core and burned @xcite . this would be impossible if the star was not metal - poor , as stellar mixing is expected to be inhibited by a high metal content @xcite . studies of grb hosts demonstrate that these galaxies tend to have a low metallicity @xcite . compared to considerations of galactic luminosity observations alone @xcite , direct spectroscopic measurements reveal metallicities that are significantly lower than expected @xcite . while the metallicity of the grb progenitor may not be directly measured , the metallicity of the galaxy itself should be indicative . a connection between grbs and metallicity implies that the cosmological grb rate should be dependent upon the formation rate of metal - poor stars . at higher redshifts , the universe is less metal - enriched than at the present @xcite , resulting in an increase in the expected evolution of the grb rate density ( compared to the sfh ) , as shown by langer and norman @xcite ( see also refs . @xcite ) . accounting for stellar evolution effects , yoon , langer , and norman calculated the expected ratio of grbs to core - collapse supernovae as a function of redshift , as seen in fig . 7 of ref . @xcite , which we approximate as @xmath10 . this is consistent with estimates of this ratio @xcite and can be used to calculate the increase in the absolute grb rate density ( of importance to cosmic - ray studies ) . as the progenitors of core - collapse supernovae are very short lived , @xmath11 is expected to closely follow the sfh ( assuming an unchanging imf @xcite ) . utilizing constraints from the diffuse supernova neutrino background @xcite ( in addition to direct observations ) , an updated sfh was derived by hopkins and beacom @xcite , which is well - described by the piecewise - linear fit : @xmath12 , where @xmath13 ( 3.4 , 0 , -7 ) when ( @xmath14 , @xmath15 , @xmath16 ) . combining this sfh with the parametrized form of the grb / sn ratio , we find the source evolution term , @xmath17 , to have the form @xmath18 with @xmath19 . in fig . [ fig : cr_evol ] , we present this evolution model , in comparison to the sfh alone , as well as the quasar ( qso ) evolution model used in refs . @xcite . note that our rate history evolves as @xmath20 for @xmath14 , which is substantially steeper than the models commonly considered . in particular , the qso model , which has been extensively used , only rises as @xmath21 . previous studies have found an evolution in grb luminosity @xmath22 @xcite , which would result in a similar effect on cosmic - ray evolution . while this may just be due to stronger beaming , we will not consider this scenario and instead focus only upon changes in the grb rate density . grb rates that evolve more strongly than the standard sfh have , in fact , recently been determined using swift data @xcite . to directly check for compatibility with observations , we calculate the expected redshift distribution ( assuming swift @xcite detection sensitivity ) for the metallicity - dependent evolution , following the grb model presented by le and dermer @xcite , and compare to a sample of 46 long bursts discovered by swift ( compiled from the updated lists of ref . @xcite ) , all of which have reliable redshifts . details of this calculation are contained in appendix [ appa ] . in particular , we assume an opening angle that varies between @xmath23 and @xmath24 ( with a power law distribution of the form @xmath25 where the index is @xmath26 ) and an absolute gamma - ray energy per grb of @xmath27 released in @xmath28 s , all of which are well within the reasonable ranges of these parameters @xcite assuming a uniform jet ( see ref . @xcite for alternative jet models ) . in fig . [ fig : cr_sw ] , we compare the distribution of swift bursts to the predictions of the metallicity - biased model and the standard sfh . the top panel suggests that the cumulative distribution of observed bursts ( shaded region ) can be approximately described by either the metallicity - dependent evolution or the sfh . however , another perspective is presented in the bottom panel , where the differential event distribution is shown ( in redshift bins of width 0.5 ) . this allows for a more direct comparison with redshift data , which reveals that most bursts are discovered between @xmath29 , in good agreement with our strongly evolving rate . this stronger evolution allows for a better fit to the data , when compared to the sfh . a more detailed discussion of the uncertainties involved in grb detection ( particularly the observed bursts that lack a reliable redshift ) , as well as the various degeneracies that exist between parameters in grb modeling , is beyond the scope of this work . we refer the interested reader to ref . @xcite ( and references therein ) . ) . the excess of low-@xmath30 events may be due to a llgrb class ( as discussed in section [ ratesect ] ) . [ fig : cr_sw],width=312 ] while the metallicity - biased model is in generally good agreement with swift observations at high redshifts , an excess of events seems to be present at low redshifts , which affects the cumulative distribution . this might be explained by a separate class of low - luminosity grbs ( llgrb ) @xcite . while these events are typically weaker than cosmological bursts , they may be several orders of magnitude more abundant in the local universe @xcite . one possibility is that these llgrbs are the result of progenitors that had a higher metal content than typical grbs , leading to a less luminous ( and less beamed ) gamma - ray output . in the context of the collapsar model , this is not an unreasonable conclusion , as higher progenitor metallicity should lead to increased loss rates of both mass and angular momentum . one consequence of the strong evolution that we present is a predicted burst rate density of @xmath31 gpc@xmath32 yr@xmath33 in the local universe ( when normalized to the observed redshift distribution ) , which quickly rises to @xmath34 gpc@xmath32 yr@xmath33 by redshift @xmath35 . the average emissivity from these bursts ( at @xmath36 ) is @xmath37 gpc@xmath32 yr@xmath38 erg mpc@xmath32 yr@xmath33 , which is comparable to the emissivity required to account for the @xmath8 ev uhecr flux , @xmath39 erg mpc@xmath32 yr@xmath33 ( as found in ref . @xcite ) . considering that there is no _ a priori _ reason for these numbers to be so similar , along with the fact that the region of the cosmic - ray spectrum of greatest interest ( @xmath8 ev ) must arise from sources at @xmath36 ( as we will discuss in section [ uhecr ] ) , this result is quite interesting . grbs may also be more luminous in cosmic rays due to a baryon loading factor ( @xmath40 ) that may be @xmath41 @xcite , which could account for any difference . these considerations , along with the isotropy of the measured cosmic - ray spectrum , allow for grbs to be further examined as a candidate to produce the observed uhecr . to be considered as a viable source of uhecr , gamma - ray bursts must have the ability to both accelerate protons to energies @xmath7 ev and generate a cosmic - ray flux adequate to explain the observed spectrum . in conventional grb models , a portion of the kinetic energy of a relativistically expanding fireball ( with lorentz factor @xmath42 few hundred ) is converted into internal energy @xcite . electrons are accelerated inside this jet by internal shocks and subsequently produce gamma - rays through synchrotron and inverse - compton processes @xcite . protons should also be shock - accelerated in a similar fashion . in the internal shock model , the shocks that accelerate protons are expected to be only mildly relativistic in the wind rest frame , resulting in an @xmath43 spectrum . in order to efficiently accelerate protons to ultrahigh energies , the time scale of acceleration should be shorter than both the wind expansion time ( to allow for an adequate period of confinement in shocked regions ) and the proton energy loss time scale . the former sets the ratio of magnetic field and electron energy densities to order unity , which is necessary in order to account for gamma - ray emission from synchrotron emission boosted to the observer s frame . the latter imposes an upper limit on magnetic field strength ( and lower limit on lorentz factor ) . see refs . @xcite for details ( and alternative models ) . cosmic rays mainly lose energy through three processes during propagation . at very high energies ( @xmath4410@xmath45 ev ) , uhecr energy loss is dominated by photopion production on the cmb , @xmath4 , which has a maximum cross section at the @xmath46(1232 ) resonance @xcite . below the photopion threshold ( and with @xmath47 ev ) , electron - positron pair production on diffuse photons , @xmath48 , becomes important @xcite . while the cross section for this process is high , each interaction results in only a small energy loss . finally , the expansion of the universe results in adiabatic energy loss , which is independent of energy . we can account for the total proton energy loss through the characteristic timescales associated with each process ( as in refs . @xcite ) : @xmath49 . the rate of energy loss for propagating protons is then @xmath50 . with the redshift - dependent energy loss rate determined , the injection energies of cosmic rays , @xmath51 , can be calculated as a function of detected energy ( @xmath52 ) and originating redshift ( @xmath30 ) through the differential equation @xmath53 where @xmath54^{1/2}$ ] ( with @xmath55 , @xmath56 , and @xmath57 ( km / s)/mpc ) . the top panel of fig . [ fig : cr_eloss ] shows cosmic - ray energy loss with redshift . the lines correspond to the injection energy needed at redshift @xmath30 in order for a cosmic ray to be observed with a given energy at earth ( @xmath9 ) . for example , a cosmic - ray proton with a measured energy of @xmath1 ev must have been produced at @xmath58 . this effect is analogous to the fazio - stecker relation for gamma rays @xcite . as energy losses above 3@xmath5910@xmath45 ev are quite severe , the observed uhecr must be produced in the local universe . this , along with the necessary conditions required to produce uhecr at all , strongly constrains the population of prospective sources . in order to be detected at a given energy at @xmath9 . shown for illustrative purposes are the realms of photopion ( dark - shaded region ) and pair - production ( light - shaded region ) losses . bottom : cosmic - ray spectra expected from the grb ( solid ) , qso ( dashed ) , and sfh ( dotted ) source evolution models , assuming @xmath60 and normalization at @xmath61 ev . all curves are well within range of experimental data above @xmath1 ev ( light - shaded region , dark - shaded when normalized to the spectral dip ) . [ fig : cr_eloss],width=312 ] in evaluation of the cosmic - ray spectrum , we have assumed that , at these energies , the cosmic - ray spectrum is entirely composed of protons with an injection spectrum @xmath62 ( per unit comoving volume per unit energy , per unit time ) , which is cut - off at a chosen @xmath63 and normalized to account for the observed spectrum of uhecr . cosmic rays injected at @xmath64 , will experience energy losses and be detected at @xmath52 . taking into account the evolution of sources , @xmath65 , we can calculate the uhecr spectrum as , @xmath66 we calculate the partial derivative @xmath67 numerically from eq . [ eq : eloss ] . in the bottom panel of fig . [ fig : cr_eloss ] , we present the expected cosmic - ray spectrum for the three source evolution models presented in fig . [ fig : cr_evol ] , with an injection spectrum of the form @xmath68 ( plotted as @xmath69 to emphasize spectral features @xcite ) . we choose the normalization of @xmath70 , @xmath71 , such that @xmath72 ev@xmath73 @xmath74 s@xmath33 sr@xmath33 at 10@xmath45 ev for these spectra ( and in our subsequent results ) , as the uncertainties at higher energies await the resolution that will be delivered by auger @xcite . this usually necessitates the local emissivity of sources between @xmath75 ev to be on the order of @xmath7610@xmath77 erg mpc@xmath32 yr@xmath33 @xcite . note that the data from the agasa , hires and yakutsk experiments @xcite differs by almost a factor of 2 in overall normalization in our @xmath78 plot ( light shaded region ) , however , the overall spectra do not significantly disagree @xcite and model - independent information can be extracted from the shape alone , especially when various experimental data are normalized to the spectral dip feature at @xmath79 ev @xcite ( dark shaded region ) . when considering the cosmic - ray spectrum resulting from grb evolution , we have followed the general form of waxman by assuming that the entire uhecr spectrum above @xmath1 ev can be accounted for by grbs , with an @xmath68 injection spectrum @xcite . however , for an injection spectrum of this form , it is challenging to account for cosmic rays with energies @xmath80 ev @xcite . strong source evolution with redshift offers a certain degree of assistance . it has been assumed previously that an extension of the galactic cosmic - ray spectrum may account for any remaining deficiency @xcite . contributions from extragalactic sources with a lower energy reach may also be considered ( e.g. , agn , cluster shocks @xcite , or llgrbs ) . if llgrbs are able to produce cosmic rays ( as suggested in ref . @xcite ) , they may be able to contribute to the flux of lower energy cosmic rays ( since they have a high local rate @xcite ) and alleviate the need for an extension of the galactic component . a recent galactic grb has also been proposed as a potential source of lower energy cosmic rays @xcite . while the milky way is generally metal - rich ( contrary to observations of typical grb hosts ) , it possesses a gradient in its metallicity distribution which may have allowed for a low - luminosity burst in the past . alternatively , one can consider an @xmath81 injection spectrum ( as might be expected from relativistic shocks ) , which may offer a better fit at lower energies @xcite . this would require a larger @xmath82 , which itself may explain grbs with low radiative efficiencies @xcite . it is important to note that the cosmic - ray spectrum itself depends only mildly upon evolution ( since the observed uhecr with @xmath83 ev must be produced at @xmath36 ) and may be explained by a combination of grbs and a lower - energy component . however , the strong evolution implied by the metallicity - biased grb model results in an increased flux of cosmogenic neutrinos . this neutrino flux , which must be present if grbs are to account for the observed uhecr , provides an independent test of the grb uhecr model , which we will now examine . the flux of cosmogenic neutrinos produced via the gzk process is quite sensitive to cosmic - ray source evolution @xcite . this is due to the unique ability of neutrinos to propagate through large distances at very high energies without appreciable energy loss ( unlike cosmic rays ) , combined with a lower threshold for photopion production at larger @xmath30 ( since @xmath5 ) . in fact , a significant portion of the expected neutrino flux originates between from @xmath84 . near the photopion production threshold , about 20% of the original proton energy is lost in each interaction . neutrinos are subsequently produced through the decay chain , @xmath85 , with each daughter neutrino ( @xmath86 ) receiving @xmath871/20 of the parent proton energy @xcite . at higher energies , multi - pion production dominates @xcite ; however , this approximation remains viable , as the inelasticity of each interaction increases ( approaching 50% ) . , width=312 ] waxman and bahcall ( wb ) have presented an upper bound on cosmogenic neutrino production ( shown in figs . [ fig : gzk1 ] and [ fig : gzk2 ] as a shaded band ) based on the assumption of an @xmath68 injection spectrum , with normalization chosen between @xmath1 ev to @xmath88 ev to produce the observed cosmic - ray spectrum @xcite . this yields an energy - dependent rate of cosmic - ray generation of @xmath89 erg mpc@xmath32 yr@xmath33 ( with @xmath90 ) . the total @xmath91 energy flux at earth ( not corrected for oscillations ) is estimated as @xmath92 where @xmath93 yr is the hubble time and the factor @xmath94 arises from the assumption that only one quarter of the energy lost is carried away by muon neutrinos . adiabatic redshift losses and the effect of source evolution are taken into account by @xmath95 ( estimated to be @xmath96 0.6 with no source evolution , @xmath87 3 with qso - like evolution ) @xcite . the neutrino spectrum produced in the gzk process may be better approximated for a general cosmic - ray injection spectrum and source evolution through a somewhat more sophisticated approach . as the energy loss distance at these energies is relatively short , we assume that cosmic rays lose all of their energy rapidly . the fraction of the original proton energy that is lost to neutrinos can then be parametrized with a gradual step function , @xmath97 , where 0.45 is the asymptotic fraction of injected cosmic - ray energy transferred to neutrinos above @xmath98 ev ( as shown in fig . 1 of ref . @xcite ) and @xmath99 ev governs the onset of photopion production above @xmath100 ev . the total neutrino flux at earth can be cast as @xmath101 \frac{de^\prime_p}{de_\nu } \frac{\mathcal{w}(z)}{dz / dt } dz\ , , \label{eq : nuflux}\ ] ] where @xmath102 and the factor @xmath103 reflects the approximation of each daughter neutrino receiving about @xmath104 of the injected proton energy . a more detailed derivation of this formula is given in appendix [ appb ] . the additional factor of @xmath105 in @xmath106 accounts for the lowering of the photopion energy threshold as the cmb temperature increases at higher redshift . this simple formulation is similar to the notation based on neutrino yield functions often used in prior studies @xcite and provides a reasonably accurate neutrino spectrum which agrees rather well with the literature ( in the energy range most interesting to uhe neutrino detectors ) , with a deviation from the simulated spectra not larger than the variations typically seen between such simulations . the overall normalization ( set by @xmath70 ) is again chosen such that the corresponding predicted cosmic - ray flux agrees with the measured cosmic - ray data at @xmath1 ev ( where it is well - determined ) , as discussed in section [ uhecr ] . given an uhecr source evolution , @xmath65 , we can calculate the expected flux of cosmogenic neutrinos produced through photopion production on the cmb . while our simple , analytic method allows for a more transparent look at the effects of source evolution , it does not fully encompass the particle physics involved , particularly the low and high energy tails of the distributions of particle decays ( which affect the low- and high - energy ends of the neutrino spectrum ) . we utilize the publicly - available simulation package crpropa @xcite , which uses the sophia @xcite code to handle particle processes , for this purpose . we have made use of the analytical estimate described above and an extensive comparison to previous results presented in the literature ( with similar parameters ) , to verify the results . at the redshifts of greatest interest , @xmath107 , only cosmic rays with @xmath108 ev can ever have the ability to produce neutrinos through the gzk process ( even with a decreased photopion threshold ) . at lower energies , photopion production can be facilitated by the cosmic infrared background ( irb ) , resulting in additional neutrinos of correspondingly lower energy @xcite . additionally , extragalactic magnetic fields may increase the path length of cosmic - ray propagation @xcite . however , due to the various uncertainties involved , we have chosen not include these effects . , width=312 ] fig . [ fig : gzk1 ] compares the resulting neutrino flux from the grb model to those expected from qso - like evolution and the sfh , assuming an @xmath68 injection spectrum ( with a sharp cutoff at @xmath88 ev ) . the flux shown is the sum of all neutrino flavors ( both electron and muon types are produced at the source ) , as a detector such as anita has a nearly equal taste for each flavor . as can be clearly seen , the stronger evolution of the grb model produces a larger neutrino flux than other models . also shown are the expected sensitivities of anita @xcite ( which began taking data in late 2006 ) and the proposed arianna / salsa detectors @xcite . the flux predicted by strong grb evolution , with parameters given by our metallicity - biased model , extends well into the reach of anita . this presents an opportunity ( within the next few years ) to either confirm or significantly constrain the parameters of this model in a manner that would not be possible by cosmic - ray observations alone . we can also consider the effect of varying the assumed parameters in the injection spectrum . since only cosmic rays with energies greater than @xmath1 ev will ever contribute significantly ( even at high redshifts ) to the neutrino flux , harder cosmic - ray spectra will result in larger fluxes . shown in fig . [ fig : gzk2 ] are the neutrino fluxes arising from cosmic - ray injection spectra of @xmath109 , @xmath110 , and @xmath111 , assuming the strong evolution . in particular , a @xmath112 spectrum might arise from relativistic shocks ( see ref . @xcite for details ) . since @xmath113 would be quite difficult to reconcile with cosmic - ray observations , this range can be regarded as an upper bound , which will soon be constrained by anita . assuming a lower cutoff energy would be similar in effect to a softening of the spectral index . another interesting scenario is producing a softer spectral index ( @xmath114 ) , even if each grb possesses an intrinsic @xmath68 spectrum . as proposed in ref . @xcite , if the distribution of the high - energy cutoffs of cosmic - ray sources follows a power law , then the overall spectrum that is observed will follow a power law with a different slope . such an argument may be particularly applicable to grbs , as cosmic rays might be produced by a spectrum of sources ranging from llgrbs to strong grbs with very metal - poor progenitors . the increase in our understanding of gamma - ray bursts can be traced to the improved capabilities now available to study these phenomena . the ability to quickly and accurately localize a grb has led to the establishment of a grb - supernova connection and allowed for the study of the host galaxies in which these events occur . observations indicate that these hosts tend to be underluminous , star - forming and metal - poor . a connection between a grb and its host galaxy metallicity is not surprising in the context of the collapsar model , which requires rapidly - rotating stars that lack a h / he envelope ( in order to be in accord with supernova observations ) . these requirements can be satisfied by a metal - poor progenitor star . introducing a metallicity bias leads to an accelerated evolution of the cosmic grb rate density ( relative to the sfh ) , which allows for a better fit to recent swift data . in models that attribute ultrahigh energy cosmic rays to gamma - ray bursts , this evolution provides the history of uhecr production in the universe . if grbs are to account for the observed cosmic - ray spectrum , they _ must _ generate a flux of cosmogenic neutrinos . a broad approach which utilizes all available means of observation will continue to be of great utility in unveiling the mysteries of gamma - ray bursts , particularly a relation to uhecr . further observations of grb host galaxies , along with those of core - collapse supernovae ( particularly type ib / ic ) , will provide invaluable information concerning grb progenitors . a systematic study of host metallicities would be important for firmly establishing the grb - metallicity anti - correlation , which , along with more accurate measurements of cosmic metallicity evolution @xcite , would allow for improved grb rate - related calculations . direct observations of neutrinos from grbs ( e.g. , by icecube @xcite or a km@xmath115 mediterranean detector @xcite ) would confirm the acceleration of protons to high ( though not necessarily ultrahigh ) energies . in order to produce a detectable signal , a particularly strong burst , with measured gamma - ray fluence @xmath116 erg @xmath74 ( e.g. , the recent grbs 060928 @xcite and 061007 @xcite , which were observed to have fluences very near this threshold ) is required @xcite . while the needed neutrino telescopes ( which may also reveal the sources of galactic cosmic rays @xcite ) are still under construction , future observations by swift @xcite should reveal similar bursts , in addition to determining the redshift - dependent grb rate with unprecedented precision . cosmogenic neutrinos present a unique tool to examine the grb - as - proton accelerator conjecture . while the cosmic - ray spectrum that will be measured by auger will allow for further assessments of the viability of prospective source models , the combined measurements of neutrinos and uhecr would break degeneracies between the various models , as discussed by seckel and stanev @xcite . it is well - known that strong source evolution can lead to observable cosmogenic neutrino signals ; however , a physically motivated model was lacking . the enhanced cosmogenic neutrino fluxes expected to result from strong grb evolution will allow for this model to be tested in a novel fashion . we have produced this strong evolution naturally through metallicity dependence , however , our general result is also applicable to _ any _ grb model that attempts to explain the observed swift redshift distribution with additional evolution . the sensitivity afforded by anita will allow for near - term examination which may either affirm or , if the expected flux is not found , place substantial constraints upon the model parameters . the lower detection threshold achieved by an arianna - like detector will allow for a realistic opportunity to discriminate between evolution models . measurements of fluxes consistent with that expected from grb evolution would provide compelling evidence for this model . as the relatively young field of particle astrophysics continues to progress , it is quite an exciting possibility that metals , once relegated to the realm of pure astronomy , may , in fact , hold the key to the production of the highest energy particles in the universe . * * acknowledgments.**we would like to thank john beacom for many illuminating discussions ; shinichiro ando , john cairns , charles dermer , kris stanek , floyd stecker , casey watson , and eli waxman for helpful suggestions ; and baha balantekin , jim beatty , and terry walker for useful comments on the manuscript . this work is supported by the national science foundation under career grant no . phy-0547102 to jb , the department of energy grant de - fg02 - 91er40690 , and the ohio state university . in calculating the redshift distribution of gamma - ray bursts expected to be observable by swift , we follow the model of le and dermer @xcite , using similar , but somewhat simplified , notation . this requires the assumption of a characteristic grb gamma - ray energy output ( @xmath117 ) and a duration in the grb rest frame ( @xmath118 ) , additionally assuming flat temporal and spectral burst profiles . each burst is assigned a jet opening angle , @xmath119 , which is selected from a power law distribution of the form @xmath120 , between an assumed @xmath121 and @xmath122 , normalized with @xmath123 $ ] . the observed energy flux from a grb at a luminosity distance , @xmath124 , then takes the form @xmath125 where @xmath126 is a bolometric correction factor , which accounts for the fraction of the burst spectrum within the energy band of the grb detector . solving this equation for the detector trigger sensitivity threshold , @xmath127 , yields the beaming angle , @xmath128 , required for detection . in effect , by assuming a constant @xmath117 we will use the opening angle as a proxy for the isotropic equivalent energy ( @xmath129 ) often used in grb studies ( e.g. , ref . @xcite ) . for a comoving grb rate , @xmath130 , the detectable grb rate is reduced to @xmath131 by beaming through angle , @xmath119 . we then find the detectable grb fraction as @xmath132}^{\cos \theta_{\rm min } } ( 1-\cos \theta)\ , g_0\ , ( 1-\cos \theta)^s \ , d\cos \theta\ ] ] @xmath133)^{2+s } - ( 1-\cos \theta_{\rm min } ) ^{2+s}}{(1-\cos \theta_{\rm max})^{1+s } - ( 1-\cos \theta_{\rm min})^{1+s}}\ , , \label{a1}\ ] ] we then compute the number of detected bursts per unit redshift as @xmath134 where @xmath135 takes into account cosmic time dilation , and the comoving volume element is given as @xmath136 substitution of eqs . ( [ a1 ] ) and ( [ a3 ] ) into eq . ( [ a2 ] ) agrees with the formulation presented in ref . we describe a simple , analytic method to compute the cosmogenic neutrino spectrum observed at earth , based upon the assumption that each daughter neutrino resulting from a photopion interaction receives @xmath137 of the parent proton energy . that is , a neutrino _ detected _ with energy , @xmath138 , was produced with energy , @xmath139 , at the source , from a parent proton which had energy , @xmath140 . we predict the observed ( @xmath9 ) neutrino spectrum , @xmath141,by integrating over the contributions to the flux from redshifts up to @xmath142 as @xmath143 where @xmath144 accounts for the fact that neutrinos are observed in a narrower energy bin then they were originally produced ( due to redshifting ) . evolution in the density of sources is accounted for through @xmath65 . the neutrino spectrum at production , @xmath145 , can be directly related to the cosmic - ray injection spectrum @xmath146 ( per unit comoving volume per unit energy , per unit time ) , through the conservation of the transferred energy , as @xmath147 \right ) % \frac{de_p^{\prime}}{de_\nu^\prime } \\ % % \frac{dn_{\nu}}{de_\nu^\prime } \frac{de_\nu^\prime}{de_\nu } & = & 20 \,\varphi(e_p^\prime ) \ , { \cal s}\ ! \left[(1+z ) \times e_p^{\prime } \right ] \ , \frac{d e_p^{\prime } } { d e_\nu } \label{b2}\end{aligned}\ ] ] with @xmath106 as defined in section [ nu ] . substitution of eq . ( [ b2 ] ) into eq . ( [ b1 ] ) yields eq . ( [ eq : nuflux ] ) . t. k. gaisser , arxiv : astro - ph/0501195 . m. takeda _ et al . _ , astropart . * 19 * , 447 ( 2003 ) ; r. u. abbasi _ et al . _ [ high resolution fly s eye collaboration ] , phys . b * 619 * , 271 ( 2005 ) ; a. v. glushkov _ et al . _ [ yakutsk collaboration ] proc . of 28th int . cosmic ray conf . ( tsukuba , japan ) , * 1 * , 389 ( 2003 ) . j. abraham _ et al . _ [ pierre auger collaboration ] , nucl . instrum . a * 523 * , 50 ( 2004 ) ; p. sommers [ pierre auger collaboration ] , arxiv : astro - ph/0507150 . a. m. hillas , ann . astrophys . * 22 * , 425 ( 1984 ) ; d. f. torres and l. a. anchordoqui , rept . phys . * 67 * , 1663 ( 2004 ) . k. greisen , phys . * 16 * , 748 ( 1966 ) ; g. t. zatsepin and v. a. kuzmin , jetp lett . * 4 * , 78 ( 1966 ) . f. w. stecker , phys . rev . * 21 * , 1016 ( 1968 ) . t. stanev , r. engel , a. mucke , r. j. protheroe and j. p. rachen , phys . d * 62 * , 093005 ( 2000 ) . v. s. beresinsky and g. t. zatsepin , phys . b * 28 * , 423 ( 1969 ) ; f. w. stecker , astrophys . space sci . * 20 * , 47 ( 1973 ) ; f. w. stecker , astrophys . j. * 228 * , 919 ( 1979 ) . c. t. hill and d. n. schramm , phys . d * 31 * , 564 ( 1985 ) . d. de marco and t. stanev , phys . rev . d * 72 * , 081301 ( 2005 ) . r. engel , d. seckel and t. stanev , phys . d * 64 * , 093010 ( 2001 ) ; t. stanev , nucl . suppl . * 136 * , 103 ( 2004 ) . p. l. biermann and p. a. strittmatter , astrophys . j. * 322 * , 643 ( 1987 ) ; j. p. rachen and p. l. biermann , astron . astrophys . * 272 * , 161 ( 1993 ) . e. waxman , phys . lett . * 75 * , 386 ( 1995 ) ; e. waxman , astrophys . j. * 606 * , 988 ( 2004 ) . m. vietri , astrophys . j. * 453 * , 883 ( 1995 ) ; m. vietri , d. de marco and d. guetta , astrophys . j. * 592 * , 378 ( 2003 ) . c. d. dermer , astrophys . j. * 574 * , 65 ( 2002 ) ; c. d. dermer and m. humi , astrophys . j. * 556 * , 479 ( 2001 ) . k. z. stanek _ et al . _ , astrophys . j. * 591 * , l17 ( 2003 ) ; j. hjorth _ et al . _ , nature * 423 * , 847 ( 2003 ) ; s. e. woosley and j. s. bloom , arxiv : astro - ph/0609142 . b. paczynski , astrophys . j. * 308 * , l43 ( 1986 ) ; t. piran , rev . mod . phys . * 76 * , 1143 ( 2005 ) ; p. meszaros , rept . phys . * 69 * , 2259 ( 2006 ) . e. waxman , astrophys . j. * 452 * , l1 ( 1995 ) . e. waxman and j. n. bahcall , phys . d * 59 * , 023002 ( 1998 ) ; j. n. bahcall and e. waxman , phys . rev . d * 64 * , 023002 ( 2001 ) . a. m. hopkins and j. f. beacom , arxiv : astro - ph/0601463 . t. totani , astrophys . j. * 511 * , 41 ( 1999 ) ; a. w. blain and p. natarajan , mon . not . soc . * 312 * , l35 ( 2000 ) ; v. bromm and a. loeb , astrophys . j. * 575 * , 111 ( 2002 ) . e. le floch _ et al . _ , astron . astrophys . * 400 * , 499 ( 2003 ) . e. le floch , v. charmandaris , w. j. forrest , f. mirabel , l. armus and d. devost , astrophys . j. * 642 * , 636 ( 2006 ) . j. p. u. fynbo _ et al . _ , astron . astrophys . * 406 * , l63 ( 2003 ) ; j. x. prochaska _ et al . _ , astrophys . j. * 611 * , 200 ( 2004 ) ; j. gorosabel _ et al . _ , astron . astrophys . * 444 * , 711 ( 2005 ) ; p. jakobsson _ et al . _ , mon . not . 362 * , 245 ( 2005 ) ; j. sollerman , g. ostlin , j. p. u. fynbo , j. hjorth , a. fruchter and k. pedersen , new astron . * 11 * , 103 ( 2005 ) . k. z. stanek _ et al . _ , acta astron . * 56 * , 333 ( 2006 ) . a. s. fruchter , _ et al . _ , nature * 441 * , 463 ( 2006 ) . s. e. woosley , astrophys . j. * 405 * , 273 ( 1993 ) ; a. macfadyen and s. e. woosley , astrophys . j. * 524 * , 262 ( 1999 ) ; s. woosley and a. heger , astrophys . j. * 637 * , 914 ( 2006 ) . s. c. yoon and n. langer , astron . astrophys . * 443 * , 643 ( 2005 ) . n. langer and c. a. norman , astrophys . j. * 638 * , l63 ( 2006 ) . s. c. yoon , n. langer and c. norman , astron . astrophys . * 460 * , 199 ( 2006 ) . t. le and c. d. dermer , arxiv : astro - ph/0610043 . f. daigne , e. m. rossi and r. mochkovitch , mon . not . soc . * 372 * , 1034 ( 2006 ) . b. j. boyle and r. j. terlevich , mon . not . * 293 * , l49 ( 1998 ) ; p. c. hewett , c. b. foltz , and f. h. chaffee , astrophys . j. * 406 * , l43 ( 1993 ) ; m. schmidt , d. p. schneider , and j. e. gunn , astron . j. * 110 * , 68 ( 1995 ) . d. seckel and t. stanev , phys . lett . * 95 * , 141101 ( 2005 ) . s. w. barwick _ et al . _ [ anita collaboration ] , phys . lett . * 96 * , 171101 ( 2006 ) ; p. miocinovic _ et al . _ [ anita collaboration ] , econf * c041213 * , 2516 ( 2004 ) ; s. w. barwick _ et al . _ [ anita collaboration ] , arxiv : hep - ex/0611008 . p. w. gorham , d. saltzberg , r. c. field , e. guillian , r. milincic , d. walz and d. williams , phys . d * 72 * , 023002 ( 2005 ; p. gorham , d. saltzberg , p. schoessow , w. gai , j. g. power , r. konecny and m. e. conde , phys . e * 62 * , 8590 ( 2000 ) . d. saltzberg _ et al . _ , phys . lett . * 86 * , 2802 ( 2001 ) ; s. w. barwick , arxiv : astro - ph/0610631 . k. s. capelle , j. w. cronin , g. parente and e. zas , astropart . * 8 * , 321 ( 1998 ) ; x. bertou , p. billoir , o. deligny , c. lachaud and a. letessier - selvon , astropart . * 17 * , 183 ( 2002 ) . c. d. dermer , j. m. holmes , ultra - high energy cosmic rays , gamma - rays and neutrinos from grbs " , in preparation ; c. d. dermer , arxiv : astro - ph/0611191 ; arxiv : astro - ph/0610195 . e. waxman and j. n. bahcall , phys . rev . lett . * 78 * , 2292 ( 1997 ) ; e. waxman and j. n. bahcall , astrophys . j. * 541 * , 707 ( 2000 ) ; c. d. dermer and a. atoyan , phys . rev . lett . * 91 * , 071102 ( 2003 ) ; m. vietri , phys . * 80 * , 3690 ( 1998 ) ; b. paczynski and g. h. xu , astrophys . j. * 427 * , 708 ( 1994 ) ; j. p. rachen and p. meszaros , phys . d * 58 * , 123005 ( 1998 ) ; j. k. becker , m. stamatikos , f. halzen and w. rhode , astropart . phys . * 25 * , 118 ( 2006 ) ; j. alvarez - muniz , f. halzen and d. w. hooper , phys . rev . d * 62 * , 093015 ( 2000 ) ; f. halzen and d. w. hooper , astrophys . j. * 527 * , l93 ( 1999 ) ; z. g. dai and t. lu , astrophys . j. * 551 * , 249 ( 2001 ) ; s. nagataki , k. kohri , s. ando and k. sato , astropart . phys . * 18 * , 551 ( 2003 ) ; s. razzaque , p. meszaros and e. waxman , phys . lett . * 90 * , 241103 ( 2003 ) ; s. razzaque , p. meszaros and e. waxman , phys . d * 69 * , 023001 ( 2004 ) ; s. ando and j. f. beacom , phys . lett . * 95 * , 061103 ( 2005 ) ; k. asano , astrophys . j. * 623 * , 967 ( 2005 ) ; k. asano and s. nagataki , astrophys . j. * 640 * , l9 ( 2006 ) ; k. murase and s. nagataki , phys . rev . d * 73 * , 063002 ( 2006 ) ; k. murase and s. nagataki , phys . lett . * 97 * , 051101 ( 2006 ) . j. ahrens _ et al . _ [ icecube collaboration ] , astropart . phys . * 20 * , 507 ( 2004 ) . a. cuoco , g. mangano , g. miele , s. pastor , l. perrone , o. pisanti and p. d. serpico , arxiv : astro - ph/0609241 , http://www.km3net.org/. d. kocevski and e. liang , astrophys . j. * 642 * , 371 ( 2006 ) ; b. e. schaefer , m. deng , and d. l. band , astrophys . j. * 563 * , l123 ( 2001 ) ; n. m. lloyd - ronning , c. l. fryer , and e. ramirez - ruiz , astrophys . j. * 574 * , 554 ( 2002 ) ; d. yonetoku , t. murakami , t. nakamura , r. yamazaki , a. k. inoue , and k. ioka , astrophys . j. * 609 * , 935 ( 2004 ) ; c. firmani , v. avila - reese , g. ghisellini and a. v. tutukov , astrophys . j. * 611 * , 1033 ( 2004 ) . a. heger , c. l. fryer , s. e. woosley , n. langer and d. h. hartmann , astrophys . j. * 591 * , 288 ( 2003 ) ; c. l. fryer , s. e. woosley and d. h. hartmann , astrophys . j. * 526 * , 152 ( 1999 ) ; b. paczynski , astrophys . j. * 494 * , l45 ( 1998 ) . r. sari , t. piran and j. halpern , astrophys . j. * 519 * , l17 ( 1999 ) ; w. zhang , s. e. woosley and a. i. macfadyen , astrophys . j. * 586 * , 356 ( 2003 ) ; d. a. frail _ et al . _ , astrophys . j. * 562 * , l55 ( 2001 ) ; x. y. dai _ et al . _ , arxiv : astro - ph/0609269 . p. a. crowther , l. dessart , d. j. hillier , j. b. abbott and a. w. fullerton , astron . astrophys . * 392 * , 653 ( 2002 ) ; j. s. vink and a. de koter , astron . astrophys . * 442 * , 587 ( 2005 ) ; p. a. crowther , arxiv : astro - ph/0610356 . g. rockefeller , c. l. fryer and h. li , arxiv : astro - ph/0608028 . j. petrovic , n. langer , s. c. yoon and a. heger , astron . astrophys . * 435 * , 247 ( 2005 ) . a. heger , n. langer and s. e. woosley , astrophys . j. * 528 * , 368 ( 2000 ) ; m. h. pinsonneault , s. d. kawaler , s. sofia , and p. demarque , astrophys . j. * 338 * , 424 ( 1989 ) . c. wolf and p. podsiadlowski , arxiv : astro - ph/0606725 ; e. berger , d. b. fox , s. r. kulkarni , d. a. frail and s. g. djorgovski , arxiv : astro - ph/0609170 . r. jimenez and z. haiman , nature * 440 * , 501 ( 2006 ) ; h. a. kobulnicky and l. j. kewley , astrophys . j. * 617 * , 240 ( 2004 ) . p. jakobsson _ et al . _ , astrophys . * 447 * , 897 ( 2006 ) ; p. jakobsson _ et al . _ , aip conf . proc . * 836 * , 552 ( 2006 ) , http://www.astro.ku.dk/@xmath87pallja/grbsample.html , http://www.mpe.mpg.de/@xmath87jcg/grbgen.html . p. natarajan , b. albanna , j. hjorth , e. ramirez - ruiz , n. tanvir and r. wijers , mon . not . * 364 * , l8 ( 2005 ) ; r. s. priddey , n. r. tanvir , a. j. levan , a. s. fruchter , c. kouveliotou , i. a. smith and r. a. m. wijers , mon . not . soc . * 369 * , 1189 ( 2006 ) ; e. ramirez - ruiz , d. lazzati and a. w. blain , astrophys . j. * 565 * , l9 ( 2002 ) . podsiadlowski , p. a. mazzali , k. nomoto , d. lazzati and e. cappellaro , astrophys . j. * 607 * , l17 ( 2004 ) . e. e. salpeter , astrophys . j. * 121 * , 161 ( 1955 ) . l. e. strigari , j. f. beacom , t. p. walker and p. zhang , jcap * 0504 * , 017 ( 2005 ) ; h. yuksel , s. ando and j. f. beacom , phys . c * 74 * , 015803 ( 2006 ) ; s. ando , j. f. beacom and h. yuksel , phys . * 95 * , 171101 ( 2005 ) . d. n. burrows _ et al . _ , space sci . rev . * 120 * , 165 ( 2005 ) . g. ghirlanda , g. ghisellini and c. firmani , arxiv : astro - ph/0610248 . e. rossi , d. lazzati and m. j. rees , mon . not . soc . * 332 * , 945 ( 2002 ) ; b. zhang , x. y. dai , n. m. lloyd - ronning and p. meszaros , astrophys . j. * 601 * , l119 ( 2004 ) ; e. ramirez - ruiz and n. m. lloyd - ronning , new astron . * 7 * , 197 ( 2002 ) . j. p. norris , astrophys . j. * 579 * , 386 ( 2002 ) . e. liang , b. zhang and z. g. dai , arxiv : astro - ph/0605200 . e. pian , _ et al . _ , nature * 442 * , 1011 ( 2006 ) ; a. m. soderberg , _ et al . _ , nature * 442 * , 1014 ( 2006 ) . c. d. dermer and a. atoyan , arxiv : astro - ph/0606629 . c. d. dermer , j. chiang and m. bottcher , astrophys . j. * 513 * , 656 ( 1999 ) . p. meszaros , ann . astrophys . * 40 * , 137 ( 2002 ) . t. piran , phys . rept . * 314 * , 575 ( 1999 ) ; r. sari , t. piran and r. narayan , astrophys . j. * 497 * , l17 ( 1998 ) . m. j. rees and p. meszaros , astrophys . j. * 430 * , l93 ( 1994 ) ; b. zhang _ et al . _ , astrophys . j. * 642 * , 354 ( 2006 ) . a. mucke , j. p. rachen , r. engel , r. j. protheroe and t. stanev , publ . austral . * 16 * , 160 ( 1999 ) . g. r. blumenthal , phys . d * 1 * , 1596 ( 1970 ) . v. berezinsky , a. z. gazizov and s. i. grigorieva , phys . d * 74 * , 043005 ( 2006 ) ; r. aloisio , v. berezinsky , p. blasi , a. gazizov , s. grigorieva and b. hnatyk , arxiv : astro - ph/0608219 . g. g. fazio and f. w. stecker , nature * 226 * , 135 ( 1970 ) ; t. m. kneiske , t. bretz , k. mannheim and d. h. hartmann , astron . astrophys . * 413 * , 807 ( 2004 ) ; f. w. stecker , m. a. malkan and s. t. scully , astrophys . j. * 648 * , 774 ( 2006 ) . v. s. berezinsky and s. i. grigoreva , astron . astrophys . * 199 * , 1 ( 1988 ) . d. de marco , p. blasi and a. v. olinto , astropart . phys . * 20 * , 53 ( 2003 ) . s. t. scully and f. w. stecker , astropart . * 16 * , 271 ( 2002 ) . j. n. bahcall and e. waxman , phys . b * 556 * , 1 ( 2003 ) . c. a. norman , d. b. melrose , and a. achterberg , astrophys . j. * 454 * , 60 ( 1995 ) ; h. kang , d. ryu and t. w. jones , astrophys . j. * 456 * , 422 ( 1996 ) ; s. gabici and p. blasi , astrophys . j. * 583 * , 695 ( 2003 ) . n. gupta and b. zhang , arxiv : astro - ph/0606744 ; k. murase , k. ioka , s. nagataki and t. nakamura , arxiv : astro - ph/0607104 . s. d. wick , c. d. dermer and a. atoyan , astropart . phys . * 21 * , 125 ( 2004 ) ; c. d. dermer and j. m. holmes , astrophys . j. * 628 * , l21 ( 2005 ) . v. berezinsky , arxiv : astro - ph/0509675 . m. ave , n. busca , a. v. olinto , a. a. watson and t. yamamoto , astropart . * 23 * , 19 ( 2005 ) ; d. hooper , a. taylor and s. sarkar , astropart . * 23 * , 11 ( 2005 ) . o. e. kalashev , v. a. kuzmin , d. v. semikoz and g. sigl , phys . d * 66 * , 063004 ( 2002 ) ; z. fodor , s. d. katz , a. ringwald and h. tu , jcap * 0311 * , 015 ( 2003 ) ; d. v. semikoz and g. sigl , jcap * 0404 * , 003 ( 2004 ) ; m. ahlers , l. a. anchordoqui , h. goldberg , f. halzen , a. ringwald and t. j. weiler , phys . d * 72 * , 023001 ( 2005 ) . e. armengaud , g. sigl , t. beau and f. miniati , arxiv : astro - ph/0603675 , http://apcauger.in2p3.fr/crpropa/index.php . a. mucke , r. engel , j. p. rachen , r. j. protheroe and t. stanev , comput . commun . * 124 * , 290 ( 2000 ) . t. stanev , phys . b * 595 * , 50 ( 2004 ) ; e. v. bugaev , a. misaki and k. mitsui , astropart . * 24 * , 345 ( 2005 ) ; d. de marco , t. stanev and f. w. stecker , phys . rev . d * 73 * , 043003 ( 2006 ) . m. kachelriess and d. v. semikoz , phys . b * 634 * , 143 ( 2006 ) . k. hurley , _ et al . _ , gcn circ . * 5684 * , ( 2006 ) ; e. bellm , _ et al . _ , gcn circ . * 5685 * , ( 2006 ) ; k. l. page , _ et al . _ , gcn circ . * 5686 * , ( 2006 ) ; m. ohno , _ et al . _ , gcn circ . * 5688 * , ( 2006 ) ; s. golenetskii , _ et al . _ , gcn circ . * 5689 * , ( 2006 ) .

we present our diagnosis of the role that massive stars play in the formation of low- and intermediate - mass stars in ob associations ( the ori region , the oriob1 , and lacob1 associations ) . we find that the classical t tauri stars and herbig ae / be stars tend to line up between luminous o stars and bright - rimmed or comet - shaped clouds ; the closer to a cloud the progressively younger they are . our positional and chronological study lends support to the validity of the radiation - driven implosion mechanism , where the lyman continuum photons from a luminous o star create expanding ionization fronts to evaporate and compress nearby clouds into bright - rimmed or comet - shaped clouds . implosive pressure then causes dense clumps to collapse , prompting the formation of low - mass stars on the cloud surface ( i.e. , the bright rim ) and intermediate - mass stars somewhat deeper into the cloud . these stars are a signpost of current star formation ; no young stars are seen leading the ionization fronts further into the cloud . young stars in bright - rimmed or comet - shaped clouds are likely to have been formed by triggering , which would result in an age spread of several myrs between the member stars or star groups formed in the sequence .

astro-ph0509315

most o and b stars are congregated into ob associations @xcite in which young low- ( classical t tauri stars , or cttss ) and intermediate - mass ( herbig ae / be stars , or haebe ) stellar groups are also found ( see the review by * ? ? ? * ) . what is the relationship between the formation of massive stars and that of low - mass stars ? does star formation in an ob association proceed in a bimodal manner for massive and for low - mass stellar groups ? if so , which group would form first ? it is noted that massive stars have a profound influence on the surrounding molecular clouds . on the one hand , the radiation and energetic wind from a massive star could cause the evaporation of nearby clouds , hence terminating the star formation processes . on the other hand , the massive star could provide `` just the touch '' needed to prompt the collapse of a molecular cloud which otherwise may not contract and fragment spontaneously . do massive stars play primarily a destructive or promotional role in star formation in a molecular cloud ? @xcite suggests that low- and intermediate - mass stars form first in an ob association , but soon after massive o stars appear , the cloud is disrupted , which hinders further star formation . alternatively , @xcite and @xcite propose that low - mass stars form first out of cloud fragments and are distributed throughout the entire molecular cloud . once the o stars form , their expanding ionization fronts ( i - fronts ) then play a constructive role in inciting a sequence of star formation in neighboring molecular clouds . the triggering of star formation by massive stars appears to take place on different length scales @xcite . the scoob2 association might be one example of triggered star formation @xcite . in this case the upper centaurus lupus subgroup was formed first in the middle of the molecular cloud complex , which then prompted star formation on both sides , eventually becoming the upper scorpius and lower centaurus crux associations . @xcite propose a similar mechanism , but with a series of supernova explosions as the triggering sources . the star formation activities can be sustained as long as stars massive enough are produced in the sequence and there is enough surrounding material . this sequential formation process leads naturally to an age spread among member stars or subgroups @xcite , and the stellar aggregates thus formed ( out of separate clouds ) tend to be sparsely distributed and gravitationally unbound because of the expanding i - fronts or an initially unbound giant molecular cloud ( see * ? ? ? * ) . on a smaller scale , the signature of ongoing star formation , such as young stellar jets , evaporated gaseous globules ( eggs ) and water masers have been found in the periphery of regions @xcite . there are two kinds of triggering mechanisms ( e.g. , * ? ? ? * ) , collect - and - collapse " and radiation - driven implosion ( rdi ) . in the collect - and - collapse scenario first proposed by @xcite and recently demonstrated observationally by @xcite , @xcite , and @xcite , the expanding i - fronts from an region piles up a shell of dense gas and dust , in which clumps fragment and collapse to form the next generation of stars . in the rdi scenario @xcite , the formation sequence begins with photoionization of a nearby molecular cloud by a massive star . the shock fronts embracing the surface of the cloud compress the cloud until reaching the critical density for gravitational collapse resulting in the formation of new stars . the latest star formation , as traced by protostellar cores @xcite or water masers @xcite , takes place at the compressed layer of a cloud . @xcite propose a scenario in which an egg appears when a dense clump is impinged upon by the i - fronts . the photoevaporation then erodes the circumstellar disk into a protoplanetary disk , or a `` proplyd '' @xcite . subsequently formed massive stars can carve out their own cavities to continue the triggering process @xcite . the exposure of the protoplanetary disk in such environments would , in addition to being truncated in size , contain short - lived radio nuclides from the ejecta from one or more nearby supernovae , such as has been observed in meteorites in the solar system @xcite . our results substantiate the above sequential star formation scenario by providing clear chronological and positional evidence that massive stars prompt the birth of lower - mass stars out of molecular clouds . in the orion star - forming region for example , there is concrete evidence of triggered star formation as manifested by the bright - rimmed clouds ( brcs ) in the vicinity of o stars ( * ? ? ? * hereafter paper i ) . these brcs are considered the remnant of molecular clouds that have been photoionized by a nearby massive star @xcite . according to paper i , only brcs that are associated with strong @xmath1 100 @xmath2 emission ( tracer of high density ) and h@xmath3 emission ( tracer of the ionization front ) show signs of ongoing star formation . furthermore , cttss are more likely seen between the o stars and the brcs , with those closer to the brcs being progressively younger , and there are no cttss far ahead of the i - fronts . in this paper we extend the study to the lacob1 association as well as include intermediate - mass young stars in our sample . we describe in 2 the archival data and our spectroscopic and imaging observations . in addition to the lacob1 sources , some of the stars in oriob1 considered in paper i to be young star candidates have been spectroscopically confirmed . these results are also presented here . finally , we discuss star - formation activities and histories in @xmath0 ori , oriob1 , and lacob1 in 3 , and consider star formation in general in ob associations in 4 . the conclusions are summarized in 5 . cttss are young stellar objects characterized by their infrared excess . usually cttss are more likely to be spatially closer to a star - forming region than are the weak - line t tauri stars ( wttss ) . the latter are also pms stars , but are more evolved than cttss in terms of clearing of their inner circumstellar disks . thus the cttss trace more recent star formation . in paper i , we proposed an empirical set of criteria to select ctts candidates from the 2mass point source catalog @xcite . in this paper , we apply the same selection procedure ( e.g. , 2mass colors , good photometric qualities , and exclusion of extended sources ) but include young intermediate - mass stars , the haebe stars into our sample . different young stellar populations , wttss , cttss , and haebe stars occupy distinctly different regions in the 2mass color - color diagram ( see paper i ) . the haebe stars in general exhibit larger infrared excess than cttss do . therefore we select as haebe star candidates 2mass point sources with colors redder than the line defined by @xmath4 ; ctts candidates are selected by the same method described in paper i , namely between the two parallel lines , @xmath5 and @xmath4 , and above the dereddened ctts locus @xcite , @xmath6 . table 1 shows the fields in the @xmath0 ori region , oriob1 , and lacob1 studied in this paper , which include 7 brcs , one comet - shaped cloud and two control regions . in addition to the 2mass point - source database from which we select our ctts and haebe candidates , we also make use of the h@xmath3 emission survey data @xcite , @xmath7 reddening @xcite , @xmath1 100 @xmath2 , and co @xcite emission to trace , respectively , the distribution of the ionization fronts , cloud extinction , ir radiation , and molecular clouds with respect to the spatial distribution of our young star sample . the spectra of bright ctts and haebe candidates were taken at the beijing astronomical observatory ( bao ) and at the kitt peak national observatory ( kpno ) . at the bao , low - dispersion spectra with a dispersion of 200 mm@xmath8 , corresponding to 4.8 pixel@xmath8 , were taken with the 2.16 m optical telescope from 2003 october 31 to november 3 , and on 2004 september 56 . an omr ( optomechanics research , inc . ) spectrograph was used with a tektronix 1024@xmath91024 ccd detector covering 40009000 . these spectra were used to confirm the young stellar nature ( e.g. , the h@xmath3 and other characteristic emission lines ) of the pms star candidates selected on the basis of the 2mass colors . medium - dispersion spectra for a selected set of sample stars were taken with the kpno 2.1 m telescope on 2004 january 25 . the goldcamera spectrometer , with a ford 3 k@xmath91 k ccd with 15 @xmath2 pixels , was used with the grating # 26new , giving a dispersion of 1.24 pixel@xmath8 . these medium - dispersion spectra allowed us to identify the lithium absorption at 6708 , the spectral signature of a low - mass pms star . all the spectroscopic data were processed with the standard noao / iraf packages . after correction for bias and flat - fields , the iraf package kpnoslit was used to extract and to calibrate the wavelength and flux of each spectrum . to check the legitimacy of our selection criteria , we also observed two control fields , in addition to the star - forming clouds . all the fields included in this study are summarized in table 1 . the brcs were imaged on 2004 november 38 using the 1 m telescope at the lulin observatory in taiwan ( table 2 ) . a pi1300b ( roper scientific ) ccd camera was used , which has @xmath10 pixels , each 20 @xmath2 square , yielding a @xmath11 field of view . h@xmath3 ( @xmath12 , @xmath13(fwhm)=30 ) images were taken for all brcs . in addition , lbn437 was observed with an [ ] ( @xmath14 , @xmath13(fwhm)=80 ) filter . for every target field tens of images were taken , each with an exposure time of 120 to 300 s. the images were processed for bias , dark and flat - fielding corrections with the standard procedures . the main purpose of the spectral observations was to identify pms star candidates and to validate the selection criteria for haebe stars . the imaging observations can help us to trace the distribution of the i - fronts in the brcs . by combining the spectral and imaging observations , we can study the spatial distribution of pms stars relative to i - fronts in brcs . tables 3 , 4 , and 5 list , respectively , the cttss ( plus some ctts candidates ) , haebe stars , and non - pms sources identified from spectroscopic observations . in table 3 , stars 131 cttss are in orion and 3240 cttss are in lacerta . in table 4 , stars 4148 are haebe stars in the orion region , whereas the others are in the lacerta region . we derive the h@xmath3 , [ ] , and [ ] equivalent widths of the cttss . some of the cttss listed in table 3 do not show lithium absorption , but exhibit other ctts characteristics , such as the h@xmath3 , , and/or forbidden [ ] and [ ] emission line(s ) in their spectra . since most of these spectra show veiling , their li absorption line might be veiled by continuum radiation . thus they are included in the ctts sample ( table 3 ) even though the li line is not readily discernible . cttss without a li absorption line are not unusual ; recently @xcite also found a lithium - depleted ctts , st34 , in the taurus - auriga t association . figure [ fig : spectra ] presents an example of the spectra of a ctts and a haebe star . no pms stars were found in any of the two control fields ; most of the sources there are either carbon stars or m giants . figure [ fig : oriob1 ] and figure [ fig : lamori ] show , respectively , the trapezium and the @xmath0ori regions in orion , with the cttss ( stars 131 in table 3 ) and haebe stars ( stars 4148 in table 4 ) being marked . the boxes mark the fields of the h@xmath3 images presented in fig . [ fig : lot1 ] . it is clear that the brcs are outlined by the h@xmath3 emission , and that some pms stars are spatially close to the i - fronts . figure [ fig : lac ] displays the @xmath1 100 @xmath2 , h@xmath3 and co emission maps of the lacob1 association . the pms stars in table 3 and table 4 are again marked . the box indicates the lbn437 region shown in figure [ fig : lot2 ] . lbn437 is a comet - shaped brc @xcite . the haebe star v375lac ( star 52 in table 4 ) associated with this cloud is believed to be the exciting source of the parsec - scale herbig - haro outflow hh398 @xcite . in paper i it was shown that cttss exhibiting continuous or veiled spectra with [ ] and/or [ ] forbidden lines , originating from jets or winds seen commonly in class i sources @xcite , tend to be redder , which is suggestive of a younger age , than those without . a color - color diagram of the pms stars in tables 3 and 4 is plotted as fig . [ fig : colors ] ; the results agree with our previous work . this correlation extends to haebe stars , in that haebe stars with forbidden line(s ) are mostly located on the upper right of the 2mass color - color diagram . as an alternative to being younger , a ctts with forbidden lines could be the result of reduced photoevaporation of the circumstellar disk , e.g. , by being away from a luminous star or shielded by a molecular cloud . only 14 of the 40 cttss and 4 of the 13 haebe stars in our sample show forbidden line(s ) in their spectra . in other words , about one third of the pms stars with strong infrared excess exhibit forbidden line(s ) . typical ctts ages are a few myr @xcite , with those with forbidden lines representing an even younger sample , probably no more than a couple myr old . the [ ] line is only present in star 2 , a ctts with a strong infrared excess and strong [ ] ( equivalent width @xmath15 10.5 ) . in our sample of cttss there is no correlation between the h@xmath3 equivalent widths and the presence of forbidden lines , or between the h@xmath3 equivalent widths and the 2mass colors . the success rate of spectroscopic confirmation of ctts and haebe candidates is extremely high for @xmath0 ori , oriob1 or lacob1 . candidates closely associated with star - forming regions all turned out to be bona fide young stars with essentially no exception , whereas the regions away from molecular clouds are mostly populated by evolved stars ( e.g. , carbon stars or m giants ) . the 2mass database enables us to effectively trace recent star formation on a large scale , without any a priori bias toward prominent or reflection nebulae which are obvious targets to search for young stellar objects . for example , stars 3538 in our sample are confirmed to be young stars . they are located away from prominent nebulosity , so it might otherwise be difficult to recognize them as young stars in a targeted survey . star formation triggered by the rdi mechanism has several characteristics which can be diagnosed observationally : ( 1 ) the remnant cloud is extended toward , or pointing to , the massive stars . ( 2 ) the young stellar groupings in the region are roughly lined up between the remnant clouds and the luminous star . ( 3 ) stars closer to the cloud , which have formed later in the sequence , are younger in age , with the youngest stars being in the interacting region , i.e. , along the bright rim of the cloud . ( 4 ) no young stars exist far behind the brc . in particular , ( 3 ) and ( 4 ) are in distinct contrast to the case of spontaneous star formation , which conceivably would not have left such distinguishing temporal and positional signposts . in table 6 we summarize the different outcomes of the triggered versus spontaneous star formation processes . in paper i , we presented evidence supporting the induced star formation in six orion brcs , namely b30 , b35 , orieast , ic2118 , ldn1616 , and ldn1634 . here , we present further spectroscopic observations of the orion sources , classified as young star `` candidates '' in paper i and extend our sample to include the lacob1 region . combined with the earlier oriob1 results , this reinforces the links between massive stars , brcs , and the formation of low - mass stars . furthermore , our young star sample now contains not only cttss , but also young intermediate - mass stars , rendering a more comprehensive understanding of the origin of stellar masses in an ob association . ic2118 , ldn1616 , and ldn1634 are three isolated brcs around the trapezium to the west of the oriona . another brc , orieast , can be found to the north - east of the trapezium . all these brcs point roughly to the trapezium ( fig . [ fig : oriob1 ] ) , indicative of the trapezium and/or the orion - eridanus superbubble being the shaping source of these brcs @xcite . in this region most cttss with forbidden line(s ) , i.e. , those of younger ages , are spatially close to the brcs , e.g. , stars 1 , 2 , and 30 in relation to ldn1616 , ic2118 , and orieast , respectively . star 7 is also associated with a remnant molecular cloud @xcite ( their cloud 6 ) . b30 and b35 are two brcs associated with an region excited by the o8 iii star @xmath0 ori and surrounded by a ring - shaped molecular cloud @xcite . @xcite find some 80 h@xmath3 stars in the @xmath0 ori region , most of which are distributed as a barlike structure extending from either side of @xmath0ori to b30 and to b35 . @xcite present photometric and spectroscopic studies of the young stellar population in the @xmath0 ori region . they suggested the ring - shaped molecular cloud to be caused by a supernova explosion that terminated recent star formation in the vicinity . it is likely that @xmath0 ori is the triggering source responsible for the star formation in b30 and b35 . it is found that photoevaporative flows @xcite stream out of the surfaces of them ; this is a demonstration of the interaction between a massive star and a molecular cloud ( fig . [ fig : lamori ] ) . here again we see that stars with forbidden lines , i.e. , stars 20 , 22 , and 44 in relation to b30 , and stars 25 and 26 to b35 , are all physically close to a brc . it is suggested that star formation is triggered by the o stars and/or by the superbubbles in ldn1616 @xcite and in ic2118 @xcite . all brcs in oriob1 which show evidence of star formation being triggered by nearby massive stars are found associated with strong @xmath1 100 @xmath2 and h@xmath3 emission ( fig . [ fig : lot1 ] ) . in every case a sequential process that pms stars closer to the triggering stars are older than those closer to the brcs can be clearly witnessed ( paper i ) . the same phenomena are also seen near @xmath0 ori . initially , the b30 and b35 clouds might have extended toward @xmath0 ori , perhaps forming a barlike structure . the i - fronts from @xmath0 ori then propagated through the clouds , prompting star formation on both sides , thereby resulting in the lining up of the pms stars , in an age sequence , between @xmath0 ori and the b30 and b35 clouds . in paper i we show that the cttss that are spatially close to brcs are among the brightest , just revealing themselves on the birthline and beginning to descend down the hayashi tracks . we also find no young stars far behind the i - fronts , i.e. , embedded in the brcs . these photoevaporated clouds typically have low extinction so that any pms stars can not have escaped the 2mass detection . in both oriob1 and the @xmath0 ori region therefore , we see the predomination of triggered star formation , as evinced in the cloud morphology , star grouping orientation , and star formation sequence . the lacob1 association , at a distance of @xmath16 pc @xcite , is one of the nearest ob associations . @xcite divides lacob1 into 2 subgroups , a " and b " , on the basis of stellar proper motions and radial velocities . the entire lacob1 covers the region of the sky from @xmath17 and @xmath18 @xcite . lacob1b occupies an area with a radius of @xmath19 centered around @xmath20 and lacob1a occupies the remaining area . the lacob1b harbors the only o star ( o9 v ) , 10lac , in the lacob1 association . our study discusses two regions in lacob1 known to have current star - forming activities , lbn437 and gal110 - 13 , a brc and a comet - shaped cloud , respectively . lbn437 is at the edge of an elongated molecular cloud @xcite and on the border of the region s126 excited by the nearby o star , 10lac . hereafter we call this elongated molecular cloud the lac molecular cloud " ( fig . [ fig : lac ] ) . between 10lac and lbn437 there is a small stellar group ( fig . [ fig : group ] ) which includes 5 cttss ( stars 3538 in table 3 , plus the ctts candidate 2massj22354224@xmath213959566 , for which we do not have spectroscopic observations ) and one haebe star ( star 53 in table 4 ) . the haebe star is an @xmath1 source , iras22343@xmath213944 . we can identify iras22343@xmath213944 as the counterpart of star 53 , because this star shows a near - infrared excess and is located within the positional error for iras22343@xmath213944 . hereafter we refer to this 6-star system as the iras22343@xmath213944 group ( fig . [ fig : lac ] ) . the size of the iras22343@xmath213944 group is about 24@xmath22 , which corresponds to @xmath23 pc at 360 pc . gal110@xmath2413 is an isolated and elongated molecular cloud ( fig . [ fig : gal ] ) at a distance of @xmath25 pc @xcite . its head - tail , comet - like shape suggests compression by ram - pressure , perhaps as a result of a recent cloud collision @xcite . star formation takes place on the compressed side of gal110@xmath2413 , e.g. , the location of the ctts star 40 ( bmand ) and the nebula vdb158 reflecting light from the b9.5v star hd222142 @xcite . in addition to hd222142 there are two other late b - type stars in the vicinity , hd222046 and hd222086 . all three b stars and star 40 share common proper motions ( * ? ? ? * data extracted from the second u.s . naval observatory ccd astrograph catalog ) , which are consistent with those of the lacob1 groups @xcite , as summarized in table 7 . gal110@xmath2413 is located near the border of the lacob1 association , at a distance not very different from that of lacob1 . gal110@xmath2413 was not included as part of lacob1 by @xcite , but our analysis suggests that the cloud , together with the young stars associated with it , is likely part of lacob1a . gal110@xmath2413 is elongated and roughly points toward 10lac ( see fig . [ fig : lac ] ) . this implies that lacob1b or 10lac alone is responsible for shaping the cloud . either shock fronts from a supernova or ionization fronts from a massive star could have caused the shape of this cloud as well as the spatial distribution of young stars in gal110@xmath2413 . in the supernova scenario a star in lacob1b more massive than 10lac exploded , and , assuming that lacob1b and 10lac are at the same distance from us ( i.e. , 358 pc ) , it would take a few hundred thousand years for the supernova shock waves ( at a speed of hundreds of km s@xmath8 ) to propagate across the 126 pc separation to arrive , compress , and finally prompt the formation of stars within gal110@xmath2413 . additional evidence in support of this supernova scenario comes from the b5v star , hd201910 , which is supposed to be a runaway star kicked out from a binary system in lacob1b , when one of the component stars became a supernova @xcite . if this is so , the kinematic time scale of the star , 2.7 myr , suggests that a supernova explosion occurred some 2.7 myr ago and the associated shocks subsequently caused gal110@xmath2413 to develop its present cometary shape . an alternative explanation is due to compression by ionization fronts from a massive star , which would be a less destructive method for star formation than a supernova explosion @xcite . we propose a scenario in which 10lac still in existence now was born at the edge of the lac molecular cloud , similar to that presented in figure [ fig : lac ] , but with the cloud originally being more extended toward 10lac . soon after its birth , 10lac ionized the surrounding molecular clouds , exposing itself to the intercloud medium . assuming that most of the uv photons of 10lac shortward of the lyman limit were used to ionize the intercloud medium , then given a typical intercloud material density @xmath26 @xmath27 @xcite , the i - fronts would travel the 126 pc distance from 10lac to gal110@xmath2413 in about 2 myr , a time scale still shorter than the main sequence life time of @xmath28 myr of 10lac @xcite . regardless of which scenario actually happened , a supernova shock front or an ionization front , lacob1b is likely responsible for the creation of gal110@xmath2413 and the associated stellar group . in lbn437 , star 52 ( v375lac ) is the only young star located at the edge of the lac molecular cloud , and interestingly there is no ctts or haebe candidate behind the interaction region . to check whether any pms stars could have escaped the 2mass detection limit of @xmath29 mag as a result of excessive dust extinction , we created the @xmath7 map of the lac molecular cloud @xcite ( as done for the brcs in oriob1 ; see paper i ) . it is assumed that the @xmath30-band luminosities of the embedded pms stars are the same as those of the visible pms stars outside the cloud , namely the iras22343@xmath213944 group . like other brcs we have analyzed the overall extinction in the lac molecular cloud is low , and the probability of nondetection is 0.014 . this means that there are indeed no embedded pms stars and hence no ongoing star formation in the lac molecular cloud . the elongated lac molecular cloud associated with lbn437 ( fig . [ fig : lac ] ) may be just the remnant of a molecular cloud , which was originally perhaps larger , extending as far as to 10lac . upon the birth of 10lac , its energetic photons evaporated and compressed the cloud , shaping the cloud into a pillar , similar to the case of gal110@xmath2413 . the iras22343@xmath213944 group and star 52 were then born on the compressed side of the cloud . at least 3 stars in the iras22343@xmath213944 group exhibit forbidden lines , which is suggestive of their youth . star 52 is likely even younger because it is the exciting source of an hh outflow . apparently star 52 is the latest product in the star formation sequence by 10lac in this cloud . @xcite derived a 30% star formation efficiency for gal110@xmath2413 . this is much higher than that of the few percent typical in star - forming regions @xcite . extinction is low in gal110@xmath2413 , with an a@xmath31 less than 0.48 mag , as estimated from its @xmath7 values , so the cloud is insufficiently dense to hide from our detection any embedded young stars similar to star 40 . as in the case for oriob1 ( paper i ) , the brcs in lacob1 also tend to have a relatively low dust extinction . such a low density condition is unfavorable for spontaneous , global cloud collapse . star formation however could take place at the interaction layer ( the bright rim ) of a molecular cloud . a stellar group could form , such as witnessed in the iras22343 + 3944 and gal110@xmath2413 groups . @xcite and @xcite derived the ages of lacob1a and lacob1b , on the basis of stellar proper motions and radial velocities : 1625 myr and 1216 myr , respectively . both these ages are too old to be consistent with the existence of 10lac ( with a lifetime of less than @xmath28 myr ) and the cttss ( typically aged a few myr ) in the region . thus lacob1a and lacob1b could not have formed at the same place and at the same time , because with a typical velocity dispersion of a few kilometers per second for an ob association @xcite , the two subgroups could not traverse the distance of 3080 pc now between them . we propose that both lacob1a and lacob1b are no more than a few myr old , and lacob1a is younger than lacob1b . figure [ fig : cmd ] shows the color - magnitude diagrams reconstructed from @xcite for the two subgroups . it can be seen that the stars in lacob1b form a clear main sequence , whereas those in the subgroup lacob1a are widely scattered to the right of the sequence . some stars in lacob1a may well still be in the pms phase , hence we postulate a younger age for lacob1a than for lacob1b . it is possible that lacob1b was formed first after which the expanding i - fronts from lacob1b triggered new generations of stars along the lac molecular cloud , the iras22343 + 3944 group and star 52 . a subsequent supernova or i - front then initiated the formation of stars in lacob1a ; eventually the birth wave " reached gal110@xmath2413 . triggered star formation has been suggested to have occurred close to regions @xcite . our study finds clear chronological and positional causality of such processes on larger scales . in @xmath0 ori , oriob1 and lacob1 , we see supporting evidence of triggered star formation . the uv photons from an o star create expanding i - fronts which evaporate and compress nearby molecular clouds , thereby shaping the clouds into brcs or comet - shaped clouds . the next generation of stars can then form efficiently , perhaps in groups , out of the compressed material . the resulting newly formed stars would line up between the massive star and the molecular coulds in a formation and hence age sequence . stars at least as massive as late herbig be types could be formed via this process ( see table 4 ) . these stars would reach the main sequence with even earlier spectral types . triggered star formation could therefore produce not only low - mass stars , but also intermediate - mass or even massive stars . in our sample , the haebe stars and cttss seem to be distributed spatially differently relative to a brc , in the sense that the cttss tend to be located near the surface of a brc , whereas the haebe stars appear preferentially to reside deeper into a brc(e.g . , star 44 in b30 , star 52 in lbn437 , and star 41 in ldn1616 ) . what we see in @xmath0 ori , oriob1 , and lacob1 is in contrast to the scenario proposed by @xcite and @xcite for which massive stars are formed in shocked cloud layers by triggering , whereas low - mass stars are formed spontaneously via cloud collapse and fragmentation . a global cloud collapse would lead to starbirth spreading throughout the cloud , but this was not observed in our study . instead , no young stars are found embedded in clouds far behind the i - fronts . more than mere fossil " molecular clouds , the brcs present convenient snapshots of how star formation must have proceeded in an ob association . when prompted to form , massive stars appear to favor denser environments where photoevaporation is relatively weak . in comparison , when a dense molecular core near the ionization layer ( i.e. , current cloud surface ) collapses , the accretion process has to compete with the mass loss arising from photoevaporation , leading to the formation of less massive stars or even substellar objects @xcite . as the i - fronts progress , the remnant cloud is eventually dispersed , with stars of different masses remaining in the same volume . low- and intermediate - mass young stars in bright - rimmed or comet - shaped clouds on the border of an ob association are more likely to be formed by triggering . assuming a shock speed of @xmath32 km s@xmath8 , this would result in an age spread of several myrs between member stars or star groups formed in the sequence . if the velocities of the shocks are higher , as in the case of a supernova explosion , the age spreads would be less . we first selected cttss and haebe stars in @xmath0 ori , oriob1 , and lacob1 based on the 2mass colors . these pms stars are then utilized to trace recent star - forming activities . the young stars are found to be lined up in an age sequence between massive stars and comet - shaped clouds or bright - rimmed clouds , with the youngest stars located near the cloud surfaces . there are no pms stars far behind the i - fronts . these results support the scenario by which the lyman continuum photons of a luminous o star create expanding i - fronts that would cause the evaporation and compression of nearby clouds to form brcs or comet - shaped clouds , thereby inducing the birth of low- and intermediate - mass stars . the brcs provide us with a convenient setting in which to see that the haebe stars tend to form in the inner , denser parts of a cloud , whereas the cttss are formed near the photoevaporating cloud layers . young stars in bright - rimmed or comet - shaped clouds near a massive star are likely to have been formed by triggering . assuming a shock speed of @xmath32 km s@xmath8 , this would result in an age spread of several myrs between member stars or star groups formed in the sequence . we want to particularly thank richard f. green , director of kpno , who kindly provided us the director s discretionary time to accomplish this work . we are also grateful to the staff of the beijing astronomical observatory for their assistance during our observation runs and to the referee , hans zinnecker , for his suggestions helpful in improving the quality of this paper . this research 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 ( nsf ) . we also used the southern h - alpha sky survey atlas ( shassa ) , supported by the nsf . we acknowledge the financial support of grant nsc92 - 2112-m-008 - 047 from the national science council and 92-n - fa01 - 1 - 4 - 5 from the ministry of education of taiwan . alcal , j. m. , wachter , s. , covino , e. , sterzik , m. f. , durisen , r. h. , freyberg , m. j. , hoard , d. w. , & cooksey , k. 2004 , , 416 , 677 bertoldi , f. 1989 , , 346 , 735 bertoldi , f. & mckee , c. f. 1990 , , 354 , 529 blaauw , a. 1958 , , 63 , 186 blaauw , a. 1961 , , 15 , 265 blaauw , a. 1964 , , 2 , 213 blaauw , a. 1991 , in the physics of star formation and early stellar evolution , ed . c. j. lada & n. d. kylafis ( dordresht : kluwer ) , 125 briceo , c. , et al . 2006 , in protostars and planets v , ed . reipurth , b. , jewitt , d. , & keil , k. ( tucson : univ . arizona press ) brown , a. g. a. , de geus , e. j. , & de zeeuw , p. t. 1994 , , 289 , 101 clark , p. c. , bonnell , i. a. , zinnecker , h. , & bate , m. r. 2005 , , 359 , 809 cutri , r. m. , et al . 2003 , 2mass all sky catalog of point sources ( pasadena : ipac ) dame , t. m. , hartmann , d. , & thaddeus , p. 2001 , , 547 , 792 dennison , b. , simonetti , j. h. , & topasna , g. a. 1998 , , 15 , 147 de bruijne , j. h. j. 1999 , , 310 , 585 de geus , e. j. , de zeeuw , p. t. , & lub , j. 1989 , , 216 , 44 deharveng , l. , zavagno , a. , & caplan , j. , 2005 , , 433 , 565 de zeeuw , p. t. , hoogerwerf , r. , de bruijne , j. h. j. , brown , a. g. a. , & blaauw , a. 1999 , , 117 , 354 dolan , c. j. , & mathieu , r. d. 1999 , , 118 , 2409 dolan , c. j. , & mathieu , r. d. 2001 , , 121 , 2124 dolan , c. j. , & mathieu , r. d. 2002 , , 123 , 387 duerr , r. , imhoff , c. l. , & lada , c. j. 1982 , , 261 , 135 dyson , j. e. , & williams , d. a. 1997 , the physics of the interstellar medium ( 2nd ed . ; bristol : institute of physics publishing ) elmegreen , b. g. & lada , c. j. 1977 , , 214 , 725 elmegreen , b. g. 1998 , in asp conf . ser . 148 : origins , ed . c. e. woodward , j. m. shull , & h. a. thronson , jr . ( san francisco : asp ) , 150 esa . 1997 , the hipparcos and tycho catalogs , 1997 , esa sp-1200 finkbeiner , d. p. 2003 , , 146 , 407 gaustad , j. e. , mccullough , p. r. , rosing , w. , & van buren , d. 2001 , , 113 , 1326 gies , d. r. , & bolton , c. t. 1986 , , 61 , 419 haffner , l. m. , reynolds , r. j. , tufte , s. l. , madsen , g. j. , jaehnig , k. p. , & percival , j. w. 2003 , , 149 , 405 healy , k. r. , hester , j. j. , & claussen , m. j. 2004 , , 610 , 835 herbig , g. h. 1962 , , 135 , 736 hester , j. j. , et al . 1996 , , 111 , 2349 hester , j. j. , desch , s. j. , healy , k. r. , & leshin , l. a. 2004 , science , 304 , 1116 hester , j. j. , & desch , s. j. 2005 , asp conf . ser . 341 : chondrites and the protoplanetary disk , 341 , 107 karr , j. l. , & martin , p. g. 2003 , , 595 , 900 kenyon , s. j. , & hartmann , l. 1995 , , 101 , 117 kenyon , s. j. , brown , d. i. , tout , c. a. , & berlind , p. 1998 , , 115 , 2491 kessel - deynet , o. , & burkert , a. 2003 , , 338 , 545 kun , m. , aoyama , h. , yoshikawa , n. , kawamura , a. , yonekura , y. , onishi , t. , & fukui , y. 2001 , , 53 , 1063 kun , m. , prusti , t. , nikoli , s. , johansson , l. e. b. , & walton , n. a. 2004 , , 418 , 89 lada , c. j. 1987 , in iau symp . 115 , star forming regions , ed . j. jugaku & m. peimbert ( dordrecht : reidel ) , 1 lang , w. j. , masheder , m. r. w. , dame , t. m. , & thaddeus , p. 2000 , , 357 , 1001 larosa , t. n. 1983 , , 274 , 815 lee , h .- chen , w. p. , zhang , z. w. , & hu , j. y. 2005 , , 624 , 808 ( paper i ) lefloch , b. & cernicharo , j. 2000 , , 545 , 340 lefloch , b. , cernicharo , j. , rodrguez , l. f. , miville - deschnes , m. a. , & cesarsky , d. 2002 , , 581 , 335 magakian , t. y. 2003 , , 399 , 141 mcgroarty , f. , ray , t. p. , & bally , j. 2004 , , 415 , 189 meyer , m. r. , calvet , n. , & hillenbrand , l. a. 1997 , , 114 , 288 odell , c. r. , wen , z. , & hu , x. 1993 , , 410 , 696 odenwald , s. , fischer , j. , lockman , f. j. , & stemwedel , s. 1992 , , 397 , 174 ogura , k. & sugitani , k. 1998 , , 15 , 91 olano , c. a. , walmsley , c. m. , & wilson , t. l. 1994 , , 290 , 235 preibisch , t. & zinnecker , h. 1999 , , 117 , 2381 preibisch , t. , & zinnecker , h. 2006 , astro - ph/0610826 sanchawala , k. , chen , w. p. , lee , h. t. , nakajima , y. , tamura , h. , daisuke , b. , sato , s. , & chu , y. h. , 2006 , , in submission , astro - ph0603043 schaerer , d. & de koter , a. 1997 , , 322 , 598 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 spitzer , l. 1998 , physical processes in the interstellar medium ( new york : wiley - interscience ) stanke , t. , smith , m. d. , gredel , r. , & szokoly , g. 2002 , , 393 , 251 sugitani , k. , fukui , y. , & ogura , k. 1991 , , 77 , 59 sugitani , k. , & ogura , k. 1994 , , 92 , 163 whitworth , a. p. , & zinnecker , h. 2004 , , 427 , 299 white , g. j. , casali , m. m. , & eiroa , c. 1995 , , 298 , 594 white , r. j. , & hillenbrand , l. a. 2005 , , 621 , l65 zacharias , n. , urban , s. e. , zacharias , m. i. , wycoff , g. l. , hall , d. m. , monet , d. g. , & rafferty , t. j. 2004 , , 127 , 3043 zavagno , a. , deharveng , l. , comern , f. , brand , j. , massi , f. , caplan , j. , & russeil , d. , , 446 , 171 ccc @xmath33 & @xmath34 to @xmath35 & @xmath36 to @xmath37 + @xmath38 & ra @xmath39 to @xmath40 & dec @xmath41 to @xmath42 + @xmath43 & ra @xmath44 to @xmath45 & dec @xmath46 to @xmath47 + & ra @xmath48 to @xmath49 & dec @xmath50 to @xmath51 + control field 1 & @xmath52 to @xmath53 & @xmath54 to @xmath55 + control field 2 & ra @xmath56 to @xmath57 & dec @xmath58 to @xmath59 ccccc b30 & 05:29:51.4 & + 12:13:58 & h@xmath3 & 5400 + b35 & 05:44:20.0 & + 09:10:40 & h@xmath3 & 5400 + ori east & 05:53:58.6 & + 01:40:37 & h@xmath3 & 3600 + ldn1616 & 05:07:06.0 & @xmath60 & h@xmath3 & 7200 + ldn1634 & 05:20:16.0 & @xmath61 & h@xmath3 & 3600 + ic2118 & 05:07:44.0 & @xmath62 & h@xmath3 & 2400 + lbn437 & 22:34:31.0 & + 40:37:44 & h@xmath3 & 3600 + lbn437 & 22:34:31.0 & + 40:37:44 & [ ] & 7200 + 1 & j05065464 - 0320047 & h(-50.6 ) , o(-0.6 ) , ca , he & a & k & , associated with ldn1616 + 2 & j05073016 - 0610158 & h(-92.5 ) , o(-10.8 ) , s(-1.4 ) , fe , ca , he & a & k & associated with ic2118 + 3 & j05073060 - 0610597 & h(-23.3 ) , ca ? & a & k & associated with ic2118 + 4 & j05122053 - 0255523 & h(-12.1 ) & a & k & + 5 & j05141328 - 0256411 & h(-210.6 ) , o(-2.5 ) , fe , ca , he & n & k & + 6 & j05152683 - 0632010 & h(-0.3 ) & a & k & h@xmath3 emission is week , could be a wtts + 7 & j05162251 - 0756503 & h(-37.7 ) , o(-1.0 ) , ca , he & a & k & + 8 & j05181685 - 0537300 & h(-57.3 ) , o(-2.3 ) , fe , ca , he & n & k & + 9 & j05191356 - 0324126 & h(-52.1 ) , ca , he & a & k & + 10 & j05191549 - 0204529 & h(-10.7 ) , o(-3.6 ) , ca & a & k & + 11 & j05201945 - 0545553 & h(-26.9 ) , ca , he & a & k & , iras05178 - 0548 , associated with ldn1634 + 12 & j05202573 - 0547063 & h(-100.2 ) , o ? , fe , ca , he & a & k & , associated with ldn1634 + 13 & j05203142 - 0548247 & h(-19.3 ) , ca , he & a & k & , associated with ldn1634 + 14 & j05253979 - 0411020 & h(-138.9 ) , fe , ca , he & n & k & + 15 & j05262158 + 1131339 & h(-15.4 ) , o(-1.8 ) , fe , ca & - & b & + 16 & j05292393 + 1151576 & h(-40.5 ) , ca , he & a & b , k & , associated with b30 + 17 & j05300203 + 1213357 & h(-33.6 ) , ca , he & a & b , k & , iras05272 + 1211 , associated with b30 + 18 & j05301313 + 1208458 & h(-5.7 ) , ca & a & b , k & , associated with b30 + 19 & j05311615 + 1125312 & h(-21.9 ) & n & b , k & + 20 & j05315128 + 1216208 & h(-127.3 ) , o(-10.2 ) , he & a & b , k & associated with b30 + 21 & j05323207 + 1044178 & h(-96.7 ) , ca , he & - & b & + 22 & j05324305 + 1221083 & h(-13.9 ) , o(-1.3 ) , ca , he & a & b , k & , iras05299 + 1219 , associated with b30 + 23 & j05330207 + 1137114 & h(-176.8 ) , fe , ca , he & - & b & + 24 & j05391268 + 0915522 & h(-215.6 ) , ca , he & - & b & + 25 & j05432091 + 0906071 & h(-21.8 ) , o(-0.5 ) , ca , he & a & b , k & , iras05406 + 0904 , associated with b35 + 26 & j05440899 + 0909147 & h(-44.5 ) , o(-2.1 ) , fe , ca , he & a & b , k & , iras05413 + 0907 , associated with b35 + 27 & j05451493 + 0721223 & h(-5.6 ) & n & b , k & + 28 & j05452235 + 0904123 & & - & b & , iras05426 + 0903 , associated with b35 + 29 & j05515035 + 0821066 & h(-181.9 ) , ca & a & b , k & + 30 & j05534090 + 0138140 & h(-29.0 ) , o(-2.7 ) , he & - & b & , irasf05510 + 0137 , associated with ori east + 31 & j05535869 + 0144094 & h(-37.0 ) , ca , he & - & b & , irasf05513 + 0143 , associated with ori east + 32 & j21370366 + 4321172 & h(-174.8 ) , fe , ca , he & a & b , k & + 33 & j21395545 + 4313082 & h(-83.6 ) , ca & - & b & + 34 & j21535750 + 4659443 & h(-51.7 ) & - & b & + 35 & j22361978 + 4006273 & h(-63.9 ) , o(-1.8 ) , ca & - & b & associated with iras22343 + 3944 group + 36 & j22362779 + 3954066 & h(-18.4 ) & - & b & associated with iras22343 + 3944 group + 37 & j22370328 + 4005185 & h(-10.3 ) , ca ? , he ? & a & b , k & associated with iras22343 + 3944 group + 38 & j22371683 + 3952260 & h(-130.8 ) , o(-4.5 ) , ca & - & b & associated with iras22343 + 3944 group + 39 & j23104483 + 4508511 & h(-7.9 ) & n & b , k & + 40 & j23373847 + 4824119 & h(-19.2 ) & a & b , k & , associated with gal110@xmath2413 + 41 & j05042998 - 0347142 & h & a3e & k & , iras 05020 - 0351 , associated with ldn1616 + 42 & j05113654 - 0222484 & h & a3e & k & + 43 & j05305472 + 1421524 & h & f2e & k & + 44 & j05312805 + 1209102 & h , o & a2e & k & , iras05286 + 1207 , associated with b30 + 45 & j05313515 + 0951553 & h & b9e & k & + 46 & j05315724 + 1117414 & h & a0e & b & , iras05291 + 1115 + 47 & j05350960 + 1001515 & h , o ? & b9e & b & , iras05324 + 0959 + 48 & j05390921 + 0925301 & h & f7e & b , k & + 49 & j21462666 + 4744154 & h , o & b9e & k & + 50 & j21514726 + 4615115 & h & a9e & k & + 51 & j22154039 + 5215559 & h & a2e & b & + 52 & j22344101 + 4040045 & h , o , s & a2e & k & + 53 & j22363511 + 4000156 & h , o & b8e & b & associated with iras22343 + 3944 group + 54 & j05232026 + 0934432 & a0 & b , k & + 55 & j05285405 - 0606063 & me & k & , iras05264 - 0608 + 56 & j05413010 + 1418225 & c & k & + 57 & j05442880 + 0652019 & m & b & + 58 & j05464207 + 0643469 & c & b & + 59 & j05480851 + 0954012 & ce & b , k & , iras05453 + 0953 + 60 & j07323273 + 2647156 & c & k & objectfbs 0729 + 269 + 61 & j07475919 + 2052254 & ce & k & + 62 & j08231037 - 0153257 & c & k & + 63 & j08292902 + 1046241 & c & k & + 64 & j08423302 + 0621195 & m & k & + 65 & j08541870 - 1200541 & ce & k & + 66 & j09111450 - 0922053 & me & k & + 67 & j09333061 - 2216282 & m & k & + 68 & j20245404 + 2609115 & m & b & + 69 & j20291739 + 2617284 & me & b & + 70 & j20304177 + 2812340 & m & b & , iras20285 + 2802 + 71 & j20311267 + 2612270 & m & b & + 72 & j20415136 + 2752525 & m & b & + 73 & j20532040 + 2516196 & c & b & + 74 & j20551307 + 3254065 & m & b & + 75 & j20555284 + 2640515 & m & b & , iras20537 + 2629 + 76 & j21040556 + 2632111 & m & b & , iras21019 + 2620 + 77 & j21244172 + 4437134 & ce & b & , iras21228 + 4424 + 78 & j21383182 + 4542469 & ce & k & , iras21366 + 4529 + 79 & j21595030 + 3313596 & m & b & + 80 & j22024329 + 4216400 & bl lac&k & + 81 & j22055958 + 3530057 & m & b & + 82 & j22070988 + 2828374 & m & b & , irasf22048 + 2813 + 83 & j22075421 + 4105113 & m & b & , iras22057 + 4050 + 84 & j22084406 + 4855248 & m & b & , iras22067 + 4840 + 85 & j22121336 + 4646065 & c & b & iras22101 + 4631 + 86 & j22135091 + 2447203 & m & b & + 87 & j22213857 + 3335586 & c & b & + 88 & j22261658 + 4221089 & a0 & b & + 89 & j22295650 + 4546539 & ce & b & + 90 & j22313443 + 4816005 & c & k & , iras22294 + 4800 + 91 & j22314368 + 4748038 & pn & k & , iras22296 + 4732 + 92 & j22451504 + 5051534 & ce & b & , iras22431 + 5036 + 93 & j22491976 + 5154487 & m & b & iras22472 + 5138 + 94 & j22514566 + 4921137 & c & b & iras22495 + 4905 + 95 & j22521809 + 3413364 & m & b & iras22499 + 3357 + 96 & j22592372 + 4811589 & me & k & + 97 & j23023314 + 4649483 & m & b & , iras23002 + 4633 + 98 & j23113005 + 4702525 & m & b & iras23092 + 4646 + 99 & j23175960 + 4645122 & m & b & , iras23156 + 4628 + sequential star formation & yes . pms stars close to triggering & no + & sources are older than those close to brcs & + stellar distribution & pms stars are located between triggering & pms stars can be anywhere , including being far + & sources or around surfaces of brcs with no & behind the surface of a brc + & young stars embedded much behind the i - fronts & + star formation efficiency & high & low , less than a few percent . + lcccccc hd222142 & b9.5 v & 0.3 & -3.1 & 0.6 & 0.6 & ucac2 + hd222086 & b9 v & 0.5 & -2.8 & 1.0 & 1.1 & ucac2 + hd222046 & b8 vp & 0.4 & -2.7 & 1.0 & 1.0 & ucac2 + star 40 & continuum & 3.4 & -7.8 & 2.7 & 2.6 & ucac2 + lacob1a & - & -0.3 & -3.7 & - & - & hipparcos + lacob1b & - & -0.5 & -4.6 & - & - & hipparcos + and [ ] , 6300 and 6363 emission lines . star 53 is an haebe star and shows the h@xmath3 in emission but the other balmer lines in absorption.,title="fig : " ] and [ ] , 6300 and 6363 emission lines . star 53 is an haebe star and shows the h@xmath3 in emission but the other balmer lines in absorption.,title="fig : " ] 100 @xmath2 image of the trapezium region . the dots indicate cttss ( white ) and haebe stars ( black ) , labeled with the identification numbers from tables 3 and 4 . pms stars with forbidden line(s ) are bracketed . the boxes mark the fields of the h@xmath3 images shown in ( fig . [ fig : lot1 ] ) . east is to the left and north to the top . ] 100 @xmath2 and co images of the @xmath0 ori region . the symbols are the same as in figure [ fig : oriob1 ] . the distribution of pms stars extends from @xmath0 ori to b30 and b35 . east is to the left and north to the top.,title="fig : " ] 100 @xmath2 and co images of the @xmath0 ori region . the symbols are the same as in figure [ fig : oriob1 ] . the distribution of pms stars extends from @xmath0 ori to b30 and b35 . east is to the left and north to the top.,title="fig : " ] images of the ori ob1 brcs . the stars in table 3 are marked . east is to the left and north to the top . the field of view of each image is @xmath11.,title="fig : " ] images of the ori ob1 brcs . the stars in table 3 are marked . east is to the left and north to the top . the field of view of each image is @xmath11.,title="fig : " ] images of the ori ob1 brcs . the stars in table 3 are marked . east is to the left and north to the top . the field of view of each image is @xmath11.,title="fig : " ] images of the ori ob1 brcs . the stars in table 3 are marked . east is to the left and north to the top . the field of view of each image is @xmath11.,title="fig : " ] images of the ori ob1 brcs . the stars in table 3 are marked . east is to the left and north to the top . the field of view of each image is @xmath11.,title="fig : " ] images of the ori ob1 brcs . the stars in table 3 are marked . east is to the left and north to the top . the field of view of each image is @xmath11.,title="fig : " ] 100 @xmath2 , h@xmath3 and co images of the lacob1 association using the orientation of the galactic coordinates , i.e. , north is to the top and the galactic longitude increases to the left . the white plus signs indicate the 3 late - b stars , hd222142 , hd222086 , and hd222046 , in gal110@xmath2413 and the cross marks indicate the ctts candidate in the iras22343 + 3944 group , 2massj22354224 + 3959566 . the other symbols are the same as in fig . [ fig : oriob1].,title="fig : " ] 100 @xmath2 , h@xmath3 and co images of the lacob1 association using the orientation of the galactic coordinates , i.e. , north is to the top and the galactic longitude increases to the left . the white plus signs indicate the 3 late - b stars , hd222142 , hd222086 , and hd222046 , in gal110@xmath2413 and the cross marks indicate the ctts candidate in the iras22343 + 3944 group , 2massj22354224 + 3959566 . the other symbols are the same as in fig . [ fig : oriob1].,title="fig : " ] 100 @xmath2 , h@xmath3 and co images of the lacob1 association using the orientation of the galactic coordinates , i.e. , north is to the top and the galactic longitude increases to the left . the white plus signs indicate the 3 late - b stars , hd222142 , hd222086 , and hd222046 , in gal110@xmath2413 and the cross marks indicate the ctts candidate in the iras22343 + 3944 group , 2massj22354224 + 3959566 . the other symbols are the same as in fig . [ fig : oriob1].,title="fig : " ] and [ ] images of lbn437 . star 52 is associated with the nebulosity hh398 . east is to the left and north to the top . the field of view of each image is @xmath11,title="fig : " ] and [ ] images of lbn437 . star 52 is associated with the nebulosity hh398 . east is to the left and north to the top . the field of view of each image is @xmath11,title="fig : " ] image of the iras22343 + 3944 group . cttss 35 , 36 , 37 , and 38 from table 3 ( in white ) and haebe star 53 ( in black ) are labeled . the cross indicates the ctts candidate , 2massj22354224@xmath213959566 . the other symbols are the same as in fig . [ fig : oriob1 ] . east is to the left and north to the top . ]

in this paper we identify objectively and analyze groups of galaxies in the recently completed esp survey ( and ; ) . we find 231 groups above the number overdensity threshold = 80 in the redshift range 5000 60000 . these groups contain 1250 members , 40.5 % of the 3085 esp galaxies within the same range . the median velocity dispersion ( corrected for measurement errors and computed at the redshift of the group ) is = 194 . we show that our result is reliable in spite of the particular geometry of the esp survey ( two rows of tangent circular fields of radius arcmin ) , which causes most systems to be only partially surveyed . in general , we find that the properties of esp groups are consistent with those of groups in shallower ( and wider ) catalogs ( e.g. cfa2n and ssrs2 ) . as in shallower catalogs , esp groups trace very well the geometry of the large scale structure . our results are of particular interest because the depth of the esp survey allows us to sample group properties over a large number of structures . we also compare luminosity function and spectral properties of galaxies that are members of groups with those of isolated galaxies . we find that galaxies in groups have a brighter with respect to non member galaxies ; the slope is the same , within the errors , in the two cases . we find that 34% ( 467/1360 ) of esp galaxies with detectable emission lines are members of groups . the fraction of galaxies without detectable emission lines in groups is significantly higher : 45% ( 783/1725 ) . more generally , we find a gradual decrease of the fraction of emission line galaxies among members of systems of increasing richness . this result confirms that the morphology - density relation found for clusters also extends toward systems of lower density . 0v 0d = = = 1=1=0pt = 2=2=0pt

astro-ph9811150

the study of groups of galaxies as dynamical systems is interesting not only _ per se _ , but also because groups can be used to set constraints on cosmological models ( e.g. frenk 1990 ; weinberg & cole 1992 , zabludoff 1993 ; zabludoff & geller 1994 , nolthenius 1994 , 1997 ) and on models of galaxy formation ( frenk 1996 ; kaufmann 1997 ) . groups are also interesting sites where to look for interactions of galaxies with their environment , in order to obtain information on galaxy evolution processes ( postman & geller 1984 , allington - smith 1993 ) . group catalogs identified in redshift space are increasing both in number and size ( cfa2n , rpg ; ssrs2 , ramella 1998 ; perseus - pisces , trasarti battistoni 1998 ; lcrs , tucker 1997 ) . at the same time , cosmological n - body simulations are reaching the resolution required to allow replication of the observational techniques for the identification of groups . in particular , frederic ( 1995a , b ) uses n - body simulations to evaluate and compare the performances of commonly used group finding algorithms . among the main properties of groups , the velocity dispersion is of particular interest . it is easy to measure and it is well suited for comparison with the predictions of cosmological n - body models ( frenk 1990 ; moore 1993 ; zabludoff 1993 ) . distributions of velocity dispersions of nearby groups are now well determined with rather small statistical uncertainties given the large size of the samples . ramella 1995 and ramella 1996 survey the redshifts of candidate faint members of a selection of nearby groups and find that the velocity dispersion of groups is stable against inclusion of fainter members . in other words , the velocity dispersion estimated on the basis of the fewer original bright members is a good indicator of the velocity dispersion obtained from a better sampling of the same group . in this paper we identify and analyze groups of galaxies in the recently completed esp survey ( vettolani 1997 ) . the esp group catalog is interesting because of its depth ( @xmath9 ) and because it samples a new independent region of the universe . esp is a nearly bi - dimensional survey ( the declination range is much smaller than the right ascension range ) , five times deeper than either cfa2 ( geller & huchra 1989 ) or ssrs2 ( da costa 1994 ) . the volume of the survey is @xmath10 at the sensitivity peak of the survey , @xmath11 , and @xmath12 at the effective depth of the sample , @xmath13 . even if the volume of esp is of the same order of magnitude of the volumes explored with the individual cfa2 , ssrs2 , and perseus - pisces samples , it intercepts a larger number of structures . in fact , the strip geometry is very efficient for the detection of large scale structures within redshift surveys ( de lapparent 1988 ) . in particular we determine the distribution of the velocity dispersions of groups and show that our result is reliable in spite of the particular geometry of the esp survey ( two rows of adjacent circular fields of radius @xmath14 arcmin , see figure 1 of vettolani 1997 ) . an important aspect that distinguishes the esp group catalog from the other shallower catalogs is that we have the spectra of all galaxies with measured redshift . it is already well known that emission line galaxies are rarer in rich clusters than in the field ( biviano 1997 ) . the relation between the fraction of emission line galaxies and the local density is a manifestation of the morphology density relationship observed for clusters ( dressler 1980 ) , a useful tool in the study of galaxy evolution . with the esp catalog we explore the extent of the morphology density relationship in the intermediate range of densities that are typical of groups at a larger depth than in previous studies . we note that preliminary results of a search for groups in the las campanas redshift survey ( shectman 1996 ) have been presented by tucker ( 1997 ) . the properties of these groups , as distant as ours , are difficult to compare with those of our esp groups and with those of shallower surveys because lcrs a ) is a red band survey ( esp and shallower surveys are selected in the blue band ) , b ) it is not simply magnitude limited , and c ) it does not uniformly sample structures in redshift space . in particular , the different selection criteria could have a strong impact on the results concerning the morphology - density relation , the luminosity segregation , and the possible differences between the luminosity functions of member and non - member galaxies . in section 2 ) we briefly describe the data ; in section 3 ) we analyze the effect of the esp geometry on the estimate of the velocity dispersion of groups ; in section 4 ) we summarize the group identification procedure ; in section 5 ) we present the esp group catalog ; in section 6 ) we analyze properties of groups that are relevant to a characterization of the large scale structure ( lss ) ; in section 7 ) we analyze the properties of galaxies in groups and compare them to the properties of `` field '' galaxies ( _ i.e. _ galaxies that have not been assigned to groups ) and `` cluster '' galaxies ; in section 8) we identify esp counterparts of aco and/or edcc clusters ( lumsden 1992 ) . finally , we summarize our results in section 9 ) . the eso slice project ( esp ) galaxy redshift survey is described in vettolani ( 1997 ) . the data of the full sample together with a detailed description of the instrumental set - up and of the data reduction can be found in vettolani ( 1998 ) . here we only briefly describe the survey . the esp survey extends over a strip of @xmath15 ( strip a ) , plus a nearby area of @xmath16 ( strip b ) , five degrees west of the main strip , in the south galactic pole region ( @xmath0 and @xmath1 respectively ; @xmath2 ) . each of the two strips is covered with two rows of slightly overlapping circular fields of angular radius @xmath17 arcmin , the separation between the centers of neighboring circles being 30 arcmin . each field corresponds to the field of view of the multifiber spectrograph optopus at the 3.6 m eso telescope that was used to obtain almost all of the spectra ( the mefos spectrograph was used in the last esp run ) . throughout this paper we will assume that the circular fields are tangent , with an angular radius of 15 arcmin : this simplification has no consequences on the galaxy sample . the total solid angle of the spectroscopic survey is 23.2 square degrees . the galaxy catalog consists of all ( candidate ) galaxies brighter than the limiting magnitude @xmath18 listed in the edinburgh durham southern galaxy catalogue ( heydon dumbleton et al . 1988 , 1989 ) . the spectra cover the wavelength range 3730 to 6050 , with an average pixel size of 4.5 . galaxy redshifts are measured by cross correlating sky - subtracted spectra with a set of 8 template stars observed with the same instrumental set - up used to obtain the galaxy spectra . in this paper we use emission line redshifts only for galaxies with no reliable absorption line redshift . the median internal velocity error is of the order of @xmath19 . from a comparison of our 8 templates with three sao radial velocity standard stars we estimate that the zero point error should be smaller than @xmath20 . the total number of confirmed galaxies with reliable redshift measurement is 3342 . the completeness of strip a and strip b are estimated to be 91% and 67% respectively . to all practical purposes , the projection of the esp survey on the sky consists of two rows of adjacent circular optopus fields of radius 15 arcmin and a separation of 30 arcmin between adjacent centers . the angular extent of groups and clusters at the typical depth of the survey ( @xmath21 ) are comparable , or even larger , than the size of the optopus fields . therefore , most systems falling into the survey s volume are only partially surveyed . the main effect of the `` mask '' of optopus fields is to hide a fraction of group members that lie within or close to the strip containing the mask ( the optopus fields cover 78% of the area of the `` un - masked '' strip ) . because of the hidden members , several poor groups may not appear at all in our catalog . on the contrary , our catalog might include parts of groups that are centered outside the esp strip . these problems notwithstanding , we expect to derive useful information on the most important physical parameter of groups , the velocity dispersion , . our estimate of the parent velocity dispersion , @xmath22 , is based upon the sample standard deviation @xmath23 . the sample standard deviation defined as @xmath24 is a nearly unbiased estimate of the velocity dispersion ( ledermann , 1984 ) , independent of the size @xmath25 of the sample . we make the standard assumptions that a ) barycentric velocities of members are not correlated with their real 3d positions within groups , and that b ) in each group the distribution of barycentric velocities is approximately gaussian . because the position on the sky of the optopus mask is not related to the positions of groups , its only effect is to reduce at random @xmath25 . therefore , using an unbiased estimate of the velocity dispersion , the mask has no effect on our determination of the average velocity dispersions of groups . the effect of the mask is to broaden the distribution of the sample standard deviations . the variance of the distribution of sample standard deviations varies with @xmath25 approximately as @xmath26 ( ledermann , 1984 ) . this distribution , proportional to the @xmath27 distribution , is skewed : even if the mean of the distribution is unbiased , is more frequently underestimated than overestimated . while it is easy to predict the effect of the mask on the determination of the velocity dispersion of a single group , it is rather difficult to predict the effect of the mask on the observed distribution of velocity dispersions of a sample of groups with different `` true '' velocity dispersions and different number of members . in order to estimate qualitatively the effect of the mask on the shape of the distribution of velocity dispersions , we perform a simple monte carlo simulation . we simulate a group by placing uniformly at random @xmath25 points within a circle of angular radius @xmath28 corresponding , at the redshift of the group , to the linear projected radius @xmath29 = 0.5 . this radius is the typical size of groups observed in shallow surveys ( e.g. rpg ) . we select the redshift of the group , @xmath30 , by random sampling the observed distribution of esp galaxy redshifts . in order to start from reasonably realistic distributions , we set @xmath25 and the velocity dispersion , , by random sampling the relative histograms obtained from our esp catalog . we limit the range of to 3 @xmath31 @xmath32 18 and the range of to 0 @xmath32 @xmath31 1000 . we lay down at random the center of the simulated group within the region of the sky defined by extending 15 arcmin northward and southward the `` un - masked '' limits of the esp survey . we then assign to each of the points a barycentric velocity randomly sampled from a gaussian with dispersion centered on @xmath30 . we compute the velocity dispersion , @xmath33 , of the velocities . finally , we apply the mask and discard the points that fall outside the mask . we discard the whole group if there are fewer than 3 points left within the mask ( @xmath34 ) . if the group `` survives '' the mask , we compute the dispersion @xmath35 of the @xmath36 members . on average , 78% of the groups survive the mask ( this fraction corresponds to the ratio between the area covered by the mask and the area of the `` un - masked '' strip ) . the percentage of surviving groups depends on the exact limits of the region where we lay down at random groups and on the projected distribution of members within @xmath29 . for the purpose of the simulation , the fraction of surviving groups is not critical . we repeat the procedure 100 times for @xmath37 = 231 simulated groups ( the number of groups identified within esp ) . at each run we compute the histograms n(@xmath36 ) and n(@xmath35 ) . in figure 1 we plot the input distribution n(@xmath25 ) thin line together with the average output distribution , @xmath38n(@xmath36)@xmath39 thick line. errorbars represent @xmath40 one standard deviation derived in each bin from the distribution of the 100 histograms n(@xmath36 ) ; for clarity we omit the similar errorbars of n(@xmath25 . the factor @xmath41 normalizes the output distribution to the number of input groups . the two histograms in figure 1 are clearly very similar since n(@xmath25 ) is within one sigma from @xmath38n(@xmath36)@xmath39 for all values of . we point out here that the similarity between the input and output histograms does not mean that the surviving groups have not changed . in fact only about 63% of the triplets survive the mask while , for example , 88% of the groups with 5 members and 98% of those with 10 members `` survive '' the mask . figure 2 shows the results of our simple simulation for the velocity dispersion . the thin histogram is the input `` true '' distribution n(@xmath42 ) . the dotted histogram is the average `` observed '' distribution obtained without dropping galaxies that lie outside the optopus mask , n(@xmath33 ) . this is the distribution we would observe if the geometry of the survey would be a simple strip . the third histogram ( thick line ) is the average output distribution in presence of the optopus mask , @xmath38n(@xmath35)@xmath39 ( errorbars are @xmath40 one - sigma ) . the input distribution n(@xmath42 ) , the distribution n(@xmath33 ) , and the distribution @xmath38n(@xmath35)@xmath39 are all within one - sigma from each other . in particular the two distributions we observe with and without optopus mask are undistinguishable ( at the 99.9% confidence level , according to the ks test ) . the low velocity dispersion bins are slightly more populated in the `` observed '' histograms because the estimate of the `` true '' is based on small . note that in the case of real observations , some groups in the lowest bin will be shifted again to the next higher bin because of measurement errors . our results do not change if we take into account the slight dependence of from observed within our esp catalog : also in this case the effect of the mask is negligible . in conclusion , the simulation confirms our expectation that the optopus mask has no significant effect on the shape of the distribution of velocity dispersions . we identify groups with the so - called friend - of - friend algorithm ( fofa ; huchra & geller , 1982 ) as described in rpg . we implement here the cosmological corrections required by the depth of the sample ( @xmath43 ) . throughout this paper we use h@xmath44 = 100 mpc@xmath45 and @xmath46 . for each galaxy in the magnitude limited esp catalog , the fofa identifies all other galaxies with a projected comoving separation @xmath47 and a line - of - sight velocity difference @xmath48 here @xmath49 and @xmath50 are the velocities of the two galaxies in the pair . all pairs linked by a common galaxy form a `` group '' . the two parameters , are scaled with distance in order to take into account the decrease of the magnitude range of the luminosity function sampled at increasing distance . the scaling is @xmath51 and @xmath52 where @xmath53^{1/3 } , \eqno ( 5)\ ] ] @xmath54 and @xmath55 is the absolute magnitude corresponding to @xmath56 at a fiducial velocity @xmath57 . we compute @xmath58 with the mattig ( 1958 ) expression , where @xmath59 . finally , @xmath60 is the @xmath61correction `` weighted '' with the expected morphological mix at each redshift as in zucca ( 1997 ) . the scaling is the same for both and and it is normalized at the fiducial velocity @xmath57 = 5000 , where 0 = @xmath62 ) and 0 = @xmath63 ) . in particular , a given value of 0 corresponds to a minimum number overdensity threshold for groups , . the luminosity function we use is the schechter parametrization with @xmath64 , @xmath65 , and @xmath66 mpc@xmath67 computed for esp galaxies by zucca ( 1997 ) . we do not consider galaxies with velocities @xmath68 because the linear extension of the survey in the direction of the declination is smaller than the typical size of a group for @xmath695000 . we also limit the maximum depth of our group catalog to @xmath70 . beyond this limit the accessible part of the luminosity function becomes very small and the scaling of the fofa linking parameters excessively large . the main characteristics of the distribution of galaxies within the volume of the universe surveyed by esp ( vettolani 1997 ) are very similar to those observed within shallower , wider angle magnitude limited redshift surveys . for this reason we expect that the conclusions on the fine - tuning of fofa reached by ramella 1989 , frederic 1995a , and rpg will hold true also for esp . in particular , rpg show that within the cfan2 redshift survey the choice of the fofa parameters is not critical in a wide region of the parameter space around (= 80 , 0 = 350 ) . with our luminosity function and fiducial velocity , we obtain = 80 for 0 = 0.26 mpc , a value comparable to the 0 value used for cfan2 . it is therefore reasonable to expect that the same results of the exploration of the parameter space will hold also for the esp survey . in order to verify our expectation , we run fofa with the following five pairs of values of the linking parameters selected among those used by rpg : (= 80 , 0 = 250 ) , (= 80 , 0 = 350 ) , (= 80 , 0 = 600 ) , (= 60 , 0 = 350 ) , (= 100 , 0 = 350 ) . based on rpg , these pairs of values are sufficient to give an indication of the stability of the group catalogs in the parameter space ( , ) . the number of groups in the five cases is n@xmath71 = 217 , 231 , 253 , 239 , and 217 respectively . we plot in figure 3 the observed distributions of the velocity dispersions of the five group catalogs . we compare the distribution obtained for (= 80 , 0 = 350 ) thick histogram in figure 3 with the other four distributions and find that the only significant difference ( 99.9 % level , according to the ks test ) occurs with the distribution obtained using the largest velocity link , 0 = 600 ( dotted histogram ) . this value of the velocity - link produces an excess of high velocity dispersion systems . frederic 1995a , ramella 1989 , and rpg argue that these high velocity dispersion systems are likely to include a significant number of interlopers ( galaxies with high barycentric velocity that are not physically related to the group in real space ) . on the basis of the results of our tests , we choose the catalog obtained with = 80 ( 0 = 0.26 ) and 0 = 350 as our final esp group catalog . this choice offers the advantage of a straightforward comparison between the properties of esp catalog and those of the cfa2n ( rpg ) , and ssrs2 ( ramella 1998 ) catalogs . we identify 231 groups within the redshift limits @xmath72 . these groups contain 1250 members , 40.5 % of the 3085 esp @xmath73 galaxies within the same range . in table 1 we present our group catalog . for each group we list the i d number ( column 1 ) , the number of members ( column 2 ) , the coordinates @xmath74 and @xmath75 ( columns 3 and 4 respectively ) , the mean radial velocity in corrected for virgo infall and galactic rotation ( column 5 ) , and the velocity dispersion ( column 6 ) . we compute the velocity dispersion following the prescription of ledermann ( 1984 ) for an unbiased estimator of the dispersion ( see previous section ) . we also take into account the cosmological expansion of the universe and the measurement errors according to the prescriptions of danese ( 1980 ) . the errors we associate to the redshifts are those output by the rvsao cross - correlation procedure multiplied by a factor 1.6 . this factor brings the cross - correlation error in rough agreement with the external error estimated from repeated observations ( vettolani 1998 here we do not distinguish between emission and absorption line redshifts ) . table 1 is available only in electronic form via anonymous ftp to cdsarc.u-strasbg.fr ( 130.79.128.5 ) or via http://cdsweb.u-strasbg.fr/abstract.html . in the case of 24 groups , the correction of for the measurement errors leads to a negative value . in column 6 of table 1 we give the error as an upper limit to for these groups . not all galaxies in the region of the sky covered by the esp survey have a measured redshift . of the original target list , 444 objects are not observed , and 207 objects have a noisy spectrum , insufficient for a reliable determination of the redshift . in table 1 we give , for each group , the ratio of these objects to the number of members ( column 7 ) . in computing these rates , we assign to each group the objects without redshift whose ( @xmath7 , @xmath76 ) coordinates fall within the angular region defined by the union of the circular regions obtained by projecting on the sky circles of linear radius @xmath77 centered on all group members . there are groups that are separated along the line - of - sight but that overlap once projected on the sky . if an object without redshift lies within the overlap region , we assign the object to both groups . there are 67 groups that do not contain any object of the photometric catalog without measured redshift . on the other hand , in the case of 51 groups the number of objects without redshift equals , or exceeds , the number of members . these groups are mostly triplets and quadruplets . only 14 out of the 51 ( possibly ) highly incomplete groups have @xmath78 5 . most of these groups are located in the relatively small region b of the redshift survey ( vettolani 1998 ) , which is the least complete ( completeness level = 71% ) . finally , we estimate that only 8 out of 231 groups are entirely contained within one optopus field . by `` entirely contained '' we mean that none of the circles of projected linear radius @xmath79 centered on the member galaxies crosses the edges of the optopus fields . in this section we discuss properties of esp groups that can be used to characterize the lss and that set useful constraints to the predictions of cosmological n - body models . the first `` global '' property of groups we consider is the ratio , @xmath80 , of their number to the number of non - member galaxies within the survey . for esp we have @xmath81 = 0.13 @xmath40 0.01 , for cfa2n rpg find @xmath82 = 0.13 @xmath40 0.01 , for ssrs2 ramella 1998 find @xmath83 = 0.12 @xmath40 0.01 . clearly the proportion of groups among galaxies is the same in all three independent volumes of the universe surveyed with esp , cfa2n and ssrs2 . because cfa2n and ssrs2 mostly sample only one large structure while esp intercepts several large structures , our result means that the clustering of galaxies in groups within the large scale structure is homogeneous on scales smaller than those of the structures themselves . we point out that , on the basis of our simple simulation , we do not expect the optopus mask to affect the determination of @xmath80 . we now consider the ratio of member to non - member galaxies , @xmath84 . within esp we have @xmath85 = 0.68 @xmath40 0.02 ; within cfan and ssrs2 , the values of the ratio are @xmath86 = 0.81 @xmath40 0.02 and @xmath87 = 0.67 @xmath40 0.02 respectively . quoted uncertainties are one poissonian standard deviation . according to the poissonian uncertainties , @xmath85 and @xmath87 are undistinguishable . the value of @xmath86 is significantly different from the other two ratios . however , the real uncertainty in the ratio of members to non - members is higher than the poissonian estimate because the fluctuations in the number of members is dominated by the fluctuations in the smaller number of groups . moreover , the total number of members is strongly influenced by few very rich systems . in fact , it is sufficient to eliminate two clusters , virgo and coma , from cfa2n in order to reduce the value of @xmath86 to @xmath86 = 0.70 @xmath40 0.02 , in close agreement with the ratio observed within esp and ssrs2 . in conclusion , groups are a remarkably stable property of the large - scale distribution of galaxies . once the richest clusters are excluded , the abundances of groups and of members relative to that of non - member or``field '' galaxies are constant over several large and independent regions of the universe . we plot in the top panel of figure 4 the cone diagram ( @xmath7 vs @xmath88 ) for the 3085 esp galaxies within 5000 @xmath89 60000 . in the bottom panel of figure 4 we plot the cone diagram of the 231 esp groups . figures 4 shows that groups trace very well the galaxy distribution , as they do in shallower surveys ( @xmath90 12000 ) . note that in figure 4 we project adjacent beams , not a strip of constant thickness . the topology of the galaxy distribution in redshift space has already been described by vettolani ( 1997 ) and will be the subject of a forthcoming paper . the most striking features are the voids of sizes @xmath91 50 and the two density peaks at @xmath92 18000 and @xmath93 30000 . these features are also the main features of the group distribution . in figure 5 we plot the redshift distributions of groups ( thick histogram ) and galaxies ( thin galaxies ) , divided by the total number of groups and by the total number of galaxies , respectively . in figure 6 we plot number densities of galaxies in redshift bins . number densities are computed using the @xmath94 estimator of davis & huchra ( 1982 ) : all the details about the density estimates are given in zucca 1997 . we vary the size of the redshift bins in order to keep constant the number of galaxies expected in each bin based on the selection function . the top panel is for member galaxies , the middle panel is for isolated and binary galaxies , and the bottom panel is for all esp galaxies . the dashed lines represent the @xmath40 one sigma corridor around the mean galaxy density . it is clear from figure 5 that the redshift distributions of groups and galaxies are undistinguishable ( 98 % confidence level ) . not surprisingly , the number density in redshift bins of members and all galaxies are highly correlated ( figure 6a and 6c ) . more interestingly , the number density distribution of non - member galaxies is also correlated with the number density distribution of all galaxies ( figure 6b and 6c ) . in particular , the two density peaks at @xmath93 18000 and @xmath93 30000 of the number density distribution of all galaxies are also identifiable in the number density distribution of non - member galaxies , even if with a lower contrast . we know from the previous section that groups are a very stable global property of the galaxy distribution within the volume of the esp and within other shallower surveys . here we show that a tight relation between non - member galaxies and groups exists even on smaller scales . our result is particularly interesting in view of the depth of the esp survey and of the number of large structures intercepted along the line - of - sight . we now discuss the velocity dispersions of esp groups . according to our simulation in section 2 , the effect on the velocity dispersions of the optopus mask is statistically negligible . the median velocity dispersion of all groups is @xmath4 = 194 ( 106,339 ) . the numbers in parenthesis are the 1st and 3rd quartiles of the distribution . poor groups with @xmath38 5 have a median velocity dispersion @xmath95=145 ( 65,254 ) , richer groups have @xmath96=272 ( 178,399 ) . for comparison , the median velocity dispersions of cfa2n and ssrs2 are @xmath97 = 198 ( 88,368 ) and @xmath98 = 171 ( 90,289 ) . we take the values of the velocity dispersions for the cfa2 and ssrs2 groups from ramella ( 1997,1998 ) . in order to compare these velocity dispersions with ours , we correct them for a fixed error of 50 ( corresponding to an rvsao error of @xmath91 35 ) and multiply them by @xmath99 . we note that , because of the optopus mask , the comparison of the velocity dispersions of `` rich '' and `` poor '' groups within esp with those of similar systems within cfa2n and ssrs2 is not meaningful . a fraction of esp `` poor '' groups may actually be part of `` rich '' groups . in figure 7 we plot ( thick histogram ) the distribution of the velocity dispersions , @xmath100 , normalized to the total number of groups . errorbars are one sigma poissonian errors . we also plot the normalized distributions of cfa2n and ssrs2 . according to the ks test , differences between @xmath100 and the other two distributions are not significant ( p@xmath101 = 0.3 and 0.2 for the comparison between esp and cfa2n and ssrs2 respectively ) . it is interesting to point out that @xmath102 and @xmath103 do differ significantly ( 97% level ) , @xmath102 being richer of high velocity dispersion systems ( marzke , 1995 ) . groups with dispersion velocities @xmath104 700 are rare and the fluctuations from survey to survey correspondingly high . the abundance of these high systems is the same within both esp and ssrs2 ( 2% ) but it is higher within cfa2n ( 5% ) . if we disregard these few high velocity dispersion systems , the difference between @xmath102 and @xmath103 ceases to be significant . from this result we conclude that each survey contains a fair representation of groups . the distribution of velocity dispersions is an important characteristic of groups because it is linked to the group mass . therefore @xmath105 constitutes an important constraint for cosmological models . furthermore , is a much better parameter for the classification of systems than the number of members ( even more so in the case of the present catalog where the optopus mask affects the number of members much more than velocity dispersions ) the esp survey provides a new determination of the shape of @xmath105 in a much deeper volume than those of existing shallower surveys . we find that , within the errors , @xmath100 is very similar to both @xmath102 and @xmath103 . in this section we examine the luminosities and the spectral features of galaxies in different environments : the `` field '' , groups , and clusters . the dependence of these properties on the environment offers insights into the processes of galaxy formation and evolution and on the dynamical status of groups . here we investigate the possible difference between the luminosity functions of member and non - member galaxies . we compute the luminosity function with the sty method ( sandage , tamman & yahil 1979 ) . we assume a schechter ( 1976 ) form for the luminosity function and follow the procedure described in detail in zucca ( 1997 ) . we find that galaxies in groups have a brighter @xmath6 with respect to non member galaxies ; the slope @xmath7 does not change significantly in the two cases . in particular the parameters we obtain are @xmath106 and @xmath107 for the 1250 members , and @xmath108 and @xmath109 for the 1835 non members . in figure 8 we draw ( dotted lines ) the confidence ellipses of the @xmath7 and @xmath6 parameters obtained in the two cases of member and non - member galaxies . the two luminosity functions differ at the @xmath110 level . in figure 8 we also plot the confidence ellipses for the parameters of the total sample ( solid lines ) derived in the same volume of esp where we identify groups . the fact that galaxies in groups are brighter than non - member galaxies is a clear demonstration of the existence of luminosity segregation in the esp survey , a much deeper survey than those where the luminosity segregation has been previously investigated ( park 1994 , willmer 1998 ) . our finding is consistent with the results of lin ( 1996 ) , who find evidence of a luminosity bias in their analysis of the lcrs power spectrum . in further support of the existence of a luminosity segregation , we also find that @xmath6 becomes brighter for members of groups of increasing richness . as before , the parameter @xmath7 remains almost constant . only in the case of the richest groups , @xmath111 , we find a marginally significant steepening of the slope @xmath7 . one interesting question is whether the environment of a galaxy has a statistical influence on the presence of detectable emission lines in the galaxy spectrum . because emission lines galaxies are mostly spirals ( kennicutt 1992 ) , the answer to this question is relevant to the investigation of the morphology - density relation in systems of intermediate density . the fraction of esp galaxies with detectable emission lines within the redshift range @xmath112 is 44% ( 1360/3085 ) . of these , ( 34 @xmath40 2)% ( 467/1360 ) are members of groups . the fraction of galaxies without detectable emission lines , , that are members of groups is significantly higher : 783/1725 or ( 45 @xmath40 2)% . we note that our detection limit for emission lines correspond to an equivalent width of about 5 @xmath113 we consider three types of environments : a ) the `` field '' , i.e. all galaxies that have not been assigned to groups , b ) poor goups , i.e. groups with 3 @xmath31 @xmath31 4 , and c ) rich groups with 5 @xmath31 . we find that the fraction of decreases as the environment becomes denser . in the `` field '' the fraction of is @xmath114 = 49% , it decreases to @xmath114 = 46% for poor groups and to @xmath114 = 33% for richer groups . in figure 9 we plot @xmath114 as a function of . we also indicate the values of @xmath114 of the two richest abell clusters in our catalog , a2840 ( @xmath114 = 21% ) and a2860 ( @xmath114 = 19% ) . the significance of the correlation between environment and @xmath114 can be investigated with a 2-way contingency table ( table 2 ) . for simplicity , we do not consider triplets and quadruplets . lrrr & @xmath115 & @xmath116 & @xmath117 `` field '' & 893 & 942 & 1835 + rich groups & 266 & 549 & 815 + total & 1159 & 1491 & 2650 + the contingency coefficient is c=0.15 and the significance of the correlation between environment and frequency of emission line galaxies exceeds the 99.9% level . our result indicates that the morphology - density relation extends over the whole range of densities from groups to clusters . previous indications of the existence of the morphology - density relation for groups are based either on very local samples ( postman & geller 1984 ) or on samples that are not suitable for statistical analysis ( allington - smith 1993 ) . very recently , hashimoto ( 1998 ) also confirm the existence of a morphology - density relation over a wide range of environment densities within lcrs . examining our result in more detail , we note that the fraction of , @xmath114 , in triplets and quadruplets is very similar to the value of @xmath114 for isolated galaxies . triplets and quadruplets are likely to correspond , on average , to the low - density tail of groups . moreover , ramella ( 1997 ) and frederic ( 1995a ) estimate that the fofa could produces a large fraction of unbound triplets and quadruplets . these `` pseudo - groups '' dilute the properties of real bound triplets and quadruplets with `` field '' galaxies , artificially increasing the value of @xmath114 . this effect , in our survey , is partially counter - balanced by the triplets and quadruplets that are actually part of richer systems cut by the optopus mask . considering that rich systems are significantly rarer than triplets and quadruplets , we estimate that the value of @xmath114 we measure for triplets and quadruplets should be considered an upper limit . our catalog also includes esp counterparts of 17 clusters listed in at least one of the aco , acos ( abell 1989 ) or edcc ( lumsden 1992 ) catalogs ( section 8 , table 3 ) . for these clusters @xmath118 = 0.25 ( 63 out of 256 galaxies ) . the number of members of these systems is not a direct measure of their richness because of the apparent magnitude limit of the catalog and because of the optopus mask . however , because they include all the richest systems in our catalog and because they are counterparts of 2-d clusters , it is resonable to assume that they are intrinsically rich . we remember here that biviano ( 1997 ) find @xmath114 = 0.21 for their sample of enacs clusters.the fact that for esp counterparts of clusters we find a lower value of @xmath114 than for the other rich groups ( @xmath119 = 0.36 without clusters ) , further supports the existence of a morphology - density relation over the whole range of densities from clusters to the `` field '' . many systems of our catalog are not completely surveyed , therefore the relationship between @xmath114 and the density of the environment we find can only be considered qualitative . however , while incompleteness certainly increases the variance around the mean result , we do not expect severe systematic biases . in order to verify that incompleteness does not affect our results , we consider the subsample of 67 groups that contain no esp objects without measured redshift . we obtain for this subsample the same relationship between group richness and fraction of emission line galaxies we find for the whole catalog . within esp we identify 12 seyfert 1 galaxies and 9 seyfert 2 galaxies . we identify type 1 seyferts visually on the basis of the presence of broad ( fwhm of a few @xmath120 km s@xmath45 ) components in the permitted lines . our list is complete with the possible exception of objects with weak broad lines which are hidden in the noise . the identification of type 2 seyferts is not straightforward , because it is based on line ratios and usually requires measurements of emission lines which fall outside our spectral range : only the f([o iii]@xmath1215007)/f(h@xmath122 ) ratio is available from our spectra , and it is therefore impossible to draw a complete diagnostic diagram ( baldwin 1981 , veilleux & osterbrock 1987 ) . we classify tentatively as type 2 seyferts all emission line galaxies with @xmath123$\lambda$5007)}/\hbox{f(h$\beta$)})\ge 0.7 $ ] : this threshold cuts out almost all non - active emission line galaxies , but also many narrow - line agn with a medium to low degree of ionization . thus the list of possible seyfert 2 galaxies is almost free of contamination , but should by no means be considered complete . the origin of the seyfert phenomenon could be related to the interaction with close companions ( balick & heckman 1982 , petrosian 1982 , dahari 1984 , mackenty 1989 ) , or to a dense environment ( kollatschny & fricke 1989 , de robertis 1998 ) . observational evidence is , however , far from conclusive . for example , seyfert 1 and seyfert 2 galaxies have been found to have an excess of ( possibly ) physical companions compared to other spiral galaxies by rafanelli ( 1995 ) . laurikainen & salo ( 1995 ) agree with rafanelli ( 1995 ) about seyfert 2 galaxies , but reach the opposite conclusion about seyfert 1 galaxies . in our case , 7 ( 33% ) out of 21 seyferts are group members . for comparison , 460 ( 34% ) emission line galaxies ( not including seyfert galaxies ) are group members and 879 are either isolated or binaries . clearly , within the limits of our relatively poor statistics , we find that seyfert galaxies do not prefer a different environment than that of the other emission line galaxies . in order to test the dependence of the seyfert phenomenon on the interaction of galaxies with close companions rather than with the general environment , we compute for all seyferts and emission line galaxies the projected linear distance to their nearest neighbor , the @xmath124-distance . we limit the search of companions to galaxies that are closer than 3000 along the line of sight . we find that the distribution of @xmath124-distances of the sample of seyfert galaxies is not significantly different from that of all emission line galaxies . we also consider the frequency of companions at projected linear distances @xmath125 mpc . we have 7 seyfert galaxies with such a close companion ( 33% ) and 315 ( 23% ) emission lines galaxies . one of the 7 seyferts is a member of a binary systems , the remaining six seyferts are members of groups . even if , taken at face value , the higher frequency of close companions observed among seyfert galaxies supports a causal connection between gravitational interaction and the seyfert phenomenon , these frequencies are not significantly different . we note that members of close angular pairs ( @xmath126 arcsec ) in the original target list for esp , are more frequently missing from the redshift survey than other objects ( vettolani 1998 ) . this bias , due to optopus mechanical constraints , could hide a possible excess of physical companions of seyfert galaxies . in order to estimate how strongly our result could be affected by this observational bias , we identify the nearest neighbors of seyfert and emission line galaxies from a list including both galaxies with redshift and objects that have not been observed . when we compute projected linear distances to objects that have not been observed , we assume that they are at the same redshift of their candidate companion galaxy . as before , we do not find significant differences between the new @xmath124-distributions of seyferts and . this result demonstrates that the higher average incompleteness of close angular pairs does not affect our main conclusions : a ) seyfert galaxies within esp are found as frequently within groups as other emission line galaxies , b ) seyfert galaxies show a small but not significant excess of close physical companions relative to the other emission line galaxies . we point out again that the sample of seyferts is rather small and the statistical uncertainties correspondingly large . crrcccccc i d & esp & & @xmath74 & @xmath75 & r & @xmath127 & @xmath128 & @xmath129 + & & & ( @xmath130 ) & ( @xmath131 ) & & & & + e0163 & 6 & 3 & 22 32 40 & -40 38 41 & & & 0.13535 & 121 + e0169 & 7 & 11 & 22 34 12 & -39 50 51 & & & 0.06294 & 282 + s1055 & 21 & 9 & 22 39 43 & -40 16 07 & 0 & & 0.02901 & 102 + a4068 & 88 & 9 & 23 57 08 & -39 46 59 & 0 & 0.07151 & 0.10261 & 700 + e0435 & 121 & 8 & 00 17 31 & -40 40 37 & & & 0.15073 & 334 + a2769 & 126 & 6 & 00 21 45 & -39 53 49 & 0 & 0.15708 & 0.14020 & 419 + a2771 & 128 & 18 & 00 21 50 & -40 26 49 & 0 & 0.06260 & 0.06876 & 268 + a2828 & 176 & 5 & 00 49 10 & -39 50 54 & 0 & 0.13133 & 0.19676 & 468 + a2840 & 183 & 34 & 00 52 01 & -40 04 19 & 1 & 0.10460 & 0.10618 & 339 + a2852 & 192 & 12 & 00 57 00 & -39 54 19 & 0 & 0.17581 & 0.19845 & 235 + e0113 & 196 & 16 & 00 58 21 & -40 31 05 & & & 0.05449 & 372 + a2857 & 200 & 8 & 01 00 06 & -40 12 42 & 1 & 0.19092 & 0.19755 & 504 + e0519 & 205 & 43 & 01 02 36 & -40 03 02 & & & 0.10637 & 319 + s0127 & 213 & 25 & 01 05 27 & -40 21 08 & 0 & & 0.10498 & 505 + a2874 & 216 & 25 & 01 06 08 & -40 36 01 & 1 & 0.15812 & 0.14191 & 817 + e0529 & 218 & 17 & 01 07 40 & -40 40 34 & & & 0.10483 & 282 + e0546 & 231 & 7 & 01 19 27 & -39 53 07 & & & 0.11909 & 424 + within our survey lie the centers of 9 aco clusters , 5 acos clusters and 12 edcc clusters . several entries of the three lists correspond to the same cluster . taking into account multiple identifications , there are 20 clusters listed within one or more of the three catalogs that lie within esp . in our catalog we find at least one counterpart for 17 out of the 20 clusters . the three clusters that do not correspond to any of our systems are aco 2860 , acos 11 , and acos 32 , all of abell richness r = 0 . aco2860 is a very nearby object with a redshift , @xmath132 = 0.0268 , close to our minimum redshift . acos 11 and acos 32 are both distance class d = 6 objects that may be either projection effects or real clusters located beyond our redshift limit . we select the esp counterparts among the groups that are close to the clusters on the sky and that have a redshift compatible with the distance class and/or magnitude of the cluster . if more esp groups are counterparts of a cluster , we identify the cluster with the richest counterpart . in table 3 we list the name ( column 1 ) , and the coordinates ( columns 4 , 5 ) of the 17 clusters with esp counterpart together with their richness ( column 6 ) and , if available , their redshift as estimated by zucca ( 1993 ) ( column 7 ) . in the case of clusters listed in both edcc and aco or acos , we give the aco or acos identification number . in the same table 3 we also list the i d number of the cluster counterparts within our catalog ( column 2 ) , their number of members ( column 3 ) , redshift ( column 8) and velocity dispersion ( column 9 ) . there are 8 clusters with redshift estimated by zucca ( 1993 ) the measured redshifts of 6 of these clusters are in good agreement with the estimated redshift : the difference between the two redshifts is of the order of 10% , less than the 20% uncertainty on the estimated redshifts . for the remaining 2 clusters , aco 2828 and aco 4068 , the estimated redshift is significantly smaller than our measured redshift . the projection of the foreground systems esp 175 and esp 178 within the abell radius of aco 2828 could explain the inconsistency between estimated and measured redshift for this cluster . in the case of aco 4068 we do not find any foreground / background system within esp . aco 4068 is very close to the northern declination boundary of the esp strip . an inspection of the cosmos catalog just outside the boundary of the optopus field containing aco 4068 shows that a significant part of this cluster lies outside our redshift survey and therefore background / foreground projection could still be the cause of the inconsistency between its estimated and measured redshifts . we also note that edcc163 and acos1055 are among the most incomplete systems in our catalog . in the fields of edcc163 (= 3 ) and acos1055 (= 9 ) the number of objects without redshift is 16 and 63 respectively . we will not consider these two clusters in what follows . in panel a ) and panel b ) of figure 10 we plot , respectively , and as a function of . as expected , clusters ( represented by large dots ) populate the highest part of both diagrams at all redshifts . in both diagrams , mixed with clusters , there are also esp groups that have not been identified as clusters . the completeness of bidimensional cluster catalogs is an important issue for cosmology ( van haarlem 1997 ) since the density of these clusters and their properties are used as constraints on cosmological models ( e.g. frenk 1990 , bahcall 1997 , coles 1998 ) . it is therefore interesting to determine whether there are other systems selected in redshift space that have properties similar to those of the cluster counterparts but that have escaped 2-d identification . we limit our search for `` cluster - like '' groups to the velocity range 25000 @xmath38 @xmath38 45000 . within this range the selection function is rather stable and relatively close to its maximum . in our catalog we identify the counterparts of 8 2-d clusters within this redshift range . two of the eight clusters are of richness class r=1 ( aco2840=esp183 and aco2874=esp216 ) . the minimum number of members of these clusters is 6 and the lowest velocity dispersion of the 8 clusters is 280 . apart from the counterparts of the 8 clusters , we find 11 additional esp groups that satisfy all three conditions 25000 @xmath38 @xmath38 45000 , @xmath133 , and @xmath134 . these groups are @xmath91 10% of all groups in this redshift interval and we list them in table 4 . rrcccc esp & & @xmath135 & @xmath136 & @xmath88 & @xmath137 + & & @xmath138 & @xmath139 & & + 48 & 7 & 23 26 03 & -39 52 45 & 30231 & 333 + 96 & 8 & 00 02 04 & -40 29 17 & 29329 & 386 + 124 & 17 & 00 20 33 & -40 33 37 & 39102 & 283 + 130 & 8 & 00 23 02 & -40 15 47 & 41898 & 284 + 155 & 9 & 00 40 39 & -40 00 46 & 39280 & 349 + 186 & 7 & 00 54 16 & -40 25 00 & 30163 & 319 + 190 & 13 & 00 55 29 & -40 34 41 & 31033 & 399 + 195 & 8 & 00 58 10 & -40 06 00 & 31728 & 305 + 201 & 11 & 01 00 34 & -40 03 30 & 27225 & 429 + 203 & 10 & 01 02 28 & -40 18 29 & 27298 & 361 + 226 & 12 & 01 14 57 & -39 52 45 & 36298 & 432 + in a plane , figure 11 , the eleven `` cluster - like '' groups occupy a `` transition region '' between clusters and groups . first we note that , in this plane , the two counterparts of the r=1 aco clusters ( esp183 and esp216 ) are very distant from the `` cluster - like '' groups . the same holds true for the only rich edcc cluster that is not an aco cluster , edcc519 . we conclude that no rich cluster is missing from 2-d catalogs in the region of the sky covered by the esp survey . this conclusion is reassuring , even if it does not allow us to discuss the problem of the completeness of rich 2-d clusters in general because it is based on a small number of objects . in the case of the more numerous poorer clusters , figure 11 shows that several systems could be missing from the 2-d list . the boundaries of the cluster and group regions in the plane are blurred by the optopus mask and by the narrow width of the esp survey . it is therefore difficult to give a precise estimate of how many `` cluster - like '' groups should be considered `` missing '' from bidimensional catalogs . that poor 2-d clusters and `` cluster - like '' 3-d groups are probably the same kind of systems is confirmed by the fact that they have the same fraction of , a higher value than it is typical of richer clusters . the 11 `` cluster - like '' groups have a total of 110 members , 43 of which are : @xmath140 = 0.39 . the 4 poor clusters that have @xmath32 17 include 39 members and have @xmath141 = 0.41 . we remember here that for all esp counterparts of clusters we find @xmath118 = 0.25 . in conclusion , the comparison of esp systems with aco , acos and edcc clusters indicates that the `` low mass '' end of the distribution of clusters is poorly represented in 2-d catalogs ; on the other hand , the 2-d catalogs appear reasonably complete for high mass clusters . in this paper we search objectively and analyze groups of galaxies in the recently completed esp survey ( @xmath0 and @xmath142 ; @xmath2 ) . we identify 231 groups above the number overdensity threshold = 80 in the redshift range 5000 @xmath3 60000 . these groups contain 1250 members , 40.5 % of the 3085 esp galaxies within the same range . the median velocity dispersion of esp groups is @xmath4 = 194 ( at the redshift of the system and taking into account measurement errors ) . we verify that our estimates of the average velocity dispersions are not biased by the geometry of the esp survey which causes most systems to be only partially surveyed . the groups we find trace very well the geometry of the large scale distribution of galaxies , as they do in shallower surveys . because groups are also numerous , they constitute an interesting characterization of the large scale structure . the following are our main results on the properties of groups that set interesting `` constraints '' on cosmological models : * the ratio of members to non - members is @xmath85 = 0.68 @xmath40 0.02 . this value is in very close agreement with the value found in shallower surveys , once the few richest clusters ( e.g. coma and virgo ) are neglected . * the ratio of groups to the number of non - member galaxies is @xmath81 = 0.13 @xmath40 0.01 , also in very close agreement with the value found in shallower surveys . * the distribution of velocity dispersions of esp groups is not distinguishable from those of cfa2n and ssrs2 groups . these results are of particular interest because the esp group catalog is five times deeper than any other wide - angle shallow survey group catalog and the number of large scale features explored is correspondingly larger . as a consequence , the properties of esp groups are more stable with respect to possible structure - to - structure variations . the fact that the properties of esp groups agree very well with those of cfa2n and ssrs2 groups indicates that structure - to - structure variations are not large and that the properties of groups we find can be considered representative of the local universe . as far as the richest systems ( clusters ) are concerned , we identify esp counterparts for 17 out of 20 2-d selected aco and/or edcc clusters . because the volume of esp is comparable to the volume of individual shallower surveys , it is not big enough to include a fair sample of clusters . the variations from survey to survey in the number and properties of clusters are large . * the schechter luminosity function of galaxies in groups has a brighter @xmath6 ( -19.80 ) with respect to non member galaxies ( @xmath143 ; the slope @xmath7 ( @xmath144 ) does not change significantly between the two cases . * @xmath6 becomes brighter for members of groups of increasing richness . the parameter @xmath7 remains almost constant ; only in the case of the richest groups we find a marginally significant steepening of the slope @xmath7 . * 34% ( 467/1360 ) of esp galaxies with detectable emission lines are members of groups . the fraction of galaxies without detectable emission lines in groups is significantly higher : 45% ( 783/1725 ) . * the fraction of in the field is @xmath114 = 49% ; it decreases to @xmath114 = 46% for poor groups and to @xmath114 = 33% for richer groups . for the esp counterparts of aco and / edcc clusters @xmath114 = 25% . we conclude that luminosity segregation is at work in the esp survey : galaxies in the dense environment of groups are , on average , brighter than `` field '' galaxies . galaxies in groups are also less likely to have detectable emission lines in their spectra . in fact , we find a gradual decrease of the fraction of emission line galaxies among members of systems of increasing richness : the morphology - density relation clearly extends over the whole range of densities from groups to clusters . as a final note , we identify 12 seyfert 1 galaxies and 9 seyfert 2 galaxies . we find that : a ) seyfert galaxies within esp are members of groups as frequently as other emission line galaxies , and b ) seyfert galaxies show a small but not significant excess of close physical companions relative to the other emission line galaxies . we point out again that the sample of seyferts is rather small and the statistical uncertainties correspondingly large . we thank the referee for his careful reading of the manuscript and his helpful suggestions . this work has been partially supported through nato grant crg 920150 , eec contract erb ct920033 , cnr contract 95.01099.ct02 and by institut national des sciences de lunivers and cosmology gdr . abell g.o . , corwin jr . h.g . , olowin r.p . , 1989 , apjs 70 , 1 ( aco , acos ) . allington - smith j.r . , ellis r. , zirbel e.l . , oemler a. jr . , 1993 , apj 404 , 521 . bahcall n.a . , fan x. , & cen r. , 1997 , apjl 485 , 53 . baldwin j.a . , phillips m.m . , & terlevich r. , 1981 , pasp , 93 , 5 . balick b. , & heckman t.m . , 1982 , ara&a 20 , 431 . biviano a. , katgert p. , mazure a. , 1997 , a&a 321 , 84 . coles p. , pearson r.c . , borgani s. , plionis m. , & moscardini l. , 1998 , mnras 294 , 245 . da costa l.n . , geller m.j . , pellegrini p.s . 1994 , apj 424 , l1 . dahari o. , 1984 , aj 89 , 966 . danese l. , de zotti g. , & di tullio g. , 1980 , a&a 49 , 137 . davis , m. , & huchra , j.p . , 1982 , apj 254 , 437 . de robertis m.m . , yee h.k.c . , & hayhoe k. , 1998 , apj 496 , 93 . dressler a. , 1980,apj 236 , 351 . frenk c.s . , white s.d.m . , efstathiou g. , & davis m. , 1990 , apj 351 , 10 . frenk c.s . , evrard a.e . , white s.d . , & summers f.j . , 1996 , apj 472 , 460 . frederic , j.j . , 1995a , apjs , 97 , 259 . frederic , j.j . , 1995b , apjs , 97 , 275 . geller m.j . , & huchra j.p . , 1989 , science 246 , 897 . hashimoto y. , oemler a. jr . , lin h. , & tucker d.l . , 1998 , apj 499 , 589 . heydon dumbleton n.h . , collins c.a . , macgillivray h.t . , 1988 , in _ large scale structures in the universe _ w. seitter , h.w . duerbeck and m. tacke ( springer verlag ) , p. 71 . dumbleton n.h . , collins c.a . , macgillivray h.t . , 1989 , mnras 238 , 379 . huchra j.p . , & geller m.j . , 1982 , apj 257 , 423 . kaufmann g. , nusser a. , & steinmetz m. , 1997 , mnras 286 , 695 . kennicutt r.c . , 1992 , apj 388 , 310 . kollatschny w. , & fricke k.j . , 1989 , a&a 219 , 34 . de lapparent v. , geller m.j . , & huchra j.p . , 1988 , apj 332 , 44 . laurikainen e. , & salo h. , 1995 , a&a 293 , 683 . ledermann w. , 1984 , _ handbook of applicable mathematics _ ( wiley and sons ) , vol . vi , part a. lin h. , kirshner r.p . , shectman s.a . , landy s.d . , oemler a jr . , tucker d.l . , & schechter p.l . , 1996 apj 471 , 617 . lumsden s.l . , nichol r.c . , collins c.a . , guzzo l. , 1992 , mnras 258 , 1 ( edcc ) . mackenty j.w . , 1989 , apj 343 , 125 . marzke r.o . , geller m.j . , da costa l.n . , & huchra j.p . , 1995 , aj 110,477 . mattig w. 1958 , astron . 284 , 109 . moore b. , frenk c.s . , & white s.d.m . , 1993 , mnras 261 , 827 . nolthenius r , klypin a.a . , & primack j.r . , 1994 , apj 422 , l4 nolthenius r , klypin a.a . , & primack j.r . , 1997 , apj 480,43 . park c. , vogeley m.s . , geller m.j . , & huchra j.p . , 1994 , apj 431 , 569 . petrosian a.r . , 1982 , astrophysics 18 , 312 . postman m. , & geller m.j . , 1984 , apj 281 , 95 . rafanelli p. , violato m. , baruffolo a. , 1995 , aj 109,1546 . ramella m. , geller m.j . , & huchra j.p . , 1989 , apj 344 , 57 . ramella m. , geller m.j , & thorstensen j. , 1995 , aj 109 , 1469 . ramella m. , focardi p. , & geller m.j . , 1996 , a&a 312 , 475 . ramella m. , pisani a. , & geller m.j . , 1997 , aj 113 , 483 ( rpg ) . ramella m. 1998 , in preparation . sandage a. , tamman g. , & yahil a. , 1979 , apj 232 , 252 . schechter p. , 1976 , apj 203 , 297 . shectman s.a . , landy s.d . , oemler a. 1996 , apj 470 , 172 ( lcrs ) . trasarti - battistoni r. , 1997 , a&ass 130 , 341 . tucker d.l . , hashimoto y. , kirshner r.p . 1997 , astro - ph 9711176 . van haarlem m.p . , frenk c.s . , & white s.d . , 1997 , mnras 287 , 817 veilleux s. , & osterbrock d.e . , 1987 , apjs 63 , 295 . vettolani g. , zucca e. , zamorani g. 1997 , a&a 325 , 924 . vettolani g. , zucca e. , merighi r. 1998 , a&ass 130 , 323 . weinberg d.h . , & cole s. , 1992 , mnras 259 , 652 . willmer c.n.a . , da costa l.n . , & pellegrini p.s . , 1998 , aj 115 , 869 . zabludoff a.i . , geller m.j . , huchra j.p . , & ramella m. , 1993 , aj 106 , 1301 . zabludoff a.i . , & geller m.j . , 1994 , aj 107 , 1929 . zucca e. , zamorani g. , vettolani g . , 1997 , a&a 326 , 477 . zucca e. , zamorani g. , scaramella r. , & vettolani g. , 1993 , apj 407 , 470 .

we review methods to calculate eigenvalue distributions of products of large random matrices . we discuss a generalization of the law of free multiplication to non - hermitian matrices and give a couple of examples illustrating how to use these methods in practice . in particular we calculate eigenvalue densities of products of gaussian hermitian and non - hermitian matrices including combinations of gue and ginibre matrices .

1309.2568

products of random matrices arise in many areas of research and engineering @xcite . in studies of dynamical systems one uses products of random matrices to derive limiting laws for lyapunov exponents @xcite ; in information theory and wireless telecommunication to calculate channel capacities for serial mimo ( multiple - input multiple - output ) transmission @xcite . with the help of products of random matrices one investigates unitary evolution @xcite , matrix diffusion @xcite or phase transitions in quantum yang - mills theories @xcite . one encounters products of random matrices in studies of quantum entanglement @xcite , financial engineering @xcite , and many other fields of research @xcite . the problem of random matrix multiplication has attracted attention since the sixties @xcite . originally the studies were mostly concentrated on deriving limiting laws for products of infinitely many finite random matrices @xcite with some exceptions where also limiting laws for infinite matrices were formulated @xcite . recently the focus has shifted to products of infinitely large matrices due to the discovery of a correspondence between infinitely large invariant random matrices and free random variables @xcite which have been studied in the framework of free probability @xcite a new branch of probability theory that has been intensively developed since the eighties @xcite . many results derived in this theory have been applied to large random matrices in the limit of infinite matrix size @xmath0 . in this paper we review some of these results , and give practical recipes to calculate eigenvalue densities of products of large random matrices . instead of giving formal proofs we gradually build up intuition passing from matrix addition through multiplication of hermitian random matrices to multiplication of non - hermitian matrices . free addition of large random matrices is a counterpart of addition of independent random variables in classical probability . we start with addition because it is much simpler than multiplication but it already has essential ingredients needed to understand a special role of invariant random matrices played in the formulation of limiting laws for eigenvalue distributions in the large @xmath1 limit . the key feature of invariant random matrices is that their eigenvectors are not correlated with the eigenvalues , and that they are uniformly distributed . the last statement means that the unitary matrix constructed from the normalized eigenvectors is uniformly distributed on the unitary group . in effect , randomness of such matrices is entirely shaped by the probability distribution of the eigenvalues . the sum of independent invariant random matrices is an invariant random matrix . in the limit @xmath0 the eigenvalue density of the sum depends only on eigenvalue densities of the matrices added . this law is very similar to the law of addition in classical probability where the probability distribution of the sum of independent random variables can be calculated from the probability distributions of these variables . similarities to classical probability go far beyond that . one can write a dictionary between classical and free probability . in this dictionary , random variable corresponds to invariant random matrix for @xmath0 , probability density function to eigenvalue density , characteristic function to green s function , cumulant generating function ( logarithm of the characteristic function ) to r transform , mellin transform of the probability density function to s transform , independence to freeness , etc . using this correspondence one can find a full analogy between free addition and free multiplication of large invariant matrices and addition and multiplication of independent random variables . it is worth emphasizing that the laws of free addition and free multiplication can be independently derived using field theoretical methods , including perturbation theory of invariant matrix models @xcite , without referring to free probability . perturbation theory is a way of graphical enumeration of planar feynman diagrams that arise in the large @xmath1 limit @xcite . a great advantage of the field theoretical approach is that it also works in the case of non - hermitian matrices and it allows to derive a non - hermitian version of the multiplication law @xcite . to the best of our knowledge this result is beyond the reach of free probability so far . we shortly recall this result here . the paper is organized as follows . in section [ sec - add ] we discuss eigenvalues of a sum of random matrices . in particular we define a slightly modified version of matrix addition such that it is independent of the eigenvectors of the matrices added . this brings us to the class of invariant random matrices that are discussed in section [ sec - inv ] . in section [ sec - fadd ] we recall the definition of the green s function , the moment generating function and the r transform . these functions are used in the derivation of limiting laws for invariant matrices in the large @xmath1 limit . we employ these functions in the formulation of the law of free addition . in section [ sec - mult ] we discuss the law of free multiplication for invariant hermitian matrices . in section [ sec - iso ] we discuss consequences of this law for multiplication of infinitely large isotropic non - hermitian matrices . finally in section [ sec - nh ] we show how to calculate eigenvalue densities for products of large independent non - hermitian matrices . we illustrate calculations with examples . we conclude the paper with a short summary in section [ sec - concl ] . let @xmath2 and @xmath3 be @xmath4 hermitian matrices . we are interested in the eigenvalues of the sum @xmath5 . to calculate them it is not sufficient to know the eigenvalues of @xmath2 and @xmath3 . one would also need to know their eigenvectors . actually it would be enough to know relative positions of the eigenvectors of @xmath2 and @xmath3 : for example the decomposition of @xmath3-eigenvectors in the eigenbasis of @xmath2 . one can , however , define a non - standard addition of matrices that depends only on the eigenvalues of individual matrices in the sum and not on their eigenvectors . this new addition is defined in a probabilistic way as a sum @xmath6 where @xmath7 are independent haar unitary matrices and @xmath8 denotes the hermitian conjugate of @xmath9 . let us recall that haar unitary matrix is a random matrix distributed according to the uniform measure on the unitary group @xmath10 . the matrix @xmath11 is not a concrete matrix but it is a random matrix or equivalently an ensemble of matrices with a certain probability measure . the eigenvalues of @xmath11 are random variables . one is interested in the joint probability distribution of these random variables . the marginal distribution of the joint probability distribution is referred to as the eigenvalue distribution . one should note that the transformations @xmath12 and @xmath13 preserve the eigenvalues of the individual terms , but uniformly randomize their eigenvectors . one can therefore expect that the eigenvalue distribution of @xmath14 depends only on eigenvalues of @xmath2 and @xmath3 . moreover , since the sum @xmath15 can be written as @xmath16 , where @xmath17 is also a haar unitary matrix , the eigenvalue distribution of the matrix @xmath18 averaged over @xmath19 is the same as the eigenvalue density of @xmath11 . in other words it is sufficient to randomize the eigenvectors of one matrix against the other . one can consider a more general version of the problem in which @xmath2 and @xmath3 are independent random matrices instead of being deterministic ones . in this case the question is whether one can determine the eigenvalue distribution of @xmath14 from the eigenvalue distributions of individual matrices @xmath2 and @xmath3 . the answer is affirmative for @xmath0 . in this limit the eigenvalue density @xmath20 of the sum @xmath14 can indeed be determined solely from the eigenvalue densities @xmath21 and @xmath22 . for finite @xmath1 the answer is more complicated since it depends also on eigenvalue correlations . the correlations can be neglected only in the limit @xmath0 . the addition ( [ fa ] ) is called free addition when @xmath0 and @xmath2 and @xmath3 are independent . in the next section we shall see that the addition ( [ fa ] ) is equivalent to the standard matrix addition @xmath5 for invariant random matrices . random matrices are formally defined by matrix ensembles equipped with probability measures . let us denote the measure of an @xmath4 hermitian random matrix @xmath23 by @xmath24 . random matrix @xmath23 is called invariant if the measure is invariant under the transformation @xmath25 , where @xmath26 is an arbitrary unitary matrix : @xmath27 . let us give two standard examples . one can construct an invariant random matrix from a diagonal random matrix @xmath28 and haar unitary random matrix @xmath9 @xmath29 the invariance of the measure for @xmath2 : @xmath30 is a direct consequence of the invariance of the haar measure @xmath31 under the right multiplication by a unitary matrix @xmath26 . one can choose the diagonal elements @xmath32 s of @xmath33 to be independent , identically distributed real random variables . in such a case the joint eigenvalue density factorizes and @xmath34 where @xmath35 is the probability density function for the random variables representing diagonal elements of @xmath33 . the factorization of the joint eigenvalue density reflects the independence of the eigenvalues . the eigenvalue density is @xmath36 . in physical applications one frequently encounters invariant measures of the form @xcite @xmath37 where @xmath38 is a polynomial or a power series in @xmath2 and @xmath39 , denotes flat measure for @xmath4 hermitian matrices . the proportionality sign in ( [ v ] ) means that the right hand side is displayed up to a constant . this implicit constant ensures the probabilistic normalization @xmath40 . the joint probability eigenvalue distribution is given by a well - known dyson formula @xcite @xmath41 which can be interpreted as a partition function that describes a gas of @xmath1 particles in one dimension . we again skipped a normalization constant that ensures @xmath42 . in this picture the @xmath43-th eigenvalue @xmath32 corresponds to the position of the @xmath43-th particle in one - dimensional space . particles are in the potential @xmath44 . they repel each other with a logarithmic potential depending of the distance @xmath45 . the repulsion makes the eigenvalues interact . as a result they are correlated . the eigenvalue density can be calculated by integrating all but one eigenvalue from the joint eigenvalue distribution @xmath46 . in the limit @xmath0 one can relate the eigenvalue density to the potential by an integral equation @xcite @xmath47 obtained by applying the saddle point approximation to ( [ rhon ] ) . stands for principal value . one can use this equation to calculate the eigenvalue density for the given potential or to find the potential corresponding to a given eigenvalue density . in particular one can choose the potential to reproduce the same eigenvalue density as in the first ensemble ( [ ulu ] ) where the eigenvalue density is just equal to the probability density function for the diagonal elements of @xmath33 . the two random matrix ensembles ( [ ulu ] ) and ( [ v ] ) have completely different statistics of eigenvalues even if they have identical eigenvalue densities . let us come back to the general discussion of invariant random matrices . the standard addition @xmath5 of invariant random matrices is equivalent to free addition @xmath14 ( [ fa ] ) . this follows from the invariance of the measure , which implies that the matrix @xmath2 is identically distributed as @xmath48 . the same holds for @xmath3 and @xmath49 . for @xmath2 and @xmath3 being invariant random matrices , the two sums @xmath5 and @xmath14 have identical probability measures . they represent the same random matrix ensemble . therefore from here on we shall not distinguish between @xmath5 and @xmath14 while discussing invariant random matrices . the sum of invariant random matrices is also an invariant random matrix . an important property of the sum @xmath5 of invariant matrices is that the eigenvalue density depends in the limit @xmath0 solely on the eigenvalue densities of @xmath2 and @xmath3 , as follows from the discussion in section [ sec - add ] . independent invariant random matrices for @xmath0 are called free . there is one - to - one correspondence between them and the free random variables known from free probability @xcite . to summarize this section , the standard addition of invariant random matrices is equivalent to free addition ( [ fa ] ) . in the limit @xmath0 independent invariant random matrices are called free and the eigenvalue density @xmath50 of the free sum can be calculated from eigenvalues @xmath21 and @xmath22 of @xmath2 and @xmath3 . the resulting density is independent of the detailed statistics of eigenvalues . for example , it does not depend on whether eigenvalues are correlated or not or whether they are generated using the ensemble of type ( [ ulu ] ) , ( [ v ] ) or any other . the only thing that matters is the eigenvalue densities of the individual matrices in the sum . in the next section we briefly summarize the algorithm to calculate @xmath50 . for @xmath0 the eigenvalue density of an invariant random matrix @xmath2 can be described by an eigenvalue density @xmath21 having support @xmath51 on the real axis . typically @xmath51 is a finite interval . in calculations it is convenient to use the stieltjes transform of the density @xmath52 that is a complex function defined on the complex plane outside @xmath51 . this function is often called green s function . the density can be reconstructed from the green s function @xmath53 by calculating the imaginary part of the green s function close to the real axis @xmath54 the power expansion @xmath53 at infinity @xmath55 generates moments of the eigenvalue distribution @xmath56 these moments are equivalent to the trace moments @xmath57 calculated with respect to the probability measure @xmath58 for the entire matrix @xmath2 . the zeroth moment is fixed @xmath59 by the normalization of the measure . one often defines another moment generating function as a power series at zero @xmath60 it is related to the green s function @xmath61 the eigenvalue density @xmath21 , the moment generating function @xmath62 , and the green s function @xmath53 contain roughly speaking the same information about the eigenvalue distribution of the invariant matrix @xmath2 and one can reproduce these functions from each other . the idea is to calculate these functions for @xmath5 when the corresponding functions for @xmath2 and @xmath3 are given . let us first calculate the moments of @xmath5 @xmath63 the measure factorizes , since @xmath2 and @xmath3 are independent by assumption . for @xmath64 the last equation gives @xmath65 in the second equation we replaced @xmath66 by @xmath67 . the equality @xmath68 follows from the equation @xmath69 , which holds in the limit @xmath0 as a result of invariance and independence of @xmath2 and @xmath3 . the relations between moments ( [ mab2 ] ) can be written in a compact way using free cumulants which are certain ( specific ) combinations of moments @xcite : @xmath70 , @xmath71 , @xmath72 , etc . equations ( [ mab2 ] ) correspond to @xmath73 for @xmath64 . the concept of free cumulants is analogous to the standard cumulants in classical probability . let us recall that cumulants for a real random variable are generated as coefficients of the power series expansion of the logarithm of the characteristic function . the logarithm of the characteristic function for a sum of independent random variables is equal to the sum of logarithms of the characteristic functions of these random variables . therefore cumulants for the sum of independent variables are additive , exactly as free cumulants ( [ kab ] ) . free cumulants are , however , generated in a different way . the generating function for free cumulants is called r transform @xcite . free cumulants are equal to coefficients of the power series expansion @xmath74 note that there is a shift between the power of @xmath75 and the index of the corresponding cumulant . the r transform is related to the green s function which generates moments of the eigenvalue distribution . the relation reads @xmath76 we give it without a proof @xcite . one can find a diagrammatic interpretation of this equation @xcite using field theoretical methods for planar graphs enumeration @xcite . using equation ( [ gr ] ) one can find @xmath77 if @xmath78 is known . one can also invert equation ( [ gr ] ) @xmath79 this form is useful when one wants to determine the r transform for the given green s function . the r transform for the sum of free matrices @xmath5 is a sum of r transforms for @xmath2 and @xmath3 @xcite @xmath80 exactly as the logarithm of the characteristic function for the sum of independent real random variables in classical probability . the additivity of r transform reflects the invariance and independence of the measures for @xmath2 and @xmath3 in ( [ mab ] ) . for @xmath0 these two properties correspond to freeness . explicit relations between free cumulants @xmath81 and moments @xmath82 can be found by inserting the power series @xmath83 and @xmath84 into equation ( [ rg ] ) . the first three free cumulants are identical as the standard ones @xmath85 , @xmath86 , @xmath87 . from the fourth one on they are different . the fourth free cumulant is @xmath88 while the standard one is @xmath89 . the r transform was first introduced in free probability by voiculescu @xcite . later it was observed that free cumulants are related to a combinatorial problem of non - crossing lines on a plane @xcite . for this reason free cumulants are sometimes called non - crossing cumulants . for completeness we mention that equation ( [ rg ] ) can be deduced @xcite using field theoretical methods of planar graph enumeration @xcite . the green s function and the r transform correspond to sums over certain classes of diagrams , and equations like ( [ gr ] ) - to dyson - schwinger equations @xcite . the combinatorics for planar feynman diagrams for matrix models is equivalent to the combinatorial approach to freeness @xcite . we close this section by giving a standard example . recall that in classical probability the normal random variable with the probability density function @xmath90 has all cumulants equal zero except @xmath91 . what is the corresponding random matrix eigenvalue density in free probability that has all free cumulants equal zero except the second one @xmath91 ? we want to find the eigenvalue density corresponding to the r transform equal @xmath92 ( [ r ] ) . using ( [ gr ] ) we first find an equation for the green s function @xmath93 . it can be solved for @xmath77 : @xmath94 . we select the branch of the square root to match the asymptotic behavior @xmath95 for large @xmath75 . the eigenvalue density is ( [ rhog ] ) @xmath96 for @xmath97 $ ] and zero otherwise . it is the wigner semicircle law . odd moments of this distribution are equal zero while even moments @xmath98 are equal to the catalan numbers . one can use equation ( [ vrho ] ) to find that the corresponding potential is quadratic @xmath99 , so the invariant measure ( [ v ] ) is gaussian @xmath100 . let us slightly modify the r transform and consider @xmath101 . without repeating the calculations one can easily argue on general grounds that the corresponding eigenvalue density is shifted as compared to the previous case by the first cumulant @xmath102 and rescaled by the second one @xmath103 leading to @xmath104 and @xmath105 . let us now calculate the eigenvalue density of the sum of two independent large ( @xmath0 ) invariant matrices @xmath2 and @xmath3 that have eigenvalue densities given by the wigner semicircle law with different shift and scale parameters . denote these parameters by @xmath106 , @xmath107 for @xmath2 and @xmath108 , @xmath109 for @xmath3 , respectively . the r transforms are @xmath110 and @xmath111 . to find the eigenvalue density of the sum @xmath112 we first calculate r transform that is equal to @xmath113 with @xmath114 and @xmath115 . the r transform is exactly of the same form as for @xmath2 and @xmath3 . therefore the eigenvalue density for @xmath116 is also given by the wigner semicircle law . the situation is analogous to that in classical probability where the sum of two random gaussian variables is known to be a gaussian random variable . one says that the semicircle law is stable with respect to free addition in the same way as the gaussian law is stable with respect to addition in classical probability . actually there exist a bijection between stable laws in classical and free probability that allows one to classify all stable laws in free probability @xcite . just as a curiosity we mention that the only distribution that is stable with respect to addition ( random variables ) and free addition ( random matrices ) is the cauchy distribution . as we have seen the law of addition for independent large invariant matrices can be concisely expressed in terms of the r transform ( [ rr ] ) . before discussing the law of multiplication it is convenient to recall the corresponding law in classical probability . let @xmath23 and @xmath117 be independent non - negative real random variables . what is the distribution of the random variable @xmath118 ? a way to derive this distribution is via the mellin transform of the probability density function . the mellin transform @xmath119 for the product @xmath118 is the product of mellin transforms @xmath120 for @xmath23 and @xmath117 . clearly this multiplication law is analogous to the addition law discussed in the previous section ( [ rr ] ) . it remains to find the corresponding law for products of large invariant matrices . let @xmath2 and @xmath3 be independent invariant matrices . for the moment we also assume @xmath2 and @xmath3 to be positive semi - definite . later we will weaken this restriction . the first trivial observation is that even if @xmath2 and @xmath3 are hermitian , the product @xmath121 is generically non - hermitian , so its eigenvalues are complex . therefore , it is convenient to introduce a slightly modified version of the multiplication @xcite @xmath122 we use dot in the notation to distinguish @xmath123 from the standard matrix multiplication @xmath121 . square root of a positive semi - definite matrix is defined as @xmath124 where @xmath26 is a unitary matrix that diagonalizes @xmath2 : @xmath125 . the product @xmath123 of hermitian matrices is hermitian . the product @xmath123 of invariant random matrices is an invariant random matrix . moreover , the probability measure @xmath126 for the product @xmath123 is identical as for @xmath127 : @xmath128 so this product is in a sense commutative . one should also note that the trace moments for the product @xmath129 ( [ cprod ] ) are identical as for the standard product : @xmath130 . the law of free multiplication was formulated by voiculescu in free probability @xcite . later it was shown that it applies also to multiplication ( [ cprod ] ) of invariant random matrices in the limit @xmath0 @xcite . the multiplication law is expressed in terms of s transform @xmath131 where @xmath132 is a complex function . the s transform @xmath132 can be constructed from the moment generating function @xmath62 ( [ phi ] ) or actually the inverse function of @xmath133 that is usually denoted by @xmath134 : @xmath135 . the relation reads @xcite @xmath136 the s transform can be viewed as a series in @xmath75 @xmath137 that is obtained by inserting the inverse series in place of @xmath62 in equation ( [ s_chi ] ) . the coefficients of this series are related to the moments @xmath138 . the coefficients are singular for @xmath139 , in which case the s transform is ill - defined . the s transform is related to the r transform as follows @xcite @xmath140 if one knows the r transform then the last equation can be used to calculate the s transform . one can invert the last equation @xcite @xmath141 which allows one to calculate the r transform if the s transform is known . using these relations one can show that the multiplication law ( [ sab ] ) can be alternatively written in the r transform form @xcite : @xmath142 this set of equations looks more complicated than ( [ sab ] ) but in some situations it is more advantageous . for example , it can be used when the first moments vanish , that is when the s transforms are ill - defined . moreover , this form can be generalized to the case of non - hermitian matrices , as we discuss in section [ sec - nh ] . the two forms of the multiplication law are in a sense complementary . the original s transform form of ( [ sab ] ) has a very simple and appealing structure that reflects important properties of free products , for example that the multiplication @xmath123 ( [ cprod ] ) of invariant random matrices is commutative and associative : @xmath143 . this follows from the fact that multiplication on the right hand side of ( [ sab ] ) is just multiplication of complex numbers which is commutative and associative . let us give an example to illustrate how these formal relations work in practice . consider the product @xmath123 of two independent identical wishart matrices @xcite . the wishart matrix can be thought of as a square of a gaussian matrix from the gue ensemble @xcite . the matrices @xmath2 and @xmath3 have the eigenvalue densities @xmath144 @xmath145 for @xmath146 $ ] and @xmath147 for @xmath148 outside this interval . the green s function ( [ green ] ) corresponding to this density is @xmath149 and the moment - generating function ( [ phig ] ) @xmath150 the inverse function of @xmath151 fulfills the equation @xmath152 which can be solved for @xmath153 @xmath154 thus the s transform is ( [ s_chi ] ) @xmath155 we now want to find the s transform for the product @xmath156 from @xmath157 . using the multiplication law ( [ sab ] ) we obtain @xmath158 we now proceed to find the eigenvalue density @xmath159 that corresponds to the s transform given above . first we write the equation for the inverse function of the moment generating function @xmath160 using ( [ s_chi ] ) . we invert it for @xmath161 : @xmath162 it is a cubic equation for @xmath161 . replacing @xmath161 with the green s function ( [ phig ] ) we have @xmath163 this equation has three solutions given by the cardano formulas . we select the one which asymptotically behaves as @xmath164 for large @xmath75 . using ( [ rhog ] ) we find the density @xmath165 for @xmath166 $ ] and zero otherwise . we close this section with a few general comments . the multiplication law ( [ sab ] ) or ( [ rprod ] ) can be derived using solely perturbation expansion for invariant matrix models in the limit @xmath0 without referring to free probability . we refer the interested reader to @xcite . the second comment concerns the multiplication law for matrices which are not positive semi - definite . in the discussion above we restricted the class of matrices to positive semi - definite ones because we wanted to use @xmath167 in the definition of the product @xmath123 ( [ cprod ] ) . for ( invariant ) hermitian matrices the product @xmath129 is ( invariant ) hermitian . one can also consider the multiplication law for the standard product @xmath121 of hermitian matrices . in this case @xmath2 and @xmath3 do not have to be positive semi - definite . the matrix @xmath121 is non - hermitian and thus has complex eigenvalues . its trace moments @xmath168 are , however , real . since the s transform depends only on the moments , all calculations are identical as discussed before for @xmath129 except that the final result @xmath169 can not be interpreted as an eigenvalue density function , but merely as a function to calculate the trace moments of the matrix @xmath121 : @xmath170 the eigenvalue density has in this case generically a two - dimensional support on the complex plane . for example , the product of two independent hermitian matrices from the gue ensemble is a non - hermitian matrix . each gue matrix has real eigenvalues with the eigenvalue distribution given by the wigner semicircle law . the r transform @xmath171 ( we consider standardized gue with the mean zero and unit variance ) . using the r transform form of the multiplication law ( [ rprod ] ) one finds that @xmath172 and thus @xmath173 . this means that all the moments @xmath174 . it does not mean , however , that eigenvalue density of the product is zero . to calculate the eigenvalue density in this case one has to use methods that allow one to treat non - hermitian matrices . we discuss them in the subsequent sections . here we only mention that the eigenvalue density for the product is @xmath175 where @xmath75 is a complex variable . this result was first derived in @xcite with the help of field theoretical methods . note that the result is invariant under rotations around the origin of the complex plane . the extension from hermitian to non - hermitian random matrices is similar to the extension from real to complex random variables in classical probability . in this section we employ this similarity to define isotropic random matrices @xcite being probably the simplest extension of invariant random matrices to non - hermitian ones . consider a complex random variable with a probability measure that is invariant under rotation around the origin of the complex plane @xmath176 . the probability density is a function of the distance from the origin @xmath177 . such a random variable @xmath178 can be constructed as a product of a real non - negative random variable @xmath179 and a uniformly distributed random phase @xmath180 . when one knows the probability density function for the radial variable @xmath179 , one can easily derive the probability density for @xmath75 on the complex plane . isotropic random matrices are constructed in a way that imitates this construction . a random matrix @xmath181 is called isotropic if the probability measure @xmath182 is invariant under the left and right multiplication by unitary matrices @xmath183 : @xmath184 . in math and engineering community such random matrices are called bi - unitarily invariant ( bui ) and in general they can be rectangular . here we concentrate on square matrices and use the name isotropic in reference to them . the simplest way of constructing isotropic random matrices is to build them from real semi - positive diagonal random matrices @xmath2 and haar unitary matrices @xmath7 independent of @xmath2 ( free of @xmath2 for @xmath0 ) @xmath185 clearly , the diagonal matrix @xmath2 corresponds to the radial variable @xmath179 of @xmath186 and the haar unitary matrices @xmath187 and @xmath9 to the uniform phase @xmath151 . there are two of them to ensure the invariance under multiplication by a unitary matrix on both sides . alternatively one could consider a one - sided invariance , in which case the last equation would take the form @xmath188 . most of the results given below hold in either case . another way to construct such matrices is to define them by a probability measure of the form @xcite @xmath189 where @xmath190 is flat measure for @xmath4 complex matrices and @xmath191 is a polynomial or a power series in @xmath23 . there are many other ways of defining isotropic matrices . an interesting property of isotropic random matrices is that independently of the way they are defined , in the limit @xmath0 there is a one - to - one correspondence between the eigenvalue density of the matrix @xmath181 and the eigenvalue density of the hermitian matrix @xmath192 associated with @xmath181 @xcite . by construction the matrix @xmath192 is an invariant positive semi - definite random matrix . we denote the square root of this matrix by @xmath193 . note that the eigenvalues of @xmath2 correspond to singular values of @xmath181 . we can use the whole arsenal of methods discussed in the previous section to calculate the eigenvalue distribution of such matrices or their products . the correspondence between the eigenvalue distribution of a large isotropic matrix and the distribution of its singular values was found by feinberg and zee @xcite who applied field theoretical perturbative methods to solve the matrix model ( [ isov ] ) and independently by haagerup and larsen @xcite who used methods of free probability . below we quote the result in the form given by haagerup and larsen @xcite . in the terminology of free probability isotropic matrices in the limit @xmath0 correspond to variables that are called r - diagonal @xcite . here we prefer to call them large isotropic matrices . the invariance of the measure implies that the eigenvalue density of an isotropic matrix is invariant under rotation around the origin of the complex plane . for an isotropic matrix @xmath181 it is convenient to define radial cumulative density function as the probability that a random eigenvalue of the isotropic matrix lies on the complex plane within the distance @xmath23 from the origin : @xmath194 the eigenvalue density is related to the radial cumulative function as @xmath195 where @xmath196 is a possible point mass located at the origin . it corresponds to the probability of finding zero eigenvalues in the eigenvalue spectrum of @xmath181 . for @xmath197 the number of zero eigenvalues is a finite fraction of all eigenvalues . knowing @xmath198 one can calculate the density @xmath199 and vice versa . with the help of this function one can write an equation which relates the eigenvalue distributions of @xmath181 and @xmath193 @xcite @xmath200 for @xmath201 . this equation does not fix the value of the radial cumulative function for @xmath202 . it is fixed by an additional equation @xmath203 where @xmath204 is the cumulative eigenvalue density function for @xmath2 . this equation merely means that the number of zero eigenvalues of @xmath181 is equal to the number of zero eigenvalues of @xmath2 . the value @xmath205 is generically zero except the situation when there is a point measure ( dirac delta ) at zero in the eigenvalue density @xmath21 , as for instance in the distribution @xmath206 with @xmath207 , @xmath208 and the support being @xmath209 $ ] . this distribution naturally arises in the context of random matrices and it is called anti - wishart distribution in physics literature and free poissonian in math and engineering literature . the first term on the right hand side means that the fraction of zero eigenvalues of the corresponding random matrix @xmath2 is @xmath210 . the corresponding isotropic random matrix @xmath188 ` inherits ' all zero eigenvalues from @xmath2 and thus the corresponding fraction of zero eigenvalues of @xmath181 is also @xmath211 . equations ( [ hltheorem ] ) and ( [ hl2 ] ) provide an implicit relation between @xmath199 and @xmath21 that allows one to calculate one from another . a few comments are in order . equation ( [ hltheorem ] ) was first published in physical literature in a slightly different ( less transparent ) albeit equivalent form @xcite . in this paper it was also observed that for @xmath0 the support of the eigenvalue density of an isotropic random matrix has the shape of a single ring on the complex plane and that the radii of the ring can be derived from the eigenvalue density of @xmath2 . the external radius is equal to @xmath212 and the internal one to @xmath213 . the ring reduces to a disk when @xmath214 . in the paper @xcite the discussion was formalized and all statements were proven rigorously using concepts of free probability . in addition to that the s transform form of equation ( [ hltheorem ] ) and the discussion of the point measure for zero eigenvalues @xmath215 ( [ hl2 ] ) were given . from the discussion @xcite it follows that in general there are three different types of solutions for @xmath216 . the ring solution that corresponds to @xmath198 growing monotonically from @xmath217 to @xmath218 for @xmath219 $ ] and for @xmath220 , the disk solution that corresponds to @xmath198 growing monotonically from @xmath217 to @xmath218 for @xmath221 $ ] and the disk solution with a finite fraction of zero eigenvalues @xmath197 that corresponds to @xmath198 growing monotonically from @xmath222 to @xmath218 for @xmath221 $ ] . let us illustrate how the haagerup - larsen theorem ( [ hltheorem ] ) works in practice by giving an example . consider a random matrix @xmath181 from the ginibre ensemble @xcite . it is an @xmath223 random matrix whose elements are independent identically distributed , normally distributed , centered complex random variables . the probability measure for @xmath181 is @xmath224 where @xmath225 is flat measure for @xmath4 complex matrices . obviously @xmath181 is an isotropic random matrix . the limiting eigenvalue density for this matrix for @xmath0 is given by the uniform distribution on the unit disk @xcite @xmath226 it is zero outside the disk . the cumulative density @xmath227 is @xmath228 inserting it into equation ( [ hltheorem ] ) we obtain an equation for the s transform @xmath229 next we find ( [ s_chi ] ) @xmath230 , the moment generating function ( the inverse function for @xmath231 ) and the green s function ( [ phig ] ) @xmath232 finally we calculate the eigenvalue density of @xmath233 ( [ rhog ] ) @xmath234 \ , \ ] ] and correspondingly of @xmath2 @xmath235 \ .\ ] ] this is the density of singular values of the ginibre matrix . we see that it is given by the quarter - circle law . let us now discuss the product of @xmath236 independent isotropic matrices @xmath237 , @xmath238 , @xmath239 , @xmath240 @xmath241 where @xmath242 s are independent isotropic matrices @xcite . the goal is to calculate the eigenvalue density and the singular value density in the limit @xmath0 having the knowledge of eigenvalue densities for the matrices in the product . since the product itself is an isotropic matrix we can use equation ( [ hltheorem ] ) to write @xmath243 the s transform for @xmath244 can be calculated using the multiplication law ( [ sab ] ) @xmath245 since the right - hand side of this equation is a product of complex numbers , it does not depend on the order of multiplication . if instead of @xmath246 we considered a product of matrices permuted with a permutation @xmath247 @xmath248 we would obtain the s transform @xmath249 identical to @xmath250 because multiplication on the right hand side of ( [ paaa ] ) is commutative . therefore @xmath251 are equal for any permutation @xmath247 . in other words , the product of isotropic matrices is spectrally commutative in the limit @xmath0 . isotropic random matrices possess a striking property that has no counterpart in classical probability . consider the product @xmath252 of independent identically distributed isotropic random matrices . this can be realized by picking up @xmath236 independent matrices from the same ensemble and multiplying them . it turns out that in the limit @xmath0 such a product has exactly the same eigenvalue density as the @xmath236-th power @xmath253 of one matrix picked up at random from this ensemble @xcite . actually it was already observed in @xcite that singular values of @xmath254 and @xmath246 are identically distributed . this observation can be trivially extended to eigenvalues if one additionally assumes that @xmath255 ( [ hl2 ] ) . this is a sort of spectral self - averaging of multiplication in the limit @xmath0 because a single matrix from the ensemble is sufficiently representative in the @xmath236-fold product to give the same result as @xmath236 independent random realizations . the proof is short so we recall it here in the form given in @xcite where it was tacitly assumed that @xmath255 . all constituents needed for this proof can be found already in @xcite . we assume that @xmath256 ( [ hl2 ] ) . the s transform for the product of independent identically distributed matrices is the @xmath236-th power of the s transform for a single one @xmath257 as follows from equation ( [ paaa ] ) in which all @xmath258 are replaced by @xmath259 . inserting @xmath260 in place of @xmath250 in equation ( [ hlp ] ) and taking the @xmath236-th root of both sides we get @xmath261 comparing the last equation to equation ( [ hltheorem ] ) we see that @xmath262 . we can now argue that the radial cumulative density @xmath263 for the power @xmath254 is given by the same expression . indeed , using directly the definition of radial cumulative density we have @xmath264 for @xmath201 , and for @xmath265 for @xmath202 and hence @xmath266 . this completes the proof . if @xmath267 then the proof breaks down at @xmath202 . recall that @xmath222 is the probability of drawing at random a zero - eigenvalue from the spectrum of random matrix @xmath181 . the corresponding probability for the product of @xmath236 independent @xmath181 matrices is @xmath268 , while for the @xmath236-th power @xmath269 , so we see that @xmath270 for @xmath197 . in particular , one can use the product - power equivalence to calculate the eigenvalue density of the product @xmath271 of @xmath236 independent ginibre matrices in the limit @xmath0 . we know that @xmath272 for @xmath273 $ ] for one ginibre matrix ( [ fg ] ) thus for the product of @xmath236 @xmath274 \ . \label{ggg}\ ] ] the eigenvalue density for the product is ( [ rhof ] ) @xmath275 and zero outside the unit circle . this calculation reproduces in a very simple way the result that was originally obtained by different methods @xcite . this result can be generalized also to products of rectangular gaussian matrices @xcite or products @xcite of wigner matrices @xcite with non - gaussian entries . one can also find the distribution of singular values for the product of ginibre matrices @xcite . inserting radial cumulative density function ( [ ggg ] ) to the haagerup - larsen formula ( [ hltheorem ] ) one can find the green s function for @xmath276 @xcite @xmath277 and calculate the density function @xmath278 by using ( [ rhog ] ) . for @xmath279 it gives the wishart distribution ( [ wish ] ) ; for @xmath280 the distribution given in equation ( [ ww ] ) . one can find an explicit expression for any @xmath236 @xcite . the last step is to change variables @xmath281 to go from @xmath278 to @xmath282 . to summarize , large isotropic random matrices are exceptional in many respects . there exists a one - to - one correspondence between the densities of eigenvalues and singular values @xcite . multiplication of independent isotropic matrices is spectrally commutative and self - averaging in the large @xmath1 limit @xcite . one can also show that if the eigenvalue density for an isotropic matrix has a power - law singularity at zero @xmath283 with @xmath284 , then the density of singular values behaves at zero as @xmath285 @xcite . so far we have discussed invariant and isotropic random matrices . we are now going to discuss the more general case of products of non - hermitian matrices . the multiplication law for non - hermitian matrices was derived in reference @xcite by field theoretical methods . to build some intuition let us discuss gaussian non - hermitian random matrices . one can construct such matrices from independent identically distributed hermitian gaussian matrices from the gue ensemble . the probability measure for gue matrices is @xmath286 , where @xmath287 is flat measure for hermitian matrices . let us take two independent such matrices @xmath2 and @xmath3 . the following combination @xmath288 is a complex random matrix ; @xmath2 gives rise to hermitian sector of @xmath289 and @xmath290 to the anti - hermitian one . the resulting matrix @xmath289 is a ginibre random matrix @xcite with the uniform eigenvalue distribution on the unit disk ( [ ginibre ] ) . one can also define elliptic random matrices @xcite @xmath291 being linear combinations of hermitian and anti - hermitian random matrices with different coefficients controlled by a mixing parameter @xmath292 . for @xmath293 the random matrix @xmath289 reduces to the hermitian matrix @xmath2 , for @xmath294 to the ginibre matrix and for @xmath295 to the anti - hermitian matrix @xmath290 . the eigenvalue density of the matrix @xmath289 is constant on the support being an ellipse whose oblateness depends on @xmath292 . the probability measure for matrix @xmath289 can be calculated from the measure @xmath296 that factorizes since @xmath2 and @xmath3 are independent . changing variables to @xmath297 and @xmath298 one finds the probability measure for @xmath289 @xcite @xmath299 where @xmath300 . for this measure one can calculate two - point correlation functions @xmath301 . there are four such functions : @xmath302 that can be compactly written in the matrix form @xmath303 the matrix on the right hand side of the last equation defines the correlation structure of the model @xcite . it plays an important role , as we shall see below . the parameter @xmath304 takes values between @xmath305 and @xmath218 . for ginibre matrix it is @xmath217 , and for gue it is @xmath218 . without going into details we quote the law of non - hermitian matrix multiplication and refer the interested reader to @xcite . we present it in the r transform form ( [ rprod ] ) . the main difference as compared to the hermitian case ( [ rprod ] ) is that the r transform is not a complex function but a quaternionic one . we denote quaternions by calligraphic letters . the r transform maps quaternions @xmath306 onto quaternions @xmath307 . the multiplication law reads @xmath308^l \left[{\cal r}_b({\cal g}_a)\right]^r \ , \\ \left[{\cal g}_a\right]^r & = { \cal g}_{ab } \left[{\cal r}_a({\cal g}_b)\right]^l \ , \\ \left[{\cal g}_b\right]^l & = \left[{\cal r}_a({\cal g}_b)\right]^r { \cal g}_{ab } \ . \end{split } \label{rab}\ ] ] here @xmath309 , @xmath310 and @xmath311 correspond to generalized quaternionic green s functions for @xmath181 , @xmath312 and @xmath313 . they have the structure : @xmath314 where @xmath315 and @xmath316 are complex functions . we suppressed the indices of the corresponding matrix ( @xmath181 , @xmath312 or @xmath313 ) in the last equation . the density is related to the upper left element @xmath316 of the generalized quaternionic green s function @xmath317 the superscripts @xmath318 and @xmath319 refer to the left and right adjoint actions of the following unitary matrix @xmath320 on matrices in the square brackets @xmath321^l = uxu^\dagger \ , \quad [ x]^r = u^\dagger x u \ . \label{lr}\ ] ] the diagonal elements of @xmath26 are constructed from @xmath75 being the argument of the green s functions ( [ qgz ] ) . the generalized r transform is related to the green s function by an equation analogous to ( [ gr ] ) @xmath322 where @xmath323 this law looks complicated but in practice it reduces to a set of algebraic equations for @xmath316 from which one can calculate the eigenvalue density by taking the derivative with respect to @xmath324 ( [ elec ] ) . before we give an example let us mention an electrostatic analogy @xcite . in this analogy eigenvalues of a random matrix represent positions of charges in two - dimensions in the given potential , the function @xmath316 is interpreted as the electric field @xmath325 with components @xmath326 and @xmath327 and equation ( [ elec ] ) as the gauss law in two dimensions @xcite . it takes a more familiar form when written in the components of @xmath328 : @xmath329 we now illustrate how to use the multiplication law ( [ rab ] ) to calculate the eigenvalue density of the product . we consider as an example the multiplication of gaussian matrices constructed from the centered elliptic matrix @xmath289 ( [ mux ] ) by shifting and rescaling @xmath330 the scale parameter @xmath331 is a real number and the shift parameter @xmath332 is a complex number . the quaternionic r transform for @xmath181 is @xmath333 there are only two terms on the right - hand side because the matrix is gaussian and thus all higher order cumulants vanish ( [ r ] ) . the second term on the right - hand side is directly related to the matrix on the right hand side of equation ( [ tau ] ) that defines the two - point correlation structure of the model @xcite . the simplest example is @xmath334 , @xmath335 and @xmath336 ( [ aqs ] ) . it corresponds to the ginibre matrix . let us calculate the eigenvalue density for the product of two such matrices . the first equation in the multiplication law ( [ rab ] ) gives @xmath337 we used a shorthand notation @xmath338 . because @xmath181 and @xmath312 are identically distributed , we search a symmetric solution @xmath339 and @xmath340 . @xmath341 now we can use the two remaining equations ( [ rab ] ) . because of the symmetry it is enough to use only one of them , for example the first one . replacing @xmath342 by @xmath343 ( [ gr ] ) we can write this equation as @xmath344^r = \left[{\cal r}_a({\cal g}_b)\right]^l \ .\ ] ] we can now eliminate @xmath345 in the last equation using ( [ raa ] ) . writing also @xmath346^r$ ] and @xmath347^l$ ] in the explicit matrix form we have @xmath348 where @xmath338 , as before . it is a set of equations for @xmath349 and @xmath350 . it has a trivial solution @xmath351 and a non - trivial one @xmath352 and @xmath353 . the two solutions coincide at the unit circle @xmath354 . the trivial solution holds for @xmath355 and the nontrivial one for @xmath356 . note that for both of them @xmath352 . inserting these solutions to ( [ raa ] ) we obtain @xmath345 and then using ( [ gr ] ) we derive the final result for the generalized green s function of the product @xmath357 inside the unit circle @xmath358 , and @xmath359 outside . applying the gauss law ( [ elec ] ) to the upper left element of @xmath360 ( denoted by @xmath361 ) we find the density @xmath362 for @xmath358 , and @xmath363 otherwise . the calculations were done for @xmath364 . one , however , ( quite surprisingly ) obtains the same result for the product of matrices with arbitrary @xmath365 and @xmath366 @xcite . the reason for this is the following . looking at the r transform ( [ rqs ] ) we see that @xmath304 appears only in the combination @xmath367 ( or , writing it separately for @xmath181 and @xmath312 , in combinations @xmath368 and @xmath369 ) . the dependence on @xmath304 s disappears for @xmath370 . since in the solution discussed above we had @xmath370 , we conclude that this solution is independent of @xmath365 and @xmath366 and thus it holds for any @xmath365 and @xmath366 . in particular , the product @xmath313 of two hermitian gue matrices ( @xmath371 ) or any elliptic random matrices has exactly the same eigenvalue density ( [ rab ] ) , which is spherically symmetric . \(a ) where @xmath372 and @xmath373 are independent standardized ginibre matrices , each having the eigenvalue density uniformly distributed on the unit circle . the result is derived analytically @xcite . the horizontal axes correspond to the real and imaginary axes of the complex plane and the vertical one to the value of the eigenvalue density @xmath374 . the eigenvalue density has a support surrounded by a contour given by the pascal s limaon . ( b ) solid line represents the pascal s limaon . dots correspond to eigenvalues collected for ten matrices of size @xmath375 generated by monte - carlo simulations . a few eigenvalues lie outside the limiting contour but this effect is expected by finite size analysis . the number of outliers decreases when the matrix size increases . [ 1x1x ] , title="fig:",scaledwidth=35.0% ] ( b ) where @xmath372 and @xmath373 are independent standardized ginibre matrices , each having the eigenvalue density uniformly distributed on the unit circle . the result is derived analytically @xcite . the horizontal axes correspond to the real and imaginary axes of the complex plane and the vertical one to the value of the eigenvalue density @xmath374 . the eigenvalue density has a support surrounded by a contour given by the pascal s limaon . ( b ) solid line represents the pascal s limaon . dots correspond to eigenvalues collected for ten matrices of size @xmath375 generated by monte - carlo simulations . a few eigenvalues lie outside the limiting contour but this effect is expected by finite size analysis . the number of outliers decreases when the matrix size increases . [ 1x1x ] , title="fig:",scaledwidth=50.0% ] in a similar way one can find the eigenvalue density for the product of matrices with non - zero constants @xmath332 ( [ aqs ] ) but the calculations are tedious . in figure [ 1x1x].(a ) we show an analytic result for the eigenvalue density for @xmath376 where @xmath372 and @xmath373 are independent ginibre matrices @xcite . the eigenvalue support is surrounded by a contour @xmath377 that corresponds to the the pascal s limaon . in figure [ 1x1x].(b ) we compare this theoretical prediction to monte carlo simulations of large , finite matrices . the agreement is very good . as another example , in figure [ 1x1h ] we show a comparison between monte carlo results and the analytic prediction for the contour of the support for the eigenvalue density of the product @xmath378 , where @xmath379 is the gue hermitian matrix and @xmath289 is the ginibre matrix . again the comparison shows a good agreement . the details of the calculations and other examples can be found in @xcite . where @xmath289 is a standardized ginibre matrix and @xmath379 is an independent standardized hermitian gue matrix . the curves forming the contour are calculated analytically @xcite using the multiplication law ( [ rab ] ) . the extrapolation of these curves outside the support is shown in red dots . black dots correspond to eigenvalues collected for ten matrices of size @xmath375 generated by monte - carlo simulations . [ 1x1h ] , scaledwidth=50.0% ] we summarize this section shortly by emphasizing that there exists an algorithm to calculate eigenvalue densities for products of non - hermitian matrices . this algorithm works for large random matrices ( @xmath0 ) with probability measures of the type @xmath380 with a self - adjoint potential @xmath381 which is bounded from below @xcite . using free probability one can find an explicit formula for the moment generating function for the product of invariant random hermitian matrices in the limit @xmath0 . this formula allows to calculate trace moments @xmath382 for independent ( free ) random matrices @xmath383 and to determine the eigenvalue density of the matrix @xmath384 ( if @xmath2 is positive semi - definite ) from eigenvalue densities of @xmath2 and @xmath3 . we discussed how to extend the formalism to non - hermitian matrices to calculate the eigenvalue density of the product @xmath121 , which is generically a non - hermitian matrix . the extension has been worked out using field theoretical methods of planar graphs enumeration . the multiplication law ( [ rab ] ) is conveniently written in the r transform form ( [ rprod ] ) . the r transform is a quaternionic function . one can use this law to find eigenvalue densities for products of large non - hermitian random matrices . isotropic matrices are a special class of non - hermitian random matrices . these matrices have two exceptional features as far as matrix multiplication is concerned . isotropic random matrices are spectrally commutative and self - averaging in the limit @xmath0 . the first notion means that the eigenvalue distribution of the product of such matrices does not depend on the order of matrix multiplication . for example , the matrix @xmath385 has the same eigenvalue density as @xmath386 . the second one means that the product of @xmath236 independent isotropic identically distributed matrices has the same eigenvalue distribution as the @xmath236-th power of a single matrix . we finish by mentioning that recently a progress has been also made in the understanding of microscopic properties of eigenvalues statistics for products of random matrices . for example , an explicit form of the joint probability distribution for eigenvalues and singular values of the product of ginibre matrices has been explicitly found @xcite . using these results one can determine not only the eigenvalue density but also joint eigenvalue distribution and eigenvalue correlations for products of large random matrices . one can also infer finite size corrections to the limiting laws discussed in this paper . the author would like to thank romuald janik , andrzej jarosz , giacomo livan , maciej a. nowak , gabor papp , artur swiech and ismail zahed for many interesting discussions and a fruitful collaboration on the subject and jeremi ochab for critical reading of the manuscript . the author also wants to thank the organizers of the conference `` inference , computation , and spin glasses '' held in sapporo , july 28th30th 2013 , for a kind invitation to give a talk on this conference . the research presented in the paper was financed by grant no . dec-2011/02/a / st1/00119 of national centre of science in poland . the participation in the conference was supported by the grant : grant - in - aid for scientific research on innovative areas , mext , japan . 99 a. crisanti , g. paladin and a. vulpiani , _ products of random matrices , random matrices and their applications _ , ( berlin , heidelberg : springer - verlag , 1993 ) . newman , _ commun . phys . _ * 103 * , 121 ( 1986 ) . r. r. mueller , _ ieee trans . inf . theor . _ * 48 * , 2086 ( 2002 ) . janik and w. wieczorek , _ j. phys . a : math . gen . _ * 37 * , 6521 ( 2004 ) . e. gudowska - nowak , r.a . janik , j. jurkiewicz , m.a . nowak and w. wieczorek 2010 , _ chem . phys . _ * 375 * , 380 ( 2010 ) . r. lohmayer , h. neuberger and t. wettig , _ jhep _ * 0905 * , 107 ( 2009 ) . b. collins , i. nechita and k. zyczkowski , _ j. phys . a : math . theor . _ * 43 * , 275303 ( 2010 ) . bouchaud , l. laloux , m.a . miceli and m. potters m , _ epj b _ * 2 * , 201 ( 2007 ) . p. bougerol and j. lacroix j , _ products of random matrices with applications to schrdinger operators _ ( boston : birkjuser , 1985 ) . j. e. cohen , h. kesten and c.m . newman ( eds ) , _ random matrices and their applications , contemporary mathematics _ * 50 * , ( providence , ri : american mathematical society , 1986 ) . h. furstenberg and h. kesten , _ ann . math . statist . _ * 31 * , 457 ( 1960 ) . v. i. oseledec , moscow math . * 19 * , 197 ( 1968 ) . voiculescu , _ invent . math . _ * 104 * , 201 ( 1991 ) . r. speicher , _ math . ann . _ * 298 * , 611 ( 1994 ) . voiculescu , k.j . dykema and a. nica , _ free random variables , crm monograph series _ * 1 * , ( providence , ri : american mathematical society , 1992 ) . voiculescu , _ j. funct * 66 * , 323 ( 1986 ) . voiculescu , _ j. operator theory _ * 18 * , 223 ( 1987 ) . e. brezin , c. itzykson , g. parisi and j .- b . zuber , _ commun . phys . _ * 59 * , 35 ( 1978 ) . d. bessis , c. itzykson and j .- b . zuber , _ adv . _ * 1 * , 109 ( 1980 ) . g. t hooft , _ nucl . b _ * 72 * , 461 ( 1974 ) . z. burda , r.a . janik and m.a . nowak , _ phys . e _ * 84 * , 061125 ( 2011 ) . mehta , _ random matrices _ , ( amsterdam : elsevier 2004 ) . g. akemann , j. baik and p. di francesco ( eds ) _ the oxford handbook of random matrix theory _ , ( oxford : oxford university press , 2011 ) . f. j. dyson , _ phys . _ * 3 * , 140 ( 1962 ) . a. nica and r. speicher , _ lectures on the combinatorics of free probability , london mathematical society lecture note series _ * 335 * , ( cambridge : cambridge university press , 2006 ) . a. zee , _ nucl . b _ * 474 * , 726 ( 1996 ) . h. bercovici and v. pata , _ math . _ * 2 * , 791 ( 1995 ) . n. r. rao and r. speicher , _ elect . comm . in probab . _ * 12 * , 248 ( 2007 ) . j. wishart , _ biometrika _ 20a(1 - 2 ) , 32 ( 1928 ) . z. burda , r.a . janik and b. waclaw , _ phys . e _ * 81 * , 041132 ( 2010 ) . z. burda , m.a . nowak and a. swiech , _ phys . rev . e _ * 86 * , 061137 ( 2012 ) . j. feinberg and a. zee , _ nucl . b _ * 501 * , 643 ( 1997 ) . u. haagerup and f. larsen , _ * 176 * , 331 ( 2000 ) . a. nica and r. speicher , _ fields institute communications _ * 12 * , 149 ( 1997 ) . z. burda , g. livan and a. swiech , _ phys . rev . e _ * 88 * , 022107 ( 2013 ) . j. ginibre , _ j. math _ * 6 * , 440 ( 1965 ) . b.a . khoruzhenko and h .- j . sommers , _ non - hermitian random matrix ensembles _ , chapter 18 in reference @xcite ( preprint arxiv:0911.5645 ) . z. burda , a. jarosz , g. livan , m.a . nowak and a. swiech , _ phys . e _ * 82 * , 061114 ( 2010 ) . e. wigner e , _ ann . of math . _ * 62 * , 548 ( 1955 ) . n. alexeev , f. goetze and a. tikhomirov a , _ lith . j. _ * 50 * , 121 ( 2010 ) . s. orourke and a. soshnikov , _ electron . j. probab . _ * 16 * , 81 ( 2011 ) . k.a . penson and k. zyczkowski , _ phys e _ * 83 * , 061118 ( 2011 ) . girko , _ theor . _ * 29 * , 694 ( 1985 ) . fyodorov , b.a . khoruzhenko and h .- j . sommers , _ phys . lett . a _ * 226 * 46 ( 1997 ) . sommers , a. crisanti , h. somplinsky and y. stein , _ phys . _ * 60 * , 1895 ( 1988 ) . a. swiech , _ applications of generalized free addition and multiplication laws to large non - hermitian random matrices _ , ( msc thesis , jagiellonian university , cracow 2013 ) . g. akemann and z. burda , _ j. phys . a : math . * 45 * , 465201 ( 2012 ) . g. akemann , m. kieburg and l. wei , _ j. phys . a : math . theor . _ * 46 * , 275205 ( 2013 ) .

we have launched a search for large - scale ( kpc ) minor - axis outflows in edge - on seyfert galaxies in order to assess their frequency of occurrence and study their properties . here we present optical continuum and 2 line images and/or minor - axis long - slit spectra of 22 edge - on seyfert galaxies . six of these galaxies show at least one of the following : ( i ) bi - symmetric h halos extending along the minor axis , ( ii ) bright emission - line complexes at distances kpc ( in projection ) out of the disk , and ( iii ) double - peaked emission - line profiles from the gas along the minor - axis , suggesting that a wind - blown bubble is present . our results indicate that of seyferts have good evidence for minor - axis galactic outflows . kinetic luminosities of the galactic outflows in our sample seyferts are erg s , assuming all of the observed minor - axis emission is produced by the outflow . these values are , in general ,.1 as large as those for well - studied cases of superwinds in starburst galaxies ( heckman , armus & miley 1990 ) . however , far - infrared luminosities of our sample seyferts are also.1 as large . both starburst - driven superwinds and wide - angled outflows from the active galactic nucleus are possible explanations for the observed large - scale outflows . 2h+[nii ]

astro-ph9512169

ionized gas is known to extend ( up to @xmath81 kpc ) out of the plane in normal spiral galaxies ( e.g. dettmar 1992 ) . the origin and maintenance of this component of the interstellar medium in normal galaxies is not completely understood , but most theories support the notion that it is energized by outflows from the galaxy disk ( cf . rand 1995 ) . in powerful infrared ( ir ) galaxies , extraplanar ionized gas is much more luminous and is detectable much further ( up to tens of kpc , heckman , armus & miley 1990 ) out into the halo . the origin of these more extended structures is also explained by outflows from the galaxy . nuclear starbursts in these galaxies produce stellar winds and supernovae which rapidly heat the gas in the nuclear region . this high - pressure gas then expands rapidly and blows a wind out of the nuclear region , along the rotation axis of the disk , where the pressure gradient is lowest . this ` superwind ' interacts with clouds of dense gas in the galaxy and its halo , producing optical emission - line filaments ( see , e.g. , heckman , armus & miley 1990 ) . material swept up by the wind accumulates behind the shock front , forming a large shell ( or ` bubble ' ) along the minor axis . line emission which is observed from the shell should then have two velocity components , one from the front surface of the shell and one from the rear surface . such double - peaked emission - line profiles are indeed commonly found in minor axis spectra of superwind galaxies ( cf . heckman , lehnert & armus 1993 ) , corroborating that wind - blown bubbles are present . superwinds also manifest themselves at radio and x - ray wavelengths . the radio halos in edge - on starburst galaxies produce synchrotron emission from relativistic electrons in the wind . thermal x - rays are emitted by hot , expanding gas in the bubble and by low - density halo clouds which have been shocked by the wind . a well studied example of an extended halo is that in the edge - on starburst galaxy m82 ( radio : e.g. , seaquist & odegard 1991 ; x - ray : bregman 1994 ) . more examples are discussed in a recent review of observational and theoretical aspects of superwinds in starburst galaxies by heckman , lehnert & armus ( 1993 ) . approximately 1% of spiral galaxies have an active ( seyfert ) nucleus , which is commonly believed to be an accretion - powered supermassive black hole , i.e. , a scaled - down version of a quasar ( see , e.g. , terlevich 1992 for an alternate view ) . the emission from the active galactic nucleus ( agn ) dominates the luminosity at most wavelengths , so most studies of seyferts have focused on observing the properties of the agn itself and the environment it creates in and immediately around the nuclear region ( see , e.g. , antonucci 1993 ) . there is good evidence that nuclear outflows exist on pc scales in some seyferts ( cf . wilson 1993 ) , but it is not clear that these small - scale ( @xmath91 kpc ) outflows are connected with large - scale galactic outflows . observations of the large - scale ( @xmath01 kpc ) emission in seyferts do show that , in some cases , galactic - scale minor - axis outflows are present ( e.g. , hummel , van gorkom & kotanyi 1983 , hamilton & keel 1987 , corbin 1988 , wehrle & morris 1988 ) . in a sample of seyfert galaxies selected for extended optical or radio emission , baum ( 1993 ) found kpc - scale radio emission extending preferentially along the minor axes in seven out of ten cases and suggested that the emission comes from galactic winds blowing out of the galaxy disks . a particularly good example of a galactic outflow can be found in the edge - on seyfert galaxy ngc 5506 . diffuse radio emission is found out to @xmath3300 pc from the nucleus in the direction of the minor axis ( wehrle & morris 1987 ) and double - peaked emission - line profiles are found in minor axis spectra from regions @xmath3500 pc above and below the disk ( wilson , baldwin & ulvestad 1985 ) . the question of what powers galactic outflows in seyferts remains open , but viable sources of kinetic energy are nuclear starbursts and the active nucleus . why study galactic outflows in seyferts ? so far , only anecdotal evidence has been presented for them and they have not been systematically studied . one would like to know how galactic outflows in seyferts are different from those in more powerful agns ( e.g. radio galaxies , balqsos ) and also from those in starburst galaxies . from a broader standpoint , one would like to know what influence the outflows have on the agn and its host galaxy . galactic outflows drive large amounts of gas ( and metals ) out of the nuclear region , whereas the agn needs a constant supply of gas flowing inward . thus , galactic outflows in seyferts may regulate the power output and lifetime of the active nucleus . if the outflows are driven by starbursts , then what role , if any , do these nuclear starbursts play in the formation and evolution of the active nucleus ? many workers have proposed scenarios in which seyferts and starbursts are connected in an evolutionary sense ( e.g. scoville 1992 ) , but the issue remains open . the outflows may also have a strong influence on the physical state of the intergalactic medium . plasma which is blown out of the galaxy will increase the density of the intergalactic medium , heat it , and enhance its metallicity . in order to address these questions , we have begun a research program to systematically study a complete sample of edge - on seyfert galaxies . we shall search for observational evidence of galactic outflows , determine their physical properties and investigate possible energy sources for driving the outflows . in paper i , we present optical emission - line images and/or minor - axis spectra of the ionized gas in 22 edge - on seyfert galaxies . our future work will consist of analyzing the large - scale radio and x - ray emission from these outflows , using the same sample of edge - on seyferts . in section 2 , we introduce a statistical sample of seyfert galaxies which we shall use for continuing studies . observations and data reduction techniques are described in section 3 and images and spectra are presented for individual objects in section 4 . in section 5 , we discuss the frequency of occurrence of large - scale outflows in seyferts and discuss possible interpretations of energy sources driving the outflows . a hubble constant of 75 km s@xmath7 mpc@xmath7 is assumed throughout this paper . extended optical emission - line gas is difficult to observe in face - on disk galaxies , due to confusion with emission from the disk . we have therefore chosen to study only edge - on systems . we have defined a ` complete , ' distance - limited statistical sample of edge - on seyferts which we shall use in continuing studies . first , we selected seyferts ( liners were omitted ) from an electronic version of the seyfert catalog by huchra ( 1993 ) . this list was cross - referenced by position with an electronic version ( vers . 3.9 ) of the _ third reference catalog of bright galaxies _ ( de vaucouleurs 1991;rc3 ) . using recessional velocities ( @xmath10 ; 21 cm preferred ) and axial ratios ( @xmath11 ; major to minor ) from rc3 , we then selected objects by restricting @xmath12 5500 km s@xmath7 and @xmath13 0.4 ( @xmath14 2.5 ) . using the same method , we made another list starting from an electronic version of veron - cetty & veron s ( 1991 ) seyfert catalog . the two lists were then combined . a few additional seyferts which satisfied the restrictions listed above were selected by hand from hewitt & burbidge ( 1991 ) and from the literature . our complete sample consists of 22 objects . morphological types , positions , axial ratios , recessional velocities and seyfert types for each galaxy are listed in table 1 . we have obtained 2 images and/or long - slit spectra ( near h@xmath1 ) for 14 of these 22 objects . no bias was used to select these objects for observation . henceforth , we will refer to this sub - sample ( marked by a dagger [ ] in table 1 ) as our ` representative ' sample and shall use it for computing statistics in section 5 . we have also obtained images and/or spectra for eight additional edge - on seyferts which have @xmath15 5500 km s@xmath7 , @xmath16 0.4 , or both . morphological types , positions , axial ratios , recessional velocities and seyfert types for these galaxies are listed in table 2 . hereafter we shall refer to the sample of all objects observed as our ` extended ' sample . these 22 objects are the 14 objects marked by a dagger in table 1 plus the eight objects listed in table 2 . we shall use this extended sample for computing statistics as well ( section 5 ) . a log of new imaging and spectroscopic observations using the kpno 2.1 m telescope is given in table 3 . these data were supplemented by additional 2 images , taken at the 2.1 m telescope at kpno , the 1.5 m telescope at the cerro telolo inter - american observatory ( ctio ) , and the 2.2 m telescope at the european southern observatory ( eso ) . information about the observations for these supplemental images can be found in mulchaey , wilson & tsvetanov ( 1996 ) and tsvetanov , fosbury & tadhunter ( 1995 ) . reduction of the data was performed using the ccdproc analysis programs in iraf . the raw ccd images were processed in the normal fashion by subtracting bias frames and dividing by normalized images of the dome or blank sky . atmospheric extinction corrections were performed ( for all of the data ) using extinction curves appropriate for each observing site . during the imaging runs at the kpno 2.1 m ( august 1992 and february 1993 ) , two exposures of 1800 s were taken through narrow - band ( full width at half maximum [ fwhm ] @xmath3 75 ) filters centered near h@xmath1 . to produce an image of only the 2 line emission , an r - band image ( exposure 300 s ) was scaled and subtracted from the narrow - band image . observations of spectrophotometric standard stars were used to flux calibrate the 2 images . the limiting surface brightness levels of the 2 images are @xmath31 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . the long - slit spectra were taken in january and june 1992 at the kpno 2.1 m telescope . two exposures of 1800 s were taken with the slit oriented parallel to the galaxy s minor axis and passing through its nucleus . one exposure of 1800 s was taken along the major axis . frequent observations of a quartz lamp were used to remove the high spatial frequency structure . henear spectra were used for wavelength calibration and observations of spectrophotometric standard stars were used for absolute flux calibration . the width of the slit was 1.75 , except for ark 79 , ngc 931 , ngc 5506 and ic 1368 , for which it was 2.0 . limiting surface brightness levels at h@xmath1 are a factor of several better that those for the imaging data . the spectral coverage included h@xmath1 , [ n ii ] @xmath21 6548 , 6583 and [ s ii ] @xmath21 6716 , 6731 . continuum - free 2 images from mulchaey , wilson & tsvetanov ( 1996 ) and tsvetanov , fosbury & tadhunter ( 1995 ) were made by combining narrow - band ( fwhm @xmath350@xmath5100 ) images centered on h@xmath1 with narrow - band images of the nearby continuum . the images were flux calibrated from observations of spectrophotometric standard stars . exposure times were typically @xmath31000@xmath52000 s and the limiting surface brightness levels were @xmath32 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 ( except for the 2 image of ngc 5506 , for which the exposure time was much shorter and the limiting surface brightness is a factor of @xmath310 higher ) . descriptions of the observational data are given in table 4 . for each of the 22 objects observed , we list the type of data taken and the corresponding figure number for the images and/or spectra . we also list the total 2 luminosity from the galaxy for all objects for which we have 2 images , and the total 2 luminosity from the minor - axis emission - line gas for galaxies in which emission - line regions ( elrs ) are evident along the minor axis . the total 2 luminosity from the galaxy was calculated by summing the flux inside rectangular regions enclosing all of the emission - line nebulae from the galaxy . the 2 luminosities of the minor - axis elrs were calculated by summing flux inside rectangular regions around the individual elrs mentioned in sections 4.1 and 4.2 . in the following subsections , we present and discuss our new images and spectra and also note whether other data which has been published is suitable for searching for evidence for minor - axis outflows . seyferts which are good candidates for having such outflows are identified and statistical results for our samples are discussed in section 5 . for each of the galaxies in our complete sample , we searched the literature for published data which could be used to look for signatures of minor - axis outflows . for many of the objects ( ic 1657 , um 319 , mrk 993 , mrk 577 , ark 79 , ngc 1320 , mcg @xmath52 - 27 - 9 , ngc 4602 , eso 103-g35 , ngc 6810 , ic 1417 , ngc 7410 and ngc 7590 ) , the published data was not suitable exceptions are noted in the following subsections . our images for mrk 993 are shown in figure 1a . many elrs are scattered throughout the galactic disk , but no minor - axis emission is evident in our 2 image . we did not find double - peaked line profiles or any evidence for elrs extending out of the disk in our minor axis spectra . no extraplanar emission is noticeable in our 2 image ( figure 1b ) . the spectrum from the position 3.1 ( @xmath31 kpc ) south ( along the minor axis ) of the nucleus may have double - peaked h@xmath1 , [ nii ] and [ s ii ] emission line profiles ( consistent with what is expected from the shell of an wind - blown bubble ) , but the possible substructure in the profiles is nearly the same level of magnitude as the noise . for this reason , we have classified these features as only suggestive of a minor axis outflow . spectra with higher signal - to - noise ( s / n ) ratios would be useful for determining if these components are double peaked emission lines from a wind - blown bubble . our r - band image ( figure 1c ) shows the presence of a companion galaxy @xmath320 to the north along the minor axis . apart from emission from this companion galaxy , we find no extended 2 emission in our long - slit spectra along the minor axis . fabry - perot studies ( e.g. amram 1992 ) show that the kinematics of the emission - line gas are quite complex , but there is no evidence for an outflow from the nucleus . no minor axis emission is noticeable in our 2 image ( figure 1d ) . this galaxy forms a pair with ngc 1321 , which is positioned @xmath31.7 to the north . ngc 1386 is located in the fornax cluster , so we have used the distance to that cluster ( 20 mpc ) in our calculations . in the 2 image ( figure 1e ) , there is an elr with luminosity 3.1 @xmath17 10@xmath22 erg s@xmath7 to the northwest , at a distance of 41 ( 4.0 kpc ) from the nucleus . the continuum emission from the disk extends out to this position , so the elr could lie in the the plane of the galaxy disk . alternatively , the emission could be from gas in a halo cloud which has been ionized by a wind . weaver , wilson & baldwin ( 1991 ) have used long - slit spectroscopy to study the kinematics of the extranuclear gas within @xmath312(@xmath31 kpc ) . the line profiles and line ratios of the elrs surrounding the nucleus suggest that an outflow from the nucleus may be occurring along the _ major _ axis . ulvestad & wilson ( 1984 ) found that the morphology of the nuclear radio continuum emission is slightly extended @xmath3 400 pc toward the southwest along position angle ( p.a . ) although there is good evidence for a nuclear outflow in ngc 1386 , it does not appear to be directed along the minor axis . except for the northwest elr , all of the line emission originates from @xmath91 kpc . more solid evidence is needed to conclude that a galactic scale outflow ( i.e. one blowing out of the galactic disk ) is present in this galaxy . a dust lane separates the line emission which immediately surrounds the nucleus ( see figure 1f ) from the bright ( l@xmath23 } = $ ] 3.6 @xmath17 10@xmath4 erg s@xmath7 ) elr which extends from @xmath35@xmath515 ( [email protected] kpc ) to the northwest , in the direction of the minor axis . a number of ionized filaments of luminosity @xmath310@xmath22 erg s@xmath7 are also present beyond the northwest elr , at distances @xmath2430 ( 4.5 kpc ) . such bright filamentary structures are commonly found in luminous ir galaxies with starburst - driven superwinds ( e.g. heckman , armus & miley 1990 ) . tsvetanov , dopita & allen ( 1995 ) find evidence for outflowing gas with velocities @xmath3200 km s@xmath7 on both sides of the galaxy disk . radio continuum maps of the nuclear region of ngc 2992 show emission extending out to 2 ( 0.3 kpc ) along the minor axis which has the structure of `` striking pair of loops '' ( wehrle & morris 1988 ) . these authors interpret these loops as limb - brightened bubbles or magnetic arches . the bright minor - axis optical elrs and ` figure-8 ' radio morphology suggest that a powerful galactic outflow is occurring in ngc 2992 . our 2 image ( figure 1 g ) shows emission from the disk but no extended emission along the minor axis . no extended minor - axis emission is apparent in our 2 image ( figure 1h ) of ngc 4235 . elrs are noticeable extending from the nucleus out along the _ major _ axis , especially to the northeast ( see also pogge 1989 ) . we did not find double - peaked line profiles or any evidence for elrs extending out of the disk in our minor axis spectra . radio continuum maps of the nuclear region ( ulvestad & wilson 1989 ) are unresolved or very slightly resolved . no extended emission was noticed by hummel , beck & dettmar ( 1991 ) in their large - scale radio map . we did not obtain any new images or spectra of ngc 4388 since there is already good evidence for a galactic outflow in this galaxy . corbin , baldwin & wilson ( 1988 ) and pogge ( 1988 ) have shown that conical structures of ionized gas extend outward from the nucleus , both above and below the disk . these authors argue that some of the line - emitting gas may have been ejected from the nucleus . radio maps of this galaxy ( hummel 1983 ; stone , wilson & ward 1988 ) reveal diffuse emission extending outward from the nucleus , perpendicular to the galactic disk . much of the emission in our 2 image ( figure 1i ) comes from a ring structure in the disk ( buta & de vaucouleurs 1983 ) , but no emission is evident along the minor axis . this seyfert galaxy also houses a strong nuclear starburst . the presence of a superwind in this this galaxy has been discussed in detail by heckman , armus & miley ( 1990 ) , who argue that the wind is driven by the starburst . harnett ( 1989 ) present images of a radio halo in this galaxy , presumably produced by the wind . a bi - symmetric halo of emission - line gas can be seen in our 2 image ( figure 1j ) extending along the minor axis , @xmath310 ( 3 kpc ) on both sides of the nucleus . the h@xmath1+[nii ] luminosity of the extended emission on either side is @xmath32.5 @xmath17 10@xmath25 erg s@xmath7 . unger ( 1987 ) show that radio emission extends @xmath36 west from the nucleus and suggest that it is from material lying out of the plane . the h@xmath1 halo we observe here is probably produced by an outflow from the nuclear region and we consider ic 4329a a good candidate for a large - scale galactic outflow . emission - line nebulae extend along the minor axis , @xmath35(0.6 kpc ) north and south of the nuclear region ( see figure 1k ) . wilson , baldwin & ulvestad ( 1985 ) first noted double - peaked [ o iii ] @xmath265007 and h@xmath27 emission lines in spectra of the elrs @xmath05 north and south of the nuclear region . our spectra from these positions ( see figure 1k ) also show double - peaked h@xmath1 , [ nii ] and [ s ii ] emission lines . a ` loop ' of radio continuum emission extends to the north from the nuclear region ( wehrle & morris 1987 ) . these features imply the presence of a minor - axis wind which is blowing a shell of material northward ( and perhaps southward ) from the nucleus . thus , there is very good evidence for a minor - axis outflow in ngc 5506 . our 2 image of ic 1368 is shown in figure 1l . note that the nuclear source is elongated in the direction of the minor axis and extends @xmath35 ( 1.3 kpc ) on each side of the nucleus . such extended emission - line halos are commonly found in edge - on starburst galaxies with winds , so we have classified ic 1368 as a good candidate for a large - scale outflow . our 2 image of his galaxy shows emission from the disk , but not from regions extending along the minor axis ( figure 1 m ) . no extended emission is present in our 2 image ( figure 1n ) . much of the emission comes from clumpy h ii regions in the disk . for the following objects , there were no published data that is suitable for searching for a large - scale outflow : ugc 3255 , eso 362-g8 , ngc 4117 , ic 5169 and mrk 915 . this galaxy has an elliptical r - band morphology ( see figure 2a ) . a bright elr of luminosity @xmath28 } = $ ] 1 @xmath17 10@xmath25 erg s@xmath7 is noticeable 19.7 ( 7.4 kpc ) from the nucleus in p.a . the location of this elr is not positioned exactly on the minor axis , but the gas is located @xmath34 kpc ( in projection ) out of the disk . this suggests that it has been ejected from the disk . we have therefore classified ngc 513 as a good candidate for a large - scale outflow . we did not obtain images of ugc 3255 . the position of the slit for our minor - axis spectra is shown in a greyscale plot of a b - band image ( digitized survey plate ) in figure 2b . we did not find double - peaked line profiles or any evidence for elrs extending out of the disk in our minor axis spectra . conical elrs can be seen in the 2 image of this galaxy in figure 2c . the northeast cone extents from the nucleus out to @xmath314 ( 4.3 kpc ) along p.a . 65(roughly perpendicular to the major axis ) and has an 2 luminosity of 1.8 @xmath17 10@xmath4 erg s@xmath7 . mulchaey , wilson & tsvetanov ( 1996 ) find the gas in the conical elr to have high [ o iii]5007/h@xmath1 ratios , consistent with photoionization by the agn . the presence of gas so far out of the disk suggests that a large - scale outflow is occurring in eso 362-g8 . the presence of extranuclear elrs in mrk 10 has been previously noted by schulz ( 1982 ) , who found line emission extending westward from the nucleus . we also find extranuclear line emission to be present from our long - slit data . the orientation of the slit ( p.a . 40 ) and a graph of the relative h@xmath1 flux for positions along the slit is shown in figure 2d . emission - line gas is present along the minor axis out to 20(11.3 kpc ) to the northeast from the nuclear region , and out to 10(5.6 kpc ) to the southwest . such emission could be produced by a galactic outflow blowing along the minor axis . however , due to the limited spatial coverage of our long - slit spectra , we classify this evidence as only suggestive . a more complete study of the extranuclear optical emission - line gas in mrk 10 is clearly warranted . our 2 image ( figure 2e ) of this galaxy shows emission from the inner disk but no extended emission along the minor axis . we did not find double - peaked line profiles or any evidence for elrs extending out of the disk in our minor axis spectra . a patchy ring of h ii regions shows up well in our 2 image ( figure 2f ) . the field of view of our ccd is small ( 5 ) compared to the large optical size of the galaxy ( @xmath310 @xmath17 5 ) , so our 2 image is not very useful for looking for emission from extraplanar gas . the morphology of the 2 emission within the stellar ( r - band ) envelope is not elongated along the minor axis . we did not find double - peaked line profiles or any evidence for elrs extending out of the disk in our minor axis spectra . this galaxy has been very well studied , but we did not find any evidence in the literature for a minor axis outflow . the 2 image ( figure 2 g ) shows elrs extending along the minor axis , @xmath36.5 ( 1.3 kpc ) to the northwest and southeast of the nucleus , perpendicular to the bar . the total 2 luminosity from the northwest elrs is @xmath310@xmath25 erg s@xmath7 . the small , bright elr south of the nucleus has a h@xmath1+[nii ] luminosity of @xmath35 @xmath17 10@xmath22 erg s@xmath7 and the remaining southwest elrs emit @xmath310@xmath25 erg s@xmath7 . such emission could be produced by an outflow along the minor axis . however , the projected locations of the the elrs are inside the stellar ( r - band ) envelope , so the elrs may be located in the inner disk of ic 5169 . for this reason , we classify this evidence as only suggestive of a minor - axis outflow . no emission is evident extending along the minor axis in our 2 image ( figure 2h ) . the results from our observational data are listed symbolically in the last column of table 4 . for most of the objects we classified as good candidates for minor - axis outflows , either double - peaked emission lines are found from regions along the minor axis , or the morphology of the 2 emission resembles that of a halo extending above and below the galaxy disk . in ngc 513 and eso 362-g8 , the nebulae have conical morphologies , similar to ` ionization cones ' observed in some seyfert galaxies ( e.g. pogge 1989 , wilson & tsvetanov 1994 ) . the line emission from the extranuclear gas in these objects could be produced by ionizing radiation from the agn or by a shock from an outflowing jet . however , in both cases , the elrs are found @xmath34 kpc ( in projection ) out of the disk , which is , in general , further out than the typical maximum extent of ionization cones in seyferts ( @xmath32 kpc , wilson & tsvetanov 1994 ) . an obvious possible explanation for how the gas got so far out of the disk is that it was blown out by a minor - axis outflow . thus we consider the six galaxies ngc 2992 , ic 4329a , ngc 5506 , ic 1368 , ngc 513 and eso 362-g8 to be good candidates for having large - scale minor axis outflows . our images and spectra of ark 79 , mrk 10 and ic 5169 are suggestive of a minor axis wind , but are not entirely convincing in and of themselves . therefore , we have not included them as ` good candidates ' when calculating statistics for our samples . further optical studies of the minor axis nebulae in these galaxies would be useful for determining if galactic outflows are present . if we ignore the results in the literature and only consider results from the images and spectra we obtained , we find that for the 14 objects in our representative sample , four ( 29% ) show good evidence for minor axis outflows . including the objects from table 2 , six ( 27% ) of the 22 objects in our extended sample show good evidence for minor axis outflows . as mentioned in section 4.1 , there is evidence in the literature for minor - axis outflows in two additional seyferts in our complete sample ( ngc 4388 and ngc 4945 ) . all objects which show good evidence for minor axis outflows ( from our images and spectra or from the literature ) are marked with an asterisk in tables 1 and 2 . if we include the results from the literature , we find good evidence for minor axis outflows in four ( 29% ) of the 14 objects in our representative sample , six ( 27% ) of the 22 objects in our complete sample , and six ( 27% ) of the 22 objects in our extended sample . the results are consistent : @xmath325@xmath530% of edge - on seyferts show good evidence for large - scale galactic outflows . these results underestimate the `` true '' fraction of seyferts which have minor axis outflows . for many of the objects , data suitable for detecting emission from these outflows ( e.g. h@xmath1 images , long - slit spectra of the minor - axis elrs , deep radio - continuum images , deep x - ray images ) are not available . in addition , our images are only sensitive to surface brightnesses @xmath010@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 and we have minor axis spectra to search for double - peaked emission lines for only nine of 22 objects observed . detailed kinematic studies of extended emission - line regions in seyferts are very useful for determining if an outflow is present , but such studies have been published for very few objects in our samples . assuming our edge - on samples are unbiased subsamples of seyfert galaxies , our results suggest that large - scale galactic outflows are likely to be present in @xmath0@xmath2 of _ all _ seyfert galaxies . in table 5 , we list all seyfert galaxies known ( by us ) to have good evidence for galactic outflows . evidence comes in the form of optical emission - line nebulae and split emission lines along the minor axis , extended radio continuum emission along the minor axis , and extended x - ray halos along the minor axis . most of the outflows which have been identified are in seyferts with edge - on disks . outflows in only a few of these galaxies ( e.g. ngc 3079 , mrk 231 and ngc 4945 ) have been studied in detail . the question of what powers large - scale outflows in seyferts remains open even for the most well - studied cases . for example , in a very complete kinematic fabry - perot study of the two bi - symmetric wind - blown superbubbles in the nucleus of ngc 3079 , veilleux ( 1994 ) found that either a starburst- or an agn - driven wind was consistent . in the following subsections , we discuss two scenarios for powering the large - scale outflows in seyferts : starburst - driven superwinds and outflows from the active nucleus . superwinds in starburst galaxies have been fairly well studied ( heckman , armus & miley 1990 ; lehnert & heckman 1995 ) , so it is natural to compare properties of large - scale outflows in seyferts with those of superwinds in starburst galaxies . the superwind in the archetypical edge - on starburst galaxy m82 has an 2 luminosity of @xmath310@xmath29 erg s@xmath7 and a kinetic luminosity of 10@xmath30 erg s@xmath7 ( heckman , armus & miley 1990 ) . we can estimate kinetic luminosities for the seyfert outflows by scaling l@xmath31}$ ] for our sample seyferts ( table 4 ) by the ratio of the kinetic luminosity of m82 to the 2 luminosity of its minor - axis elrs . using this method , we find kinetic luminosities @xmath310@xmath32@xmath510@xmath33 erg s@xmath7 ( logarithmic mean 10@xmath34 erg s@xmath7 ) for the seyfert outflows , assuming all of the line emission is produced by the outflow . the implied kinetic luminosities of our seyfert outflows are , in general , a factor @xmath30.1 as large as those of superwinds from heckman , armus & miley ( 1990 ; several @xmath17 10@xmath6 erg s@xmath7 ) . however , the mean far - ir luminosity of our sample ( @xmath35 43.4 @xmath36[erg s@xmath7 ] ) is smaller than that of heckman , armus & miley s sample ( @xmath35 44.9 @xmath36[erg s@xmath7 ] ) by about the same factor , so if the seyfert outflows are powered by nuclear starbursts , smaller kinetic luminosities are to be expected . high rates of massive star formation have been inferred for the nuclear regions of several seyfert galaxies . for example , in ngc 1068 , approximately half of the ir luminosity comes from the agn and the other half comes from the starburst ( balick & heckman 1985 ) . the energy input from a possible starburst is directly proportional to the starburst component of l@xmath38 ( cf . heckman , lehnert & armus 1993 ) . however , for most seyferts , separating the starburst and agn components of l@xmath38 is not straightforward ( cf . telesco 1988 ; however , see also rodriguez - espinosa , rudy & jones 1987 , who claim that star formation produces the bulk of the far - ir emission in seyferts ) . measuring massive - star formation rates from stellar absorption lines is in general , quite difficult to do in seyferts ( cf . diaz 1992 ) . thus , it is not known how common circumnuclear starbursts are in seyferts . baum ( 1993 ) could not distinguish between starburst- or agn - driven winds using their radio data . however , they found that the luminosity of the extranuclear radio emission is comparable to that in starburst galaxies and follows the same radio - ir relation as that of starburst galaxies , suggesting that the large - scale radio emission may be of starburst origin . in order to verify that these galactic outflows are starburst - driven superwinds , one must find evidence for massive stars in the seyfert nuclei and determine if the putative starburst is powerful enough to drive a galactic wind . the agn can certainly provide enough energy to drive a galactic outflow . however , linear nuclear radio sources ( suggestive of a collimated nuclear outflow ) in seyferts are not , in general , oriented along the same position angle as the large - scale diffuse radio emission ( baum 1993 ) . therefore , it is unlikely that the large - scale minor - axis radio emission is from lobes at the end of collimated radio jets , as in powerful radio galaxies ( unless the jet outflow axis precesses ) . on the other hand , if nuclear outflows from the agn are weak , they may lose energy to the interstellar medium in the nuclear region and become poorly - collimated . if the gas continues to flow outward , it may be be diverted toward the minor axis ( where the pressure gradient is lowest ) and continue on as a wide - angled outflow . many different models have been presented for nuclear outflows from agns . theoretical models have been proposed for hydro - magnetic jets which originate from the accretion disk surrounding the black hole , or even from the black hole itself ( cf . blandford 1993 ) . another model ( krolik & begelman 1986 ; balsara & krolik 1993 ) proposes that radiation - driven winds originate at the inner edge of the proposed molecular torus . thus , estimating the amount of kinetic energy being channelled from the agn into the outflow is highly model - dependent . large - scale galactic outflows are quite common in seyfert galaxies . our results show that they are present in @xmath0@xmath2 of all seyfert galaxies . current observational studies of these outflows do not offer enough information to conclude what drives the outflows . they may be powered by circumnuclear starbursts , agn , or perhaps by both , in combination . in order to constrain the different models for the outflows , more observational work is needed to measure physical properties of these winds ( e.g. , outflow velocities , morphologies and luminosities of radio and x - ray halos , velocity and ionization structures of the outflowing gas near the nuclear region ) . thanks sylvain veilleux and the referee , r .- j . dettmar , for providing helpful comments on the manuscript . e.j.m c. acknowledges funding from the director s office of stsci . this research has made extensive use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , caltech , under contract with nasa . the work of m.d.l . at igpp / llnl was performed under the auspices of the u.s . department of energy under contract w-7405-eng-48 . rllllllrcll & ic 1657 & & ( r@xmath39)sb(s)bc & 01 14 07.3 & @xmath532 39 02 & 0.63 & 3552 & 2 & 1 & + & um 319 & mcg 0 - 4 - 112 & sb ? & 01 23 20.9 & @xmath501 58 36 & 0.43 & 4730 & 2 & 1,2,3 & + & mrk 993 & ugc 987 & sa & 01 25 31.5 & @xmath4032 08 10 & 0.52 & 4658 & 2 & 1,3 & + & mrk 577 & & s0/a & 01 49 30.1 & @xmath4012 30 32 & 0.41 & 5179 & 2 & 3 & + & ark 79 & ugc 1757 & s ? & 02 17 23.1 & @xmath4038 24 50 & 0.53 & 5157 & 2 & 2,3 & + & ngc 931 & mrk 1040 & sbc & 02 28 15.0 & @xmath4031 18 46 & 0.67 & 4993 & 1 & 1,2,3 & + & ngc 1320 & mrk 607 & sa : sp & 03 24 49.0 & @xmath503 02 31 & 0.47 & 2700 & 2 & 2,3 & + & ngc 1386 & & sb(s)0 + & 03 36 46.4 & @xmath535 59 58 & 0.42 & 864 & 2 & 1,2,3 & + * & ngc 2992 & & sa pec & 09 45 42.1 & @xmath514 19 39 & 0.51 & 2314 & 2 & 1,3 & + & mcg @xmath52 - 27 - 9 & & sb(rs)0 + pec ? & 10 35 27.4 & @xmath514 07 49 & 0.60 & 4650 & 2 & 3 & + & ngc 4235 & ic 3098 & sa(s)a & 12 17 08.9 & @xmath4007 11 31 & 0.65 & 2410 & 1 & 1,2,3 & + * & ngc 4388 & & sa(s)b : sp & 12 25 47.0 & @xmath4012 39 42 & 0.64 & 2517 & 2 & 1 & + & ngc 4602 & & sab(rs)bc & 12 40 37.1 & @xmath505 07 58 & 0.46 & 2548 & 1.9 & 3 & + * & ngc 4945 & & sb(s)cd : sp & 13 05 26.2 & @xmath549 28 15 & 0.72 & 560 & 2 & 3 & + * & ic 4329a & & sa0 + : sp & 13 49 19.4 & @xmath530 18 35 & 0.56 & 4793 & 1 & 1,2,3 & + * & ngc 5506 & & sa pec sp & 14 13 14.9 & @xmath503 12 27 & 0.52 & 1815 & 2 & 1,2,3 & + & eso 103-g35 & & sa0 & 18 38 20.3 & @xmath565 25 42 & 0.44 & 3983 & 2 & 1,3 & + & ngc 6810 & & sa & 19 43 34.2 & @xmath558 39 21 & 0.55 & 1958 & 2 & 3 & + * & ic 1368 & & sa ? sp & 21 14 12.1 & @xmath4002 10 38 & 0.44 & 3912 & 2 & 3 & + & ic 1417 & & sb ? sp & 22 00 21.6 & @xmath513 08 48 & 0.58 & 4309 & 2 & 1,4 & + & ngc 7410 & & sb(s)a & 22 55 00.7 & @xmath539 39 41 & 0.51 & 1751 & 2 & 2,3 & + & ngc 7590 & & s(r?)bc & 23 18 55.0 & @xmath542 14 17 & 0.42 & 1596 & 2 & 2,3 & + rllllllrcl & ngc 513 & ark 41 & s ? & 01 24 27.0 & @xmath4033 47 57 & 0.31 & 5949 & 2 & 1,2,3 + & ugc 3255 & & sbb ? & 05 09 48.0 & @xmath4007 29 00 & 0.63 & 5689 & 2 & 1,2,3 + * & eso 362-g8 & & s0 ? & 05 11 09.1 & @xmath534 23 36 & 0.34 & 4785 & 2 & 2,3 + & mrk 10 & & sbbc & 07 47 29.1 & @xmath4060 55 59 & 0.38 & 8753 & 1 & 1,2,3 + & ngc 4117 & & s0 : & 12 07 46.2 & @xmath4043 07 34 & 0.31 & 871 & 2 & 1,2 + & ngc 5033 & & sa(s)c & 13 13 28.0 & @xmath4036 35 38 & 0.33 & 878 & 1 & 1,2 + & ic 5169 & & ( r@xmath41)sab(r)0 + & 22 10 09.7 & @xmath536 05 20 & 0.37 & 3016 & 2 & 1,4 + & mrk 915 & mcg @xmath52 - 57 - 23 & s ? & 22 36 46.3 & @xmath512 32 40 & 0.45 & 7248 & 1 & 1,2 + llccc jan 92 & long - slit spectroscopy & goldcam spectrometer & 0.78 @xmath17 1.24 & 6200 @xmath5 7200 + & & ti 800 @xmath17 800 ccd & & + jun 92 & long - slit spectroscopy & goldcam spectrometer & 0.78 @xmath17 1.24 & 4800 @xmath5 7100 + & & ford 3k @xmath17 1k ccd & & + aug 92 & direct imaging & tek 1k @xmath17 1k ccd & 0.3 & r , h@xmath1 + feb 93 & direct imaging & tek 1k @xmath17 1k ccd & 0.3 & r , h@xmath1 + lcccccllll mrk 993 & r , h@xmath1+[nii ] & 1 & maj ( 32 ) , min ( 122 ) & 2 & 1a & 40.87& ... &43.08 & + ark 79 & r , h@xmath1+[nii ] & 1 & maj ( 87 ) , min ( 177 ) & 2 & 1b & 40.95 & ... & ... & s ? + ngc 931 & r & 1 & maj ( 73 ) , min ( 163 ) & 2 & 1c & ... & ... & 43.87 & + ngc 1320 & r , h@xmath1+[nii ] & 1 & ... & ... & 1d & 40.62& ... &43.21 & + ngc 1386 & 6939 , h@xmath1+[nii ] & 3 & ... & ... & 1e & 40.31& ... &43.15 & + ngc 2992 & 6680 , h@xmath1+[nii ] & 3 & ... & ... & 1f & 41.11&38.5&43.66 & i + mcg @xmath52 - 27 - 9 & r , h@xmath1+[nii ] & 4 & ... & ... & 1 g & 40.94& ... &43.21 & + ngc 4235 & r , h@xmath1+[nii ] & 5 & maj ( 48 ) , min ( 138 ) & 2&1h & 40.38& ... &42.36 & + ngc 4602 & 5265 , h@xmath1+[nii ] & 3 & ... & ... & 1i & 41.32& ... &43.64 & + ic 4329a & r , h@xmath1+[nii ] & 4 & ... & ... & 1j & 41.46&39.4&43.63 & i + ngc 5506 & 5265 , h@xmath1+[nii ] & 3 & maj ( 91 ) , min(1 ) & 2 & 1k & 40.74&39.7&43.43 & i , s + ic 1368 & r , h@xmath1+[nii ] & 1 & maj ( 45 ) & 6 & 1l & 40.56&39.5&43.82 & i + ic 1417 & r , h@xmath1+[nii ] & 1 & ... & ... & 1 m & 40.78& ... &43.14 & + ngc 7590 & 5159 , h@xmath1+[nii]&3 & ... & ... & 1n & 41.29& ... &43.40 & + + + ngc 513 & r , h@xmath1+[nii ] & 1 & ... & ... & 2a & 41.42&39.0&43.93 & i + ugc 3255 & ... & ... & maj ( 17 ) , min ( 107 ) & 2 & 2b & ... & ... & 43.68 & + eso 362-g8 & r , h@xmath1+[nii ] & 4 & ... & ... & 2c & 41.07&40.2&43.18 & i + mrk 10 & ... & ... & maj ( 130 ) , min ( 40 ) & 2 & 2d & ... & ... & 43.92 & s ? + ngc 4117 & 6520 , h@xmath1+[nii ] & 4 & maj ( 18 ) , min ( 108 ) & 2 & 2e & 39.38 & ... & ... & + ngc 5033 & r , h@xmath1+[nii]@xmath42 & 5 & maj ( 170 ) , min ( 80 ) & 2 & 2f & 40.68@xmath42& ... &43.22 & + ic 5169 & r , h@xmath1+[nii ] & 4 & ... & ... & 2 g & 40.57& ... &43.58 & i ? + mrk 915 & r , h@xmath1+[nii ] & 1 & ... & ... & 2h & 41.56& ... &43.49 & + lcrlll ngc 513 & 2 & 5949 & 0.31 & oi ( 1 ) & bright elr approx along min axis + & & & & & possibly ion cone + eso 362-g8 & 2 & 4830 & 0.34 & oi ( 1 ) & v bright large - scale conical elrs on min axis + & & & & & possibly ionized by agn + ngc 2992 & 2 & 2314 & 0.51 & oi ( 1),ri ( 2 ) & many bright large - scale elrs + & & & & & radio bubble , v good evidence + ngc 3079 & 2 ? & 1125 & 0.74 & oi , os , ri , xi ( 3,4,5 ) & definite wind , well studied + & & & & & could be starburst- or agn - driven + ngc 4388 & 2 & 2517 & 0.64 & oi , os , ri ( 6,7,8 ) & radial outflow of optically emitting gas + & & & & & radio finger perp to disk , good evidence + mrk 231 & 1 & 12300 & 0.13 & os ( 9 ) , ri ( 10 ) & definite wind + & & & & & could be starburst- or agn - driven + ngc 4945 & 2 & 560 & 0.72 & oi , os ( 3,4 ) & definite wind , well studied + & & & & & could be starburst- or agn - driven + ic 4329a & 1 & 4800 & 0.56 & oi ( 1 ) & bi - symmetric el halo on min axis + & & & & & + ngc 5506 & 2 & 1830 & 0.52 & oi , os , ri ( 1,11,12 ) & double - peaked els from min axis elrs + & & & & & radio bubble , v good evidence + ngc 5548 & 1 & 5149 & 0.05 & oi , os ( 13 ) , ri ( 10 ) & definite min axis outflow + & & & & & may be from agn + ngc 5728 & 2 & 2790 & 0.24 & oi , os ( 14 ) & conical elrs and outflow along min axis + & & & & & may be from agn + ic 1368 & 2 & 3912 & 0.44 & oi ( 1 ) & bi - symmetric el halo along min axis + amram , p. , marcelin , m. , bonnarel , f. , boulesteix , j. , afanasiev , v. l. , & dodonov , s. n. 1992 , a&a , 263 , 69 antonucci , r. r. j. 1993 , ara&a , 31 , 473 balick , b. , & heckman , t. m. 1985 , aj , 90 , 197 balsara , d. s. , & krolik , j. j. 1993 , , 402 , 109 baum , s. a. , odea , c. p. , dallacassa , d. , de bruyn , a. g. , & pedlar , a. 1993 , , 419 , 553 blandford , r. d. 1993 , in astrophysical jets , ed . d. burgarella , m. livio & c. p. odea ( cambridge : cambridge university press ) , p. 15 bregman , j. n. 1994 , in the soft x - ray cosmos : proc . of the rosat science symposium , ed . e. schlegel & r. petre ( new york : aip ) , p. 164 buta , r. , & de vaucouleurs , g. 1983 , , 51 , 149 corbin , m. r. , baldwin , j. a. , & wilson , a. s. 1988 , , 334 , 584 dettmar , r .- j . 1992 , fund . cosmic phys . , 15 , 143 diaz , a. i. 1992 , in relationships between active galactic nuclei and starburst galaxies , ed . a. v. filippenko ( san francisco : asp ) , p. 181 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , buta , r. j. , paturel , g. , & fouque , p. 1991 , third reference catalog of bright galaxies ( rc3 ; new york : springer - verlag ) fullmer , l. , & lonsdale , c. 1989 , cataloged galaxies and quasars in the iras survey ( jpl pub . d-1932 , version 2 , appendix b ) hamilton , d. , & keel , w. c. 1987 , , 321 , 211 harnett , j. , haynes , r. , klein , u. , & wielebinski , r. 1989 , a&a , 216 , 39 heckman , t. m. , armus , l. , & miley , g. k. 1990 , , 74 , 833 heckman , t. m. , lehnert , m. d. , & armus , l. 1993 , in the environment and evolution of galaxies , ed . j. m. shull & h. a. thronson ( dordrecht : kluwer ) , p. 455 hewitt , a. & burbidge , g. 1991 , , 75 , 297 huchra , j. 1993 , private communication ( updated electronic version of `` catalogue of seyfert galaxies and other bright agn '' ) hummel , e. , beck , r. , & dettmar , r .- j . 1991 , a&as , 87 , 309 hummel , e. , van gorkom , j. , & kotanyi , c. 1983 , , 267 , l5 keel , w. c. , & wehrle , a. e. 1993 , aj , 106 , 236 krolik , j. h. , & begelman , m. c. 1986 , , 308 , l55 lehnert , m. d. , & heckman , t. m. 1995 , , 97 , 89 maia , m. a. g. , da costa , l. n. , willmer , c. , pellegrini , p. s. & rite , c. 1987 , aj , 93 , 546 moshir , et al . 1990 , iras faint source catalog , version 2 mulchaey , j. s. , wilson , a. s. , & tsvetanov , z. 1996 , apjs , in press ( feb 10 issue ) pogge , r. w. 1988 , , 332 , 702 pogge , r. w. 1989 , , 345 , 730 rand , r. j. 1995 , in gas disks in galaxies , ed . j. m. van der hulst ( the hague : kluwer ) , in press rodriguez - espinosa , j. m. , rudy , r. j. , & jones , b. 1987 , , 312 , 555 schommer , r. a. , caldwel , n. , wilson , a. s. , baldwin , j. a. , phillips , m. m. , williams , t. b. , & turtle , a. j. 1988 , , 324 , 154 schulz , h. 1982 , a&a , 115 , 209 scoville , n. z. 1992 , in relationships between active galactic nuclei and starburst galaxies , ed . a. v. filippenko ( san francisco : asp ) , p. 159 seaquist , e. & odegard , n. 1991 , , 369 , 320 stone , j. l. , jr . , wilson , a. s. , & ward , m. j. 1988 , , 330 , 105 telesco , c. m. 1988 , ara&a , 26 , 343 terlevich , r. 1992 , in relationships between active galactic nuclei and starburst galaxies , ed . a. filippenko ( san francisco : asp ) , p. 133 tsvetanov , z. , dopita , m. , & allen , m. 1995 , b.a.a.s . 26 , no . 5 , p. 871 tsvetanov , z. i. , fosbury , r. a. e. , & tadhunter , c. n. 1995 , apjs , submitted ulvestad , j. s. , & wilson , a. s. 1984 , , 285 , 439 ulvestad , j. s. , & wilson , a. s. 1989 , , 343 , 659 unger , s. w. , lawrence , a. , wilson , a. s. , elvis , m. , & wright , a. e. 1987 , mnras , 228 , 521 veilleux , s. , cecil , g. , bland - hawthorn , j. , tully , r. b. , filippenko , a. v. , & sargent , w. l. w. 1994 , , 433 , 48 veron - cetty , m. p. , & veron , p. 1991 , electronic version of `` a catalog of quasars and active nuclei , '' 5th ed . ( eso sci . 10 ) weaver , k. a. , wilson , a. s. , & baldwin , j. a. 1991 , , 366 , 50 wehrle , a. e. , & morris , m. 1987 , , 313 , l43 wehrle , a. e. , & morris , m. 1988 , , 95 , 1689 wilson , a. s. 1993 , in astrophysical jets , ed . d. burgarella , m. livio & c. p. odea ( cambridge : cambridge university press ) , p. 121 wilson , a. s. , baldwin , j. a. , & ulvestad , j. s. 1985 , , 291 , 627 wilson , a. s. , & tsvetanov , z. i. 1994 , aj , 107 , 1227 wilson , a. s. , wu , x. , heckman , t. m. , baldwin , j. a. , & balick , b. 1989 , , 339 , 729 * figure 1 * images and minor - axis spectra for the 14 objects observed in our complete sample . levels for all of the contour plots of the images differ by a multiplicative factor of 2 . the highest level for the contour plots of the r - band and continuum images is 50% of the maximum surface brightness . slit positions for the minor - axis spectra are overlayed on the contour plots . * ( a ) mrk 993 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 0.5 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( b ) ark 79 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 0.5 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . also shown are spectra ( effective aperture 2.0 @xmath17 2.34 ) from the nucleus and from regions 3.1 ( 1 kpc ) north and south of the nucleus , along the minor axis . * ( c ) ngc 931 . * contour plot of the r - band emission ( left ) . emission was detected along the minor axis , but it is from the companion galaxy located @xmath320 to the north . * ( d ) ngc 1320 . * its companion galaxy ngc 1321 lies @xmath31.7 to the north of ngc 1320 . contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( e ) ngc 1386 . * contour plots of the red continuum emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 2.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( f ) ngc 2992 . * contour plots of the red continuum emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( g ) mcg @xmath52 - 27 - 9 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 2.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( h ) ngc 4235 . * contour plots of the r - band emission ( left ) the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( i ) ngc 4602 . * contour plots of the green continuum emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 2.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( j ) ic 4329a . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 2.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( k ) ngc 5506 . * contour plots of the green continuum emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath43 erg s@xmath7 @xmath19 arcsec@xmath20 . also shown are spectra ( effective aperture 2.0 @xmath17 2.34 ) from the nucleus and from regions 6.2 ( 0.7 kpc ) north and south of the nucleus , along the minor axis . * ( l ) ic 1368 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( m ) ic 1417 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 0.5 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( n ) ngc 7590 . * contour plots of the green continuum emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 4.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * figure 2 * images and minor - axis spectra for eight objects observed which were not in our complete sample . levels for all of the contour plots of the images differ by a multiplicative factor of 2 . the highest level for the contour plots of the r - band and continuum images is 50% of the maximum surface brightness . slit positions for the minor - axis spectra are overlayed on the contour plots . * ( a ) ngc 513 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 0.5 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( b ) ugc 3255 . * linearly - scaled greyscale plot of the b - band emission from a digitized sky survey plate obtained at stsci ( left ) . * ( c ) eso 362-g8 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( d ) mrk 10 . * linearly - scaled greyscale plot of the b - band emission from a digitized sky survey plate obtained at stsci ( left ) . we also show the relative intensity of the h@xmath1 line emission along the minor - axis slit . * ( e ) ngc 4117 . * contour plots of the red continuum emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( f ) ngc 5033 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . note that the field of view of the ccd was smaller than the extent of the emission from the galaxy . * ( g ) ic 5169 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 2.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 . * ( h ) mrk 915 . * contour plots of the r - band emission ( left ) and the 2 emission ( right ) . the lowest level displayed in the 2 contour plot is 1.0 @xmath17 10@xmath18 erg s@xmath7 @xmath19 arcsec@xmath20 .

in this work we develop a semi - analytical variational ansatz to study the properties of few photon excitations interacting with a collection of quantum emitters in regimes that go beyond the rotating wave approximation . this method can be used to approximate both the static and dynamical properties of a superconducting qubit in an open transmission line , including the spontaneous emission spectrum and the resonances in scattering experiments . the approximations are quantitatively accurate for rather strong couplings , as shown by a direct comparison to matrix - product - state numerical methods , and provide also a good qualitative description for stronger couplings well beyond the markovian regime .

1512.04244

the impressive progress in coupling single photons to single emitters , both in the microwave @xcite and in the optical domains @xcite , allows us to talk about an emerging field of propagating - photon quantum technology . this field already demonstrates new devices that foresee interesting applications , such as few - photon transistors @xcite , non - classical states of the radiation field @xcite , or photon - mediated interactions @xcite . the development of these technologies has been accompanied by numerous analytical and numerical techniques to model light - matter and light - mediated matter interactions . these include single - photon single - qubit effective boundary conditions @xcite , input - output theory @xcite or scattering theory @xcite . nonetheless , the degree of control of some methods , or their computational generality for arbitrary numbers of photons , remains an open problem . moreover , all these methods are restricted to the rotating wave approximation ( rwa ) regime , in which counterrotating terms are neglected . the rwa breaks down , for instance , when the spontaneous emission rate of a two - level quantum emitter , @xmath0 , becomes comparable to its energy gap , @xmath1 . such deep ultra - strong coupling is now within experimental reach in the superconducting world @xcite , and we expect that it will become also feasible in the near future in experiments with single emitters in contact with optical photons and plasmons . the microscopic hamiltonian describing light - matter interactions can be formally mapped , in any coupling regime , onto the so - called spin - boson model @xcite . hence , in dealing with strong and ultra - strong interactions , optics can look into condensed - matter physics for inspiration , generalising methods that work with the spin - boson hamiltonians , and adapting them in order to describe few propagating photonic excitations . some of these methods , such as the non - interacting - blip approximation ( niba ) @xcite , or wilson s numerical renormalisation group ( nrg ) @xcite , focus directly on the dynamics of the emitters , such that one can not address the scattering of photons . other methods , such as the sophisticated numerical techniques based on the density - matrix renormalisation group ( dmrg ) or matrix - product states ( mps ) , have a greater potential in dealing with these effects @xcite , at the expense of a higher computational cost . in this work , we study a third family of methods based on the lang - firsov transformation for the polaron problem @xcite . these methods consist on a variationally - optimised unitary transformation that displaces the electromagnetic field based on the state of the two - level emitter , and has already provided valuable results for the equilibrium properties of the spin - boson model @xcite . we introduce a _ dynamical polaron ansatz _ , which is a time - dependent variational wavefunction describing the two - level emitter at zero temperature , together with the scattering states of the photons . we thus upgrade this family of variational methods to address relevant non - equilibrium problems , such as the spontaneous emission or the ultra - strong spectroscopy of the emitter . additionally , these dynamical ansatzs lead to analytical results in certain regimes , which are valuable to develop a physical intuition about these complex non - equilibrium effects , and can be straightforwardly generalised to more complex situations with multiple quantum emitters coupled to propagating photons . the numerical and analytical methods are simpler than full mps simulations , and a comparison with these shows good qualitative and even quantitative agreement up to very large coupling strengths , @xmath2 in the ohmic spin - boson model . this shows the potential of these conceptually and technically simpler tools for analysing and designing future experiments in propagating - photon quantum technologies . the outline of this work is as follows . in sect . [ sec : spin - boson ] , we introduce the spin - boson model as we use it to describe light - matter interaction of few two - level emitters , also refereed to as qubits , with a low - dimensional photonic waveguide or microwave transmission line . in sect . [ sec : ansatzs ] , we describe different variational anstze . we start with two variational anstze based on the lang - firsov transformation describing a static and a time - dependent wavefunction with up to one hybrid excitation with variational weights in the qubits or in the photonic modes . we also describe briefly the mps methods that we compare with . in sect . [ sec : numerics ] , we apply all these methods to the study of a single quantum emitter in the photonic line , and its interaction with the propagating photons . we show how the ansatz properly describes the qubit polarization in the open field , the frequency renormalization , and also the transmission and reflection coefficients for an incoming low - intensity photonic wave - packet with a broad distribution among the possible photonic modes , even in the ultra - strong coupling regime . finally , in sect . [ sec : conclusions ] , we discuss our conclusions and possible extensions to treat multiple quantum emitters . the hamiltonian describing a collection of quantum emitters interacting with a one - dimensional electromagnetic ( em ) field in the ultra - strong coupling regime corresponds to the well - known _ few - impurity spin - boson model _ @xcite , unitarily transformed to a rotated spin basis , namely @xmath3 here , @xmath4 are bosonic operators that create and annihilate a quantum of the em field with frequency @xmath5 labelled by @xmath6 ( e.g. in three - dimensions , @xmath6 contains the photon wavevector and its polarisation ) . the emitters are modelled as two - level atoms , also referred to as qubits or spins , @xmath7 , @xmath8 , @xmath9 , @xmath10 , and we have introduced the transition frequency @xmath11 . in the above hamiltonian , the atom - photon interaction is defined in terms of the couplings @xmath12 , which depend on the positions of the qubits @xmath13 . as customary in the quantum theory of radiation @xcite , the em field acts as a bosonic reservoir that modifies the dynamics of the quantum emitters depending on the distribution of the atom - photon couplings for different frequencies , namely the spectral density @xmath14 for instance , when the atom - photon couplings are weak , and the correlations of the em bath decay sufficiently fast , a single excited qubit in contact with the em field will decay exponentially with a rate @xmath15 @xcite , while a pair of qubits will show collective effects due to the exchange @xcite , or the spontaneous emission @xcite , of em photons . we shall consider both discrete and continuous descriptions of a one - dimensional em reservoir , which model the photons in a transmission line . _ ( i ) discretized spin - boson model.- _ a transmission line of length @xmath16 can be divided into @xmath17 segments of length @xmath18 , which leads to a discretized momentum @xmath19 , with @xmath20 . the boson frequencies @xmath21 , which arise due to the couplings between the transmission line segments , and the spin - boson couplings @xmath22 , correspond to @xmath23 here , we have introduced the cutoff frequency @xmath24 , which increases as the number of segments is raised . in the continuum limit @xmath25 , and for energies well below the cutoff , one recovers the linear dispersion @xmath26 with @xmath27 playing the role of the speed of light in the transmission line . let us note that this discretized model will be used to perform numerical calculations . _ ( ii ) continuum spin - boson model.- _ alternatively , we can model the transmission line directly in the continuum limit @xmath28 , where @xmath29 , and the high - frequencies are exponentially cut off by substituting @xmath30 let us note that the cutoff in this model can be set to any desired value , and the dispersion is assumed to be linear for all frequencies . this continuous model will be exploited to derive a number of analytical predictions . both models of the bosonic bath lead to an ohmic spectral density at low - enough frequencies @xmath31 , which is identical for all qubits , namely @xmath32 where @xmath33 is a dimensionless spin - boson coupling strength that plays an important role . in the context of the single - impurity spin - boson model , ultra - strong couplings lead to the so - called localization - delocalization quantum phase transition @xmath34 @xcite , and the coherent - incoherent dynamical crossover at @xmath35 @xcite . as mentioned above , we shall explore the consequences of such effects in the static and dynamical properties for few quantum emitters by using two types of variational ansatz : _ ( i ) polaron variational ansatz.- _ this ansatz was originally developed to understand the groundstate properties of the kondo effect through its connection to the spin - boson model @xcite , and has the non - variational lang - firsov transformation as its precursor @xcite . the polaron ansatz builds on a variational family of spin - dependent coherent states , and correctly captures the quantum phase transition at @xmath34 . as shown in @xcite , the polaron method also agrees with some of the finite - temperature properties obtained by other methods @xcite , and can be combined with markovian master equations to study the spin dynamics , although the predicted coherent - incoherent crossover does not coincide with @xmath35 . in this work , we introduce a _ dynamical polaron ansatz _ that will allow us to overcome these limitations , and to develop a simple analytical understanding of how the archetypical quantum - optical properties of predicted within the rwa @xcite , are modified in the ultra - strong coupling regime . moreover , this ansatz will allow us to address the dynamics of the propagating photons , which is extremely important given their experimental accessibility . _ ( ii ) matrix - product - state variational ansatz.- _ introduced in ref . @xcite and connected to dmrg @xcite for one - dimensional quantum systems in @xcite , mps anstze have the non - variational valence - bond states as their precursor @xcite . the mps ansatz is defined as the product of a set of matrices , whose number is determined by the structure of the problem , and its size determines the accuracy of the variational procedure . as explained below , in this work we use static and time - dependent mps to study the spin - boson model in frequency space , generalising static and time - dependent techniques introduced in @xcite . the comparison between the quasi - exact numerical ansatz of mps and the dynamical polaron ansatz applied to eq . , will allow us to assert the regimes of validity of the later . in our effort to make the polaron ansatz more familiar to the quantum optics community interested in going beyond the weak - coupling rwa regime , we shall determine the regimes of validity of its simpler analytical predictions . in this section , we describe in detail the above variational techniques that can be used to study both static and dynamical effects for a collection of quantum emitters interacting with a 1d em field through the spin - boson model . we will describe the general approach , and leave its application to particular settings for the following sections . the static polaron ansatz is a variational method to approximate the equilibrium properties of a single - impurity spin - boson model , which has been applied to the ohmic @xcite and sub - ohmic @xcite cases with significant success , specially in light of its considerable simplicity . it can be improved by enlarging the number of variational parameters @xcite , and generalised to the two - impurity spin - boson model @xcite . the static ansatz can be defined in terms of a variational polaron transformation , which captures the relevant correlations between the spins and the bosons , and acts on a product state where the spins and the bosons are not entangled . for an arbitrary number of spins @xmath36 , it can be defined as follows @xmath37}=u_{\rm p}^\dagger\left[f_{ik}\right]\ket{0}\otimes\ket{\psi_{\rm s}[c_{\boldsymbol{\sigma}}]}.\ ] ] here , we have introduced the polaron unitary transformation @xmath38=\bigotimes_{i , k}{{\rm e}}^{\sigma_i^x\left(f^*_{ik}a_k^{\dagger}-{\rm h.c.}\right)},\ ] ] where @xmath39 $ ] is the set of all variational polaron parameters @xmath40 . the static polaron ansatz is defined in terms of the global bosonic vacuum @xmath41 , and the variational spin state @xmath42}=\hspace{-2ex}\sum_{\boldsymbol{\sigma}\in\{\uparrow,\downarrow\}^ { n_{\rm s}}}\hspace{-2ex}c_{\boldsymbol{\sigma}}\ket{{\sigma}_1}\otimes\ket{{\sigma}_2}\otimes\cdots\otimes\ket{\sigma_{n_{\rm s}}},\ ] ] which depends on the set @xmath43 $ ] of all variational spin parameters @xmath44 , fulfilling @xmath45 . the variational minimization over the spin - boson hamiltonian , defined as @xmath46}\{\bra{\psi_{\rm gs}^{\rm p}[f_{ik},c_{\boldsymbol{\sigma}}]}h\ket{\psi_{\rm gs}^{\rm p}[f_{ik},c_{\boldsymbol{\sigma}}]}\},\ ] ] can be expressed in terms of a simpler minimization , @xmath46}\{\bra{\psi_{\rm s}[c_{\boldsymbol{\sigma}}]}h_{\rm s}[f_{ik}]\ket{\psi_{\rm s}[c_{\boldsymbol{\sigma}}]}\},\ ] ] which requires diagonalising the following spin hamiltonian instead of the original spin - boson model @xmath47=\sum_i\frac{\delta_i}{2}{{\rm e}}^{-\xi_i[f_{ik}]}\sigma_i^z+\sum_{i , j}j_{ij}[f_{ik}]\sigma_i^x\sigma_j^x.\ ] ] this spin hamiltonian corresponds to a long - range version of the paradigmatic ising model in a transverse field @xcite . this model displays qubit frequencies that get exponentially renormalized through @xmath48=\sum_k f_{ik}^{\phantom{*}}f^*_{ik}+{\rm c.c.},\ ] ] and photon - mediated ising interactions with strengths @xmath49=\sum_k\left(\omega_kf_{ik}^{\phantom{*}}f_{jk}^*-g_{ik}^{\phantom{*}}f_{jk}^*-g_{ik}^*f_{jk}^{\phantom{*}}\right).\ ] ] since the variational energy is a quadratic functional of the spin parameters , their minimization is simple and can be carried out analytically for the single- or the two - impurity problem , which fixes the spin parameters in terms of the polaron ones . accordingly , the problem reduces to function minimization , and yields the optimal parameters @xmath50 $ ] , which are denoted with a star super - index , and shall be used in the following sections . for more than two impurities , since the ising interactions can have any particular pattern , an analytical solution can not be obtained in general , and one must resort to lanczos methods to extract the groundstate of the spin hamiltonian , which can be efficiently implemented for reasonably high @xmath36 . although understanding the static properties of the spin - boson model is already a non - trivial problem , specially for a few impurities , a considerably more challenging task is to develop an accurate description of non - equilibrium effects , and a number of techniques have been put forth over the years for that purpose @xcite . as already mentioned in the introduction , most of these techniques can only address qubit obervables , which is consistent with situations where the bosonic bath can not be measured . however , with the advent of the new propagating - photon quantum technologies , this situation has been reversed , as the photonic properties of the setup are now accessible . the goal of this section is to introduce an accurate , yet simple , variational ansatz that captures these dynamical effects for both the emitters and the photons . in order to introduce such a dynamical polaron ansatz , let us revisit the spin hamiltonian in eq . , specified for the optimal parameters @xmath50 $ ] obtained with the static ansatz . the eigenstates of this hamiltonian contain , in addition to the variational groundstate @xmath51}$ ] , a number @xmath52 of spin excitations @xmath53 with energies @xmath54 that can be excited if the qubits absorb a photon from the em environment . inspired by our previous works on different quantum many - body models @xcite , we can define a dynamical variational ansatz by creating such spin and photonic excitations over the polaron transformed groundstate , namely @xmath55}=u_{\rm p}^\dagger[f_{ik}^\star]w_{\rm sp}[\alpha_s(t),\alpha_k(t)]\ket{0}\otimes\ket{\psi_{\rm gs}},\ ] ] where we have introduced an operator that creates the relevant spin - photon excitation @xmath56=\sum_{s=1}^{n_{\rm e}}\alpha_s(t)\ket{\psi^{s}_{\rm e}}\bra{\psi_{\rm gs}^{\phantom{+}}}+\sum_k\alpha_k(t ) a_k^\dagger,\ ] ] with a certain set @xmath57 $ ] of time - dependent variational parameters @xmath58 . by rearranging these parameters in a complex - valued vector @xmath59 fulfilling @xmath60 , we can construct a lagrangian that leads to a time - dependent variational principle @xcite , namely @xmath61=\frac{{{\rm i}}}{2}\left(\boldsymbol{\alpha}^\dagger\partial_t\boldsymbol{\alpha}- \partial_t\boldsymbol{\alpha}^\dagger\boldsymbol{\alpha}\right)-\mathcal{e}_{\rm p}[\boldsymbol{\alpha}^\dagger,\boldsymbol{\alpha}],\ ] ] where we do not write the explicit time - dependence to ease notation @xmath62 . here , we have introduced the energy functional associated to the spin - boson hamiltonian , namely @xmath63=\bra{\psi_{\rm exc}^{\rm p}[\boldsymbol{\alpha}]}(h-\epsilon_{\rm gs}^{\rm p})\ket{\psi_{\rm exc}^{\rm p}[\boldsymbol{\alpha}]},\ ] ] which is a quadratic functional of the variational parameters . building a variational action from the above lagrangian , the principle of minimal action @xcite leads to a system of euler - lagrange equations that describe the dynamics of the system restricted to the region of the hilbert space spanned by the states parametrised by eq . . the accuracy of this variational method thus depends on our physically - motivated choice of the dynamical ansatz , and shall be benchmarked by comparing to well - known properties for qubit observables of the single - impurity spin - boson model @xcite , and to our results of photon scattering using time - dependent mps simulations . for the simple parametrisation , the lagrangian leads to a linear system of first - order differential equations @xmath64 . here , the matrix @xmath65 can be obtained by evaluating the matrix elements @xmath66(h_{\rm p}[f_{ik}^{\star}]-\epsilon_{\rm gs}^{\rm p})w_{\rm sp}[\boldsymbol{\alpha}]\ket{0}\otimes\ket{\psi_{\rm gs}},\ ] ] for the polaron - transformed spin - boson hamiltonian @xmath67&=\sum_i\frac{\delta_{i}}{2}\left(\sigma_i^z\cos\theta_i[f_{ik}^{\star}]-\sigma_i^y\sin\theta_i[f_{ik}^{\star}]\right ) \nonumber\\ & + \sum_k\omega_k a_k^{{\dagger}}a_k^{\phantom{\dagger}}+\sum_{i , j}j_{ij}[f_{ik}^{\star}]\sigma_i^x\sigma_j^x\nonumber\\ & + \sum_{ik}\sigma_i^x\left(\big(g_{ik}-\omega_kf_{ik}^\star\big)a_k^{\phantom{\dagger}}+{\rm h.c.}\right),\end{aligned}\ ] ] which depends on the operator @xmath68=-2{{\rm i}}\sum_k(f_{ik}^\star a_k^{\phantom{\dagger}}-{\rm h.c.}).\ ] ] by diagonalizing the matrix @xmath65 , one finds the excitation energies and eigenstates , which are an admixture of the spins and photons , and can be understood as some sort of spin - photon waves . however , if one is interested in the reduced dynamics of either the spins , as customary in studies of the spin - boson model @xcite , or an incoming photonic wave - packet for transmission - reflection experiments , one may try to develop a weisskopf - wigner - type theory @xcite for the above system of first - order differential equations . these equations can always be rewritten as @xmath69 where we have introduced the energy of the spin excitations @xmath70 , and the couplings @xmath71 , and @xmath72 , which can be obtained from the matrix elements of @xmath65 . note that the couplings fulfil @xmath73 , and can be thus diagonalized by an orthogonal transformation @xmath74 . accordingly , we can transform the variational parameters @xmath75 , and the remaining couplings @xmath76 , such that @xmath77 this system of equations resembles the weisskopf - wigner equations of spontaneous emission of two - level atoms coupled to the em field @xcite . let us highlight , however , that they are valid beyond the rwa intrinsic to the standard weisskopf - wigner theory thanks to the polaronic variational methods . if one is interested in the reduced dynamics of the quantum emitters , as usual in the spin - boson model or in the theory of quantum radiation , one can develop a weisskopf - wigner - type @xcite theory by formally integrating the equations for the bosonic amplitudes , and substituting them in the equations for the qubit amplitudes . conversely , one may be interested in the scattering of propagating photons from the collection of quantum emitters , which would require the opposite process . in the following sections , we shall use both approaches , highlighting the importance of taking into account non - markovian effects in the ultra - strong coupling regime . let us note at this point that if the initial state contains some atomic coherences , the ansatz must be generalised to include also an amplitude in the groundstate @xmath78\to w_{\rm sp}[\alpha_s(t),\alpha_k(t)]+\alpha_{\rm gs}(t)$ ] . however , this amplitude does not contribute with any term in the evaluation of eq . , and thus @xmath79 , whereas the time - evolution of the remaining variational parameters is still described by eq . yet , including this groundstate amplitude may be necessary to calculate the dynamics of certain observables , such as the atomic coherences that are important for the coherent - incoherent transition of the spin - boson model . at this stage , it is worth pointing out that our dynamical ansatz differs from the application of the so - called davidov ansatzs , which arise in the study of exciton - phonon interactions , to the spin - boson model @xcite . there are two crucial differences : _ ( i ) _ our ansatz is built in two steps , such that the polaron parameters are already fixed during the computation of the dynamical properties . this contrasts the davidov ansatzs , which consists on time - dependent polaron parameters with additional time - dependent variational weights for each spin state . _ ( ii ) _ our ansatz considers also additional single - photon excitations , which are absent in the davidov ansatzs @xcite . property _ ( i ) _ will be crucial to be able to derive analytical expressions for the dynamics , whereas property _ ( ii ) _ will be crucial to describe the effect of spontaneous emission and photon scattering . the previous ansatz will be compared with a well - established method @xcite for the numerical simulation of the spin - boson model , which combines ideas from the quantum impurity ansatz @xcite , matrix product operators @xcite , and arnoldi - type evolution methods @xcite . more precisely , we write down a variational wavefunction for the qubit - photon system as @xmath80 where @xmath81 is the quantum state of the qubit , @xmath82 are the fock states of @xmath83 different photon modes in frequency space , and the @xmath84 s are different matrices with a size of up to @xmath85 , where @xmath86 is the bond dimension of the mps ansatz and @xmath87 is the maximum occupation of the bosonic modes . in addition to this encoding of the wavefunction , we efficiently write the spin - boson model as a long - range - interaction matrix product operator ( mpo ) , which has a rather small bond dimension , @xmath88 . combining the efficient representation of the hamiltonian with the mps wavefunction , we can either compute approximations to the ground state of spin - boson model , or implement a time - evolution algorithm . in the first case we work by minimizing the energy functional @xmath89 = \frac{\braket{\psi_a | h |\psi_a}}{\braket{\psi_a | \psi_a}},\ ] ] with respect to the collection of numbers in all the matrices / tensors @xmath84 . the minimization procedure is efficient thanks to the mpo representation of the hamiltonian , even when it contains long - range interactions , and it is implemented with a generalisation of the dmrg sweeping technique . the time evolution is implemented using an approximation of the exponential of the hamiltonian for short times . more precisely , we construct @xmath90 as a linear combination of mps , @xmath91 , where the vectors @xmath92 form a krylov basis built with mps themselves , as explained in ref . the use of arnoldi expansions allows us to profit from the mpo expansion of the hamiltonian and work with the long - range interactions , something which is much harder with trotter expansions . it has to be remarked that , while more accurate than the dynamical polaron ansatz , the mps method is more complex from the implementation and computational point of view . the number of parameters in an mps variational wavefunction scale as @xmath93 , while in our few photon dynamical ansatz with one qubit , we have at most @xmath94 degrees of freedom . it is therefore interesting to use the mps as a benchmark with which to assert the range of validity of the polaron ansatz , with the idea of both having a flexible and simpler tool , and also a way to implement potential analytical approximations and effective models . once the different variational ansatz have been described , let us apply them to the simplest possible scenario : a single quantum emitter ultra - strongly coupled to a 1d em field . we will exploit the analytic polaron predictions based on the continuum model to offer physical insight , and benchmark the numerical polaron results based on the discretised model with mps simulations for an identical discretisation . the main objective is to prove that the simple polaron ansatz , in comparison to the more involved mps technique , provides a sufficiently accurate description of both static and dynamical phenomena , with the hope that it will be established as a simple theoretical tool within the quantum optics community dealing with the ultra - strong coupling regime . let us consider the solution of the variational system of equations for the continuum single - impurity spin - boson model in eqs . and @xcite . for @xmath95 , the ising hamiltonian reduces to a single - spin problem , and the energy functional is @xmath96}\left\{j_{11}[f_{1k}]-\half\delta_{{\rm r}}[f_{1k}]\right \},\ ] ] where we have introduced the renormalised frequency @xmath97}.\ ] ] the function minimisation yields @xmath98},\ ] ] which amounts to a non - linear system of equations for the polaron parameters . in this case , the variational spin parameters are @xmath99),\ ] ] where @xmath100 is the heaviside step function ( i.e. if the renormalized qubit frequency vanishes , the variational groundstate corresponds to the two - fold degenerate lang - firsov transformed state ) . therefore , by solving the system of implicit equations and substituting in eq . , we can recover the variational groundstate and calculate any observable . let us note that by applying a non - variational perturbative approach in the polaron - transformed picture , the same condition has been found by imposing that the first - order perturbations vanish @xcite , as customary in spin - wave approaches @xcite . note that the system of equations can be rewritten in terms of a single implicit equation for the renormalized qubit frequency @xmath101 where the super - index @xmath102 stands for the continuum polaron model , and the exact spectral function corresponding to eq . is @xmath103 . this equation can be solved analytically in the so - called scaling limit @xmath104 , where one finds @xmath105 with @xmath106 , and @xmath0 as euler s constant . therefore , the polaron ansatz predicts that the frequency of the quantum emitter gets renormalized as a consequence of its coupling to the photonic excitations ( i.e. a photonic polaron cloud dresses the quantum emitter and leads to a renormalized transition frequency ) . this agrees with adiabatic renormalisation group arguments @xcite , and locates the localization - delocalization transition at @xmath34 , where the qubit frequency vanishes . this is an example of the of the so - called boundary quantum phase transitions that also arise in other condensed - matter contexts @xcite . using this result , we can recover the polaron and spin optimal parameters , which shall be denoted as @xmath107 $ ] , and calculate any static observable by constructing the polaron groundstate , such as @xmath108 for weak spin - boson interactions @xmath109 , @xmath110 as expected from the groundstate of a bare qubit @xmath111 . however , as the coupling to the bosonic bath is raised , the groundstate starts populating also the excited bare state @xmath112 , until @xmath113 for @xmath114 . in the usual rotated basis of the spin - boson model @xcite , this corresponds to the localised phase . as emphasised previously , the continuum model serves to gain analytical insight into the static effects . however , it is the discrete version which can be benchmarked with mps simulations and , more importantly , the model that provides a more accurate microscopic description of the bosonic bath in the physical transmission line . the solution of the variational problem in this discretised spin - boson model requires the use of numerics , and has not been considered previously in the literature . in the discretised model in eqs . and , the transmission line is divided into coupled segments , such that the cutoff frequency depends on the input parameters @xmath16 and @xmath17 , being @xmath16 the total physical length of the line , and @xmath17 the number of segments of length @xmath115 in which it is divided . we are interested in raising the number of segments @xmath17 for a fixed length @xmath16 , such that the cutoff @xmath116 can be maximised towards the above scaling limit , and the linear region of the dispersion relation @xmath26 around the qubit frequencies @xmath11 contains as many photonic modes as possible . however , we also note that the computational complexity of the polaron / mps ansatz grows with @xmath17 , as the number of bosonic modes also increases , @xmath19 with @xmath20 , and we can not consider an arbitrarily - large @xmath17 . in the discretised polaron ansatz , we numerically diagonalize the transformed hamiltonian to obtain the state of minimum energy as a function of the variational parameters @xmath117 $ ] . then , we obtain the optimal parameters @xmath118 $ ] by using a numerical optimization routine , minimizing this energy with respect to the variational parameters . once these optimal parameters are obtained , we can calculate numerically parameters such as the renormalized qubit frequency , taken directly from the displaced hamiltonian @xmath119 where the index dp " stands for the discretised polaron . in order to benchmark these static predictions with the mps simulations , we should focus on some expectation values that can also be obtained through mps . an observable of interest might be the qubit polarisation , whose calculation is simple using the polaron ground - state @xmath120}u_{\rm p}^{\dagger}[f_{1k_n}^\star]\sigma_1^z u_{\rm p}[f_{1k_n}^\star]\ket{\psi_{\rm s}[c_{\sigma_1}^\star]}$ ] . after the numerical optimisation , which yields @xmath121}$ ] for the discretised spin - boson model , we obtain the polarisation by calculating @xmath122}\sigma_1^z \ket{\psi_{\rm s}[c_{\sigma_1}^\star]}e^{-2\sum_n |f^\star_{1k_n}|^2 } .\ ] ] note that the eigenvectors are displaced by the polaron transformation , but so does the spin operator ( hence the exponential due to the renormalization ) , so the result will be on the correct frame of reference . we represent these results in fig . [ fig_sz_all ] with the same results from the mps simulations @xmath123 , where we observe a reasonably - good agreement between the polaron and mps results for the same discretisation with @xmath124 bosonic modes , which still lies far away from the scaling limit . we note that the discrete polaron results can be extended to a finer discretisation with @xmath125 , and approach the prediction of the continuum polaron in the scaling limit . such high number of bosonic modes compromises the accuracy of the mps simulation , and highlights the ultimate power of the computationally less expensive polaron methods . as a function of the spin - boson coupling @xmath126 . in dotted red line , the value from mps simulations @xmath127 for @xmath124 and @xmath128 . in blue solid line , we represent @xmath129 for the discrete ansatz , with @xmath128 and @xmath124 . ] so far , we have been concerned with the static properties of an ultra - strongly coupled qubit , although an even richer phenomenology arises out of equilibrium . in the context of the quantum theory of radiation , the typical situation is to study the evolution of an initially excited quantum emitter , whose population decays irreversibly as a consequence of the photonic reservoir . by looking at the coherences , one may find the analogue of the coherent - incoherent dynamical crossover of the spin - boson model , where the spin displays a transition between damped oscillations and over - damped decay as the spin - boson coupling strength is increased beyond @xmath35 . the correct prediction of this dynamical effect is more challenging than the localization - delocalization transition , and requires more involved techniques @xcite than the combination of the polaron static ansatz with a markovian master equation @xcite . from previous non - variational techniques @xcite , it becomes clear that the markovian assumption must be abandoned if one wants to capture the correct dynamical behaviour . the objective of this section is to study the evolution of the populations and coherences of an initially - excited quantum emitter , ultra - strongly coupled to the 1d em field by applying the dynamical ansatzs introduced previously , and we will show that the above previous limitations can be overcome with our method . for the single - impurity case @xmath95 , the situation simplifies as the ising hamiltonian reduces to a single - spin problem , and we have a simple groundstate @xmath130 together with a single spin excitation @xmath131 , provided that @xmath132 . we can thus easily build the dynamical ansatz , and obtain the corresponding differential equations , which are the variational analogue of the weisskopf - wigner theory @xcite of spontaneous emission @xmath133 here , the following couplings between the spin and photonic excitations arise @xmath134 together with the additional couplings between the bare photonic modes @xmath135 which are symmetric @xmath73 , as announced in the section introducing the dynamical ansatz . hence , this coupling matrix can be diagonalized by an orthogonal transformation leading to the general equations , which read in this case @xmath136 we integrate out the photonic modes by first changing to a rotating frame @xmath137 , @xmath138 , and then substituting the following expression @xmath139 where we have assumed that there are no photonic excitations in the initially excited state , except for those intrinsic to the polaron cloud dressing the emitter . this leads to an integro - differential equation of the convolution type @xmath140 where we have defined the following memory kernel @xmath141 since we are interested in deriving some analytical formulas , we need to replace all the sums by integrals over the spectral density . to do so , we note that the frequency shifts of the photons @xmath142 contribute at a higher - order of the spin - boson coupling , and neglecting them yields @xmath143 _ ( a ) markovian approximation : _ as customary in the weisskopf - wigner theory @xcite , we perform a change of variables @xmath144 in the convolution , and a markovian approximation to extend the integration domain to @xmath145 , and substitute @xmath146 in eq . . in this case , after using @xmath147 where @xmath148 stands for cauchy s principal value , the differential equation for the qubit can be expressed as @xmath149 where we have introduced the single - qubit decay rate @xmath150 , as well as the single - qubit lamb shift @xmath151 . the decay rate within this markovian approximation can be easily evaluated @xmath152 this yields a very sensible result : the decay rate , which is given by the value of the spectral function evaluated at the bare qubit frequency @xmath153 within the usual rwa and weak - coupling assumptions @xcite , must be substituted by the value of the spectral function at the renormalized qubit frequency according to eq . . in agreement with more involved methods @xcite , our simple dynamical polaron ansatz predicts that the localization - delocalization transition at @xmath34 , where the renormalized qubit frequency vanishes , is also accompanied by a vanishing decay rate @xmath154 . in this markovian regime , and in the scaling limit , the lamb - type shift can be calculated after solving the principal value integral , and yields @xmath155 it is interesting to note that , in analogy to the original calculation of the lamb shift of the em field where the self - energy is subtracted , the variational polaron formalism includes this self - energy directly in the groundstate energy instead of the lamb shift . this contrasts the calculations within the rwa @xcite , where the self - energy is not subtracted , and the lamb shifts diverge with the cutoff frequency @xmath156 . if we take the lamb shift together with the polaron renormalisation , the frequency of the quantum emitter becomes @xmath157 , which predicts that the evolution of coherences will stop at @xmath34 , instead of the prediction @xmath35 of the coherent - incoherent transition by other methods @xcite . as we argue below , this is an artefact of the markovian approximation , and can be cured by a more careful analysis . _ ( b ) non - markovian approximation : _ let us reconsider the integro - differential equation using resolvent - operator techniques @xcite . by laplace transform , and using the bromwich contour to invert it , one can express the solution as @xmath158 , where we have introduced the following propagator @xmath159 and we have defined the so - called level - shift operator @xmath160 obtained from the principal value integral , and the so - called level broadening operator @xmath161 note that , if one simply substitutes @xmath162 in both operators , which is known as the single - pole approximation , one can easily perform the integral recovering the markovian rate and lamb - type shift . however , if we are interested in the coherent - incoherent transition , a more careful analysis is required . in particular , by looking at solutions of @xmath163 where the propagator gets maximal , one finds @xmath164 . a taylor expansion about this solution shows that the propagator at @xmath35 only leads to an exponential damping . therefore , this non - markovian treatment is capable of locating the coherent - incoherent transition at the correct spin - boson coupling strength @xmath35 . and @xmath128 . in dotted red line and for the same input parameters we represent the mps data . ] let us now consider the dynamics under the discretised spin - boson model , which describes the spontaneous emission in the physical transmission line by using numerical means , and can be benchmarked again with mps simulations , serving thus as a test of the validity of our method . regarding the dynamical polaron ansatz with such discretised model , a clear advantage is that the vector of time - dependent variational parameters becomes finite @xmath165 , and one can directly solve the schrdinger equation @xmath64 without making any connections to the weisskopf - wigner typical approximations . from a numerical perspective , we can extend this ansatz , at the same computational effort , to consider also the amplitudes of the groundstate in the presence / absence of one additional photonic excitation for the different modes . we thus obtain a @xmath166 matrix @xmath65 by evaluating numerically eq . for the different matrix elements . then , the system of differential equations is solved numerically after specifying a particular initial condition , such as @xmath167 . in order to study the spontaneous emission rate , we know that the probability amplitude of the qubit in the excited state @xmath168 oscillates with frequency @xmath169 , and decays with @xmath170 . if we consider the expectation value of the excitation number operator @xmath171 , we are able to directly extract the spontaneous emission rate via exponential fitting of @xmath172 which we plot as a function of the coupling strength in fig . [ fig_gamma_all ] , alongside with the same quantity obtained with the mps ansatz . the agreement of both approaches is remarkable , and serves as a test of the validity of the proposed dynamical polaron ansatz . let us remark that , by solving directly the variational schrdinger equation , no markovian approximations are taken with the discretised ansatz , and it should thus give more accurate predictions than the continuum results based on this assumption . , the two models display some divergences . ] an important and crucial advantage of the discretised model is that , for the same effort , one can get information about the dynamics of the photons . as emphasized previously , this is rather unique to our method , and very important in light of the current experiments . for instance , it may be of interest to study the photon number density @xmath173 after the qubit has relaxed completely . it will give an indirect measure of the qubit frequency akin to a spectroscopy experiment , since the emitted photon is expected to have the same energy as the qubit excitation . we calculated the photon distribution as a function of the coupling strength and the frequency of the modes within our discrete polaron ansatz , and compared the predictions to another simulation with the same parameters using the mps ansatz . as shown in fig . [ fig_emission ] , both methods show clearly a peak in the distribution around the renormalized qubit frequency calculated through the static ansatz in the previous section , and thus confirm the above intuition that this photonic observable serves as a spectroscopy probe . let us note that , since we are using periodic boundary conditions for the transmission line , time must be long enough for the spin excitation to have decayed , but sufficiently short that the photon can not get around the line and scatter with the emitter again ( revival time ) . the other type of experiment that can be simulated with these tools is the scattering of one photon travelling in the transmission line with one qubit impurity . the photon distribution , after the collision has occurred , will give information about the transmission and reflection coefficients of the photons in the transmission line ( @xmath174 and @xmath175 respectively ) . in the polaron - transformed frame , we can consider the initial state as a product of its two components , the qubit in the ground state and the photon distribution specified by the probability amplitudes @xmath176 , namely @xmath177 during the collision , and some time after it has happened , photon and qubit interact such that the wavefunction is no longer in a product state . however , for long - enough times , the qubit will have decayed completely , and we can describe the state again as @xmath178 with a different photon distribution specified by the probability amplitudes @xmath179 . this enables us to define the transmission and reflection coefficients as @xmath180 being @xmath181 and @xmath182 the complex phases of the coefficients , and @xmath183 include all the phases due to the evolution . the calculation of transmission and reflection coefficients , @xmath184 and @xmath185 , can be achieved from the expectation value of the photon number operator . instead of having an initially excited qubit , this time we initialise the system with one single photon travelling in one direction of the line with the qubit in its ground state , which corresponds to eq . for a particular set of probability amplitudes . this photon will be very localized , and will have a flat frequency distribution , but only in the positive momentum . after the scattering , some of the modes will be absorbed by the qubit and re - emitted afterwards in both directions , effectively being reflected / transmitted from the qubit . this will be depicted by modes with positive momentum vanishing , and some other modes with the same energy but opposite momentum appearing . to check that our polaron predictions for this scattering experiment are consistent , we compare the photon density resonances with the renormalized qubit gap obtained through the static ansatz . we can confirm this in fig . [ fig_scattering_sh ] , which also shows that for high enough couplings , the qubit stops interacting with the photons because its dynamics gets frozen . this analysis of the probability density of the photons gives the norm of the transmission and reflection coefficients . additionally , we can also calculate the phase shift as the difference between the final and initial phases , by subtracting the phase due to the time evolution . to perform this task , we run two simulations , the first one with the qubit connected to the line and the second one without it ( equivalent to setting the coupling strength to zero ) . we then extract the phase shift of each mode as the division between the amplitude of that mode in the scattered and in the free case as in eq . . [ fig_scattering_sh ] * ( b ) * shows this scattering phase for different values of the coupling strength . as expected , the phase jump occurs around the renormalized qubit frequency . in this work , we have introduced a simple and general technique based on a time - dependent variational principle to describe dynamical aspects of a system of quantum emitters ultra - strongly coupled to the 1d em field beyond the rwa regime . the dynamics of the system is described within a region of the hilbert space that is relevant for the typical experimental situations encountered in spontaneous emission of initially - excited emitters , or scattering of an incoming photon wave packet by the quantum emitters . by a physically - motivated parametrisation of this relevant region of the hilbert space , these dynamical variational anstze become computationally less expensive than matrix - product - state simulations , and allow us to develop some physical insight in certain regimes where analytical results can be derived . more importantly , they allow to address the dynamics of the photonic degrees of freedom , which becomes very relevant in light of the recent experimental progress . in order to benchmark the accuracy of our variational techniques , we have performed a detailed comparison of static and dynamical predictions for a single quantum emitter ultra - strongly coupled to the em field . in this context , we have compared the predictions of these simple polaron methods with the numerical results of the quasi - exact mps methods for both equilibrium and non - equilibrium scenarios , showing a remarkable agreement in typical situations of spontaneous emission and photon scattering . given the computational simplicity of the introduced ansatz , as compared to the complexity of mps methods , and its demonstrated reasonable accuracy , we believe that it can become a useful theoretical tool for the quantum optics community interested in the ultra - strong coupling regime . the computational advantage of the introduced ansatz becomes even more important as the number of quantum emitters is increased . although we have focused in the present manuscript on single - emitter applications , the general scheme presented in sec . [ sec : ansatzs ] can be applied to any number of emitters . for instance , we can start from the static polaron ansatz for a couple of quantum emitters @xmath186 @xcite , and build our dynamical ansatz to analyse photon - mediated interactions and collective effects in the spontaneous emission . to illustrate some the power of our dynamical ansatz , let us advance some of the predictions that will be detailed elsewhere . the static variational problem for two identical emitters can be reduced to a couple of implicit equations , one of the renormalised qubit frequencies @xmath187 , and another one for the photon - mediated ising interactions @xmath188 , which display an interesting dependence with the inter - qubit distance @xcite . let us discuss two particularly simple limits : _ ( a ) _ at large distances , the interactions are vanishingly small , such that @xmath189 where @xmath190 coincides with the single - emitter predictions . we thus see that for @xmath34 , the frequencies of both qubits renormalize to zero in the scaling limit , such that the localization - delocalization transition of both impurities occurs at the same critical coupling as the single - impurity ohmic spin - boson model . _ ( b ) _ conversely , at short distances @xmath191 , the interactions are ferromagnetic and increase with the cutoff @xmath192 where @xmath193 , provided that @xmath194 . in this case , the renormalized frequency vanishes at a smaller spin - boson coupling @xmath195 , which was used in @xcite to locate the localisation - delocalization phase transition . turning our attention onto the dynamical effects , let us note that the qubits can also exchange real photons beyond the virtual photons associated to the polaron clouds , and this leads to collective effects in the two - impurity spin - boson model . in order to account for these effects , we note that the variational spin hamiltonian evaluated at @xmath196 $ ] leads to two spin - wave excitations that can be excited from the groundstate if the spins absorb a real photon from the em environment . in the markovian limit of the weisskopf - wigner - type theory analogous to the single - emitter case , we find that the amplitudes of the excitations fulfil @xmath197 where we have introduced the single - qubit @xmath198 and collective relaxation rates @xmath199 , as well as the single - qubit @xmath200 lamb shifts , and collective interactions @xmath201 . let us comment on the particular expressions for these parameters in order . these expressions allow us to study how the spontaneous emission of the pair of qubits gets modified as the coupling to the em field increases , eventually entering the ultra - strong coupling regime . regarding the individual decay rates , their change with respect to the weak - coupling value is caused by both the renormalization of the qubit frequency @xmath206 , and the ising interactions due to virtual photon exchange @xmath207 . as a consequence , the individual decay rates will also depend on the inter - qubit distance . in the scaling limit @xmath208 , using the results form the previous section - , we find @xmath209 where we have introduced @xmath210 , and @xmath211 are the above parameters evaluated at @xmath212 . let us first comment on the regime where the qubits are so far apart that the ising interaction is negligible . when the spin - boson coupling is sufficiently weak @xmath109 , we recover the result expected from the usual weisskopf - wigner theory of spontaneous emission @xmath213 which rests on the rotating wave approximation ( rwa ) . regarding the collective spontaneous emission in this weak - coupling regime , we observe that it gets suppressed for distances @xmath214 with @xmath215 , which coincides again with the predictions based on the rwa @xcite . on the contrary , the collective spontaneous emission attains a maximum at @xmath216 with @xmath215 , leading to sub / superadiant channels related to the singlet / triplet bell states being dark states @xcite . these predictions are modified as the spin - boson coupling is raised , or as the qubits get closer . by raising the spin - photon couplings , still at large distances , we see that the individual emission rates depend on the spectral density evaluated at the renormalized qubit frequencies @xmath217 . therefore , the individual and collective spontaneous emission get totally suppressed in the localized regime @xmath218 at sufficiently large distances . in the delocalized regime , @xmath219 , both individual and collective incoherent decay channels contribute to the dynamics . in contrast to the rwa , the collective rates get suppressed ( enhanced ) at distances that are also controlled by the renormalized frequency @xmath220 ( @xmath221 ) with @xmath215 . therefore , in order to exploit such collective decay to engineer entangled states dissipatively , as proposed in @xcite , it is very important to estimate the correct distance dependence by a careful calculation of the renormalized frequency . let us now move onto the other regime where the standard weisskopf - wigner predictions are modified : short inter - qubit distances . even in the weak - coupling regime where @xmath222 , the strength of the individual decay rates , @xmath223 , is different from the rwa predictions @xmath213 as a consequence of the ferromagnetic ising interactions @xmath224 . the distance where the collective decay rates get suppressed ( enhanced ) is also modified by the presence of interactions @xmath225 ( @xmath226 ) with @xmath215 . hence , in order to exploit these collective decay to engineer entangled states dissipatively , it is also important to estimate the correct distance by a careful calculation of the interactions due to virtual photon exchange . the differences with respect to the standard weisskopf - wigner theory become more important as the spin - boson coupling is increased . as shown in eqs . and , the individual and collective spontaneous emission are suppressed in the localized regime , which at short distances occurs for a weaker coupling @xmath227 . in addition to the above incoherent decay rates , the qubits will also suffer a frequency shift and a coherent interaction due to the exchange of real photons ( i.e. on - shell contribution ) . the frequency shifts ( i.e. lamb shifts ) of the qubits due to the photonic environment can be expressed as @xmath228 where @xmath229 is defined in terms of the exponential integral @xmath230 . in the scaling limit , and according to eqs . - , the lamb shifts become @xmath231 as occurs for the single - emitter case , a calculation beyond the markovian approximation should be performed to locate exactly the coherent - incoherent transition . let us now focus on the coherent photon - mediated interactions @xmath232 , which have two contributions . as argued in the previous section , @xmath233 is caused by the exchange of virtual off - shell photons . the remaining contribution corresponds to the exchange of real on - shell photons . in particular , again in the scaling limit , we find @xmath234 where we have introduced @xmath235 and we have used the following functions @xmath236 at large distances , where the ising contribution @xmath188 due to virtual photon exchange vanishes , all the qubit - qubit interaction is due to the exchange of real photons . moreover , @xmath237 also vanishes at large distances , and we obtain @xmath238 let us first consider the weak - coupling limit @xmath239 , where the qubit frequencies approach the bare value @xmath222 . in this case , we obtain @xmath240 , in accordance with the results based on the rwa @xcite . we thus recover the result that the photon - exchange interactions are suppressed for the distances @xmath241 with @xmath215 , where the collective spontaneous decay is maximal , and viceversa @xcite . as the spin - boson coupling is increased , these distances are changed as a consequence of the renormalization of the qubits frequency @xmath242 with @xmath215 . however , when the spin - photon coupling is sufficiently large , the interactions are totally suppressed since @xmath243 and no coherent swap of the excitation can occur . we thus see that in the localized phase , all coherent and incoherent processes are inhibited . the situation changes considerably at very short distances , where the coherent ising part @xmath188 given by eq . becomes the leading term in the scaling limit . in this case , the interactions @xmath244 become independent of the renormalized qubit frequency , and diverge with the cutoff frequency . such behaviour is consistent with the results based on the rwa approximation , and predict that the excitation can always be coherently swapped between the qubits , provided that they are close enough . in the localized regime , @xmath245 , the decay channels are suppressed , such that the qubits continue swapping the excitation indefinitely . let us finally highlight that , just as the simple markovian approximation leading to eqs . can be improved to predict the coherent - incoherent transition by taking into account non - markovian effects , it can also be improved by taking into account retardation times for the exchange of photons @xcite , thus making an interesting connection with the emerging causality for spin - boson models discretised on a lattice @xcite . t. niemczyk , f. deppe , h. huebl , e. p. menzel , f. hocke , m. j. schwarz , j. j. garcia - ripoll , d. zueco , t. hmmer , e. solano , a. marx , and r. gross , http://www.nature.com/nphys/journal/v6/n10/full/nphys1730.html [ nat . phys . * 6 , * 772 ( 2010 ) ] . a. j. leggett , s. chakravarty , a. t. dorsey , m. p. a. fisher , a. garg , and w. zwerger , http://journals.aps.org/rmp/abstract/10.1103/revmodphys.59.1 [ rev . * 59 , * 1 ( 1987 ) ] , _ and references therein_. r. bulla , n. tong , and m. vojta , http://journals.aps.org/prl/abstract/10.1103/physrevlett.91.170601[phys . rev . lett . * 91 , * 170601 ( 2003 ) ] ; f. b. anders , r. bulla , m. vojta , http://journals.aps.org/prl/abstract/10.1103/physrevlett.98.210402[phys . * 98 , * 210402 ( 2007 ) ] . i. g. lang and y. a. firsov , zh . * 43 , * 1843 ( 1962 ) , ( see also y. a. firsov , http://link.springer.com/book/10.1007/978-1-4020-6348-0/page/1[small polarons : transport phenomena ] , in polarons in advanced materials , ed . a. s. alexandrov , springer verlag , bristol , 2007 ) . r. silbey , and r. a. harris , http://scitation.aip.org/content/aip/journal/jcp/80/6/10.1063/1.447055[j . * 80 , * 2615 ( 1984 ) ] ; r. a. harris , and r. silbey , http://scitation.aip.org/content/aip/journal/jcp/83/3/10.1063/1.449469[j . * 83 , * 1069 ( 1985 ) ] . c. cohen - tannoudji , j. dupont - roc , and g. grynberg , http://onlinelibrary.wiley.com/doi/10.1002/9783527617197.fmatter/summary[atom-photon interactions : basic process and applications ] ( wiley - vch , weinheim , 2004 ) . s. stlund and s. rommer , http://journals.aps.org/prl/abstract/10.1103/physrevlett.75.3537[phys . * 75 , * 3537 ( 1995 ) ] ; j. dukelsky . martn - delgado , t. nishino and g. sierra , http://iopscience.iop.org/0295-5075/43/4/457[europhys . * 43 , * 457 ( 1998 ) ] . a. w. chin , j. prior , s. f. huelga , and m. b. plenio , http://journals.aps.org/prl/abstract/10.1103/physrevlett.107.160601[phys . lett . * 107 , * 160601 ( 2011 ) ] ; s. bera , a. nazir , a. w. chin , h. u. baranger , and s. florens , http://journals.aps.org/prb/abstract/10.1103/physrevb.90.075110[phys . b * 90 , * 075110 ( 2014 ) ] . a. kurcz , a. bermudez , and j. j. garca - ripoll , http://journals.aps.org/prl/abstract/10.1103/physrevlett.112.180405[phys . * 112 , * 180405 ( 2014 ) ] ; a. kurcz , j. j. garca - ripoll , and a. bermudez , http://link.springer.com/article/10.1140%2fepjst%2fe2015-02378-x[eur . j. special topics * 224 , * 483 ( 2015 ) ] . n. wu , l. duan , x. li , and y. zhao , http://scitation.aip.org/content/aip/journal/jcp/138/8/10.1063/1.4792502[j . * 138 , * 084111 ( 2013 ) ] ; y. yao , l. duan , z. l , c .- q . wu , and y. zhao , http://journals.aps.org/pre/abstract/10.1103/physreve.88.023303[phys . e * 88 , * 023303 ( 2013 ) ] . h. zheng , http://link.springer.com/article/10.1140%2fepjb%2fe2004-00152-7[eur . j. b * 38 , * 559 ( 2004 ) ] ; q. wang , a .- hu , and h. zheng , http://journals.aps.org/prb/abstract/10.1103/physrevb.80.214301[phys . rev . b * 80 , * 214301 ( 2009 ) ] . l. davidovich and h. m. nussenzveig , in foundations of radiation theory and quantum electrodynamics , ed . a. o. barut , ( springer 1980 , pp http://link.springer.com/chapter/10.1007/978-1-4757-0671-0_7[83-108 ] ) a. gonzalez - tudela , d. martin - cano , e. moreno , l. martin- moreno , c. tejedor , and f. j. garcia - vidal , http://journals.aps.org/prl/abstract/10.1103/physrevlett.106.020501[phys . * 106 , * 020501 ( 2011 ) ] .

the theory of the magnetic catalysis of chiral symmetry breaking in qed is developed . an approximation for the schwinger - dyson equations describing reliably this phenomenon is established , _ i.e. _ , it is shown that there exists a consistent truncation of those equations in this problem . the equations are solved both analytically and numerically , and the dynamical mass of fermions is determined . epsf.sty * keywords : * quantum electrodynamics , chiral symmetry , magnetic field .

hep-ph9908320

recently the magnetic catalysis of dynamical chiral symmetry breaking has been established as a universal phenomenon in 2 + 1 and 3 + 1 dimensions : a constant magnetic field leads to the generation of a fermion dynamical mass even at the weakest attractive interaction between fermions @xcite . the essence of this effect is the dimensional reduction @xmath0 in the dynamics of fermion pairing in a magnetic field : at weak coupling , this dynamics is dominated by the lowest landau level ( lll ) which is essentially ( @xmath1)-dimensional @xcite . the effect may have interesting applications in condensed matter physics @xcite and cosmology @xcite . in particular , this phenomenon was considered in 3 + 1 dimensional qed @xcite . since the dynamics of the lll is long - range ( infrared ) , and the qed coupling constant is weak in the infrared region , one may think that the rainbow ( ladder ) approximation is reliable in this problem . as was shown in refs . @xcite , in all the covariant gauges , the dynamical mass of fermions in this approximation is @xmath2 , \label{m_dyn}\ ] ] where @xmath3 is a magnetic field , the constant @xmath4 is of order one and @xmath5 is the renormalized coupling constant related to the scale @xmath6 . are higher order contributions indeed suppressed in this problem ? the answer is no " . as was shown in refs . @xcite , because of the ( 1 + 1)-dimensional form of the fermion propagator of the lll fermions , there are relevant higher order contributions . in particular , considering this problem in the improved rainbow approximation ( with the bare vertex in the schwinger - dyson equations for both the fermion propagator and the polarization operator ) , it was shown that , in all the covariant gauges , the fermion mass @xmath7 is given by eq . ( [ m_dyn ] ) but with @xmath8 @xcite . as we wrote in the paper @xcite , it is a challenge to define the class of all those diagrams in qed in a magnetic field that give a relevant contribution in this problem " . in this paper , we will solve the problem . ( a brief outline of our results was given in ref . we will show that there exists a ( non - covariant ) gauge in which the schwinger - dyson equations written in the improved rainbow approximation are reliable : in other words , in that gauge , there exists a consistent truncation of the schwinger - dyson equations for this _ non - perturbative _ problem . the expression for @xmath7 takes the following form , @xmath9 , \label{m}\ ] ] where @xmath10 is the number of fermion flavors,@xmath11 , @xmath12 and the constant @xmath13 is of order one . this expression for @xmath7 is essentially different from that in the rainbow approximation ( [ m_dyn ] ) . as we will see , this reflects rather rich and sophisticated dynamics in this problem . the paper is organized as follows . in section [ secii ] the schwinger - dyson equations in qed in a magnetic field are discussed . in section [ seciii ] we define a ( non - covariant ) gauge in which the improved rainbow approximation for these equations is reliable . in section [ seciv ] the loop expansion for the schwinger - dyson equations is considered and it is confirmed that the improved rainbow approximation is indeed reliable in the gauge introduced in section [ seciii ] . in section [ secv ] the schwinger - dyson equations in this approximation are solved both analytically and numerically . in section [ secvi ] we summarize the main results of the paper . the details of our analysis and some useful formulas and relations are presented in appendices [ seca ] , [ secb ] and [ secc ] . the lagrangian density of massless qed in a magnetic field is @xmath14+j^\mu a_\mu , \label{lag}\ ] ] where the covariant derivative @xmath15 is @xmath16 and the source @xmath17 provides a chosen external field : @xmath18 . as we will see below , in the case of a constant magnetic field @xmath3 , with @xmath19 the source @xmath20 . notice that the vector potential @xmath21 ( [ extpotential ] ) corresponds to the so called symmetric gauge for the external vector potential , and the magnetic field is in the @xmath22 direction . besides the dirac index ( n ) , the fermion field carries an additional flavor index @xmath23 . then the lagrangian density in eq . ( [ lag ] ) is invariant under the chiral @xmath24 symmetry ( we will discuss the anomalous @xmath25 in sec . [ secvi ] ) . the schwinger - dyson ( sd ) equations in qed in external fields were derived by schwinger and fradkin ( for a review , see ref . the equations for the fermion propagator @xmath26 are @xmath27 here @xmath28 is the bare fermion propagator in the external field @xmath29 , @xmath30 is the fermion mass operator , and @xmath31 , @xmath32 are the full photon propagator and the full amputated vertex . the full photon propagator satisfies the equations @xmath33 where @xmath34 is the free photon propagator and @xmath35 is the polarization operator . the equation for the external photon field @xmath36 in a covariant gauge is @xmath37 where the vacuum current @xmath38 is @xmath39 and @xmath40 is the gauge parameter . notice that for @xmath21 in eq . ( [ extpotential ] ) , corresponding to a constant magnetic field , the source @xmath41 is zero . indeed , in this case , @xmath42 , and @xmath43 because of the symmetry @xmath44 , with @xmath45 and @xmath46 corresponding to rotations in the @xmath47-plane and lorentz transformations in the @xmath48-hyperplane , respectively . the bare fermion propagator @xmath28 in a constant magnetic field was calculated by schwinger @xcite . in the symmetric gauge ( [ extpotential ] ) , it has the form @xmath49 where the fourier transform of @xmath50 is @xmath51 \nonumber \\ & \cdot & \left[(p^0\gamma^0-p^3\gamma^3+m)(1+\gamma^1\gamma^2\tan(ebs ) ) -p_{\perp}\gamma_{\perp}(1+\tan^2(ebs))\right ] . \label{sfourier}\end{aligned}\ ] ] here a transverse vector @xmath52 and @xmath53 is a fermion mass . then , using the identity @xmath54 for the vector potential ( [ extpotential ] ) , it is not difficult to show directly from the sd equations ( [ g-1 ] ) , ( [ sd - fer ] ) , ( [ sd - pho ] ) and ( [ pi_munu ] ) that @xmath55 [ 14 ] in other words , in a constant magnetic field , the schwinger phase is universal for green functions containing one fermion field , one antifermion field , and any number of photon fields , and the full photon propagator is translation invariant . our aim is to show that there exists a gauge in which the approximation with a bare vertex , @xmath56 is reliable for the description of spontaneous chiral symmetry breaking in a magnetic field . in this section , we will show that there is a ( non - covariant ) gauge in which the approximation with a bare vertex ( [ bare - ver ] ) ( the improved rainbow approximation ) is reliable . we begin by recalling the following facts concerning the problem of the magnetic catalysis of chiral symmetry breaking @xcite : 1 . at weak coupling , there is the lll dominance in the dynamics of fermion pairing . it is because of the presence of the large landau gap of order @xmath57 , which is much larger than the dynamical fermion mass @xmath7 ( for weak coupling ) . in other words , higher landau levels decouple from the infrared dynamics ( with @xmath58 ) in the same way as it happens with kaluza - klein ( kk ) modes in kk theories of gravity . this fact was explicitly shown in the nambu - jona - lasinio model @xcite and in qed @xcite . the propagator @xmath59 ( [ sfourier ] ) can be expanded over the landau levels @xcite . the contribution from the lll is @xmath60 where the magnetic length @xmath61 , @xmath62 , @xmath63 , and @xmath64 . the matrix @xmath65 is the projection operator on the fermion states with the spin polarized along the magnetic field . this point and eq . ( [ tildes ] ) clearly reflect the ( 1 + 1)-dimensional character of the dynamics of fermions in the lll . this property is preserved also in the case when the fermion mass is generated dynamically ( @xmath66 ) . in the fermion propagator in a magnetic field , there is a genuine spontaneous chiral symmetry breaking in this problem : the mermin - wagner - coleman ( mwc ) theorem @xcite , forbidding spontaneous breakdown of continuous symmetries at @xmath67 , is not applicable to this case . the point is that the mwc theorem is based on the fact that gapless nambu - goldstone ( ng ) bosons can not exist in @xmath68 dimensions . on the other hand , since the ng bosons connected with spontaneous chiral symmetry breaking are neutral , there is no dimensional reduction of their propagator in a magnetic field @xcite . therefore their propagator is ( @xmath69)-dimensional and there are no obstacles for their existence in this case . ] the presence of the projection operator @xmath70 implies that the bare vertex for fermions from the lll is @xmath71 . therefore the lll fermions couple only to the longitudinal @xmath72 components of the photon field . 4 . in the one - loop approximation , with fermions from the lll , the photon propagator takes the following form in covariant gauges @xcite , @xmath73 , \label{cov - gauge}\end{aligned}\ ] ] where the symbols @xmath74 and @xmath75 in @xmath76 are related to the @xmath77 and @xmath72 components , respectively , and @xmath40 is a gauge parameter . the explicit expression for @xmath78 \pi ( q_{\parallel}^2)$ ] is given in refs . @xcite . for our purposes , it is sufficient to know its asymptotes , @xmath79 where @xmath80 . + notice that the polarization effects are absent in the transverse components of @xmath81 . this is because , as was already pointed out in item 3 above , fermions from the lll couple only to the longitudinal components of the photon field . , there is a strong screening effect in the @xmath82 component of the photon propagator . ( [ pi - uv ] ) implies that @xmath83 with @xmath84 for @xmath85 and @xmath86 in the photon propagator ( [ screenedphoton ] ) . however , the tensor and the spinor structure of this dynamics are exactly the same as in the schwinger model : indeed , the lll fermion propagator ( [ tildes ] ) and the vertex @xmath87 are two - dimensional , and only the longitudinal @xmath72 components of a photon field are relevant here . this point will be crucial for finding a gauge in which the improved rainbow approximation ( with the bare vertex ( [ bare - ver ] ) ) is reliable . we recall that , as was shown in ref . @xcite , despite the smallness of @xmath5 , the expansion in @xmath5 is broken in covariant gauges in this problem . the reason is that , because of the smallness of @xmath7 in eq . ( [ m_dyn ] ) as compared to @xmath57 , there are mass singularities , @xmath88 , in infrared dynamics . in appendix [ seca ] , we analyze these singularities in the vertex . calculating the one - loop correction to the vertex , one finds that , when external momenta are of order @xmath7 or less , there are contributions of order @xmath89 . they come from the term @xmath90 in @xmath91 in eq . ( [ cov - gauge ] ) . how can one avoid such mass singularities ? a solution is suggested by the schwinger model . it is known that there is a gauge in which the full vertex is just the bare one @xcite . it is the gauge with a bare photon propagator @xmath92 with the ( non - local ) gauge function @xmath93 , where the polarization function @xmath94 in the schwinger model ( of course , here @xmath95 ) . then , the full propagator is proportional to @xmath96 , @xmath97 the point is that since now @xmath98 and since the fermion mass @xmath99 in the schwinger model , all loop contributions to the vertex are proportional to @xmath100 in that gauge and , therefore , disappear follows from the two identities for the two - dimensional dirac matrices : @xmath101 and @xmath102 ( @xmath103 , @xmath104 , @xmath105 ) . ] . let us return to the present problem . as it was emphasized above , the tensor and the spinor structure of the lll dynamics is ( 1 + 1)-dimensional . now , take the bare propagator @xmath106 with @xmath107 + q_{\parallel}^2/q^2 $ ] . then , the full propagator is @xmath108 } \nonumber \\ & = & -i\frac{g^{\parallel}_{\mu\nu}}{q^2+q_{\parallel}^2 \pi(q_{\perp}^2 , q_{\parallel}^2 ) } -i \frac{g^{\perp}_{\mu\nu}}{q^2 } + i\frac{q^{\perp}_{\mu}q^{\perp}_{\nu } + q^{\perp}_{\mu } q^{\parallel}_{\nu } + q^{\parallel}_{\mu}q^{\perp}_{\nu } } { ( q^2)^2}. \label{non - l}\end{aligned}\ ] ] the crucial point is that , as was pointed out above , the transverse degrees of freedom decouple from the lll dynamics . therefore only the first term in @xmath81 , proportional to @xmath109 , is relevant . notice now that dangerous mass singularities in loop corrections to the vertex might potentially occur only in the terms containing @xmath110 from a numerator ( @xmath111 ) of _ each _ fermion propagator in a diagram ( all other terms contain positive powers of @xmath7 , coming from at least some of the numerators and , therefore , are harmless ) . however , because of the same reasons as in the gauge ( [ gauge - s ] ) in the schwinger model , all those potentially dangerous terms disappear in the gauge ( [ non - l ] ) . therefore all the loop corrections to the vertex are suppressed by positive powers of @xmath5 in this gauge . this in turn implies that those loop corrections may result only in a change @xmath112 in expression ( [ m ] . ) in other words , in gauge ( [ non - l ] ) there exists a _ consistent _ truncation of the sd equations and the problem is essentially soluble in that gauge . ) is unique in that . in other gauges , there is an infinite set of diagrams giving relevant contributions to the vertex . therefore , in other gauges , one needs to sum up an infinite set of diagrams to recover the same result for the fermion mass . ] in the next section , we will consider the loop expansion for the sd equations in this problem in more detail . the consideration of mass singularities in loop corrections given at the end of the previous section was , though general , somewhat heuristic . first of all , one has to define more rigorously the perturbative expansion for the sd equations which is used in this problem . it is the loop expansion based on the cjt ( cornwall , jackiw , and tomboulis ) effective action @xmath113 for composite operators @xcite ( for a review see ref . the conditions for extrema of @xmath114 yield the sd equations , @xmath115 in the loop expansion for @xmath114 , the _ full _ photon and fermion propagators are used in two - particle irreducible diagrams for @xmath114 . in qed , the problem is essentially reduced to the loop expansion ( with the full photon and fermion propagators ) for the vertex . the full photon propagator is given by eq . ( [ non - l ] ) , and the full propagator for fermions from the lll has the form @xmath116 ( compare with eq . ( [ tildes ] ) and see below ) . here @xmath117 is a dynamical mass function of fermions . in this section we will derive the sd equations for the fermion propagator in the one - loop and two - loop approximations . it will be shown , that while the one - loop approximation , coinciding with the improved rainbow approximation , is reliable in non - covariant gauge ( [ non - l ] ) , it is not reliable in covariant gauges ( [ cov - gauge ] ) . from eqs . ( [ g-1 ] ) and ( [ sd - fer ] ) one gets the following equation for the fermion propagator @xmath26 in the two - loop approximation ( by using the vertex function in the one - loop approximation , see eq . ( [ 14b ] ) and eq . ( [ a3 ] ) in appendix [ seca ] ) , @xmath118 here @xmath28 is the bare fermion propagator of massless fermions ( @xmath99 ) . the graphic form of this equation is shown in fig . [ fig - sd ] . after extracting the schwinger phase factors in the full and bare fermion propagators [ see eq . ( [ 14a ] ) ] , @xmath119 eq . ( [ b1 ] ) reads as @xmath120 where @xmath21 is given in eq . ( [ extpotential ] ) and the shorthand @xmath121 stands for @xmath122 . first , let us show that the solution to the above equation , @xmath123 , allows the factorization of the dependence on the parallel and perpendicular coordinates , @xmath124 notice that this form for @xmath123 is suggested by a similar expression for the bare propagator , @xmath125 with @xmath126 ( see eq . ( [ tildes ] ) ; in the chiral limit , as in the present problem , the bare mass @xmath99 ) . in order to perform the integrations over the perpendicular components of @xmath127 and @xmath128 in eq . ( [ b3 ] ) , it is convenient to make use of the photon propagator in the momentum representation , @xmath129 after substituting this representation along with those in eqs . ( [ b4 ] ) and ( [ b5 ] ) into the sd equation ( [ b3 ] ) and performing the straightforward , though tedious , integrations over @xmath130 , @xmath131 , @xmath132 and @xmath133 we arrive at @xmath134 -iq_1^{\parallel}(x_1^{\parallel}-y_2^{\parallel } ) -iq_2^{\parallel}(x_2^{\parallel}-y_1^{\parallel } ) \right ) \nonumber\\ & & \times s(x^{\parallel}-x_1^{\parallel})\gamma^{\mu}_{\parallel } g(x_1^{\parallel}-x_2^{\parallel})\gamma^{\sigma}_{\parallel } g(x_2^{\parallel}-y_2^{\parallel})\gamma^{\nu}_{\parallel } g(y_2^{\parallel}-y_1^{\parallel})\gamma^{\rho}_{\parallel } g(y_1^{\parallel } ) { \cal d}_{\mu\nu}\left(q_1^{\parallel},q_1^{\perp}\right ) { \cal d}_{\sigma\rho}\left(q_2^{\parallel},q_2^{\perp}\right ) , \label{b8}\end{aligned}\ ] ] where @xmath135\equiv \epsilon_{3ij}q_{1i}q_{2j}$ ] . since no dependence on @xmath136 has left , we conclude that the form of @xmath123 in eq . ( [ b4 ] ) is indeed consistent with the structure of the sd equation . regarding this equation , it is necessary to emphasize that the perpendicular " components of the @xmath137-matrices are absent in it . indeed , because of the identity @xmath138 , all those components are killed by the projection operators coming from the fermion propagators . substituting now the photon propagator in the feynman - like ( non - covariant ) gauge ( [ non - l ] ) into the sd equation , we see that only the first term in eq . ( [ non - l ] ) , proportional to @xmath109 , leads to a nonvanishing contribution . in other words , the photon propagator is effectively proportional to @xmath109 ( justifying the name of the gauge ) . by switching to the momentum space , we obtain @xmath139 \right ) \nonumber\\ & & \times \gamma^{\mu}_{\parallel } g(p^{\parallel}-q_1^{\parallel})\gamma^{\sigma}_{\parallel } g(p^{\parallel}-q_1^{\parallel}-q_2^{\parallel } ) \gamma^{\nu}_{\parallel } g(p^{\parallel}-q_2^{\parallel})\gamma^{\rho}_{\parallel } { \cal d}_{\mu\nu}\left(q_1^{\parallel},q_1^{\perp}\right ) { \cal d}_{\sigma\rho}\left(q_2^{\parallel},q_2^{\perp}\right ) . \label{b11}\end{aligned}\ ] ] the general solution to this equation is given by the ansatz , @xmath140 where @xmath141 and @xmath142 . making use of this as well as of the explicit form of the photon propagator , the previous equation splits into the system of two coupled equations , @xmath143 \right ) } { \left[q_1 ^ 2+(q_1^{\parallel})^2 \pi \left((q_1^{\perp})^2,(q_1^{\parallel})^2\right)\right ] \left[q_2 ^ 2+(q_2^{\parallel})^2 \pi \left((q_2^{\perp})^2,(q_2^{\parallel})^2\right)\right ] } , \label{b13}\\ b_{p}&=&-i\frac{\alpha}{2\pi^3 } \int \frac{d^2 q_{\parallel}b_{p - q } } { \left(p_{\parallel}-q_{\parallel}\right)^2-b_{p - q}^2 } \int \frac{d^2 q_{\perp } \exp\left(-(q_{\perp}l)^2/2\right ) } { q^2+q_{\parallel}^2\pi\left(q_{\perp}^2,q_{\parallel}^2\right ) } \nonumber\\ & & -\left(\frac{\alpha}{2\pi^3}\right)^2 \int \frac{d^2 q_1^{\parallel } d^2 q_2^{\parallel } ( p_{\parallel}^2+q_1^{\parallel}\cdot q_2^{\parallel } ) a_{p - q_1}b_{p - q_1-q_2 } a_{p - q_2 } } { \left(a_{p - q_1}^2 ( p^{\parallel}-q_1^{\parallel})^2-b_{p - q_1}^2\right ) \left(a_{p - q_1-q_2}^2 ( p^{\parallel}-q_1^{\parallel } -q_2^{\parallel})^2 -b_{p - q_1-q_2}^2\right ) \left(a_{p - q_2}^2 ( p^{\parallel}-q_2^{\parallel})^2 -b_{p - q_2}^2 \right ) } \nonumber\\ & & \times \int \frac{d^2 q_1^{\perp } d^2 q_2^{\perp } \exp\left(-(q_1^{\perp}l)^2/2-(q_2^{\perp}l)^2/2 + il^2[q_1^{\perp}\times q_2^{\perp } ] \right ) } { \left[q_1 ^ 2+(q_1^{\parallel})^2 \pi \left((q_1^{\perp})^2,(q_1^{\parallel})^2\right)\right ] \left[q_2 ^ 2+(q_2^{\parallel})^2 \pi \left((q_2^{\perp})^2,(q_2^{\parallel})^2\right)\right]}. \label{b14}\end{aligned}\ ] ] this concludes our derivation of the two - loop sd equations for @xmath144 and @xmath145 which define the fermion propagator through eq . ( [ b12 ] ) . let us now show that the improved rainbow ( one - loop ) approximation is reliable in gauge ( [ non - l ] ) . in that approximation , the sd equations ( [ b13 ] ) and ( [ b14 ] ) are @xmath146 equation ( [ oneloopb ] ) was solved both numerically and analytically ( see the next section and appendix [ secb ] ) . in particular , it was shown that the approximation with @xmath147 for @xmath148 and @xmath149 rapidly decreasing for @xmath150 is a very good one ( see fig . [ fig - b ] in the next section ) . moreover , as it is shown in sec.[secv ] and appendix [ secb ] , in the improved rainbow approximation , it is sufficient to use the constant photon mass approximation for the polarization function : @xmath151 [ see eq . ( [ pi - uv ] ) ] . then , equation ( [ b14 ] ) at @xmath152 reduces to @xmath153 where the two terms in the right hand side come out as the estimates of the following two integrals , @xmath154 by assuming that the first term on the right hand side in eq . ( [ b15 ] ) is of order one , we see that the second term , corresponding to the two - loop correction , is indeed suppressed : it is of order @xmath155 . note that we should have expected this result taking into account the calculations in appendix [ seca ] . as is shown there , the higher order corrections to the vertex are suppressed in the feynman - like gauge , and the two - loop contribution in the sd equation occurs as a result of the one - loop insertion in the vertex ( see fig . [ fig - sd ] ) . if we repeat the same analysis in the case of the covariant gauge ( [ cov - gauge ] ) , we end up with the following estimate , @xmath156 where the main contribution to both terms , as is easy to check , comes from the @xmath157 component in the photon propagator . in contrast to what we had in the feynman - like gauge , after assuming that the first term in eq . ( [ b16 ] ) is of order of one , we see that the second term would be also of order one . in other words , there is no suppression of higher order terms in the covariant gauge . similarly , there is no suppression in all other gauges in which the @xmath157 component is non - zero . therefore , in agreement with the general arguments of the previous section , the feynman - like gauge ( [ non - l ] ) is special : in this gauge there exists a consistent truncation of the sd equations . since the dynamical mass is a gauge invariant quantity , in other gauges one needs to sum up an infinite set of diagrams to recover the same result for it . obviously it is rather difficult to classify all the relevant diagrams in those gauges , and so the existence of the feynman - like gauge ( [ non - l ] ) is the key point that allows to treat the present problem reliably . as was shown in the previous section , in the gauge ( [ non - l ] ) , the sd equations ( [ g-1 ] ) , ( [ sd - fer ] ) , ( [ sd - pho ] ) and ( [ pi_munu ] ) with the bare vertex ( [ bare - ver ] ) are reliable . they form a closed system of integral equations . in euclidean space eq . ( [ oneloopb ] ) has the form @xmath158 where the polarization function @xmath159 is defined from eq . ( [ pi_munu ] ) with the bare vertex ( [ bare - ver ] ) . equation ( [ sd - b ] ) was solved by using both analytical and numerical methods . the analytical approach is considered in appendix [ secb ] . here we will describe the numerical solution . the results of the numerical analysis of eq . ( [ sd - b ] ) are shown in figs . [ fig - m ] , [ fig - b ] and [ fig - fit ] . the crucial points in the analysis are the following : \a ) the polarization function @xmath160 is in principle a complicated functional of the fermion mass function @xmath117 . however , as it is shown in appendix [ secb ] , the leading singularity , @xmath161 , in @xmath162 in eq . ( [ m ] ) is induced in the kinematic region with @xmath163 . in that region , the fermions can be treated as massless , and therefore the polarization function is @xmath164 [ see eqs . ( [ pi - uv ] ) and ( [ m_gamma ] ) ] . in other words , in this approximation , the photon propagator is a propagator of a free massive boson with @xmath165 . \b ) in the numerical solution , the following ansatz for @xmath166 was used , @xmath167 for small @xmath5 ( @xmath168 ) and different @xmath10 ( @xmath169 ) the best fit was found with @xmath170 and @xmath171 ( see fig . [ fig - fit ] ) . this fit corresponds to the expression @xmath9 , \label{dynmass}\ ] ] where @xmath172 , @xmath12 and @xmath173 . \c ) the numerical solution shows that the function @xmath117 is essentially constant for @xmath174 , @xmath175 , and rapidly decreases for @xmath176 , see fig . [ fig - b ] . therefore this approximation is self - consistent : the ward - takahashi identity for the vertex is satisfied in the relevant kinematic region of momenta ( see appendix [ secc ] ) , and the pole of the fermion propagator appears at @xmath177 . it is instructive to clarify the origin of the difference between the dynamical mass @xmath7 ( [ m_dyn ] ) , calculated in the rainbow approximation , and the expression for @xmath7 ( [ dynmass ] ) . the integral equation ( [ sd - b ] ) for @xmath149 can be rewritten in the form , @xmath178 with @xmath179 and @xmath180 in the rainbow approximation and in the improved rainbow approximation , in gauge ( [ non - l ] ) , respectively ( henceforth we will omit the symbol @xmath75 ; the two - dimensional vector @xmath181 ) . the numerical analysis of this integral equation shows that the so called linearized approximation , with @xmath182 in the denominator of eq . ( [ inteq : b ] ) , is an excellent approximation . then we get @xmath183 this equation is equivalent to a two - dimensional schrdinger - like differential equation . indeed , introducing the function @xmath184 we get the equation @xmath185 where @xmath186 and the potential @xmath187 is @xmath188 at @xmath179 ( the rainbow approximation ) , the last integral is expressed through the integral exponential function @xmath189 , @xmath190 using the asymptotic relations for @xmath191 @xcite , we get @xmath192 [ 60 ] equation ( [ 60a ] ) implies that the potential is long - range in the rainbow approximation . using now the asymptotic relations for @xmath193 @xcite , we get the following asymptotes for @xmath187 at @xmath194 , @xmath195 [ 61 ] therefore , while the short - distance behavior of the potential is independent of @xmath196 ( compare eqs . ( [ 60b ] ) and ( [ 61b ] ) ) , its long - distance behavior in the rainbow approximation and in the improved rainbow approximation is essentially different : while the former potential is long - range [ see eq . ( [ 60a ] ) ] , the latter is short - range [ see eq . ( [ 61a ] ) ] . this point yields the physical origin of the difference of expressions ( [ m_dyn ] ) and ( [ dynmass ] ) for @xmath7 in those two approximations . it is also instructive to discuss this point using results proved in the literature for the two - dimensional ( @xmath197 ) schrdinger equation @xcite . the form of equation ( [ eq : schre ] ) implies that @xmath198 plays the role of energy @xmath199 in the schrdinger equation . the results of ref.@xcite ensure that a ) there is at least one bound state for any attractive potential for @xmath197 , _ i.e. _ , there is at least one solution with @xmath200 ; b ) for short - range potentials of the form @xmath201 , where @xmath202 is independent of @xmath5 , the energy @xmath203 of the lowest bound state is @xmath204 , with @xmath205 , as @xmath206 . if the mass @xmath196 were independent of @xmath5 , the potential ( [ 61a ] ) would satisfy the constraints in item b ) above . however , because @xmath207 as @xmath206 , we get an additional logarithmic factor in the power of the exponent in equation ( [ dynmass ] ) . on the other hand , since the potential @xmath187 is long - range in the rainbow approximation [ see eq . ( [ 60a ] ) ] , it clearly does not satisfy the constraints in item b ) above , and @xmath208 in that approximation . , @xmath7 would be indeed @xmath209 , with @xmath210 , in the case of @xmath211 independent of @xmath5 . notice that the transition from @xmath212 ( the improved rainbow approximation ) to @xmath179 ( the rainbow approximation ) corresponds to changing @xmath213 to @xmath214 . then , expression ( [ dynmass ] ) for @xmath7 transforms into expression ( [ m_dyn ] ) . ] the magnetic catalysis of chiral symmetry breaking in qed is a phenomenon with rather rich and sophisticated dynamics . it yields a ( first , to the best of our knowledge ) example in which there exists a consistent truncation of the schwinger - dyson equations in the problem of dynamical symmetry breaking in a ( 3 + 1)-dimensional gauge theory without fundamental scalar fields . it is instructive to compare this problem with dynamical symmetry breaking in ( 3 + 1)-dimensional gauge theories without external fields ( for a review , see refs . @xcite ) . in the case of non - abelian gauge theories , such as qcd , dynamical chiral symmetry breaking is generated in the infrared region , where the effective coupling constant is strong . this prevents to elaborate a consistent truncation of the schwinger - dyson equations in those theories . in abelian gauge theories , on the other hand , a solution with a non - zero fermion mass exists in ladder approximation for any value of the coupling constant @xcite , _ if _ there is no ultraviolet cutoff in the schwinger - dyson equations ( at finite cutoff , in the ladder approximation , dynamical chiral symmetry breaking takes place only if the coupling constant is large enough @xcite ) . this fact implies that the ultraviolet region is responsible for chiral symmetry breaking in that case . since the running coupling is strong in the ultraviolet region in abelian gauge theories , this again prevents to elaborate a consistent truncation of the schwinger - dyson equations in those theories either . in contrast , the dynamics of chiral symmetry breaking in qed in a magnetic field is long - range , and the qed coupling constant is small in infrared . as a result , a consistent truncation of the sd equations exists in the present problem in the gauge ( [ non - l ] ) . the crucial point in the present analysis is the dimensional reduction in the dynamics of the fermion pairing . however , there is an essential difference between qed and the nambu - jona - lasinio model in a magnetic field : in qed , there is an additional neutral field , the photon field . as a result , the dynamics in qed in a magnetic field is much more sophisticated than that in the njl model . in particular , the photon propagator includes the four - dimensional @xmath215 and is not reduced to the two - dimensional form . however , the tensor and the spinor structure in this model in the infrared region is exactly the same as in the schwinger model . this point is crucial for making this problem essentially soluble . what is the chiral symmetry , @xmath216 or @xmath217 , in this problem ? it is known that in massless qed , without external fields , the chiral symmetry is @xmath217 : though the singlet axial current @xmath218 is not conserved , the corresponding charge @xmath219 is conserved . the latter is connected with the absence of instanton - like configurations in qed . in the presence of an external magnetic field , the situation however might be different . indeed , the dynamics in qed in a magnetic field is intimately connected with the schwinger model where the @xmath25 symmetry is explicitly broken . we believe that this issue deserves further study . an important ingredient of the dynamics in qed in a magnetic field is the pseudo - higgs effect . it is not a genuine higgs effect since there is no complete screening of the electric charge : eq . ( [ pi - ir ] ) implies that in the deep infrared region with @xmath220 , there are ordinary coulomb - like forces . still , the pseudo - higgs effect is manifested in creating a resonance in the photon channel , with @xmath221 . this resonance provides the dominating forces leading to chiral symmetry breaking . thus , in this problem , the region primarily responsible for chiral symmetry breaking is the region of intermediate momenta : @xmath222 and @xmath223 . this point is noticeable as an example for a possibility discussed for the qcd dynamics @xcite : the dynamics of spontaneous chiral symmetry breaking might be provided by forces essentially independent of the dynamics of confinement ( the infrared dynamics ) . another noticeable point is the existence of a special gauge in which the description of the nonperturbative dynamics is essentially simplified . recall that the conventional viewpoint now is that there is a particular gauge in qcd ( the maximal abelian gauge ) which is the most appropriate one for the description of confinement and chiral symmetry breaking @xcite . at last , there are arguments in support of a dimensional reduction in the dynamics of chiral symmetry breaking in qcd @xcite . the present model yields an example of such a mechanism . we hope that the dynamics of the magnetic catalysis of chiral symmetry breaking in qed will provide insight into the non - perturbative dynamics of more complicated theories , such as quantum chromodynamics . we are grateful to anthony hams for his generous help in numerical solving the schwinger - dyson equations . we thank h. minakata , a. smilga , l.c.r . wijewardhana and k. yamawaki for useful discussions . the work of v.p.g . was supported by swiss national science foundation grant no . ceec / nis/96 - 98/7 ip 051219 and foundation of fundamental researches of ministry of science of ukraine under grant no.2.5.1/003 . the work of i.a.s . was supported by u.s . department of energy grant # de - fg02 - 84er40153 . in this appendix we will calculate the one and two - loop corrections to the vertex in the gauge ( [ non - l ] ) ( see fig.[fig - ver ] ) . in agreement with the general arguments in sec . [ seciii ] , it will be shown that they are small : of order @xmath5 and @xmath224 in one - loop and two - loop approximations , respectively . on the other hand , we will show that in covariant gauges ( [ cov - gauge ] ) , there is a large , @xmath225 , correction to the vertex already in the one - loop approximation . as was pointed out in sec . [ seciii ] , the loop expansion we use is based on the effective action of cornwall , jackiw , and tomboulis . in this expansion , the full fermion and photon propagators are used and only two - particle irreducible graphs have to be taken into account @xcite . our aim is to show that in the gauge ( [ non - l ] ) , the improved rainbow approximation is reliable , _ i.e. _ , the loop corrections to the vertex , with the photon and fermion propagators calculated in that approximation , are small . in this connection , notice that in the improved rainbow approximation the function @xmath226 in the fermion propagator equals one [ see eq . ( [ oneloopa ] ) ] and @xmath227 for all @xmath228 ( see fig . [ fig - b ] ) . therefore one can use the bare propagator ( [ tildes ] ) with @xmath229 in the calculations of the loop corrections to the vertex : indeed , as was indicated in sec . [ seciii ] , the perturbative expansion in @xmath5 potentially might be destroyed by mass singularities coming from infrared region , where @xmath227 . after getting rid of the schwinger phase factors according to the prescription in eq . ( [ 14 ] ) , we get the following expression for the one - loop correction to the vertex function , @xmath230 ( see fig . [ fig - ver]a ) . by performing the fourier transform in both @xmath231 and @xmath232 , we arrive at @xmath233 where we first integrated over @xmath234 and @xmath235 , and then shifted @xmath236 by @xmath237 . substituting the fermion propagator as in eq . ( [ tildes ] ) , but with @xmath229 , into the last expression , we obtain @xmath238 + il^2 [ p_{\perp}\times k_{\perp}]\right ] } { \left[(q_{\parallel}+p_{\parallel})^2-m_{dyn}^2\right ] \left[(q_{\parallel}+k_{\parallel})^2-m_{dyn}^2\right ] } \nonumber\\ & \times&\gamma^{\lambda } ( \hat{q}_{\parallel}+\hat{p}_{\parallel}+m_{dyn } ) o^{(- ) } \gamma_{\parallel}^{\mu } ( \hat{q}_{\parallel}+\hat{k}_{\parallel}+m_{dyn } ) o^{(- ) } \gamma^{\nu } { \cal d}_{\lambda\nu}(q_{\parallel},q_{\perp } ) , \label{a5}\end{aligned}\ ] ] where @xmath239\equiv \epsilon_{3ij}q_{i}p_{j}$ ] ( with @xmath240 ) . as is clear from this expression , the dependence on the perpendicular momenta is weak in infrared , @xmath241 , and exponentially suppressed in ultraviolet . for our purposes , it is sufficient to use @xmath242 . regarding the parallel components , we keep them non - zero so far . after substituting the photon propagator ( [ non - l ] ) into eq . ( [ a5 ] ) , we notice that the result contains two kinds of terms , namely , terms proportional to @xmath243 and terms proportional to @xmath244 . as we discuss in sec . [ seciv ] , the former are completely irrelevant for the sd equation and we drop them . the latter are relevant and has to be carefully analyzed . as is easy to check , they come exclusively from the @xmath245 term in the photon propagator , and the explicit expression ( in euclidean space ) reads @xmath246 o^{(- ) } } { \left[(q_{\parallel}+p_{\parallel})^2+m_{dyn}^2\right ] \left[(q_{\parallel}+k_{\parallel})^2+m_{dyn}^2\right ] \left(q_{\parallel}^2+q_{\perp}^2+m_{\gamma}^2\right)}. \label{a5a}\end{aligned}\ ] ] in a standard way , we introduce the feynman parameters and perform the integration over @xmath247 . the result reads , @xmath248 ^ 2}. \label{a5b}\end{aligned}\ ] ] note that this is zero for @xmath249 , and that the integral in the right hand side is finite for any finite @xmath250 . in the most interesting region of momenta , @xmath251 , this integral contains the logarithmic contribution , @xmath252 . while estimating the effect of the one - loop correction in eq . ( [ a5b ] ) to the sd equation in sec . [ seciv ] , we could check that the most important contribution comes from the region of momenta ( up to the exchange of @xmath253 and @xmath254 ) @xmath255 and @xmath256 . in this particular case , from eq . ( [ a5b ] ) we obtain @xmath257 this , in its turn , results in a suppressed two - loop correction to the sd equation . now , let us consider the correction to vertex function at two - loop order . there are two diagrams in this order ( see fig . [ fig - ver ] ) . here we shall explicitly describe the correction connected with the diagram with two crossed photon lines ( fig . [ fig - ver]b ) ; the analysis of another correction , connected with the diagram with two parallel photon lines ( fig . [ fig - ver]c ) can be done similarly . in coordinate space , the expression corresponding to the diagram with two crossed photon lines reads @xmath258 \gamma^{\sigma}\tilde{s}^{(l)}(x - x_1 ) \gamma^{\nu}\tilde{s}^{(l)}(x_1 ) \nonumber\\ & \times&\gamma^{\mu}\tilde{s}^{(l)}(-y_1 ) \gamma^{\rho}\tilde{s}^{(l)}(y_1-y ) \gamma^{\lambda}{\cal d}_{\nu\lambda}(x_1-y ) { \cal d}_{\sigma\rho}(x - y_1 ) . \label{a7}\end{aligned}\ ] ] in the momentum space , this becomes @xmath259 performing the straightforward integrations over @xmath127 and @xmath128 , and then over @xmath260 ( by making use of the explicit form of the fermion propagator ) , at the end we arrive at the following expression for the two - loop vertex correction at @xmath261 ( the case @xmath262 is not expected to give a very different estimate , but it is much harder to work with ) , @xmath263 -2 l^2 q_{1}^{\perp}\cdot q_{2}^{\perp}\right ] } { \left((q_{1}^{\parallel})^2-m_{dyn}^2\right ) \left((q_{2}^{\parallel})^2-m_{dyn}^2\right ) \left[(q_{1}^{\parallel}+q_{2}^{\parallel})^2-m_{dyn}^2\right]^2 } \nonumber\\ & & \times \gamma^{\sigma } ( \hat{q}_2^{\parallel}+m_{dyn } ) o^{(- ) } \gamma^{\nu } ( \hat{q}_{1}^{\parallel } + \hat{q}_2^{\parallel}+m_{dyn } ) o^{(- ) } \gamma^{\mu } ( \hat{q}_{1}^{\parallel } + \hat{q}_2^{\parallel}+m_{dyn } ) o^{(-)}\nonumber\\ & & \times \gamma^{\rho } ( \hat{q}_1^{\parallel}+m_{dyn } ) o^{(-)}\gamma^{\lambda } { \cal d}_{\nu\lambda}(q_{1}^{\parallel},q_{1}^{\perp } ) { \cal d}_{\sigma\rho}(q_{2}^{\parallel},q_{2}^{\perp}).\end{aligned}\ ] ] switching to the euclidean space and substituting the photon propagator in the feynman like gauge ( [ non - l ] ) , we find that the only nonzero contribution is proportional to @xmath70 , @xmath264 -2 l^2 q_{1}^{\perp}\cdot q_{2}^{\perp}\right ] } { \left(q_{1}^2+m_{dyn}^2\right ) \left(q_{2}^2+m_{dyn}^2\right ) \left[(q_{1}+q_{2})^2+m_{dyn}^2\right]^2 \left(q_{1}^2+(q_{1}^{\perp})^2+m_{\gamma}^2\right ) \left(q_{2}^2+(q_{2}^{\perp})^2+m_{\gamma}^2\right ) } \nonumber\\ & & \leq \mbox{const}\gamma^{\mu}_{\parallel } o^{(- ) } \alpha^{2 } \ln^2\left(\frac{1+(m_{\gamma}l)^2 } { ( m_{\gamma}l)^2}\right ) . \label{a8}\end{aligned}\ ] ] if we assume that @xmath10 is not too large , so that @xmath265 , then the above estimate for the two - loop correction becomes @xmath266 thus , we conclude that , as expected , the feynman like non - local gauge leads to suppressed higher order corrections . the latter , in its turn , means that the solution to the sd equation with the bare vertex and the photon propagator in eq . ( [ non - l ] ) is reliable , and that the result presumably approaches the exact one when the fine structure constant is very small . at this point it is instructive to explicitly demonstrate that the observed suppression in higher orders is the exclusive property of the special gauge in eq . ( [ non - l ] ) . to prove this , let us consider the one - loop vertex correction in the case of the covariant gauge ( [ cov - gauge ] ) . after substituting the photon propagator into eq . ( [ a5 ] ) , we arrive at the following estimate @xmath267 where , as is easy to check , the leading term with double logarithm comes from the @xmath268 component in the photon propagator in eq . ( [ cov - gauge ] ) . since in the rainbow approximation the solution to the sd equation in the covariant gauge yields @xmath269 [ see eq . ( [ m_dyn ] ) ] , we conclude that the one - loop correction in eq . ( [ a17 ] ) is of the same order as the bare vertex , _ i.e. _ , @xmath270 . as a result , such an approximation is not self - consistent . notice that if @xmath211 were equal to zero ( _ i.e. _ , @xmath271 in eq . ( [ cov - gauge ] ) and the @xmath272 term would be absent there ) , the first two terms in eq . ( [ a17 ] ) would cancel , and the one - loop correction @xmath273 would be small ( @xmath274 ) . as a result , expression ( [ m_dyn ] ) for @xmath7 would be correct . therefore the origin of the deviation of expressions ( [ m_dyn ] ) and ( [ m ] ) for @xmath7 is the generation of a non - zero @xmath196 . we emphasize that the @xmath275 component in the photon propagator , responsible for breaking the consistency of the rainbow approximation to the sd equation , is absent in the gauge eq . ( [ non - l ] ) . as a result , in this gauge , there exists a consistent truncation of the sd equation for small values of the coupling constant . since the dynamical mass is a gauge invariant quantity , in other gauges one needs to sum up an infinite set of diagrams to recover the same result . obviously it is rather difficult to classify all the relevant diagrams in those gauges , and so the existence of the special gauge is the key point that allows to solve the problem of the magnetic catalysis in qed reliably . in this appendix we will describe the analytical solution of the sd equation ( [ sd - b ] ) . first of all , let us show that the leading singularity , @xmath276 , in @xmath277 in eq . ( [ dynmass ] ) is induced in the kinematic region with @xmath278 and @xmath279 ( in that region , fermions can be treated as massless ) . as was shown in sec . [ secv ] , the approximation with @xmath147 for @xmath148 and @xmath149 rapidly decreasing for @xmath150 is reliable in this problem ( see fig . [ fig - b ] ) . then , taking @xmath280 in eq . ( [ sd - b ] ) , we arrive at the equation @xmath281 matching now the asymptotes ( [ pi - ir]),([pi - uv ] ) at @xmath282 in euclidean space , we get @xmath283 . \label{c2}\end{aligned}\ ] ] it is clear that , because of @xmath250 in @xmath284 , the first term in the square bracket on the right hand side of this equation is of order @xmath225 and can be neglected : it can not give a contribution of order @xmath276 to @xmath250 . then we arrive at the estimate , @xmath285 the double logarithmic contribution comes from the region @xmath286 , @xmath287 . therefore one can write @xmath288 to calculate the last integral with double logarithmic accuracy , we write @xmath289\nonumber\\ & \simeq&\frac{\alpha}{2\pi } \left[\log\frac{2eb}{m_\gamma^2}\log\frac { m_\gamma^2 } { m_{dyn}^2}+{1\over2}\log^2\frac{2eb } { m_\gamma^2}\right]= \frac{\alpha}{4\pi}\log\frac{2eb}{m_\gamma^2 } \log\left[\frac{2|eb| m_\gamma^2 } { m_{dyn}^4}\right ] . \label{c5}\end{aligned}\ ] ] this equation implies that @xmath290 comparing this expression with eq . ( [ dynmass ] ) , one can see that this estimate is quite reasonable . the origin of that is a rather simple form of the fermion mass function @xmath149 : @xmath291 for @xmath292 and @xmath149 rapidly decreases for @xmath293 . therefore the dominant contribution to the sd equation ( [ sd - b ] ) comes from the region with @xmath294 and @xmath295 , where fermions can be treated as massless . this in turn justifies the approximation with the polarization function @xmath296 . now we proceed at solving analytically sd equation ( [ sd - b ] ) for a mass function in the improved ladder approximation . it is @xmath297 where @xmath298 and we shifted the momentum of integration ( also we omit the symbol @xmath75 in the rest of this appendix ) . after the integration over the angular coordinate , eq . ( [ masseq ] ) becomes @xmath299 with the kernel @xmath300 to study eq . ( [ scalareq ] ) analytically , we break up the momentum integration into two regions and expand the kernel appropriately for each region ( compare with refs . @xcite ) @xmath301 . \label{approxeq}\end{aligned}\ ] ] introducing dimensionless variables @xmath302 and also the dimensionless mass function @xmath303 , we rewrite the last equation in the form @xmath304 , \label{inteq}\ ] ] where @xmath305 the solutions of the integral equation ( [ inteq ] ) satisfy the second - order differential equation @xmath306 where the prime denotes derivative with respect to @xmath307 . the boundary conditions are @xmath308 @xmath309 the function @xmath310 has asymptotic behavior @xmath311 we consider now the linearized version of eq . ( [ diffeq ] ) when the term @xmath312 in denominator is replaced by a constant @xmath313 ( @xmath314 ) : the numerical analysis shows that it is an excellent approximation . the two independent solutions of that equation near the point @xmath315 behave as @xmath316 and @xmath317 , and the uvbc selects the last one . in the region @xmath318 , the equation takes the form @xmath319 introducing the variable @xmath320 , eq . ( [ lineareq ] ) can be rewritten in the form of an equation for the hypergeometric function , @xmath321 the general solution to eq . ( [ hyper ] ) has the form @xmath322 where @xmath323 @xmath324 @xmath325 . from the infrared boundary condition ( [ irbc ] ) , one gets @xmath326 equating now the logarithmic derivatives of solution ( [ genersol ] ) ( at @xmath318 ) and @xmath327 ( at @xmath328 ) at the point @xmath329 , we arrive at the equation determining the quantity @xmath330 ( _ i.e. _ , the dynamical mass @xmath7 ) : @xmath331 where @xmath332 since the variable @xmath333 is @xmath334 ( we suppose @xmath335 ) , in what follows we need asymptotic behavior of @xmath336 at small and large negative values of its argument @xmath333 . using corresponding formulas from @xcite , we find for small values of @xmath333 ( @xmath337 ) @xmath338+o(z^2\ln z)\right\},\\ u_1^\prime & \simeq & 1+(1+\nu^2 ) z + o(z^2 ) , \\ u_2^\prime&\simeq&2re\left\{\frac{\gamma(1 + 2i\nu ) } { \gamma^2(1+i\nu ) } \nu^2(\ln(-z)+1 -h_0)+o(z\ln z)\right\},\end{aligned}\ ] ] where @xmath339 at large @xmath340 we have @xmath341+o(z^{-1}),\\ u_2 & \simeq & 2 \cos\left(\nu\ln ( -z)\right ) + o(z^{-1}),\\ u_1^\prime & \simeq & -\frac{1}{z}\sqrt{\frac{\tanh(\pi\nu)}{\pi\nu } } \cos\left(\nu\ln(-z)+\phi(\nu)\right)+o(z^{-2 } ) , \\ u_2^\prime & \simeq & -\frac{2\nu}{z } \sin\left(\nu\ln(-z)\right ) + o(z^{-2}),\end{aligned}\ ] ] where @xmath342 by making use of these asymptotes , we obtain the following expressions for @xmath343 , @xmath344 , \\ a_2 & = & ( \frac{du_2}{dx}+ u_2)|_{x=1 } \simeq2 \cos\left(\nu\ln\frac{\pi } { \bar\alpha}\right)+2\sin\left(\nu\ln \frac{\pi}{\bar\alpha}\right),\\ b_1 & = & \frac{du_1}{dx } |_{x=0 } \simeq -\frac{\pi}{\bar\alpha } , \\ b_2&= & \frac{du_2}{dx } |_{x=0 } \simeq -\ln\frac{\pi a^2}{\bar\alpha}.\end{aligned}\ ] ] and , finally , the solution to eq . ( [ phieq ] ) reads @xmath345 \simeq\left(n\alpha \over\pi\right)^{2/3 } \exp\left[-\frac{2\pi}{\alpha\ln ( \pi / n\alpha)}\right ] , \quad \mbox{as } \quad \alpha\to 0 . \label{analdynmass}\ ] ] the obtained analytical expression ( [ analdynmass ] ) for the dynamical mass is close both to estimate ( [ c6 ] ) and to the numerical solution ( [ dynmass ] ) . the ratio of the values of @xmath346 in the analytical solution and in the numerical one is @xmath347 . this , rather mild , discrepancy reflects the approximations made in the kernel of the integral equation ( [ scalareq ] ) in reducing it to the differential equation ( [ diffeq ] ) . in this appendix , we discuss the simplest ward - takahashi ( wt ) identity relating the fermion propagator and the vertex in the problem of magnetic catalysis ( for another approach see ref . @xcite ) . let us start from the wt identity in coordinate space : @xmath348 after multiplying this expression with @xmath349 on the left and @xmath350 on the right and integrating over @xmath351 and @xmath352 , we arrive at another representation , @xmath353 by taking into account the universal schwinger phase factors of the propagators and the vertex [ see eq . ( [ 14 ] ) in sec . [ secii ] ] , this identity ( in the momentum representation ) reads @xmath354 now , by substituting the bare vertex along with the solution for the fermion propagator as in eq . ( [ gen - sol ] ) with @xmath355 and @xmath356 ( recall that the latter is a very good approximation in the most important region of momenta , @xmath357 ) , we obtain the following relation @xmath358 o^{(- ) } = e^{-(p_1^{\perp } l)^2/2-(p_2^{\perp } l)^2/2 + i l^2 [ p_1^{\perp}\times p_2^{\perp } ] } ( \hat{p}_1^{\parallel}-\hat{p}_2^{\parallel } ) o^{(-)}. \label{wt - id}\end{aligned}\ ] ] once again , the approximation for the fermion propagator with @xmath355 and @xmath356 is reliable only in the dynamical region @xmath359 ) in powers of @xmath360 and keeping only the leading order contribution , we see that eq . ( [ wt - id ] ) indeed turns into an identity . in the end , we remind that the same conclusion about the wt identity was reached in ref . @xcite by making use of a different approach . k. farakos and n.e . mavromatos , int . j. mod b * 12 * , 809 ( 1998 ) ; g.w . semenoff , i.a . shovkovy and l.c.r . wijewardhana , mod . a * 13 * , 1143 ( 1998 ) ; w.v . liu , cond - mat/9808134 ( to appear in nucl . b ) . a.i . akhiezer and v.b . berestetsky , quantum electrodynamics " ( interscience , ny , 1965 ) ; d. m. gitman , e.s . fradkin and sh.m . shvartsman , in quantum electrodynamics with unstable vacuum " , edited by v.l . ginzburg ( nova science , commack , ny , 1995 ) . loskutov and v.v . skobelev , a * 56 * , 151 ( 1976 ) ; w. dittrich and m. reuter , effective lagrangians in quantum electrodynamics " ( springer , berlin , 1985 ) ; g. calucci and r. ragazzon , j. phys . a * 27 * , 2161 ( 1994 ) .

we investigate lepton flavor violation in the scotogenic model proposed by ma in which neutrinos acquire non - zero masses at the 1-loop level . although some works exist in this direction , they have mainly focused on the radiative decay . motivated by the promising new projects involving other low - energy processes , we derive complete analytical expressions for and conversion in nuclei , and numerically study their impact on the phenomenology . we will show that these processes can actually have rates larger than the one for , thus providing more stringent constraints and better experimental perspectives .

1312.2840

the search for lepton flavor violation ( lfv ) is going to live an unprecedented era with great experimental efforts in many different fronts . in addition to the well - known searches for the radiative decay @xmath3 , new projects involving other low - energy processes , such as @xmath1 or @xmath2 conversion in nuclei , are going to look for a positive lfv signal . for many years , the experiment leading to the most stringent constraints has been meg @xcite . this experiment , which searches for the radiative decay @xmath4 , recently published a new limit , @xmath5 , obtained with an updated analysis of the 2009 - 2010 data sample together with the analysis of the new data collected in 2011 @xcite . the expectation is that meg can reduce the current bound by another order of magnitude , with sensitivities of about @xmath6 after 3 years of acquisition time @xcite . however , the most impressive improvements in the next few years are expected in @xmath7 and @xmath2 conversion in nuclei . for the former , the mu3e experiment is expected to reach a sensitivity of @xmath8 ( after upgrades @xmath9 ) @xcite . this would imply an improvement of 3 - 4 orders of magnitude with respect to the current bound . for @xmath2 conversion in nuclei several project will compete in the next few years . these include mu2e @xcite , deeme @xcite , comet @xcite and prism / prime @xcite . the expected sensitivities for the conversion rate range from a modest @xmath10 to an impressive @xmath11 . finally , the limits for @xmath12 observables are less stringent , although significant improvements are expected at @xmath13 factories @xcite . table [ tab : sensi ] summarizes the current experimental bounds and future sensitivities for the low - energy lfv observables . .current experimental bounds and future sensitivities for some low - energy lfv observables . [ cols="^,^,^ " , ] different observables may have very different rates for a given model . for example , the rates for @xmath7 and @xmath2 conversion in nuclei are typically suppressed with respect to @xmath4 in models where the dominant lfv contributions are induced by dipole operators , like the minimal supersymmetric standard model . however , there are many frameworks where this is not the case . for this reason , one needs to fully understand the _ anatomy _ of lfv in each model in order to determine the expected hierarchies among observables , which then become indirect tests of the model . in this paper we pursue this goal in the context of a model proposed by ma in which neutrinos acquire masses at the 1-loop level @xcite . the same symmetry that forbids the tree - level contribution to dirac neutrino masses , a @xmath14 parity , also gives rise to a dark matter candidate . this simple extension of the standard model ( sm ) , usually called _ scotogenic model _ , constitutes a very simple framework to address the most important motivations to go beyond . although some works have been already done regarding lfv in this model @xcite , they have either focused on @xmath15 or neglected contributions beyond the photonic dipole . to the best of our knowledge , this is the first time @xmath16 and @xmath2 conversion in nuclei are fully considered . as we will see , these processes might actually have rates larger than the one for @xmath4 , thus providing better bounds and experimental perspectives . the rest of the paper is organized as follows : in sec . [ sec : model ] we describe the model and its basic features . in sec . [ sec : analytical ] we present our analytical results , whereas sec . [ sec : pheno ] contains a numerical discussion addressing some phenomenological issues of interest . finally , we summarize our results and conclude in sec . [ sec : conclusions ] . the model under consideration @xcite adds three right - handed neutrinos @xmath17 ( @xmath18-@xmath19 ) and one @xmath20 doublet @xmath21 to the sm particle content . in addition , a @xmath14 parity is imposed , under which the new particles are odd and the sm ones are even symmetry , the left - handed neutrinos in the sm lepton doublet do not form a dirac pair with the ` right - handed ' neutrinos @xmath17 . for this reason , strictly speaking , it is not correct to call the @xmath17 singlets right - handed neutrinos . nevertheless , this has become common practice in the literature and we will stick to this denomination . [ foot : n ] ] . the interaction of the right - handed neutrino sector is described by the lagrangian @xmath22 note that one can always write the right - handed neutrino mass term as a diagonal matrix without loss of generality . the scalar potential @xmath23 is given by @xmath24 \ , .\end{aligned}\ ] ] we assume that the parameters in the scalar potential are such that the doublet @xmath21 does not get a vacuum expectation value . this is fundamental in order to keep the @xmath14 symmetry unbroken . after electroweak symmetry breaking , the masses of the charged component @xmath25 and neutral component @xmath26 are split to @xmath27 where the mass difference between @xmath28 and @xmath29 is @xmath30 . after symmetry breaking , the light neutrino masses are generated at the 1-loop level discrete symmetry . ] . the neutrino mass matrix can be expressed as @xmath31\nonumber\\ & \equiv&\left(y^{t}\lambda y\right)_{\alpha\beta } , \label{eq : nu - mass}\end{aligned}\ ] ] where @xmath32 and @xmath33 are the masses of @xmath28 and @xmath29 respectively , and the @xmath34 matrix is defined as @xmath35 \ , .\ ] ] in particular , when @xmath36 ( @xmath37 ) , the mass matrix gets the simplified form @xmath38.\ ] ] this neutrino mass matrix is diagonalized as @xmath39 where @xmath40 is the pmns ( pontecorvo - maki - nakagawa - sakata ) matrix . here @xmath41 , @xmath42 , @xmath43 is the dirac phase and @xmath44 , @xmath45 are the majorana phases . the yukawa matrix @xmath46 can be written using an adapted casas - ibarra parametrization @xcite as @xmath47 where @xmath48 is an complex orthogonal matrix which satisfies @xmath49 . in this section we present our analytical results for the lfv processes @xmath50 , @xmath1 and @xmath2 conversion in nuclei . before we proceed to the analytical discussion a comment is in order . it is well - known that the rates for lfv processes get greatly enhanced in models with right - handed neutrinos at the electroweak scale @xcite . this is due to the fact that the gim suppression at work in the sm contribution is spoiled by the mixing between left- and right - handed neutrinos . one could naively think that this is also the case in the scotogenic model . however , the unbroken @xmath14 symmetry forbids this mixing , ( see footnote [ foot : n ] ) , and thus the enhancement in the @xmath51 loops is not present . we will show that the enhancement is still possible , but with @xmath52 loops instead . . ] the most popular searches for lfv have focused on the radiative process @xmath50 . this is described by the effective lagrangian @xmath53 where @xmath54 is a transition magnetic moment . it proves convenient to define it in terms of the dipole form factor @xmath55 as @xmath56 , where terms proportional to @xmath57 have been neglected and @xmath58 is the electromagnetic coupling , related to the electromagnetic fine structure constant as @xmath59 . in the model under consideration , @xmath55 gets contributions at the 1-loop level from the feynman diagrams in fig . [ fig : meg ] . they lead to the following expression @xmath60 where the @xmath61 parameters are defined as @xmath62 and the loop function @xmath63 is given in appendix [ sec : appendix1 ] . finally , the branching fraction for @xmath0 is calculated as @xmath64 where @xmath65 is the fermi constant . next we consider the process @xmath1 ( more precisely denoted as @xmath66 ) . although this has attracted less attention , important projects are going to be launched in the near future , with the mu3e experiment as the leading one . there are four types of 1-loop diagrams that contribute to @xmath1 . these are @xmath67-penguins , @xmath68-penguins , higgs - penguins and box diagrams . in our computations we did not consider higgs - penguins , since we are mostly interested in processes involving the first two charged lepton generations , whose small yukawa couplings suppress higgs contributions . notice that this assumption would not be valid for lfv processes involving @xmath12 leptons . however , the experimental limits in this case are not as stringent as those found for processes involving the first two generations , and thus their consideration would not change the phenomenological picture . . the wavy line represents either a photon or a z - boson . ] . ] let us consider the momentum assignment @xmath69 . then , the @xmath67-penguin diagrams shown in fig . [ fig : m3e - g ] lead to the amplitude @xmath70 where @xmath71 is the photon momentum . other operators turn out to be suppressed by charged lepton masses and thus they are neglected in eq . . the coefficient @xmath55 was given in eq . , whereas the coefficient @xmath72 , which corresponds to the photonic non - dipole contributions , is given by @xmath73 where the loop function @xmath74 is given in appendix [ sec : appendix1 ] . similarly , we now consider the contributions from @xmath68-penguin diagrams , also shown in fig . [ fig : m3e - g ] . neglecting sub - dominant terms proportional to @xmath75 , @xmath76 being the 4-momentum of the @xmath68-boson , the resulting amplitude can be written as @xmath77 where @xmath78 are the tree - level @xmath68-boson couplings to a pair of charged leptons . here @xmath79 is the @xmath20 gauge coupling and @xmath80 is the weak mixing angle . the coefficient @xmath81 is given by @xmath82 equation shows that @xmath68-penguins are suppressed by the charged lepton masses @xmath83 and @xmath57 . therefore , although we fully derived and included them in our computation , we found that they always have negligible contributions to the lfv processes considered in this paper . for this reason , the total decay width for @xmath1 will be mainly given by the @xmath67-penguins and the box contributions , whose relative size will determine the phenomenology . finally , the box diagrams contributing to the process @xmath1 are shown in fig . [ fig : m3e - box ] . one finds the following amplitude @xmath84 \left [ \bar u ( k_1 ) \gamma_\mu p_l u(p ) \right].\end{aligned}\ ] ] the coefficient @xmath13 is given by . the rest of box contributions are clearly suppressed in the scotogenic model . ] @xmath85,\end{aligned}\ ] ] where the loop functions @xmath86 and @xmath87 are given in appendix [ sec : appendix1 ] . the branching ratio for @xmath1 is given by @xmath88\nonumber\\ & & \times \ , \mathrm{br}\left(\ell_{\alpha}\to\ell_{\beta}\nu_{\alpha } \overline{\nu_{\beta}}\right ) \ , , \label{eq : l3lbr}\end{aligned}\ ] ] where @xmath89 and @xmath90 are given by @xmath91 in eq . , the mass of the charged lepton in the final state , @xmath92 , is kept only in the logarithmic term , where it plays the role of regulating the infrared divergence that would appear otherwise . the most remarkable experimental projects in the near future will be devoted to searches for @xmath2 conversion in nuclei . the great sensitivities announced by the different collaborations might make this observable the most stringent one in most neutrino mass models . we will present our results using the notation and conventions of refs . the conversion rate , relative to the the muon capture rate , can be expressed as @xmath93 here @xmath68 and @xmath94 are the number of protons and neutrons in the nucleus , @xmath95 is the effective atomic charge ( see @xcite ) , @xmath96 is the nuclear matrix element and @xmath97 represents the total muon capture rate . the values of these parameters for the nuclei used in experiments can be found in @xcite and references therein . furthermore , @xmath98 and @xmath99 ( taken to be @xmath100 in the numerical evaluation ) are the momentum and energy of the electron and @xmath101 is the muon mass . in the above , @xmath102 and @xmath103 ( with @xmath104 and @xmath105 ) are given by @xmath106 the numerical values of the @xmath107 coefficients can be found in @xcite . as for @xmath1 , the @xmath2 conversion rate receives contributions from @xmath67- , @xmath68- and higgs - penguins . note , however , the absence of box contributions ( besides the tiny sm contribution ) . this is due to the unbroken @xmath14 symmetry , which forbids the coupling between the @xmath108 scalars and the quark sector . moreover , we neglect again the higgs - penguin contributions due to the smallness of the involved yukawa couplings . therefore , the corresponding couplings are @xmath109 conversion in nuclei . the wavy line represents either a photon or a z - boson . ] the photon and @xmath68-boson couplings can be computed from the feynman diagrams in fig . [ fig : mueconversion ] . one finds that the relevant ( non - negligible ) couplings are @xmath110 the form factors @xmath72 , @xmath55 and @xmath81 are given in section [ subsec : l3l ] , see equations , and . furthermore , @xmath111 is the electric charge of the corresponding quark and @xmath112 are the tree - level @xmath68-boson couplings to a pair of quarks . in this section we present and discuss our numerical results . we will explore the parameter space and highlight some relevant phenomenological issues which , to the best of our knowledge , have not been discussed in the existing literature . in the numerical evaluation of our results we considered both hierarchies for the light neutrino spectrum for normal hierarchy and @xmath113 for inverted hierarchy , although we will denote it by @xmath114 in general . ] , normal hierarchy ( nh ) and inverted hierarchy ( ih ) , and randomly chose the neutrino oscillation parameters in the @xmath115 ranges found by the global fit @xcite ( _ free fluxes + rsbl _ results ) . we note that these ranges are in good agreement with the ones found by other fits , see refs . @xcite . for @xmath116 , the atmospheric angle , we selected the local minimum in the first octant , in agreement with @xcite . unless explicitly expressed otherwise , all our numerical results were obtained for a degenerate right - handed neutrino spectrum , assuming a random real @xmath48 matrix and @xmath117 . this value was found in @xcite to be compatible with a correct right - handed neutrino dm relic density due to the resulting size of the yukawa couplings . moreover , note that it is natural for @xmath118 to be very small since , in case it was exactly zero , a definition of a conserved lepton number would be possible @xcite . most lfv phenomenological studies focus on the radiative decay @xmath120 , ignoring other lfv observables . there are two reasons for this . first , the great performance of the meg experiment , that recently set the quite impressive bound @xmath121 . and second , the dipole dominance in many models of interest . when the dipole contributions originated in photon penguin diagrams dominate , the rate for @xmath7 is correlated with the rate for @xmath4 . in this case a simple relation can be derived @xcite @xmath122 since the proportionality factor is much smaller than one , @xmath123 is suppressed with respect to @xmath4 and the latter becomes the process leading to the most stringent constraints . this assumption has been present in all previous works on lepton flavor violation in the scotogenic model @xcite . they have either assumed explicitly that photon penguin diagrams dominate or simply ignored 4-fermion observables ( like @xmath7 ) and concentrated on @xmath120 ( an approach consistent with the assumption that the photonic dipole contributions dominate ) . here we want to study under what conditions that is a bad simplification of the phenomenology . in order to do so , we consider the ratio decays due to the better experimental bounds and perspectives . similar results are obtained for @xmath12 decays . ] @xmath124 in those regions of parameter space where @xmath125 , the observable that provides the most stringent limits is @xmath126 , whereas @xmath127 would be the most relevant observable in regions where @xmath128 . as a function of @xmath129 . for the definition see eq . . ] since the photonic dipole operators contribute to both observables , the only way to obtain @xmath125 is to have dominant contributions from box and/or photonic non - dipole diagrams in @xmath130 ( @xmath68-penguins are suppressed by charged leptons and thus their contribution is always negligible ) . since the photonic non - dipole diagrams , given by the @xmath72 form factor , never exceed the dipole ones as much as to compensate the large factor that multiplies @xmath131 in the branching ratio formula ( see eq . ) , they are never dominant . we are therefore left with a _ competition _ between photonic dipole operators and box diagrams . assuming box dominance in @xmath7 and a degenerate right - handed neutrino spectrum one can estimate @xmath132 where @xmath133 is the average size of the yukawa coupling and the function @xmath134 is defined as @xmath135 the function @xmath134 is shown in fig . [ fig : hxi ] . notice the cancellation for @xmath136 . this pole is caused by an exact cancellation between the contributions from the loop functions @xmath137 and @xmath138 . however , for @xmath139 and @xmath140 one always has @xmath141 . and @xmath126 as a function of @xmath129 . a degenerate right - handed neutrino spectrum has been assumed , see text for details . to the left for nh , whereas to the right for ih . the horizontal dashed lines show the current upper bounds.,title="fig : " ] and @xmath126 as a function of @xmath129 . a degenerate right - handed neutrino spectrum has been assumed , see text for details . to the left for nh , whereas to the right for ih . the horizontal dashed lines show the current upper bounds.,title="fig : " ] it is clear from eq . and fig . [ fig : hxi ] that in order to increase the value of @xmath142 one requires large yukawa couplings and a large mass difference between the right - handed neutrinos and the @xmath21 scalars ( in order to be far from @xmath143 ) . this is illustrated in fig . [ fig : brs - xi ] , where we show @xmath127 ( blue ) and @xmath126 ( red ) as a function of @xmath129 . the horizontal dashed lines represent the current upper bounds on the branching ratios . fixed values @xmath144 tev and @xmath145 ev ( lightest neutrino mass ) are taken . on the left - hand side we show our results for nh , whereas the right - hand side shows our results for ih . a random dirac phase @xmath43 has been taken . as can be derived from the spread of the points , this parameter has a much larger influence for nh . as expected from our previous estimate , one can in principle have @xmath125 ( or equivalently , @xmath146 ) for @xmath147 values far from @xmath148 . although the region with @xmath149 is already excluded for this value of @xmath118 , the region with @xmath150 is compatible with all experimental constraints . note that in this figure all points have @xmath151 yukawa couplings . larger values for @xmath118 would decrease the size of the yukawa couplings ( see eq . ) , which in turn would imply a reduction of all lfv rates . another parameter that turns out to be very relevant in the determination of the ratio @xmath142 is @xmath114 , the mass of the lightest neutrino . in order to illustrate this fact , we consider two scenarios : ( i ) scenario a : @xmath152 tev and @xmath153 tev , and ( ii ) scenario b : @xmath154 tev and @xmath144 tev . in both cases we assume a degenerate right - handed neutrino spectrum , a random dirac phase and a random real @xmath48 matrix . our numerical results for scenario a are shown in figs . [ fig : brs - m1-a ] and [ fig : r - m1-a ] . the left - hand side of these figures were obtained with nh , whereas the right - hand side shows our results for ih . we see that large values of the lightest neutrino mass may lead to large variations in the @xmath127 and @xmath126 branching ratios , and thus in the ratio @xmath142 . we conclude that the lfv rates in the scotogenic model are very sensitive to the absolute scale for neutrino masses . figure [ fig : brs - m1-a ] demonstrates that in scenario a the neutrino mass hierarchy also has a clear impact on the lfv rates . while for low @xmath114 , @xmath126 is clearly below the upper bound for nh , it is largely excluded for ih since it exceeds it . similarly , while for low @xmath114 the ratio @xmath142 is @xmath155 in nh ( as expected from dipole domination ) , the contributions from box diagrams already lead to a small increase for ih , where @xmath156 . these figures can be understood by analyzing how the yukawa couplings depend on @xmath114 . in particular , we must study the combinations of yukawa couplings that contribute to the lfv processes considered here . let us suppose that box diagrams dominate @xmath16 ( otherwise we would be in a dipole dominated scenario where the ratio @xmath142 would not deviate significantly from @xmath157 ) . then we have the relations @xmath158 assuming degenerate right - handed neutrinos and a real @xmath48 matrix , we can use the casas - ibarra parametrization in eq . to obtain @xmath159 analytical results for the relevant elements of the matrix combinations ( or _ flavor structures _ ) in the previous expressions can be found in appendix [ sec : appendix2 ] . and @xmath126 as a function of the lightest neutrino mass . scenario a is assumed , see text for details . to the left for nh , whereas to the right for ih . the horizontal dashed lines show the current upper bounds.,title="fig : " ] and @xmath126 as a function of the lightest neutrino mass . scenario a is assumed , see text for details . to the left for nh , whereas to the right for ih . the horizontal dashed lines show the current upper bounds.,title="fig : " ] as a function of the lightest neutrino mass . scenario a is assumed , see text for details . to the left for nh , whereas to the right for ih.,title="fig : " ] as a function of the lightest neutrino mass . scenario a is assumed , see text for details . to the left for nh , whereas to the right for ih.,title="fig : " ] let us first focus on the nh case . notice that in scenario a we have @xmath160 . with such a small value for @xmath147 , we expect the @xmath137 term in the box contribution to dominate over the @xmath138 term . therefore , we must inspect the expressions for @xmath161 and @xmath162 or , as shown above , @xmath163 and @xmath164 . on the one hand , we see that @xmath163 depends only on differences of mass eigenvalues . therefore , it decreases for higher values of the lightest neutrino mass . this can be easily understood from the expansion @xmath165 , where @xmath166 is the corresponding squared mass difference . this expansion is valid for @xmath167 . on the other hand , eq . clearly shows that @xmath168 increases for higher values of the lightest neutrino mass . this explains why @xmath127 decreases with @xmath114 while @xmath126 increases . the resulting behavior for the ratio @xmath142 is then trivially deduced from these considerations . notice that this quantity can reach values as high as @xmath169 . in this case it is obvious that one can not ignore @xmath126 , but in fact this branching ratio becomes the most relevant lfv observable . the discussion for ih would be a bit more involved . in this case we find a larger relevance of the @xmath138 piece . in fact , for @xmath170 ev this term competes with the @xmath137 term , leading to the feature observed on the right - hand sides of figs . [ fig : brs - m1-a ] and [ fig : r - m1-a ] . let us now consider our results for scenario b , shown in figs . [ fig : brs - m1-b ] and [ fig : r - m1-b ] . again , we present our results for nh on the left - hand side and our results for ih on the right - hand side . regarding nh , it is already clear at first sight that the results are qualitatively very similar to those found in scenario a. although the lfv rates are very different ( much lower in this case ) , the dependence on @xmath114 is very similar . notice that all points in these figures are actually allowed by the current limits . this was expected , since it is well - known that lfv constraints are more easily satisfied in scenarios with @xmath171 @xcite . on the other hand , the difference between nh and ih found in scenario a is not present in scenario b , in which both cases show the same behavior . and @xmath126 as a function of the lightest neutrino mass . scenario b is assumed , see text for details . to the left for nh , whereas to the right for ih . the horizontal dashed lines show the current upper bounds.,title="fig : " ] and @xmath126 as a function of the lightest neutrino mass . scenario b is assumed , see text for details . to the left for nh , whereas to the right for ih . the horizontal dashed lines show the current upper bounds.,title="fig : " ] as a function of the lightest neutrino mass . scenario b is assumed , see text for details . to the left for nh , whereas to the right for ih.,title="fig : " ] as a function of the lightest neutrino mass . scenario b is assumed , see text for details . to the left for nh , whereas to the right for ih.,title="fig : " ] finally , let us briefly discuss a scenario with non - degenerate right - handed neutrinos . the spectrum in the right - handed neutrino sector has an impact on the lfv rates , as we want to illustrate here . in order to do so , we consider a spectrum of the type @xmath172 , with two fixed mass eigenvalues ( @xmath173 ) and one varying ( @xmath174 ) . although one can imagine other scenarios , this simple family of non - degenerate spectra serves to show the qualitative behavior that we want to emphasize . [ fig : nondeg ] shows a representative example of how the lfv rates can change in a non - degenerate right - handed neutrino spectrum . on the left , we show @xmath127 ( blue ) and @xmath126 ( red ) as a function of @xmath175 , where the horizontal dashed lines represent the current upper bounds on the branching ratios . on the right we show the resulting @xmath142 ratio . fixed values @xmath176 tev , @xmath177 tev , @xmath144 tev and @xmath145 ev were assumed . a nh spectrum for the light neutrinos was chosen for this figure and we allowed for a random dirac phase @xmath43 . and @xmath126 ( to the left ) and the resulting @xmath142 ratio ( to the right ) as a function of @xmath178 . normal hierarchy for the light neutrinos and a non - degenerate right - handed neutrino spectrum ( with @xmath179 tev and @xmath180 tev ) have been assumed , see text for details . the horizontal dashed lines show the current upper bounds.,title="fig : " ] and @xmath126 ( to the left ) and the resulting @xmath142 ratio ( to the right ) as a function of @xmath178 . normal hierarchy for the light neutrinos and a non - degenerate right - handed neutrino spectrum ( with @xmath179 tev and @xmath180 tev ) have been assumed , see text for details . the horizontal dashed lines show the current upper bounds.,title="fig : " ] as naively expected , low @xmath174 values enhance both branching ratios , with @xmath126 being the one that typically gets the larger enhancements . this is caused by the large box contributions induced by the lightest right - handed neutrino . on the other hand , when @xmath181 , the contribution of the heaviest right - handed neutrino ( with a mass @xmath174 ) becomes sub - dominant and the lfv rates remain barely the same as in the degenerate case . this implies that the general conclusions drawn from the numerical results shown in this section are not restricted to degenerate scenarios . besides this fact , we do not find any other remarkable feature in the lfv phenomenology for non - degenerate right - handed neutrinos . we have already shown the relevant role played by the lightest neutrino mass in the resulting lfv branching ratios . let us now extend the discussion to the other undetermined low - energy parameter ( besides the majorana phases ) , the dirac phase @xmath43 . as starting point , we discuss how a non - zero dirac phase can change the prediction for @xmath182 . in order to do that , we consider the ratio @xmath183 , where @xmath184 is the value of the branching ratio for @xmath185 . this is explicitly shown in fig . [ fig : deltaratios ] , where contours of these ratios are drawn in the @xmath186 plane . in this figure we chose normal hierarchy for the light neutrinos , a degenerate right - handed neutrino spectrum and a real @xmath48 matrix . although these results were obtained for specific values of the remaining parameters , we emphasize that the @xmath187 does not depend on them when the right - handed neutrinos are degenerate and @xmath48 is a real matrix , see eqs . and . contours in the @xmath186 plane . to the left for @xmath188 and @xmath189 , to the right for @xmath190 and @xmath189 . normal hierarchy for the light neutrinos , a degenerate right - handed neutrino spectrum and specific ( but generic ) values for the free parameters have been assumed , see text for more details.,title="fig : " ] contours in the @xmath186 plane . to the left for @xmath188 and @xmath189 , to the right for @xmath190 and @xmath189 . normal hierarchy for the light neutrinos , a degenerate right - handed neutrino spectrum and specific ( but generic ) values for the free parameters have been assumed , see text for more details.,title="fig : " ] the largest variations are found for @xmath127 and @xmath191 , most directly affected by @xmath43 . for the former , we find that the branching ratio can be reduced by almost an order of magnitude , depending on the value of @xmath43 . in the latter case , the branching ratio can be increased by a factor of @xmath192 just by switching on the dirac phase . moreover , in both cases we find that @xmath114 is also determinant . we do not show our results for the remaining case , @xmath190 and @xmath193 , since we found very little dependence on the dirac phase . contours in the @xmath186 plane . to the left for @xmath194 , @xmath195 and @xmath196 , to the right for @xmath197 , @xmath193 and @xmath198 . normal hierarchy for the light neutrinos , a degenerate right - handed neutrino spectrum and specific ( but generic ) values for the free parameters have been assumed , see text for more details.,title="fig : " ] contours in the @xmath186 plane . to the left for @xmath194 , @xmath195 and @xmath196 , to the right for @xmath197 , @xmath193 and @xmath198 . normal hierarchy for the light neutrinos , a degenerate right - handed neutrino spectrum and specific ( but generic ) values for the free parameters have been assumed , see text for more details.,title="fig : " ] these results tell us that the lfv rates are highly sensitive to the low - energy neutrino parameters . the question then arises as to whether one can get information about them by measuring lfv observables . in case of @xmath182 , we have already seen that , for the specific scenario of degenerate right - handed neutrinos and a real @xmath48 matrix , the flavor dependence of the amplitude will be determined just by low energy parameters : neutrino masses , mixing angles and cp violating phases . therefore , by taking ratios of branching ratios ( what we call _ flavor ratios _ ) , the dependence on the high - energy parameters cancels out and we are left with functions of @xmath114 and @xmath43 . more precisely , we can make use of eqs . and to write @xmath199 note that there is no sum over @xmath200 , @xmath201 , @xmath202 and @xmath203 in the previous expression . our results for these ratios are presented in fig . [ fig : llgratios ] . we show @xmath204 ( to the left ) and @xmath205 ( to the right ) contours in the @xmath186 plane . under the assumptions of a degenerate right - handed neutrino spectrum and vanishing phases in the @xmath48 matrix , this figure would allow one to set important constraints on @xmath114 and @xmath43 in case two branching ratios were measured . furthermore , we see again the important dependence on these two low - energy parameters , since the ratios can change by more than one order of magnitude . the same will be true for @xmath206 when one of the two pieces , @xmath137 or @xmath138 , dominates . a particularly interesting scenario arises when the term containing the loop function @xmath207 gives the dominant contribution . as we have found numerically , this assumption is typically valid for @xmath208 or for large @xmath114 . in this case , the special dependence on the yukawa matrices , @xmath209 , implies that the @xmath48 matrix drops from the flavor ratios even when it contains complex entries , since @xmath210 . we have investigated this scenario and obtained the results in fig . [ fig : l3lratios ] . we concentrate on @xmath211 ( on the left ) and @xmath212 ( on the right ) . in the derivation of these plots , we neglected the contribution from the @xmath137 term . moreover , we assumed normal hierarchy for the light neutrinos and a degenerate right - handed neutrino spectrum . it is clear that , again , the parameters @xmath43 and @xmath114 may have a very strong impact on the 3-body branching ratios . on the left - hand side of the figure we see that ( for this parameter configuration ) @xmath126 is typically larger than @xmath213 . the ratio between these two observables is only close to @xmath148 for @xmath214 , whereas in the rest of the @xmath186 plane one has @xmath215 . on the other hand , the right - hand side of the figure shows that the ratio @xmath216 is mostly determined by @xmath114 , with @xmath43 playing a secondary role . as for the previous case , the ratio could be close to @xmath148 ( for low @xmath114 ) or much larger ( for high values of the lightest neutrino mass ) . contours in the @xmath186 plane . to the left for @xmath194 , @xmath195 and @xmath196 , to the right for @xmath197 , @xmath193 and @xmath198 . normal hierarchy for the light neutrinos and a degenerate right - handed neutrino spectrum haven assumed , see text for more details.,title="fig : " ] contours in the @xmath186 plane . to the left for @xmath194 , @xmath195 and @xmath196 , to the right for @xmath197 , @xmath193 and @xmath198 . normal hierarchy for the light neutrinos and a degenerate right - handed neutrino spectrum haven assumed , see text for more details.,title="fig : " ] our study reveals that lfv observables in the scotogenic model are highly sensitive to low - energy parameters such as the dirac phase or the lightest neutrino mass . however , it also reveals a large degeneracy , this is , the lfv rates are not correlated with a single parameter . furthermore , our results regarding flavor ratios have been obtained for a special case : degenerate right - handed neutrinos and real @xmath48 matrix . in a more general scenario one expects departures from the values of the flavor ratios obtained here . in conclusion , it is not possible to determine the value of a single parameter by measuring a flavor ratio . only the combination of measurements of the low - energy parameters with the discovery of one ( or several ) lfv processes can really put the flavor structure of the scotogenic model under experimental test . so far we have discussed our results on @xmath0 and @xmath1 . now we move on to discuss @xmath2 conversion in nuclei . in this model , we have found that @xmath68-penguins give a very little contribution to @xmath2 conversion in nuclei compared to that of the @xmath67-penguins . in this situation one could naively expect dipole operators to dominate the conversion rate . when this is the case , one expects a simple relation @xcite @xmath217 where @xmath218 is a function that depends on the nucleus and ranges from @xmath219 to @xmath220 for the nuclei of interest . however , in addition to the dipole contribution given by @xmath55 , @xmath67-penguins also have the non - dipole contribution given by @xmath72 . in fact , eqs . and tell us that , for degenerate right - handed neutrinos , one has @xmath221 . therefore , the relative weight of these two different photon contributions depends on the loop functions @xmath222 and @xmath223 . these are shown in fig . [ fig : f2g2xi ] , where one can see that @xmath224 . for @xmath139 the difference between @xmath223 and @xmath222 is small and both contributions have similar weights . however , for @xmath140 ( @xmath150 ) one has @xmath225 and @xmath72 becomes the most relevant contribution due to the additional ( large ) logarithmic factor that multiplies @xmath131 in the branching ratio formula , see eq . . moreover , even if the photonic non - dipole terms can be slightly larger than the dipole ones , box diagrams give even larger contributions in the same region of parameter space . ] . this is illustrated in fig . [ fig : mueconv - xi ] , where we show our results for @xmath127 and @xmath226 , as well as their ratio . the same parameter configuration as in fig . [ fig : brs - xi ] has been selected : fixed values @xmath144 tev and @xmath145 ev , random real @xmath48 matrix and dirac phase . these numerical results were obtained for nh , although very similar results are found for ih . we focused on @xmath2 conversion in titanium , although the same behavior is found for other nuclei . and @xmath223 as a function of @xmath129 . for the definitions see appendix [ sec : appendix1 ] . ] and @xmath227 ( to the left ) and the ratio @xmath228 ( to the right ) as a function of @xmath129 . normal hierarchy for the light neutrinos and a degenerate right - handed neutrino spectrum have been assumed , see text for details . the horizontal dashed lines show the current upper bounds.,title="fig : " ] and @xmath227 ( to the left ) and the ratio @xmath228 ( to the right ) as a function of @xmath129 . normal hierarchy for the light neutrinos and a degenerate right - handed neutrino spectrum have been assumed , see text for details . the horizontal dashed lines show the current upper bounds.,title="fig : " ] we find that for large values of @xmath147 , the @xmath2 conversion rate in titanium gets enhanced by photonic non - dipole contributions . this is a positive result , given the great experimental perspectives for @xmath2 conversion in nuclei in the near future . the scotogenic model is a popular extension of the standard model that accounts for neutrino masses and dark matter . as for most neutrino mass models , lepton flavor violation is one of the most attractive phenomenological issues , as it may reveal the underlying mechanism that leads to neutrino masses and mixings . in this work we have studied the predictions obtained in the scotogenic model for the lfv processes with the best experimental perspectives in the near future : @xmath3 , @xmath1 and @xmath2 conversion in nuclei . full analytical expressions have been derived , going beyond the usual dipole dominance approximation . our computation includes , besides the dipole photon penguin contribution , non - dipole photon contributions , @xmath68-penguins as well as box diagrams . the full consideration of all contributions to lfv processes leads to a very interesting picture . given the rich lfv phenomenology in the scotogenic model , we are sure that more complete studies can be performed . here we have explored some of the phenomenological consequences of our analytical results . this may serve as a summary of our main conclusions : * box diagrams dominate the lfv amplitudes in some parts of parameter space . this scenario leads to a deviation from the naive expectations obtained from the dipole dominance assumption and makes @xmath1 more constraining than @xmath229 . * the mass hierarchy between the right - handed neutrinos and the inert doublet scalars is of fundamental relevance for lfv observables . we have found that parameter points with large yukawa couplings and @xmath150 or @xmath149 typically have enhanced box diagrams , thus leading to @xmath230 . this is caused by the particular behavior of the loop functions . * in the scotogenic model , there are two dark matter candidates : the lightest right - handed neutrino @xmath231 and the lightest neutral @xmath21 scalar ( @xmath28 or @xmath29 ) @xcite . when @xmath232 , the lightest neutral @xmath21 constitutes the dark matter of the universe . otherwise , @xmath231 is the dark matter particle @xcite . in case of @xmath231 dm ( @xmath233 ) , the only possible annihilation channel is @xmath234 , via the yukawa interaction . for this reason , yukawa couplings of @xmath151 are required in order to obtain the observed dark matter relic density @xmath235 @xcite , and this may lead to incompatibility with the lfv bounds . it is thus clear that the dark matter phenomenology of @xmath231 and lfv are closely related . we have explicitly constructed parameter points where all the requirements for right - handed neutrino dark matter are met : @xmath236 , large yukawa couplings and @xmath237 in the appropriate range , as found in dedicated studies @xcite . our investigation reveals that although most of these points lead to violation of the lfv bounds , a small fraction of them are perfectly compatible . these valid points involve some small tuning of the parameters and could only be found due to the generality of our scans ( not limited to any fixed structure of the yukawa couplings ) . these results can be seen as a positive indication in favor of the validity of right - handed neutrino dark matter , although detailed studies are required to get a definitive and robust conclusion . these are , however , beyond the scope of this paper . on the other hand , we would like to point out that in case the dark matter is provided by the scalar @xmath21 , one can always obtain the correct relic density since , in addition to the yukawa interactions , this particle has gauge and scalar interactions @xcite , not correlated with lfv . * the lfv rates are highly sensitive to the low - energy parameters @xmath114 ( the mass of the lightest neutrino ) and @xmath43 ( the dirac phase ) . in particular , large @xmath114 typically enhances box diagrams . * in some specific scenarios ( with degenerate right - handed neutrinos ) , the ratios of branching ratios depend only on @xmath114 and @xmath43 . under some assumptions , this may allow us to test the flavor structure of the model . * interestingly , the rate for @xmath2 conversion in nuclei can also be enhanced beyond the dipole contribution in some regions of the parameter space . our study reveals that non - dipole photon contributions become very relevant for @xmath150 . this may lead to @xmath2 conversion rates in nuclei as large as the branching ratio for @xmath4 . these are good news given the promising experimental projects in @xmath2 conversion in nuclei . we would like to stress that our ( qualitative ) conclusions are not restricted to ma s scotogenic model , but should apply to a much wider class of radiative neutrino mass models . in particular , extended versions of the scotogenic model ( like the model proposed in @xcite ) should have , at least in some corners of parameter space , a similar phenomenology . the presence of tev scale particles with sizable couplings to the sm states also leads to interesting prospects at the lhc . although the direct production of the right - handed neutrinos is typically suppressed due to their singlet nature , they will be produced in the decays of the @xmath21 scalars when this is kinematically allowed . in turn , the @xmath21 scalars may have non - negligible production cross - sections provided they are light . this possibility , not related to the lepton sector , has been studied in some detail . in this case one expects multilepton final states with a significant amount of missing energy @xcite . furthermore , the scotogenic states may also modify the usual higgs boson decays , with observable implications at the lhc @xcite . to conclude , the anatomy of lepton flavor violation in the scotogenic model has been fully determined and some interesting phenomenological aspects have been explored . some definite predictions have been made , and these may be used to put the model under experimental test . the connection between neutrino masses and lepton flavor violation is a powerful test for this purpose . hopefully , a positive signal in one ( or several ) experiments in the next few years will provide valuable hints on the mechanism behind neutrino masses . we would like to thank thomas schwetz for many fruitful discussions . av also thanks nuria rius and juan racker for their comments on the manuscript and acknowledges partial support from the anr project cpv - lfv - lhc nt09 - 508531 . tt acknowledges support from the european itn project ( fp7-people-2011-itn , pitn - ga-2011 - 289442-invisibles ) . we present in this appendix the loop functions that appear in the paper , @xmath238 these loop functions do not have any poles . in the limit @xmath239 and @xmath240 , the functions become @xmath241 @xmath242 using the conventions in eq . and neglecting the majorana phases one finds @xmath244 + c_{23}^2 ( c_{12}^2 m_2+m_1 s_{12}^2 ) \label{eq : dag22 } \\ & & + 2 \ , c_{12 } c_{23 } s_{12 } s_{13 } s_{23 } \cos \delta \ , ( m_1-m_2 ) , \nonumber \\ \left ( u_{\mathrm{pmns } } \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^{\dag } \right)_{33 } & = & c_{23}^2 \left [ s_{13}^2 ( c_{12}^2 m_1+m_2 s_{12}^2)+c_{13}^2 m_3 \right ] + s_{23}^2 ( c_{12}^2 m_2+m_1 s_{12}^2 ) \\ & & + 2 \ , c_{12 } c_{23 } s_{12 } s_{13 } s_{23 } \cos \delta \ , ( m_2-m_1 ) , \nonumber \\ \left ( u_{\mathrm{pmns } } \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^{\dag } \right)_{21 } & = & c_{12 } c_{13 } s_{12 } c_{23 } ( m_2-m_1)+c_{13 } s_{13 } s_{23 } e^{-i \delta } \left [ m_3 - m_2 + c_{12}^2 ( m_2-m_1 ) \right ] , \\ \left ( u_{\mathrm{pmns } } \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^{\dag } \right)_{31 } & = & c_{12 } c_{13 } s_{12 } s_{23 } ( m_1-m_2)+c_{13 } s_{13 } c_{23 } e^{-i \delta } \left [ m_3 - m_2 + c_{12}^2 ( m_2-m_1 ) \right ] , \\ \left ( u_{\mathrm{pmns } } \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^{\dag } \right)_{32 } & = & c_{23 } s_{23 } \left [ ( s_{12}^2 - c_{12}^2 s_{13}^2)(m_2 - m_1 ) + c_{13}^2 ( m_3 - m_2 ) \right ] \\ & & - c_{12 } s_{12 } s_{13 } ( c_{23}^2 e^{i \delta } - s_{23}^2 e^{-i \delta})(m_2 - m_1 ) . \nonumber\end{aligned}\ ] ] @xmath246 + c_{23}^2 ( c_{12}^2 m_2+m_1 s_{12}^2 ) \\ & & + 2 \ , c_{12 } c_{23 } s_{12 } s_{13 } s_{23 } e^{i \delta } \ , ( m_1-m_2 ) , \nonumber \\ \left ( u_{\mathrm{pmns}}^ * \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^\dagger \right)_{33 } & = & c_{23}^2 \left [ e^{2 i \delta } s_{13}^2 ( c_{12}^2 m_1+m_2 s_{12}^2)+c_{13}^2 m_3 \right ] + s_{23}^2 ( c_{12}^2 m_2+m_1 s_{12}^2 ) \\ & & + 2 \ , c_{12 } c_{23 } s_{12 } s_{13 } s_{23 } e^{i \delta } \ , ( m_2-m_1 ) , \nonumber \\ \left ( u_{\mathrm{pmns}}^ * \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^\dagger \right)_{21 } & = & c_{12 } c_{13 } s_{12 } c_{23 } ( m_2-m_1)+c_{13 } s_{13 } s_{23 } e^{i \delta } \left [ e^{-2 i \delta } m_3 - m_2 + c_{12}^2 ( m_2-m_1 ) \right ] , \nonumber\\\\ \left ( u_{\mathrm{pmns}}^ * \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^\dagger \right)_{31 } & = & c_{12 } c_{13 } s_{12 } s_{23 } ( m_1-m_2)+c_{13 } s_{13 } c_{23 } e^{i \delta } \left [ e^{-2 i \delta } m_3 - m_2 + c_{12}^2 ( m_2-m_1 ) \right ] , \nonumber\\\\ \left ( u_{\mathrm{pmns}}^ * \ , \hat{m}_\nu \ , u_{\mathrm{pmns}}^\dagger \right)_{32 } & = & c_{23 } s_{23 } \left [ ( s_{12}^2 - e^{2 i \delta } c_{12}^2 s_{13}^2)(m_2 - m_1 ) + c_{13}^2 ( m_3 - e^{2 i \delta } m_2 ) + ( e^{2 i \delta } -1 ) m_2 \right ] , \nonumber\\ & & - c_{12 } s_{12 } s_{13 } e^{i \delta } ( c_{23}^2 - s_{23}^2)(m_2 - m_1 ) . \end{aligned}\ ] ] j. adam _ et al . _ [ meg collaboration ] , phys . lett . * 107 * ( 2011 ) 171801 [ arxiv:1107.5547 [ hep - ex ] ] . j. adam _ et al . _ [ meg collaboration ] , arxiv:1303.0754 [ hep - ex ] . a. m. baldini , f. cei , c. cerri , s. dussoni , l. galli , m. grassi , d. nicolo and f. raffaelli _ et al . _ , arxiv:1301.7225 [ physics.ins-det ] . a. blondel , a. bravar , m. pohl , s. bachmann , n. berger , m. kiehn , a. schoning and d. wiedner _ et al . _ , arxiv:1301.6113 [ physics.ins-det ] . d. glenzinski [ mu2e collaboration ] , aip conf . * 1222 * ( 2010 ) 383 . r. m. carey _ et al . _ [ mu2e collaboration ] , fermilab - proposal-0973 . m. aoki [ deeme collaboration ] , pos ichep * 2010 * ( 2010 ) 279 . y. g. cui _ et al . _ [ comet collaboration ] , kek-2009 - 10 . the prime working group , unpublished ; loi to j - parc 50-gev ps , loi-25 , http://j-parc.jp/researcher/hadron/en/pac_0606/pdf/p20-kuno.pdf b. aubert _ et al . _ [ babar collaboration ] , phys . * 104 * ( 2010 ) 021802 [ arxiv:0908.2381 [ hep - ex ] ] . u. bellgardt _ et al . _ [ sindrum collaboration ] , nucl . b * 299 * ( 1988 ) 1 . k. hayasaka _ et al . _ , phys . b * 687 * ( 2010 ) 139 [ arxiv:1001.3221 [ hep - ex ] ] . w. bertl _ et al . _ [ sindrum ii collaboration ] , eur . j. * c47 * ( 2006 ) 337 - 346 h. okada and t. toma , phys . d * 86 * , 033011 ( 2012 ) [ arxiv:1207.0864 [ hep - ph ] ] . p. s. b. dev and a. pilaftsis , phys . rev . d * 86 * ( 2012 ) 113001 [ arxiv:1209.4051 [ hep - ph ] ] . m. aoki , j. kubo and h. takano , phys . d * 87 * , 116001 ( 2013 ) [ arxiv:1302.3936 [ hep - ph ] ] . y. kajiyama , h. okada and t. toma , phys . d * 88 * , 015029 ( 2013 ) [ arxiv:1303.7356 ] . s. kanemura , t. matsui and h. sugiyama , phys . b * 727 * , 151 ( 2013 ) [ arxiv:1305.4521 [ hep - ph ] ] . s. s. c. law and k. l. mcdonald , jhep * 1309 * , 092 ( 2013 ) [ arxiv:1305.6467 [ hep - ph ] ] . m. hirsch , r. a. lineros , s. morisi , j. palacio , n. rojas and j. w. f. valle , jhep * 1310 * , 149 ( 2013 ) [ arxiv:1307.8134 [ hep - ph ] ] . d. restrepo , o. zapata and c. e. yaguna , jhep * 1311 * , 011 ( 2013 ) [ arxiv:1308.3655 [ hep - ph ] ] . e. ma , i. picek and b. radovcic , arxiv:1308.5313 [ hep - ph ] . m. lindner , d. schmidt and a. watanabe , arxiv:1310.6582 [ hep - ph ] . h. okada and k. yagyu , arxiv:1311.4360 [ hep - ph ] . j. kubo , e. ma and d. suematsu , phys . b * 642 * , 18 ( 2006 ) [ hep - ph/0604114 ] . d. aristizabal sierra , j. kubo , d. restrepo , d. suematsu and o. zapata , phys . d * 79 * ( 2009 ) 013011 [ arxiv:0808.3340 [ hep - ph ] ] . d. suematsu , t. toma and t. yoshida , phys . d * 79 * , 093004 ( 2009 ) [ arxiv:0903.0287 [ hep - ph ] ] . a. adulpravitchai , m. lindner and a. merle , phys . rev . d * 80 * ( 2009 ) 055031 [ arxiv:0907.2147 [ hep - ph ] ] . j. a. casas and a. ibarra , nucl . b * 618 * , 171 ( 2001 ) [ hep - ph/0103065 ] . a. ilakovac and a. pilaftsis , nucl . b * 437 * ( 1995 ) 491 [ hep - ph/9403398 ] . f. deppisch and j. w. f. valle , phys . d * 72 * ( 2005 ) 036001 [ hep - ph/0406040 ] . f. deppisch , t. s. kosmas and j. w. f. valle , nucl . b * 752 * ( 2006 ) 80 [ hep - ph/0512360 ] . a. ilakovac and a. pilaftsis , phys . d * 80 * ( 2009 ) 091902 [ arxiv:0904.2381 [ hep - ph ] ] . r. alonso , m. dhen , m. b. gavela and t. hambye , jhep * 1301 * ( 2013 ) 118 [ arxiv:1209.2679 [ hep - ph ] ] . d. n. dinh , a. ibarra , e. molinaro and s. t. petcov , jhep * 1208 * ( 2012 ) 125 [ arxiv:1205.4671 [ hep - ph ] ] . a. ilakovac , a. pilaftsis and l. popov , phys . d * 87 * ( 2013 ) 053014 [ arxiv:1212.5939 [ hep - ph ] ] . a. abada , d. das , a. vicente and c. weiland , jhep * 1209 * ( 2012 ) 015 [ arxiv:1206.6497 [ hep - ph ] ] . p. s. b. dev , c. -h . lee and r. n. mohapatra , phys . d * 88 * ( 2013 ) 093010 [ arxiv:1309.0774 [ hep - ph ] ] . e. arganda and m. j. herrero , phys . d * 73 * , 055003 ( 2006 ) [ hep - ph/0510405 ] . j. hisano , t. moroi , k. tobe and m. yamaguchi , phys . d * 53 * , 2442 ( 1996 ) [ hep - ph/9510309 ] . y. kuno and y. okada , rev . * 73 * ( 2001 ) 151 [ hep - ph/9909265 ] . e. arganda , m. j. herrero and a. m. teixeira , jhep * 0710 * ( 2007 ) 104 [ arxiv:0707.2955 [ hep - ph ] ] . h. c. chiang , e. oset , t. s. kosmas , a. faessler and j. d. vergados , nucl . a * 559 * ( 1993 ) 526 . t. s. kosmas , s. kovalenko and i. schmidt , phys . b * 511 * ( 2001 ) 203 [ hep - ph/0102101 ] . m. c. gonzalez - garcia , m. maltoni , j. salvado and t. schwetz , jhep * 1212 * ( 2012 ) 123 [ arxiv:1209.3023 [ hep - ph ] ] . d. v. forero , m. tortola and j. w. f. valle , phys . d * 86 * ( 2012 ) 073012 [ arxiv:1205.4018 [ hep - ph ] ] . g. l. fogli , e. lisi , a. marrone , d. montanino , a. palazzo and a. m. rotunno , phys . d * 86 * ( 2012 ) 013012 [ arxiv:1205.5254 [ hep - ph ] ] . d. schmidt , t. schwetz and t. toma , phys . d * 85 * ( 2012 ) 073009 [ arxiv:1201.0906 [ hep - ph ] ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1303.5076 [ astro-ph.co ] . s. kashiwase and d. suematsu , phys . d * 86 * , 053001 ( 2012 ) [ arxiv:1207.2594 [ hep - ph ] ] . s. kashiwase and d. suematsu , eur . j. c * 73 * , 2484 ( 2013 ) [ arxiv:1301.2087 [ hep - ph ] ] . y. farzan , phys . d * 80 * ( 2009 ) 073009 [ arxiv:0908.3729 [ hep - ph ] ] . m. gustafsson , s. rydbeck , l. lopez - honorez and e. lundstrom , phys . d * 86 * ( 2012 ) 075019 [ arxiv:1206.6316 [ hep - ph ] ] . s. -y . ho and j. tandean , phys . d * 87 * ( 2013 ) 095015 [ arxiv:1303.5700 [ hep - ph ] ] . a. arhrib , y. -l . s. tsai , q. yuan and t. -c . yuan , arxiv:1310.0358 [ hep - ph ] .

single component nonrelativistic dissipative fluids are treated independently of reference frames and flow - frames . first the basic fields and their balances , then the related thermodynamic relations and the entropy production are calculated and the linear constitutive relations are given . the usual basic fields of mass , momentum , energy and their current densities , the heat flux , pressure tensor and diffusion flux are the time- and spacelike components of the third order mass - momentum - energy density - flux four - tensor . the corresponding galilean transformation rules of the physical quantities are derived . it is proved , that the non - equilibrium thermodynamic frame theory , including the thermostatic gibbs relation and extensivity condition and also the entropy production is independent of the reference frame and also the flow - frame of the fluid . the continuity - fourier - navier - stokes equations are obtained almost in the traditional form if the flow of the fluid is fixed to the temperature . this choice of the flow - frame is the thermo - flow . a simple consequence of the theory is that the relation between the total , kinetic and internal energies is a galilean transformation rule .

1508.00121

the concept of absolute , motion independent time is the result of our experience on the surface of the earth , where the motions are slow and there is an apparently fixed background reference frame . however , the space is relative also in this case , it is different for different observers : the so - called non - relativistic space - time is galilean relativistic . with the help of the notions of special relativistic space - time , the heuristic concepts of our everyday experience can be clarified and an exact mathematical model of galilean relativistic space - time can be formulated @xcite . in spite of our seemingly self confident knowledge about slow motions of everyday life , such kind of exact space - time model is unavoidable to formulate and handle fundamental physical principles . if a physical phenomena happens without interacting with an observer , then one may find useful an adequate mathematical model , where this independent existence is reflected . furthermore , considering the tempestuous history of concepts about space and time , this kind of model seems to be unavoidable . the mathematical notions in a physical theory are not simple tools , they are the building blocks , the material of the whole building . the shape and flexibility of blocks contributes to the stability of the construction . with properly formed blocks one can spare the mortar of imagination for a better use : for the design of the building and for other conceptual questions . a good example of the importance of proper building material may be elimination of space coordinates in classical field theories with the help of more abstract notions of vectors and tensors . coordinate free concepts improve the focus on fundamental notions and are also helpful to develop better calculation methods in engineering . this work is an argument why reference frames should be avoided , and it provide tools how they can be eliminated , similarly to coordinates , from theories of continua . our example is single component fluid mechanics in this case . time is absolute in galilean - relativistic space - time , it is the same for different observers , in other words , it is independent of reference frames . in the following the adjective _ absolute _ will be used in this sense for other reference frame free physical quantities , too . other similar concepts , like objective or covariant , are mostly avoided . there are two points where our treatment is simplified . first , it has been long known that space and time are not vector spaces but better represented by affine spaces , because none of them have a canonical center @xcite . in this work the affine character of galilean space - time is not crucial , therefore we use a vector representation for it . second , we do not deal with the exact mathematical representation of physical units , nevertheless , it is interesting both from conceptual and practical points of view @xcite . the definition of the complete space - time model is given in the appendix . _ absolute time is not a subspace of the galilean relativistic space - time_. this is the critical , hidden problem of the usual superficial representations of galilean relativistic space - time . at first it seems to be less important than the mentioned center and unit dependencies . however , if we represent space - time by a cartesian product of the one dimensional time and three dimensional space , or even worse by @xmath0 , our theory is unavoidably reference frame dependent . the proper representation of the absolute time has consequences . one of them is , that in non - relativistic physical theories space - time four - covectors can not be identified with four - vectors and they transform differently when the reference frame changes . an other consequence is that the trace of a second order space - time four - tensor or a four - cotensor does not exist ; only a mixed four - tensor has a trace . in this respect the galilean relativistic space - time is not a specific low speed limit of special relativistic or general relativistic ones . the operations with space - time quantities are different from the usual relativistic operations , require care despite familiarity in special or general relativistic calculations . for example , here some components of four - quantities , and not only the complete quantity , may be reference frame independent : the timelike component of a vector and the spacelike component of a covector is absolute . several problematic aspects of non - relativistic physics are due to inappropriate space - time models : 1 . _ principle of material frame indifference . _ this principle formulates a simple and evident(looking ) property , postulating that the material must be independent of the reference frame , therefore the physical quantities , laws , equations of motions and material functions must reflect this requirement . in the usual formulations of the principle , the required invariance under various transformation rules is restricted to three dimensional spatial vectors and tensors . according to the usual explanations this restriction is due to the absolute time . for point masses one expects invariance for galilean transformations related to inertial reference frames . for continua this is not enough , therefore invariance under rigid body motions of the reference frame is required . this most accepted formulation is due to walter noll ( @xcite , see also in @xcite ) . the principle and also its formulation initiated a long and unfinished discussion among those who are interested in the fundamental aspects of continuum physics . here we give an incomplete list of the most important looking related works @xcite . + with the help of the galilean relativistic space - time model one can show that the formal invariance ( independency of the angular velocity of the rotating reference frame ) is a wrong requirement , because reference frame independency is not always invariance , sometimes it requires that the characteristics of the relative motion appear in the transformation rules @xcite , as it is evident in case of galilean transformations . moreover , from a space - time point of view the formulation of material frame indifference of noll is self - contradictory @xcite . deformation measures . _ not only the constitutive functions , but also the basic fields of continuum mechanics are expected to be frame indifferent . one of the most discussed quantities in this respect is the finite elastic deformation . one can define infinite number of deformation measures that are objective in the sense of noll , and even more that are not . it is remarkable that the space - time requirements distinguish a single concept of deformation @xcite , which turned out to be advantageous from other points of view , too @xcite . 3 . _ flow - frames . _ a different problem concerns the basic field of fluid theory , the velocity of the fluid . according to a recent idea of howard brenner , the velocities in the mass , momentum and energy balances are not evidently the same . this idea has a long history @xcite and it is analogous to the problem of flow - frames in case of relativistic fluids @xcite . what is the velocity of the fluid ? what defines the motion from a physical point of view ? the mass , the momentum , the energy or something else ? do we have a choice ? in the usual fluid equations the velocities are relative velocities , therefore one must consider the role of the space - time to get a reliable answer . compatibility with special relativity . _ one may like to see the correspondence of physical quantities of galilean relativistic fluids and special relativistic fluids beyond the usual comparison of relative quantities . in special relativity the basic quantity is the covariant second order energy - momentum density four - tensor . the energy is the time - timelike part of this tensor and the transformation properties of the energy are consequences of this fact . what could be a similar physical quantity in galilean relativity _ the origin of transformation rules . _ the energy , e.g. the kinetic energy , when expressed by the relative velocity is definitely relative and frame dependent . is there an objective physical quantity behind the energy ? do we know the transformation rules of the non - relativistic energy ? what about the transformation rules of other physical quantities , like tensors and covectors _ compatibility with kinetic theory . _ the consistency with statistical physics , more properly with kinetic theory raises some questions , too . for example momentum series expansion of kinetic theory is informative regarding transformation rules of macroscopic quantities @xcite . moreover , the method of derivation of continuum equations ( by chapman - enskog or momentum series expansion ) is informative regarding the thermodynamics of fluids . for example the internal energy is determined in a close relation with the pressure . is there an aspect in phenomenological continuum physics , where the hierarchical structure of momentum equations seems to be natural ? on the other hand , at the same time , nonrelativistic kinetic theory is considered as reference frame dependent , mostly due to macroscopic , phenomenological considerations ( e.g. transformation properties of the heat flux @xcite ) . this is a fundamental contradiction with general aspects of relativistic theories . _ thermodynamics . _ one may think , that thermodynamic relations are inherently absolute , because they express directly properties of materials . covariance of thermodynamic relations is expected in the relativistic case , this is an important question since the birth of relativity theory ( see the problem of temperature of moving bodies e.g. in @xcite ) . interestingly , in the non - relativistic theory this question is rarely treated ( but see e.g. @xcite ) . the galilean - covariance of thermodynamical relations is not evident . therefore it is not clear whether the gibbs relation is independent of reference frames or not . one may expect also that dissipation is absolute , can not depend on the motion of an external observer . _ natural philosophy . _ finally we emphasize that space - time formulation of a theory reorganizes our attitude to some fundamental concepts . in the framework of the galilean relativistic space - time model reference frames , including inertial reference frames , are secondary , derived concepts . they are not fundamental , moreover should be avoided in case of general problems , contrary to the common belief @xcite . in the following we give a reference frame independent theory of single component dissipative galilean - relativistic fluids starting with the absolute basic fields , their balances , the thermodynamic relations and finally calculating the entropy production . along the absolute treatment the usual relative formulas and the corresponding transformation rules are calculated , together with the conditions that lead to the relative continuity navier - stokes fourier system of equations . in this work a particular version of abstract indexes of penrose @xcite is introduced , that hopefully contributes to the transparency of the reference frame free meaning of the equations of galilean relativistic fluid mechanics . we introduce three different indexes . the four - vectors and tensors of the galilean relativistic space - time are denoted by upper @xmath1 indexes , the covectors and cotensors by lower @xmath1 indexes . overlined indexes from the beginning of the alphabet , @xmath2 , denote spacelike four - vectors or spacelike parts of four - tensors , when positioned upper and spacelike four - covectors or four - cotensor components when positioned lower . it is important that the upper or lower position of the @xmath1 indexes is fixed , they can not be pulled or pushed up or down , there is no canonical , observer independent identification between vectors and covectors . on the the hand the position of spacelike @xmath3 indexes can be changed , because of the euclidean sturcture of space vectors . the indexes of the usual relative vectors and tensors , those that are related either to the fluid or the external reference frame are denoted distinctively by @xmath4 . these indexes are used whenever a single relative vector ( typically the relative velocity ) is present in the formula . the space - time model , the calculation rules and the notation are explained in the appendices . the following sections require the detailed knowledge of appendix a , where the foundations of the galilean - relativistic space - time model are given and also of appendix b , where the most important calculation rules are derived and summarized . the fundamental balances of special relativistic fluids are expressed by four - divergences of the four - densities of the extensive physical quantities . in nonrelativistic physics these four - densities and four divergences are hidden , behind the usual relative formulation : the change of the extensive quantity is due to the local and simultaneous change inside the considered spatial region ( timelike part ) and the outward or inward flow ( spacelike part ) . in the following we derive the usual relative forms of the balances in a galilean relativistic framework . therefore , we express the four - vector field @xmath5 with the help of its relative parts due to the four - velocity field of the fluid @xmath6 , introducing its @xmath7-form , @xmath8 , into the balance where @xmath9 is a space - time index and @xmath10 is the spacelike index . @xmath11 and @xmath12 are the timelike and @xmath7-spacelike parts of @xmath5 , @xmath13 and @xmath14 are the @xmath7-timelike and spacelike parts of space - time derivation @xmath15 ( see appendix a ) . the absolute balance , is expressed by the @xmath7-relative parts of the four - vector @xmath5 and four - covector @xmath15 . a relative velocity field plays a central role in the local and substantial relative balances of a fluid . this velocity is the relative velocity of the fluid and an inertial observer , therefore it is the difference of the four - velocity of the fluid , @xmath6 , and the constant four - velocity field of an inertial external observer , @xmath16 . the _ local _ form of the balance is obtained with the @xmath17-form of the derivative and the four - vector field of the physical quantity : the _ substantial _ form of the balance is obtained by substituting the four - velocity of the fluid , @xmath6 , with the relative velocity of the fluid to the observer , @xmath18 : because @xmath19 and @xmath20 . in the local balances the time derivative is denoted as @xmath21 , in the substantial ones one may use a different notation , @xmath22 , or the traditional dot , therefore @xmath23 . in the following we apply a distinct notation , the indexes @xmath24 , for spacelike relative vectors and covectors , in formulas related to an external observer . both the local and the substantial balances and can be written this way . the relative form of the local balance is and the substantial follows as it is easy to check that local balances can be obtained from substantial ones and vice versa with the galilean transformation rules of covectors and vectors @xmath25 , and @xmath26 ( see appendix b ) . the substantial and local forms of the absolute balance depend on the components of the four - derivative and the four - vector splitted by the four - velocities of the fluid , @xmath6 , and the external observer , @xmath27 . it is important to remember , that the velocity field of the observer is given , the velocity field of the fluid is to be determined , we are looking for differential equations to determine it . the absolute equations are independent of _ any _ observer @xcite . the fundamental physical information is given by the four - vector @xmath5 , its four - divergence , and by the velocity field of the medium , @xmath6 . up to now , we did not say anything about the physical meaning of the four - velocity field of the medium . we will see , that the interpretation of the four - velocity field of the fluid requires thermodynamic considerations . however , first of all we should find the space - time representation of the physical quantities like mass , momentum , energy , heat flux , etc .. we will see , that actually we don not have several quantities , in galilean relativity there is one , single , absolute physical quantity that characterizes a single component fluid . in special relativistic fluid mechanics the physical quantities may be scalars , four - vectors and also higher order objects , like the second order energy - momentum tensor . it is straightforward to assume that in galilean relativistic space - time an analogous quantity exists like mass - momentum or energy - momentum . however , galilean relativistic space - time is a more restricted framework than the special relativistic one , because the space - time vectors and covectors are related only by the linear structure , there is no euclidean or pseudo - euclidean structure on the space - time level which would admit their identification . therefore , the divergence of a four - covector or four - cotensor field does not exist independently of a reference frame . similarly , there are no symmetric or antisymmetric parts of mixed four - tensors independently of reference frames . on the other hand , the observed empirical , space - time related properties of relative physical quantities are reflected in their transformation properties . in case of a reference and flow - frame independent galilean relativistic theory these transformation properties can be deduced , they are consequences of the properties of space - time quantities . this theoretical framework must be harmonized with the empirical knowledge . for example , if the galilean transformation rule of a physical quantity is like the transformation rule of the position , then this quantity maybe a spacelike part of a four vector . in this identification process the energy plays a distinguished role . one may come to the idea , that the relation of kinetic , internal and total energies is a galilean transformation rule . the total energy is the sum of internal and kinetic energies : from the point of view of a comoving reference frame , fixed to the flow of the medium , the energy is the internal energy , but for an external , inertial observer the kinetic energy have to be added . in a continuum the relation of the total energy density @xmath28 and the density of the internal energy @xmath29 is the following : due two reference frames having relative velocity @xmath30 . these preliminary considerations outline the framework toward understanding the relation of energy density and the space - time structure . in appendix b we have calculated the transformation rules of four - vectors , four - covectors and different second order tensors . scrutinizing the derived formulas one can observe , that a quadratic relative velocity in a transformation rule requires at least second order tensors . the time - timelike component of a second order four - tensor , the space - spacelike component of a second order four - cotensor and the time - spacelike or space - timelike component of a mixed second order tensor transforms quadratically with the relative velocity . however , the existence of balance form evolution equation restrict the possible choices , e.g. a second order four - cotensor field does not have a divergence in galilean relativistic space - time , therefore can not have a balance . however , energy may be also related to higher order tensors , like the time - timelike component of a third order , @xmath31 valued mixed tensor field . a further physical requirement is the compatibility to kinetic theory of gases . in the usual nonrelativistic theory the internal energy is the trace of the second order central momentum of the one particle probability distribution function by the relative velocity @xcite . kinetic theory determines the energy in this respect , using also the ideal gas equation of state . from a space - time point of view an energy connected to the space - spacelike component of a second order four - tensor is not sufficient , because the energy balance requires energy flux , this way an additional tensorial order . in a galilean relativistic continuum theory based on a second order four - tensor ( the mass - momentum density tensor ) one must introduce an independent , vectorial energy balance for the internal energy , therefore can not be obtained as a transformation rule @xcite . one may conclude , that in order to fulfill all these requirements , an @xmath32 valued , third order tensor field emerges . the components of the four - divergence of this quantity must give the fundamental balances of mass , momentum and energy together . in the next sections we will show that such a basic physical quantity leads to a consistent theory . astonishingly , one may obtain a rather similar theory assuming that the basic physical quantity is the above mentioned third order mixed tensor field . in both cases the transformation rules are the same and the entropy production is obtained in the same form ( after a long calculation ) . the compatibility with the usual energy concept of kinetic theory compels us to consider in this paper a third order four - tensor as the basic physical quantity of a galilean relativistic theory of single component fluids . looking back to the relation of total , internal and kinetic energies one may wonder about expected properties of transformation rules . considering a third reference frame , it is easy to see , that the formula is not a transitive rule . our idea of an energy transformation rule is seemingly wrong . in the following we will see , that transitivity may be expected , our formula is an oversimplified version of a more complicated relation . let us consider the tensor field @xmath33 , the _ mass - momentum - energy density - flux third order four - tensor _ of a single component fluid . we assume that the tensor field is symmetric in the second and third order , as it is indicated by the symbol @xmath34 . in the following we do not explicitly denote this symmetry , @xmath35 , only we refer to it if necessary . this tensor can be written in the following general @xmath7-form , with the components obtained by the four - velocity @xmath6 : where these components are the second order four - tensor of densities and the tensor of fluxes or current densities , that is @xmath36 is the _ mass - momentum - energy density tensor _ and @xmath37 is the _ diffusion - pressure - heat flux _ @xmath38 is the time evaluation and @xmath39 is the @xmath7-projection for taking the @xmath7-spacelike parts of vectors . the further notations are : * @xmath40 is the time - time - timelike part of the mass - momentum - energy density - flux tensor , the _ mass density _ or _ density_. * @xmath41 is the time - time - spacelike part of the mass - momentum - energy density - flux tensor , the _ momentum density_. because of the symmetry of @xmath42 it is equal to the time - space - timelike part , @xmath43 . * @xmath44 is the _ energy density tensor _ , the time - space - spacelike part of @xmath42 . * @xmath45 is the _ ( self)diffusion flux _ , the space - time - timelike part of @xmath42 . * @xmath46 is the _ pressure _ , the space - time - spacelike part of @xmath42 . because of the symmetry of the third order tensor it is equal to @xmath47 . * @xmath48 is the _ heat flux tensor _ , the space - space - spacelike part of @xmath42 . the usual energy density and heat flux can be introduced according to the kinetic theory reducing the order of the corresponding tensors by two : * @xmath49 is the _ energy density _ , * @xmath50 is the _ heat flux_. all these quantities are defined with the help of the velocity field of the medium , @xmath6 . so they are independent of any external observers . the time- and spacelike components of a physical quantity obtained by an inertial observer of constant four - velocity @xmath27 are relative quantities . the galilean transformation rule of a physical quantity means expressing these relative quantities with the aid of the components corresponding to the four velocity @xmath6 . in appendix b the transformation rules of first and second order tensors are calculated by splitting the @xmath7-form of the tensors with the four - velocity @xmath27 . the same procedure can be applied for the third order mass - momentum - energy density - flux tensor . in transformation formulas the relative velocity of the fluid to the observer , @xmath51 , appears naturally . the ( mass)density , @xmath52 , is a galilean - scalar , it is invariant : the momentum density and the mass density are components of an absolute four - vector , therefore the momentum density transforms like the spatial component of a four - vector : the ( self)diffusion flux transform like the momentum density , because together with mass density they are components of a four - vector , too : the energy density is not a galilean - scalar : the transformation rule of the pressure tensor is : the most complicated transformation rule belongs to the heat flux : finally , we summarize these rules with the help of the usual 3-index notation : two unusual quantities appeared in our reference frame free approach . one of them is the conductive current density of the mass , @xmath53 , the ( self)diffusion flux . the other one is the momentum density , @xmath54 . one can reveal their role with the help of the absolute and relative balances of the fluid . the balances of mass , momentum and energy are the components derived from the divergence of the mass - momentum - energy density - flux tensor . the @xmath7-form of the divergence is : here the dot denotes the @xmath7-time derivative : @xmath55 . the timelike part of the mass - momentum - energy balance is the mass - momentum balance : the timelike part of the mass - momentum balance ( the time - timelike part of the mass - momentum - energy balance ) is the mass balance : the @xmath7-spacelike part of is the momentum balance : the balance of energy is the @xmath7-space - spacelike part of the mass - momentum - energy balance , more properly the trace of that : one may obtain the substantial form of the balances , and by using the relative velocity @xmath51 of the fluid to the inertial reference frame . with the usual three - index notation they are these are the mass , momentum and energy balances of the fluid . the same expressions are obtained if the transformation rules are applied to exchange the inertial reference frame related @xmath17-quantities with medium related @xmath7-quantities in the local balances . the underlined terms indicate the differences between the usual balances of fluids ( see e.g. @xcite ) and the ones above . in these terms two additional quantities appear , the ( self)diffusion flux , @xmath56 , and the momentum density @xmath57 . therefore the usual closure procedure , prescribing constitutive functions for the pressure tensor , @xmath58 , and the heat flux , @xmath59 , is insufficient , further conditions are necessary to close the system of equations . to this end one should investigate the thermodynamics of fluids from the point of view of galilean relativistic space - time . the title of the section reflects the paradoxical dilemma of thermodynamics ( or thermostatics ? ) considering the motion related mechanical properties . the literature of thermodynamics rarely introduces velocity as a state variable . in rational continuum mechanical investigations this possibility is forbidden , because the usual theory does not consider classical thermodynamics as a meaningful starting point- moreover , according to the usual formulation of the principle of material frame indifference relative velocity can not be a variable in constitutive functions ( see e.g. @xcite p43 ) . from the point of view of galilean relativity these statements require further analysis @xcite . our fundamental point of view is , that classical `` equilibrium '' thermodynamics is actually time dependent : it is the homogeneous , discrete counterpart of irreversible thermodynamics . moreover , the thermodynamic relations of relativistic kinetic theory can be instructive . the essential aspects are treated in @xcite . the entropy density is a four - vector field , whose @xmath7-form is @xmath60 , with the timelike component , the entropy density , @xmath61 and the @xmath7-spacelike component , the entropy flux @xmath62 . the entropy density is the function of the mass - momentum - energy density : @xmath63 . this is an absolute relation , does not depend neither on a reference frame nor on the velocity field of the medium @xmath6 , because both the entropy density and the density tensor of the extensives , @xmath64 and @xmath65 , are absolute . the derivative of @xmath64 is the symmetric second order four - cotensor of the thermodynamic intensives , the _ chemical potential - thermovelocity - temperature four - cotensor _ , and denoted by @xmath66 in the following . therefore , @xmath67 this derivative is the _ gibbs relation _ and in the following it is treated by differentials according to thermodynamic tradition : the @xmath7-form of the chemical potential - thermovelocity - temperature cotensor is : if the tensor @xmath36 is split to components according to , then * @xmath68 is the intensive quantity related to the mass density , * @xmath69 is the intensive quantity belonging to the momentum density @xmath70 , * @xmath71 is the intensive quantity related to the energy density tensor @xmath72 . one can pull closer the treatment to the usual approach assuming in this case then the physical quantities are * the _ reciprocal temperature _ , @xmath73 : * the _ chemical potential _ @xmath74 is related to the entropic intensive of the mass density : * finally , _ thermovelocity _ is defined with the help of the momentum density related intensive therefore the @xmath7-form of the absolute gibbs relation can be calculated as : with the help of the @xmath7-split quantities finally the following form is obtained : this formula is analogous to the relativistic gibbs relation suggested in @xcite , where the compatibility with kinetic theory is considered . the enthalpy in the relativistic case is substituted by the mass density here . the legendre transformation of the entropy density four - vector defines the _ conjugated entropy _ , @xmath75 : let us give the @xmath7-form of the conjugated entropy in the following form where @xmath76 . then the absolute timelike part of the four - vector equation , the _ extensivity relation _ is this expression defines the thermostatodynamic ( thermostatic ) _ pressure _ , @xmath77 . the @xmath7-spacelike part , the entropy flux , is obtained with the help of the @xmath7-projection , @xmath78 : the extensivity relation , and the gibb relation together result in the _ gibbs duhem relation _ : where @xmath79 is the enthalpy density . the @xmath7-form of the absolute gibbs relation , may be interpreted by taking into account that the relative intensives @xmath80 are functions of the relative densities @xmath81 . therefore , one of the second order mixed partial derivatives of the internal energy @xmath82 leads to the following maxwell - relation : this is a partial differential equation for @xmath83 , whose general solution is where @xmath84 and the four velocity @xmath85 are arbitrary functions of @xmath52 and @xmath64 . the above equation of state demonstrates , that the dependence of the thermovelocity on the momentum density @xmath54 is restricted . the equation of state is very simple if @xmath86 . in this case in the following we will call this relation _ momentum condition_. the entropy density and the entropy flux are time- and spacelike components of the entropy four - vector . the components according to an external inertial observer with constant @xmath27 four - velocity lead to the corresponding transformation rules between the comoving reference frame to an inertial laboratory one : here @xmath51 is the relative velocity field . the general transformation rules of four - vectors in and the representation of @xmath75 in result in the components of the absolute chemical potential - thermovelocity - temperature four - cotensor , @xmath87 , are transformed acording to of appendix b : it is worth to analyze thermovelocity with more details . the spacelike part of a second order cotensor is absolute : @xmath88 . its @xmath17-timelike component is therefore , for the thermovelocity one obtains with the help of and : written with relative indexes this is the transformation rule . the extensivity relation is absolute and therefore galilean invariant , because it is the absolute timelike part of an four - vector equation : that can be verified also directly with the derived particular transformation rules of the physical quantities . a similar calculation shows , that the entropy flux transforms as a space vector . where the previously derived transformation rules of @xmath73 , @xmath89 , @xmath74 , @xmath56 , @xmath90 , @xmath91 , @xmath77 s @xmath92 were applied . finally , the gibbs relation is galilean invariant , too , because it is obtained applying a cotensor to a tensordifferential , where @xmath51 and @xmath93 are the relative velocities of the medium with respect to inertial observers with constant @xmath17 and @xmath94 four - velocities . the last term of the gibbs relation is galilean invariant in itself , like the @xmath95 term , because @xmath96 and @xmath97 . with the @xmath7-forms of the gibbs relation and the entropy flux , and , the entropy balance can be expressed as : substituting the balances of mass , momentum and energy , , and , then using also the gibbs - duhem relation , one obtains , that this inequality of the absolute entropy production is the second law of galilean relativistic single component fluids . this is a quadratic expression . the first term expresses the diffusion related dissipation , the constitutive quantity is the ( self)-diffusion flux , @xmath53 . in the second part of the product there is the gradient of the chemical potential divided by the temperature . the second term is the thermal part of the dissipation , where the constitutive quantity is the heat flux , @xmath59 . the gradient of the reciprocal temperature , @xmath98 is the thermodynamic force . the third term is related to mechanical dissipation , with the pressure tensor , @xmath58 , as constitutive quantity , and with the gradient of a velocity as thermodynamic force . the relevant velocity is the sum of the @xmath6 four - velocity and the @xmath83 thermovelocity . let us observe , that @xmath6 appears explicitely only here , in the velocity gradient part of the dissipation . the fourth term is new . here @xmath99 can be the constitutive quantity . therefore the inequality of the absolute entropy production seems to be solvable in the sense that in every term there is a constitutive quantity , therefore one can introduce thermodynamic fluxes and forces and assume a linear relationship with positive definite coupling between them . however , the thermodynamic conditions alone do not close the system of balances . enumerating the variables one can conclude that there is no differential equation either for the momentum or for the velocity of the continuum . the thermovelocity should be determined or fixed , too . let us recognize here , that up to know we did give any physical condition that would connect @xmath6 to the medium , therefore it is not known yet the physical meaning of the it is not known yet the physical meaning of the velocity field of the fluid . for the sake of simplicity in the following we do not investigate the last term of the entropy production , therefore the spacelike part of the conjugated entropy is considered parallel to the thermovelocity , as it is customary in special relativistic kinetic theory . in particular our assumption is , that then the entropy production simplifies to the following form : how could we reduce the number of unknowns to obtain a closed system of equations ? in addition to the usual system of basic variables , the density , the internal energy density and the velocity field , @xmath100 , we have two other fields : the momentum density and the thermovelocity , @xmath54 and @xmath83 . we have already discussed the equation of state for the thermovelocity . on the other hand up to know we did not fix what is the meaning of the four - velocity of the fluid . this choice , the physical definition of the fluid velocity is the _ flow - frame_. there are several possibilities . we may fix the velocity field @xmath6 to one of the extensive quantities of the fluids , e.g. to the mass ( @xmath101 ) , to the energy ( @xmath102 ) and also to the momentum density ( @xmath103 ) , but it can be fixed by other , more complicated ways , too . with the previous definitions relative velocity is the flow of the mass , energy or momentum related to an external observer . an example of more complicated choices is mixture of energy- and particle(mass)-flow @xcite . the different choices are not equivalent from a practical point of view . looking at the above expression of the entropy production one may recognize , that a simple form of constitutive functions is obtained fixing the four velocity of the fluid to the thermovelocity , that is @xmath104 . this flow - frame is called _ thermo - flow_. the entropy production with thermo - flow is the following : if the momentum condition equation of state , , is applied , then @xmath105 is the consequence of @xmath106 . therefore in this case a thermo - flow is necessarily a momentum - flow , too . a fluid , defined with thermo- and momentum - flows is called _ classical fluid_. then the substantial mass- , momentum- , and energy balances become simpler , too : and the entropy production is for classical fluids the velocity dependent gibb - relation , , reduces to the usual form in case of inertial observers @xmath107 and @xmath108 , therefore @xmath109 . as a consequence , the transformed form of the gibbs relation , , becomes similarly the extensivity relation , , can be written in the form these formulas show well , that for classical fluids , the so called total energy density is the @xmath17-energy density @xmath110 corresponding to the inertial reference frame , and the @xmath7-energy density @xmath82 is the internal energy . they are related by the transformation rule . similarly , in the transformation rule of chemical potential , @xmath111 , we could call @xmath112 as total and @xmath74 as internal chemical potential . in the customary extensivity relation the term of the kinetic energy is merged into pressure , instead of chemical potential . in fluid mechanics the extensivity relation appears as bernoulli equation with the so called dynamic pressure . the @xmath17-form of the entropy flux of classical fluids provides the _ total entropy flux _ : and the @xmath7-form is the _ internal entropy flux _ : the entropy production , , can be also calculated with the help of the relative balances and thermodynamic relations , - , , and . the corresponding thermodynamic forces and fluxes are given in table [ erar ] . l c c c & diffusion & thermal & mechanical + + force & -@xmath113 & @xmath114 & @xmath115 + flux & @xmath56 & @xmath89 & -@xmath116 [ erar ] according to the representation theorems of isotropic functions @xcite , for isotropic fluids the linear relationship between the thermodynamic fluxes and forces results in the following constitutive functions here @xmath117 is the ( self)diffusion coefficient , @xmath118 and @xmath119 are the ( self)soret dufour coefficients , @xmath120 is the thermodynamic coefficient of heat conduction ( @xmath121 is the fourier heat conduction coefficient ) , @xmath122 and @xmath123 are the volume and shear viscosities . the - system of basic balances , together with , and the constitutive functions is a closed system of equations , and with the notable exception of the ( self)diffusion flux , it is identical with the usual continuity - fourier - navier - stokes system of equations . our derivation shows , that the ( self)diffusion flux can not be eliminated by flow - frame choice , its neglection is a physical assumption about material properties . non - relativistic , more properly , galilean relativistic single component dissipative fluids were treated independently of reference frames and flow - frames . the particular conditions leading to the usual relative , reference frame dependent continuity - fourier - navier - stokes system of equations were given . the reference frame free theory is based on the galilean relativistic space - time model of matolcsi @xcite . our treatment here uses mostly vector spaces and introduces an adapted abstract index formalism . it was required , that usual relative physical quantities must be components of absolute ones , and that the tensorial properties should be in harmony with the momentum series expansion of the kinetic theory . the analysis of these requirements leads to our basic physical quantity , to a third order , partially symmetric four - tensor , called the mass - momentum - energy density - flux tensor of the fluid . the four - divergence of this quantity is a second order four - tensor differential equation representing unitedly the reference frame independent form of the mass , momentum and energy balances . with the help of a four - velocity field the mass balance emerges as the time - timelike component , the momentum balance as the time - spacelike component and also as the space - timelike component according to the symmetry . the energy balance is the trace of the space - spacelike component of the absolute balance . the galilean transformation rules of the particular relative physical quantities and balances were derived . one of the consequences of the theory is that the usual relation between the internal , kinetic and total energies appears as a transformation rule . the derived transformation rule of the energy density , , is more general than the usual expression with internal , total and kinetic energies , , because of the presence of ( self)momentum density . the transformation rules of the physical quantities are transitive . regarding thermodynamics our basic assumption was , that the entropy density depends on the mass - momentum - energy density . this is an absolute statement , independent of reference frames and flow - frames . the derivative of the entropy density provides the second order four - cotensor of intensive quantities , the absolute temperature - thermovelocity - chemical potential cotensor . the intensive pair of the momentum density , the thermovelocity , appears in the @xmath7-form of the gibbs relation , too . the equations of states of the relative intensives are not independent of each other , due to the absolute background . the equality of second order mixed partial derivatives of the motion related intensive quantities , the corresponding maxwell relation , can be solved and restricts the form of the thermovelocity equation of state . the particular relative form of the entropy flux follows from the basic assumptions , therefore the four - divergence of the entropy four - vector , that is the absolute entropy production , can be calculated . there is a freedom to fix the flow - frame of a fluid to the mass ( eckart flow - frame ) , to the energy ( landau - lifsic flow - frame ) or by other , different manners . the particular form of the entropy production reveals , that the most convenient choice of the flow - frame is to fix it to the temperature , and eliminate the thermovelocity . this is thermo - flow . therefore the momentum density is zero , too , because of the thermovelocity equation of state . with this condition we almost obtain the usual form of the entropy production . the deviation comes from the presence of the ( self)diffusion terms . ( self)diffusion can not be eliminated any more by changing the flow frame if we want to keep the usual form of the other terms . remarks : * representing the basic field of the continuum in a galilean relativistic space - time by a third order mixed tensor field , instead of third order tensor field , we obtain the same relative balances with the same transformation rules . then the second order spatial component of the density four - tensor will be the mass density , instead of the energy . however , the covariant parts of the mixed tensor are not compatible with the usual system of equations obtained from momentum series expansion in kinetic theory . * the system of equations - and - is different than the most investigated similar system of brenner ( see e.g. @xcite ) . therefore the arguments against the presence of this term should be reconsidered @xcite . * the stability of simple materials under simple environmental conditions is an essential experimental observation in physics . without this fundamental stability property physical phenomena may not be reproduced @xcite . the physical - mathematical representation of this stability is thermodynamics . the concepts and the structure of thermodynamics can be understood from this point of view @xcite . this approach to thermodynamics is also a benchmark , a tool of verification of a thermodynamic theory . in case of galilean relativistic fluids we expect , that the homogeneous equilibrium be asymptotically stable without any additional conditions beyond the thermodynamic requirements @xcite . for classical fluids the following statement can be proved : + the system of equations - and - is generic stable , that is its homogeneous equilibrium is linearly stable , if thermodynamic stability is fulfilled ( entropy density is concave ) , the transport coefficients are nonnegative ( second law ) and the following inequality is true : here the first two terms and the coefficient of the third term is nonnegative because of thermodynamic conditions , but the partial derivative of the last term may lead to the violation of the inequality . * the main motivation of this work comes from our attempts to clarify the relation of generic stability , thermodynamics and flow - frames in dissipative relativistic fluids : @xcite . the mathematical structure of the galilean relativistic space - time is 1 . @xmath124 _ space - time _ is an oriented four dimensional affine space of the _ world points _ _ or events _ @xmath125 , over the four dimensional vector space of _ space - time vectors _ @xmath126 . there are no euclidean of pseudo - euclidean structures on @xmath127 : the length of a space - time vector does not exist @xmath128 _ time _ is a one dimensional oriented affine space of _ instants _ @xmath129 , over the one dimensional vector space of _ durations _ @xmath131 is the _ timing _ , an affine surjection over the linear mapping @xmath132 , the _ time evaluation_. 4 . @xmath133 is the _ measure line of distances _ , which is a one dimensional oriented affine space . @xmath134 euclidean structure is a symmetric bilinear mapping , where @xmath135 is the three dimensional vector space of _ space vectors_. the empirical axiomatic foundation of this galilean relativistic space - time is given in @xcite , where the axioms are related to clear observations and measurements . this structure was given first in @xcite and further elaborated in @xcite . similar structures where suggested also in @xcite . the _ duration _ between the events @xmath136 is calculated by @xmath137 , where @xmath138 . two events are _ simultaneous _ if the duration between them is zero . the difference between two simultaneous events is a _ spacelike _ vector , an element of @xmath139 . those vectors that are not spacelike are called _ timelike_. the dual of @xmath127 , the vector space of the @xmath140 linear mappings , is denoted by @xmath141 . the elements of @xmath141 are called four - covectors and are denoted by lower indexes . similarly , the dual of @xmath139 is @xmath142 , and their elements , the spacelike vectors and covectors , are denoted by overlined upper or lower indexes @xmath143 , @xmath144 , respectively . the _ length _ of a spacelike vector is @xmath145 . there is a canonical identification of @xmath139 and @xmath142 , due to the euclidean structure . however , @xmath127 and @xmath141 can not be identified for lack of euclidean or pseudo - euclidean structures : timelike space - time vectors do not have a length . the most important elements of the model are shown on figure [ st_fig ] . time evaluation and timing introduce a foliation of space - time : the sequence of spacelike subspaces of simultaneous events . the relation of galilean relativistic space , time and space - time . ] we use an abstract index notation of vectors and covectors . @xmath146 indexes are used for absolute physical quantities in the four - dimensional space - time . the indexes are abstract in the sense , that do not refer to a particular coordinate system or reference frame , they denote the tensorial properties of the different space - time related physical quantities @xcite . upper indexes are used for vectorial ( contravariant ) , the lower indexes for covectorial ( covariant ) physical quantities . the @xmath147 indexes always refer to tensorial components of usual three dimensional relative physical quantities as seen in an inertial reference frame . these indexes indicate the presence of two absolute velocity fields , here the velocity fields of the inertial observer and that of the fluid , and can be rewritten with absolute indexes , where the corresponding four - velocities are explicitly written . we will work with a space - time notation , but for the interpretation of the traditional forms of the equations and the transformation formulas of galilean transformations the later indexes are used . the vector space @xmath139 of spacelike vectors is a subspace of @xmath127 , and its canonical embedding is denoted by @xmath148 . therefore if @xmath143 , then @xmath149 is nothing else but @xmath150 as the element of @xmath127 . if @xmath151 is a covector , that is a linear mapping @xmath152 , then its restriction to @xmath139 is an element of @xmath142 denoted by @xmath153 . the corresponding _ projection of the restriction _ is denoted by @xmath154 , therefore @xmath155 . this projection @xmath154 is the dual ( transpose ) of the canonical embedding @xmath156 . the identification of @xmath139 and @xmath142 by the euclidean structure is given by @xmath157 and its inverse @xmath158 , as @xmath159 and @xmath160 . however , one can not introduce a canonical identification of @xmath127 and @xmath141 , because of the absence of an euclidean or pseudo - euclidean structure . the above system of notations of vectors and covectors introduces a convenient formal tool of handling the space - time without embedded time . the existence of a mass point in space - time is given by _ world line functions _ , that map time into space - time @xmath161 . the structure of galilean relativistic space - time requires , that @xmath162 . therefore the time derivative of a world line function at an instant is a four - vector @xmath6 with the following property : @xmath163 . the projection to @xmath139 along an @xmath6 of gives the @xmath7-spacelike component of a vectorial physical quantity . this _ @xmath7-projection _ is @xmath164 , where @xmath165 is the identity of @xmath127 . the four - velocities may play an other role : they can map duration into space - time vector @xmath166 . let us enumerate the four basic mappings of space - time vectors : * @xmath167 , * @xmath168 , * @xmath169 , * @xmath170 , the corresponding mappings between dual spaces are : * @xmath171 , * @xmath172 , * @xmath173 , * @xmath174 . the next identities follow from the definition one can see also that @xmath175 , and @xmath176 , therefore @xmath177 . let us observe , that from the second and last equalities of result in @xmath178 and @xmath179 . these relations are summarized by the following diagrams : @xmath180 \underset{\delta^a_{\ ; \b}}\longrightarrow \end{array } \mi \begin{array}{c } \overset{\tau_a}{\longrightarrow } \\[-8pt ] \underset{u^a}{\longleftarrow } \end{array } \i , \qquad\qquad \e^*\begin{array}{c } \overset{\delta^{\;a}_{\b}}{\longleftarrow } \\[-8pt ] \underset{\pi(u)^{\;\a}_{b}}\longrightarrow \end{array } \mi^ * \begin{array}{c } \overset{u^a}{\longrightarrow } \\[-8pt ] \underset{\tau_a}{\longleftarrow } \end{array}\i^*.\ ] ] the upper lines of the above diagrams give the splittings of a vector and a covector into time- and spacelike parts . a four - vector can be split into components with respect to an @xmath6 four - velocity and can be reconstructed with the help of these @xmath7-components . if @xmath5 is a four - vector , its timelike part is @xmath11 and its @xmath7-spacelike part is @xmath181 . the timelike part of a vector does not depend on @xmath7 , therefore it is absolute . the spacelike part depends on the four - velocity of the splitting . the reconstruction is simply this formula will be called the @xmath7-form of the vector . the timelike part of an arbitrary four - velocity is @xmath182 . the spacelike part of @xmath6 by an other velocity @xmath27 , @xmath183 is the _ relative velocity _ of @xmath6 related to @xmath27 . in particular the relative velocity of @xmath6 with respect to itself is zero . the extensive quantities together with their fluxes are natural four - vectors in galilean relativistic space - time . the densities of extensive quantities are four - vector fields . the timelike part of an extensive four - vector density is the density , an @xmath7-spacelike part is the flux . the timelike parts of the four - densities are independent of any splitting velocity , their @xmath7-spacelike parts are not . we will see , that @xmath7-independency means galilean invariance . covectors can be split into @xmath7-timelike and spacelike parts and can be reconstructed with the help of these @xmath7-components . a covector @xmath184 can be written as where @xmath185 and @xmath186 . here @xmath187 , the spacelike part of @xmath184 , appears differently than the spacelike part of vectors in . one should pay attention that @xmath188 can not be decomposed additively , more properly @xmath189 is not meaningful , because @xmath190 and @xmath142 is not a subset of @xmath141 . therefore the convenient regrouping @xmath191 of the above formula is strictly speaking incorrect . however , the spacelike + timelike composition is very transparent and also helpful in the calculations . the advantage of transparent calculations is larger than the possibility of mistakes , therefore we will use this convenient decomposition , with and extra care for the presence of both parts of @xmath188 in the formulas . the space - time derivative , @xmath15 , is a covector . it can be written with its @xmath7-timelike and spacelike components as where @xmath192 is the @xmath7-timelike derivative and @xmath193 is the spacelike derivative . the spacelike derivative is absolute . the @xmath7-form of the second order tensor @xmath194 is the following where * @xmath195 is the _ time - timelike _ part of @xmath196 . it is absolute , independent of @xmath6 . * @xmath197 is the _ space - timelike _ part of @xmath196 , and @xmath198 is the _ time - spacelike _ part . * @xmath199 is the _ space - spacelike _ part of the tensor . * @xmath200 and @xmath201 . the tensor @xmath196 itself is independent of @xmath6 , therefore its left and right timelike parts are absolute . if @xmath196 is symmetric , then @xmath202 . * @xmath203 and @xmath204 are the left and right spacelike parts of @xmath196 . the @xmath7-form of the second order mixed tensor @xmath205 is : where * @xmath206 , is the _ time - timelike part _ of @xmath207 , * @xmath208 , is the _ space - timelike part _ of @xmath207 , * @xmath209 , is the _ time - spacelike part _ of the mixed @xmath207 tensor . this part is @xmath7-independent , therefore absolute , * @xmath210 , is the _ space - spacelike part _ of @xmath207 , * @xmath211 , is the _ co - timelike part _ of @xmath207 , * @xmath212 is the _ co - spacelike part _ part of the mixed tensor . it is absolute . the symmetry of a mixed tensor can not be interpreted @xmath7-independently . the @xmath7-form of the second order cotensor @xmath213 is : where * @xmath214 , is the _ time - timelike part _ of the @xmath215 cotensor , * @xmath216 , is the _ space - timelike part _ of @xmath215 , and @xmath217 , is the _ time - spacelike part _ , * @xmath218 , is the _ space - spacelike part _ of the @xmath215 cotensor . this is the @xmath7-independent part . * @xmath219 and @xmath220 are the _ left and right co - timelike parts _ of @xmath215 . if the cotensor is symmetric , then @xmath221 . * @xmath222 and @xmath223 are the @xmath7-independent left and right _ co - spacelike parts _ of the @xmath215 cotensor . the mathematical structure of the galilean relativistic space - time model reflects exactly our everyday experience , that time passes independently of the observer , but the space , the environment composed by the things around us , depends on the observer . time is absolute , space is relative . the relativity is characterized by observers . an observer is a smooth four - velocity field on the space - time ( see @xcite ) , it is not necessarily global . an inertial observer is a constant four - velocity field . previously we have given the splitting of vectors , covectors , and tensors by a four - velocity . absolute physical quantities are vector fields , covector fields , tensor fields , etc . , that is they are vector , covector , tensor , etc . valued functions interpreted on the space - time . the splitting of the fields is local , by the local observer velocity . for example if @xmath224 is a vector field and @xmath225 is an observer , then at the world - point @xmath226 the @xmath227 vector is reconstructed from its time- and @xmath228-spacelike parts . keeping in mind that here everything is related to a world - point , one can omit @xmath226 in the notation , and then our previous formulas are all valid . now we analyze the relation of time- and spacelike parts of the same absolute physical quantity by two different four - velocities . in a space - time model these _ transformation rules _ can be derived . in the following we need the projections by the two different four - velocities , @xmath7 and @xmath17 . then it is simpler if we miss the explicit notation of the velocities , denoting the @xmath7- and @xmath17-projections by @xmath229 and @xmath230 , respectively . we have seen , that the time and @xmath7-spacelike parts of a vector @xmath5 by an observer @xmath6 are @xmath231 and @xmath12 , respectively . the non - relativistic physical theories are built on these kind of relative quantities , unaware of the deeper space - time aspects . two different @xmath6 and @xmath27 four - velocities may result in different time- and spacelike parts : @xmath232 where @xmath233 denotes the splitting by @xmath6 . the @xmath7- and @xmath17-forms of the physical quantity express the absolute four - vector with the help of its time- and spacelike parts . the transformation rules give the relative quantities according to an observer with the relative quantities of an other observer . in particular @xmath234 and @xmath235 are given as a function of @xmath236 and @xmath237 and the relative velocity . in our space - time model these transformation rules can be calculated by splitting the @xmath7-form of a physical quantity , which is @xmath8 in this case , by an other observer @xmath27 . then the @xmath17-timelike part of a vector is the same , does not transform , that is galilean invariant . this is not too surprising , because the function of the splitting , the time evaluation , does not depend on the velocities . the @xmath17-spacelike part of @xmath5 is where we have denoted the relative velocity @xmath6 related to @xmath27 as the above formula is the transformation rule of the spacelike component of a four - vector . we give it also with relative indexes this is exactly the well - known _ galilean transformation_. and are the same equations written by with different notation . let us remember , that the three - indexes @xmath238 refer to the presence two four - velocities in the formula , the equation requires the presence of two observers . with this notation the absolute quantities and formulas are strictly distinguished from the usual 1 + 3 dimensional forms and the space - time based and galilean transformation based ways of thinking are separated . the complete transformation rule is in particular the transformation rule of four - velocities can be deduced directly . the splitting of a four - velocity by itself is @xmath239 therefore , the transformation rule of its spacelike part gives the expected relative velocity the complete transformation rule is : where the left hand side is @xmath6 from the point of view ( in the time and space ) of @xmath27 and the right hand side is @xmath6 from the point of view ( in the time and space ) of itself . the meaning of this velocity transformation is that observer @xmath6 is considered at rest according to itself , but is moving with the velocity @xmath30 for the observer @xmath27 . or , at the other hand , we can say with the transformations terminology that the relativ velocity @xmath30 for observer @xmath17 is transformed to zero when we change to observer @xmath7 . in contrast to the dash is used only for the timelike part , because of the accustomed notation of the relative velocity ( @xmath30 would be @xmath240 ) . in the above formulas of galilean transformations there are two _ arbitrary _ velocity fields , the formulas are not related solely to inertial observers . only the galilean invariant quantities are considered observer independent , however not only those , but a combination of properly transformed quantities may be absolute , whenever it is a component of an absolute space - time quantity . the timelike part of and absolute four - vector physical quantity ( the density of an extensive quantity ) is galilean invariant , its spacelike part ( the current density or flux ) transforms , therefore it depends on the reference frame . however , the complete four - vector is absolute . the same is valid also for the velocity , where the four - velocity is absolute , notwithstanding that the timelike part seemingly does not contain physical information . putting a physical theory in the space - time model one can consistently decide what depends on the reference frame and what does not . we have seen , that , the time- and spacelike parts of the covector @xmath184 by the velocity @xmath6 are @xmath241 and @xmath186 , respectively . these componets are represented by lower indexes and horizontal mode of writing : @xmath242 according to the previous section the transformation rules of the parts of @xmath184 is obtained by splitting the @xmath7-form of @xmath184 using the velocity field @xmath27 . for the timelike part it is where identity was applied . for the spacelike part one should use the identities : the complete transformation rule may be written with in the 1 + 3 dimensional form : a particular example is the space - time differentiation @xmath15 , which is a symbolic covector . then we obtain , that : if @xmath6 is the velocity field of a fluid , and @xmath27 is that of an observer , the @xmath243 is the _ substantial time derivative _ , @xmath30 is the relative velocity of an observer , related to the fluid and @xmath244 . the relation between the _ partial time derivative _ @xmath245 , and the substantial time derivative @xmath246 is @xmath247 . it is worth to compare the previous transformation rule based derivation and the usual method ( see e.g. @xcite ) . an observer @xmath6 splits a second order tensor , @xmath196 , into a time - timelike part , @xmath195 , into a time - spacelike part @xmath248 , into a space - timelike part , @xmath249 , and into a space - spacelike part , @xmath250 . therefore @xmath251 in general @xmath252 and also @xmath253 , because the symmetry of @xmath196 was not assumed . the corresponding transformation rules are calculated as follows . the time - timelike component is invariant the transformation rule of the time - spacelike and space - timelike components is like the transformation rule of a space vector : for the space - spacelike component one obtains a more complicated formula : here we have used . the complete transformation rule can be written also with the usual notation : the components of a second order mixed tensor @xmath207 by an observer @xmath6 observer are @xmath254 , the time - timelike , @xmath255 , the space - timelike , @xmath256 the time - spacelike and @xmath257 the space - spacelike components . that is @xmath258 it is meaningless to speak about the symmetry of mixed second order tensors . only , eventually , the observer dependent split can be symmetric . this is also clear from the galilean transformation rules , because the time - spacelike and space - timelike parts transform differently . we obtain for the time - timelike component : the time - spacelike component seems to be a three - vector , but does not transform . this can be understood using the identities : the galilean transformation rule for the space - timelike component is : finally for the space - spacelike part one derives here the usual identities were used . then the complete transformation rule can be written as : the components of the second order cotensor @xmath215 split by the observer @xmath6 are the @xmath259 time - timelike , @xmath260 space - timelike , @xmath261 time - spacelike and the @xmath262 space - spacelike parts . that can be written in a matrix form as @xmath263 if @xmath215 is not symmetric , @xmath264 and @xmath265 . the calculation of the transformation rules of the components is the following . for the time - timelike part one obtains : the time - spacelike part transforms identically with the space - timelike part : the space - spacelike part is galilean invariant here the usual identities were applied , too . finally the complete transformation rule is given in a matrix form : this work is based on the fundamental ideas of tams matolcsi who condensed space - times into mathematical models . the author is also indebted to tams flp , who was convinced all the time that there is an absolute nonrelativistic energy . in this work it does not look like a cotensor , but that possibility is not excluded yet . flp t. a tr nem abszolt - a trid , mint a galilei - fle relativitsi elv kvetkezmnye . in flp t. , editor , _ j eredmnyek a kontinuumfizikban _ , volume 8 of _ mrnkgeolgia - k ozetmechanika kisknyvtr _ , chapter 1 , pages 1135 . megyetemi kiad , budapest , 2008 . ( in hungarian ) w. noll . space - time structures in classical mechanics . in _ the foundations of mechanics and thermodynamics ( selected papers by walter noll ) _ , pages 204210 . springer verlag , berlin - heidelberg - new york , 1974 . originally : pp28 - 34 , delaware seminar in the foundations of physics , berlin - heidelberg - new york , springer , 1967 . a. i. murdoch . objectivity in classical continuum physics : a rationale for discarding the principle of invariance under superposed rigid body motions in favour of purely objective considerations . , 15:309320 , 2003 . flp t. kontinuumok kinematikjnak j rtelmezse . in flp t. , editor , _ j eredmnyek a kontinuumfizikban _ , volume 8 of _ mrnkgeolgia - k ozetmechanika kisknyvtr _ , chapter 3 , pages 5599 . megyetemi kiad , budapest , 2008 . ( in hungarian ) p. vn and t. bir . dissipation flow - frames : particle , energy , thermometer . in m. pilotelli and g. p. beretta , editors , _ proceedings of the 12th joint european thermodynamics conference _ , pages 546551 , brescia , 2013 . cartolibreria snoopy . isbn 978 - 88 - 89252 - 22 - 2 , arxiv:1305.3190 . p. vn and t.s . thermodynamics and flow - frames for dissipative relativistic fluids . in g. chacn - acosta , garca - perciante a.l . , and a. sandoval - villalbazo , editors , _ plasma physics and relativistic fluids _ , volume 1578 of _ aip conf . proceedings _ , pages 114121 , 2014 . proceedings of the v leopoldo garca coln mexican meeting on mathematical and experimental physics , el colegio nacional , september 9 - 13 , 2013 . mexico city . arxiv:1310.5976 . asymptotic stability and the second law in extended irreversible thermodynamics . in s. rionero and t. ruggeri , editors , _ 7th conference on waves and stability in continuous media , bologna , italy october 4 - 9 . 1993 _ , volume 23 of _ series on advances in mathematics for applied sciences _ , pages 384389 , singapore - new jersey - london - hong kong , october 1994 . quaderno cnr - gruppo nazionale per la fisica matematica , world scientific .

in this paper , we study a general linear networked system that contains a tunable memory subsystem ; that is , it is decoupled from an optical field for state transportation during the storage process , while it couples to the field during the writing or reading process . the input is given by a single photon state or a coherent state in a pulsed light field . we then completely and explicitly characterize the condition required on the pulse shape achieving the perfect state transfer from the light field to the memory subsystem . the key idea to obtain this result is the use of zero - dynamics principle , which in our case means that , for perfect state transfer , the output field during the writing process must be a vacuum . a useful interpretation of the result in terms of the transfer function is also given . moreover , a four - nodes network composed of atomic ensembles is studied as an example , demonstrating how the input field state is transferred to the memory subsystem and how the input pulse shape to be engineered for perfect memory looks like .

1403.1698

quantum memory is , in a wide sense , a device that stores or freezes a quantum state both spatially and in time . a highly successful example is that a light pulse is frozen in a cloud of atoms @xcite . also quantum memory is of technological importance particularly in quantum information science , such as the quantum repeater for quantum communication @xcite . because of these scientific / technological importance , the field of quantum memory has experienced significant progress in both theory and experiment . we refer to for instance @xcite for reviewing the current situation of this very active research area . basic schematic of an ideal quantum memory . @xmath0 denotes the mode of the memory subsystem used for the storage , while @xmath1 is the mode of the buffer subsystem , which transports the input state to ( from ) the memory subsystem from ( to ) the optical field with mode @xmath2 ( @xmath3 ) . the system structure can be switched from the stage ( a ) to ( b ) , or from ( b ) to ( c ) , by tuning some controllable parameters . in the stage ( b ) , @xmath0 is decoupled from @xmath1 and thus the optical field , implying that @xmath0 is decoherence free . in the stages ( a ) and ( c ) , on the other hand , @xmath0 couples to the field for state transportation . ] now let us see a basic and general schematic of an ideal quantum memory in an abstract way . first of all , we assume that the system contains a subsystem that can store any quantum state without loss in a long period of time ; we call this specific component the _ memory subsystem_. ideally , the memory subsystem should be completely decoupled from the other system components and surrounding environment during it stores the state ; i.e. , it is _ decoherence free ( df ) _ @xcite . note that , in this storage stage , the memory subsystem is decoupled even from the channel used for transferring an input state or retrieving the stored state . hence , the second assumption is that , during the writing / reading process , the system can be tuned so that the memory subsystem couples to that transportation channel . that is , the system should be the one that contains a tunable port switching the opening / closing of the memory subsystem . indeed this basic schematic is employed in each specific memory device . in the case of atoms based on the electromagnetic induced transparency ( eit ) effect , an isolated memory subsystem is served by a set of metastable collective atomic states , and an external control field ( with rabi frequency @xmath4 ) can switch on / off of the coupling between the metastable states and the optical field for state transportation @xcite . we also find successful demonstrations in optical cavity or optomechanical oscillator arrays @xcite , where the switching mechanism is served by adiabatic frequency detuning of the memory subsystem . further , a similar switching procedure is employed in the photon echo quantum memory @xcite . note that , if the system does not contain a switching mechanism , it is generally impossible to perfectly transfer an unknown state to a memory subsystem ( i.e. df subsystem ) @xcite . the above - mentioned basic schematic for quantum memory is illustrated in fig . [ df memory schematic ] . in the writing stage ( a ) , an input state is sent to the system over an input channel with mode @xmath5 . let us here assume that , by devising a certain nice procedure " , all the input state is transferred to the memory subsystem with mode @xmath0 . we then close the port , and @xmath0 becomes decoherence free . in this storage stage ( b ) , ideally , the memory subsystem can store the state for a long time . finally , in the reading stage ( c ) , by opening again the port we can retrieve the state at any later time , which is sent over the output channel with mode @xmath6 . so what is a certain nice procedure " to achieve the best or hopefully perfect state transfer ? for some typical quantum memories such as eit and off - resonant raman memories , we can explicitly formulate this problem ; that is , the question is what is the optimal temporal shape of the pulsed light field carrying the input state . this optimization problem is very important and actually has been deeply investigated in several papers ; for instance , towards the most efficient atomic memory with eit , in @xcite the input wave packet as well as the controllable rabi frequency @xmath4 ( see the second paragraph ) are carefully designed , although the method is a heuristic one based on iterative tuning of the parameters . it is also notable that the so - called _ rising exponential _ type function is found as an effective pulse shape @xcite . this is a pulse with exponential increase of e.g. the probability density of photon counting or the amplitude of a coherent field , which can be physically implemented @xcite . motivated by these results and further the fact that a large - scale quantum memory is required for constructing practical quantum communication architecture , the following questions naturally arise . is there a pulse shape for achieving perfect state transfer for general and large - scale quantum memory networks ? is there a general yet simple guideline for synthesizing such desirable pulse shape ? what is the fundamental origin of the rising exponential function as a desirable pulse shape ? is the rising exponential pulse optimal in a certain sense ? also , is it effective even for a large - scale network ? solving these problems should accelerate the progress of the quantum memory research in its deeper understanding and practical implementation . in this paper , we consider a general _ passive linear system _ , which models a wide variety of systems such as optical systems @xcite , nano - mechanical oscillators @xcite , vibration mode of a trapped particle @xcite , and atomic ensembles @xcite . as mentioned before , the system is assumed to contain a tunable memory subsystem ; that is , by tuning a certain parameter , we can switch opening / closing of the memory subsystem , which is df during the storage period . note that in our case this df subsystem corresponds to a system having the so - called _ dark mode _ @xcite . another assumption is that an unknown input state to be transferred is encoded in a continuos - mode single - photon field or a coherent field . with these setups , we give answers to all the questions posed in the above paragraph . the essential idea is the use of _ zero - dynamics principle_. this concept originates from the classical notion of zero " of a _ transfer function _ , which is a fundamental tool used in systems and control theory @xcite . more precisely , for a general linear classical system @xmath7 where @xmath8 is the input , @xmath9 is the output , and @xmath10 are matrices , its input - output relation is simply characterized by a transfer function @xmath11=d+c(si - a)^{-1}b$ ] as @xmath12=h[s]u[s]$ ] ( see section 4.3 for detailed description ) ; then , if the input is given by @xmath13 , where @xmath14 is a _ zero _ of the transfer function ( meaning @xmath15=0 $ ] ) , we have @xmath16e^{zt } = 0 $ ] under some additional conditions . that is , the _ zero - dynamics _ is a system whose output is always zero . in fact , the concept of zero - dynamics is very important in analysis and synthesis for even general nonlinear systems @xcite . actually , the zero - dynamics principle described above can be directly applied to general quantum memory problem ; if the state transfer is perfectly carried out , the input field must be completely absorbed in the system , and the output field must not contain any small pieces of the input field . that is , as a principle , the output should be zero " during the writing and storage stages . surprisingly , this simple zero - dynamics principle leads us to prove , very easily , that a rising exponential function is a unique pulse shape achieving the perfect state transfer for general ( and thus large - scale ) linear passive networks . moreover , based on this first main result , we give an explicit , simple , and general procedure for designing the wave packet carrying an unknown state that is as a result perfectly absorbed in the memory subsystem . this paper is organized as follows . section 2 provides some preliminaries , describing general passive linear systems , optical field states , and linear df subsystems . also the problem is explicitly formulated in section 2.4 . in section 3 , we study a simple single - mode oscillator as a memory system , to show the fact that a rising exponential function appears as a unique pulse shape achieving the perfect state transfer ; based on this result , the idea of zero - dynamics principle is discussed . section 4 provides our first main result ; for general passive linear systems , we derive the rising exponential pulse from the zero - dynamics principle . also we give a useful interpretation of this fact in terms of the transfer function . in section 5 , we present our second main result , showing the concrete procedure for perfect writing , storage , and reading ; this explicitly shows the pulse shape to be synthesized for perfect state transfer from the optical field to the memory subsystem . section 6 is devoted to derive the time evolution equations of some statistical quantities of the dynamics , which is useful for numerical simulation . in section 7 , we study a linear network composed of atomic ensembles trapped in a cavity , which contains a tunable memory subsystem ; this example shows how the designed pulse actually looks like and how the state transfer from the optical field to the atomic ensembles evolves in time . section 8 summarize the paper and discusses some future works . in appendix a , we briefly examine the case of an _ active _ linear memory system . appendix b provides a case study comparing the zero - dynamics principle and the so - called _ dark state principle_. * notation : * we use the following notations : for a matrix @xmath17 , the symbols @xmath18 , @xmath19 , and @xmath20 represent its hermitian conjugate , transpose , and complex conjugation in elements of @xmath21 , respectively ; i.e. , @xmath22 , @xmath23 , and @xmath24 . for a matrix of operators we use the same notation , in which case @xmath25 denotes the adjoint to @xmath26 . for a time - dependent variable @xmath27 , we denote @xmath28 . in this paper , we study a general linear open system composed of @xmath29 oscillators with mode @xmath30^\top$ ] that couples to an optical field with continuous mode @xmath5 ; hence they satisfy @xmath31=\delta_{ij}$ ] and @xmath32=\delta(s - t)$ ] . the system is driven by a quadratic hamiltonian @xmath33 with @xmath34 an @xmath29-dimensional hermitian matrix . the system and the field instantaneously couple through the interaction hamiltonian @xmath35 $ ] with @xmath36 an @xmath29-dimensional complex _ row _ vector . then the unitary operator @xmath37\ ] ] ( @xmath38 is the initial time ) produces the heisenberg equations of @xmath39 and @xmath40 as follows ; @xmath41 where @xmath42^\top$ ] and @xmath43 . the second term in @xmath21 stems from the ito - correction . note also that , due to the ideal markov property , the optical field @xmath5 does not have its own dynamical time evolution ; rather @xmath6 represents the output of the system . the above open system with input @xmath5 and output @xmath6 is called the _ passive linear system _ in the sense that it does not contain any active component such as an optical parametric amplifier in optics case . as mentioned in section 1 , a passive linear system can model a wide variety of systems . the main reason why we focus on this class of general systems is that it preserves the total energy during the interaction between the system and the field . this implies that the perfect state transfer is equivalent to the perfect energy transfer , which as a result allows us to use the zero - dynamics principle to characterize the perfect memory ; this will be discussed in detail in section 4 . also in appendix a , we show a brief case study where the system contains an active component . for a general treatment of passive linear systems , see @xcite ; the notation used in this paper follow these references , where in general @xmath36 is an @xmath44 complex matrix representing @xmath45 input optical fields . in this paper , we consider the case where the input is given by a single photon state or a coherent state , which is carried by an optical pulse field with continuous - mode @xmath5 . they are described as follows . * single photon field state : * the single photon state in a single mode system is , as is well known , produced by acting a creation operator @xmath46 to the ground state @xmath47 ; i.e. @xmath48 to describe the continuous - mode single photon field state , we define the annihilation and creation process operators @xmath49 @xmath50 is an associated function in @xmath51 , representing the shape of the optical pulse field . also @xmath50 satisfies the normalization condition @xmath52 . due to this , @xmath53 and @xmath54 satisfy the usual ccr ; @xmath55=1 $ ] . the single photon field state is , in a similar way as above , produced by acting the creation process operator on the vacuum field @xmath56 as follows @xcite : @xmath57 due to the normalization condition of @xmath50 , we find that @xmath58 . also note the relation @xmath59 ; thus , @xmath50 has the meaning of the wave function with @xmath60 the probability of photo detection per unit time . let us now assume that the pulse shape @xmath50 can be expanded as @xmath61 the coefficient @xmath62 represents ( unknown ) classical information encoded in the optical field , which satisfies @xmath63 , and @xmath64 is a set of orthonormal functions satisfying @xmath65 note that @xmath29 is the number of modes of the system . then the single photon field state can be written as @xmath66 @xmath67 is called the single photon code state with pulse shape @xmath68 @xcite ; from the condition , they are orthonormal , i.e. @xmath69 . also the field operator @xmath70 satisfies the ccr @xmath71=\delta_{jk}$ ] . * coherent field state : * another important state is a coherent state . a coherent state in a single mode system is generated by acting a displacement operator @xmath72 on the ground state as follows ; @xmath73 where @xmath74 denotes the amplitude of @xmath75 . likewise , a coherent field state is defined in terms of the creation and annihilation process operators as follows ; @xmath76 { |{0}\rangle}_f,\ ] ] where @xmath77 is a complex - valued function , representing the amplitude of the state ; that is , this is a coherent pulse field modulated with envelope function @xmath77 . note that @xmath77 is not necessarily normalized , but its power @xmath78 is finite . now we assume that @xmath77 is given , in terms of the orthonormal functions @xmath64 , by @xmath79 where @xmath80 represents ( unknown ) classical information to be stored . the power of @xmath81 , i.e. the mean photon number in unit time , is then given by @xmath82 . the coherent field state is as a result described by @xmath83 { |{0}\rangle}_f.\ ] ] note that this is not a superposition of the coherent field states @xmath84 , unlike the single photon field state . the general theory of df subsystems states that , if a df subsystem exists , then the system hilbert space @xmath89 is decomposed to @xmath90 , where any observable in @xmath91 evolves unitarily . in our case , the decomposition is of the form @xmath92 ; thus the system variables are , more precisely , represented by @xmath93 and @xmath94 . this special class of continuous - variable df subsystems appears in several situation @xcite . also for a general theory of the linear df subsystem , including a necessary and sufficient condition for a given linear system to have a df mode , see @xcite . here we describe the problem discussed throughout the paper . our system is the general passive linear system , and it is assumed to be tunable ; that is , by appropriate tuning of its parameter(s ) , a part of the system , the memory subsystem , couples or decouples to the optical field carrying the information . hence the memory subsystem can be switched to be a df or a non - df subsystem . as mentioned in section 2.3 , the memory subsystem stores the state during it is in the df mode , while it should be in the non - df mode when we transfer the input state or retrieve the stored state . in particular , we assume that the matrix @xmath95 in the writing / reading stages is _ hurwitz _ , i.e. , the real parts of all the eigenvalues of @xmath21 are negative ; as will be shown later , this condition is necessary for perfect state transfer . on the other hand , in the storage stage , the dynamical equation takes the form , thus @xmath21 is not hurwitz . the field s initial state is given by @xmath96 in the case of single photon input field or @xmath81 in the case of coherent input field . the system s initial state @xmath97 is assumed to be separable , hence it is given by @xmath98 . in particular , we will set it to be the ground state @xmath99 satisfying @xmath100 . at time @xmath38 , the system and the field start to interact , via the unitary operator @xmath101 given in eq . . the composite state at time @xmath102 is then given by @xmath103 or @xmath104 . in this writing stage , the memory subsystem is in the non - df mode , so it couples to the field . but once the state transfer has been completed , then we switch the system parameters so that the memory subsystem becomes decoherence free , and its state is preserved during the storage stage . hence , it would be desirable if the state @xmath105 is of the separable form @xmath106 and the memory subsystem s state @xmath107 contains the full information of the input field state @xmath96 or @xmath81 . therefore , our goal is to appropriately synthesize the pulse shape @xmath50 or @xmath77 , or more precisely their basis functions @xmath64 , so that the above desirable transition from the field s initial state to @xmath107 occurs . lastly we remark on the switching configuration . in general , the system matrices @xmath34 and @xmath36 ( and thus @xmath21 ) can change in time ( i.e. time varying ) in order to realize high quality quantum memory . for instance in @xcite , the authors consider the time varying system matrices depending on the control field with frequency @xmath4 , which is optimized via a heuristic method . on the other hand , in this paper , we assume that @xmath34 and @xmath36 are time varying , but they are constant in each stage of the memory procedure ; that is , they are _ piecewise constant_. in particular , we will take the same system matrices in the writing and reading stages . in this section , we examine a simple case where the memory system is given by a single - mode passive linear system . as will be shown later , this system does not contain a tunable df component , so it does not work as a perfect storage device . rather the purpose here is that , by focusing only on the writing stage , requiring perfect state transfer uniquely determines the pulse shape of the input optical field . based on this result , we then derive the explicit form of the output field , and show the notion of zero - dynamics in this case . here we study only the single - photon input case , but it is straightforward to obtain a similar result in the case of coherent field state . let us consider the following single - mode ( i.e. @xmath108 ) linear system interacting with an optical field , which is obtained by setting @xmath109 and @xmath110 in eq . : @xmath111 where @xmath112 is the interaction strength ; in optics case , this system is typically given by an optical cavity with @xmath112 proportional to the transmissivity of the coupling mirror . the goal is to send a single photon state over the input pulse field and write it perfectly down to the system . note that , however , clearly this system does not contain a tunable df component , hence our interest here is only in the state transfer . the input state is given by eq . , which is now essentially @xmath113 . thus in this case let us take a superposition of the vacuum and the single photon field state @xmath114 where @xmath115 are the encoded ( unknown ) classical information . recall that the system s initial state is set to the ground state @xmath116 . the dynamical equation has the following solution : @xmath117 where @xmath118 is the time we stop the interaction . this can be rewritten as @xmath119 where @xmath120 note that @xmath121 . equation can be further represented as @xmath122 @xmath123 is the field creation operator with pulse shape @xmath124 , which is defined in eq . , and @xmath125 is the unitary operator given in eq . . from the above equation we find that , in the limit @xmath126 , the field creation operator @xmath123 is completely mapped to the system creation operator @xmath127 . this means that the perfect state transfer from the optical pulse field to the system mode can be carried out as shown below . in the schrdinger picture , the whole state at time @xmath118 is given by @xmath128 { |{0}\rangle}_s{|{0}\rangle}_f \nonumber \\ & & \hspace*{-1.5em } = u(t_0 , t_1 ) \big [ \alpha i_s\otimes i_f + \beta i_s\otimes b^*(\xi)\big ] u^*(t_0 , t_1 ) u(t_0 , t_1 ) { |{0}\rangle}_s{|{0}\rangle}_f \nonumber \\ & & \hspace*{-1.5em } = \big [ \alpha i_s\otimes i_f + \beta u(t_0 , t_1)(i_s\otimes b^*(\xi))u^*(t_0 , t_1 ) \big ] { |{0}\rangle}_s{|{0}\rangle}_f , \nonumber\end{aligned}\ ] ] where in the last equality @xmath129 is used . let us now set the input pulse shape to be @xmath130 . then , from eq . , we have @xmath131 { |{0}\rangle}_s{|{0}\rangle}_f \nonumber \\ & & \hspace*{-1.6em } = \big [ \alpha { |{0}\rangle}_s + \frac{\beta}{\sqrt{1-e^{-\kappa(t_1-t_0 ) } } } { |{1}\rangle}_s \big ] \otimes { |{0}\rangle}_f \nonumber \\ & & \hspace*{2em } \mbox { } - \frac{\beta e^{-\kappa(t_1-t_0)/2}}{\sqrt{1-e^{-\kappa(t_1-t_0 ) } } } u(t_0 , t_1 ) a^*(t_0 ) u^*(t_0 , t_1 ) { |{0}\rangle}_s{|{0}\rangle}_f . \nonumber\end{aligned}\ ] ] therefore , in the limit @xmath126 , we have @xmath132 \otimes { |{0}\rangle}_f,\ ] ] which means that the input field state is completely transferred to the system state . in particular , in the case @xmath126 and @xmath133 , the input pulse shape is given by @xmath134 this is called the _ rising exponential pulse _ @xcite . it is clear from the above discussion that the rising exponential is the unique pulse shape for perfect state transfer from the single photon field to the system . let us further study the final state @xmath135 , where the input pulse shape is now set to @xmath136 it is possible to obtain the explicit solution : @xmath137 where @xmath138 with @xmath139 first , at @xmath133 , we have @xmath140 + \frac{2\kappa\beta}{\kappa+\gamma}\sqrt{\frac{\gamma}{\kappa } } { |{1}\rangle}_s{|{0}\rangle}_f,\ ] ] which becomes @xmath141 only when @xmath142 . that is , the frequency bandwidth of the input pulse shape has to be exactly equal to that of the memory system to attain the perfect state transfer . this is a form of the so - called _ impedance matching _ for efficient energy transfer ; in section 7 , we will discuss the matching condition in a more practical setup where the system is composed of a cavity and atomic ensembles . next , in the limit @xmath143 , the whole state again becomes separable : @xmath144 where @xmath145 this @xmath146 represents the pulse shape of the output optical field over the whole period . hence , if @xmath142 , the output field is vacuum in the writing stage @xmath147 ; this means that the single photon input field is completely absorbed into the system , and the output field does not contain any pieces of the input state . in the optics case where the system is given by a cavity , a physical meaning of this fact is that it happens destructive interference between the light field reflected at the coupling mirror and the transmissive light field leaking from the cavity ; as a result , the output field of the cavity is always in vacuum , i.e. zero " , while the system s state still dynamically changes in time . in general , the dynamics of a system whose output is always zero is called the zero dynamics @xcite . hence in this case the cavity dynamics during the writing process is exactly a zero dynamics . in this section , based on the so - called energy - balanced identity , we show the notion of zero - dynamics principle as a guideline for perfect state transfer in general passive linear systems . then , we prove that the zero - dynamics principle readily leads to the rising exponential function as a unique pulse shape . further , a useful view of the zero - dynamics principle in terms of the transfer function is provided . note that the zero - dynamics principle is essentially equivalent to the so - called _ dark state principle _ ; this idea was first employed in @xcite for the application to a lossless node - to - node state transfer in a cavity qed network , and later several applications have been developed , e.g. lossless gate operation @xcite . appendix b provides a detailed case study comparing the zero - dynamics principle and the dark state principle and then discusses their difference . first we remark that the pulse shape of the single - photon input field , @xmath50 , and that of the output field , say @xmath146 , can be connected through a dynamical equation having the same form as eq . . actually , by multiplying @xmath148 by eq . from the right hand side and using the relation @xmath149 , we find that the mean photon number of the output field is given by @xmath150 by definition this should be written as @xmath151 , hence @xmath50 and @xmath146 are related through the following dynamics : @xmath152 where @xmath153 is a @xmath29-dimensional c - number vector . note that @xmath153 does not have a particular physical meaning , unlike the vector @xmath154 appearing just below . the same classical dynamical equation holds for the case of coherent input field ; noting that @xmath155 , we readily see that the vector of mean amplitudes , @xmath156^\top$ ] with @xmath157 , follows @xmath158 where @xmath159 is the amplitude of the output field @xmath6 . this equation has the same form as eq . , hence in what follows we use eq . when discussing the input - output relation of the corresponding wave packets . to pose the zero - dynamics principle , it is important to first remember that , for the general passive linear system , the following _ energy balance identity _ @xcite holds : @xmath160 this indicates that the total energy contained in the system and the field is preserved for all time . indeed , from eq . we immediately have @xmath161 where the mean is taken for the state @xmath162 . now we assume @xmath163 . then , for the energy of the input pulse field to be completely transferred to the system , we need @xmath164 for @xmath165 $ ] with @xmath102 the stopping time of the writing process . this is a rigorous description , in the case of passive linear systems , of the zero - dynamics principle ; that is , for the general quantum memory problem with a passive system , the output field must be vacuum ( i.e. zero " ) for perfect state transfer . surprisingly , this simple condition uniquely determines the form of the input pulse shape @xmath50 as shown below . first , from the requirement @xmath166 , we have @xmath167 , which further leads to @xmath168 this has the solution @xmath169 , with @xmath170 a fixed vector . thus , again from the condition @xmath171 , we have @xmath172 note that the input is sent during the writing stage @xmath173 , and @xmath174 in the storage and reading stages in @xmath175 . taking this into account , we end up with the expression @xmath176 where @xmath177 is the heaviside step function taking @xmath178 for @xmath179 and @xmath180 for @xmath181 . since @xmath21 is hurwitz , as assumed in section 2.4 , the real parts of all the eigenvalues of @xmath182 are strictly positive . . is a generalization of the rising exponential function . in fact , this immediately recovers the result in the example , where @xmath183 , @xmath110 , and particularly @xmath133 . lastly we remark that the zero dynamics is given by eq . , which is defined up to time @xmath102 . first , to see the idea let us return to the example studied in section 3 . the input pulse shape is given by eq . , whose laplace transformation is @xmath188=\sqrt{\kappa}/(s-\kappa/2)$ ] . also in this case the transfer function is given by @xmath189 = 1 - \frac{\kappa}{s+\kappa/2 } = \frac{s-\kappa/2}{s+\kappa/2}.\ ] ] hence , the output is computed as @xmath190 = \frac{s-\kappa/2}{s+\kappa/2}\cdot \frac{\sqrt{\kappa}}{s-\kappa/2 } = \frac{\sqrt{\kappa}}{s+\kappa/2},\ ] ] and its inverse laplace transform then yields @xmath191 thus , we again see that the input field is completely absorbed in the system during @xmath192 ; i.e. the perfect state transfer has been carried out . the most notable point is clearly that the _ zero _ of @xmath187 $ ] is erased ( in general , if for a transfer function @xmath15 $ ] there exists a @xmath14 satisfying @xmath15=0 $ ] , then @xmath14 is called a zero ) . now we can generalize the above fact ; for simplicity , we set @xmath133 . the transfer function of the general passive linear system is given by eq . . also the laplace transformation of the rising exponential function is given by @xmath193 = \int_{-\infty}^0 -\eta_1^\top e^{-a^\sharp t } c^\top e^{-st } dt = c(si+a^\dagger)^{-1}\eta_1.\ ] ] therefore , the output is computed as @xmath194 = g[s]\xi[s ] = \big [ 1-c(si - a)^{-1}c^\dagger \big]c(si+a^\dagger)^{-1}\eta_1 \nonumber \\ & & \hspace*{-2.5em } = c(si+a^\dagger)^{-1}\eta_1 - c(si - a)^{-1 } \big [ ( si - a ) - ( si+a^\dagger ) \big](si+a^\dagger)^{-1}\eta_1 \nonumber \\ & & \hspace*{-2.5em } = c(si - a)^{-1}\eta_1 . \nonumber\end{aligned}\ ] ] since @xmath21 is hurwitz , the output @xmath195 $ ] does not contain a zero , implying @xmath196 . in general , if a transfer function contains a ( transmission ) zero , then there alway exists an input signal such that the corresponding output takes zero @xcite . therefore , we have the following interpretation of the zero - dynamics principle for quantum memory in terms of transfer function ; in general , a linear memory system needs to have a zero for perfect state transfer , and the input state is sent over an optical field whose pulse shape is characterized by that zero . this view would be useful particularly in the case of multi input channels . in this section , we provide a detailed procedure to achieve the perfect memory , which is composed of the following three stages ; the perfect state transfer from the input field to the memory subsystem ( writing ) , decoherence - free preservation of the transferred state ( storage ) , and the appropriate retrieving of the system state into the output field ( reading ) . the setup was described in section 2.4 ; note again that the system matrices @xmath34 and @xmath36 ( thus @xmath21 ) change depending on the memory stage , but they are piecewise constant . also , as motivated by the result obtained in section 3 , we will take @xmath126 , while keeping general @xmath102 . one of the main questions is as follows ; although we have derived the rising exponential function from the zero - dynamics principle , it still contains some parameters that should be chosen appropriately ; more precisely , it is given by @xmath197 , and we need to determine @xmath170 so that the input field state is completely transferred to the memory subsystem . in this section , we will see that this synthesis problem is clearly solved . the solution of the general linear equation is explicitly given by @xmath198 since @xmath21 is hurwitz , we can take the limit @xmath126 , which yields @xmath199 where again @xmath102 is the stopping time of the writing process . let us now define the following vector of rising exponential functions : @xmath200 then the above solution of @xmath201 can be expressed as @xmath202^\top.\ ] ] this is a vector of creation operators @xmath203 , implying that it has to satisfy the canonical commutation relation @xmath204 ; actually , we have @xmath205 \nu(\tau)^\top ds d\tau = \int_{-\infty}^{\infty } \nu(s)^\sharp \nu(s)^\top ds \nonumber \\ & & \hspace*{-3em } = e^{a t_1 } \big [ \int_{-\infty}^{t_1 } e^{-as}c{^{\dagger}}c e^{-a{^{\dagger}}s}ds \big]e^{a{^{\dagger}}{t_1 } } = e^{a t_1 } \big [ \int_{-\infty}^{t_1 } \frac{d}{ds}\big ( e^{-as}e^{-a{^{\dagger}}s}\big)ds \big]e^{a{^{\dagger}}{t_1 } } = i. \nonumber\end{aligned}\ ] ] this relation shows that @xmath206 are orthonormal ; @xmath207 . * case i : single photon state . * we consider the case where the input is the single photon field state . also the system is assumed to be in the separable ground state at the initial time @xmath38 . then , through the interaction ( see fig . 2 ( a ) ) the whole state changes to @xmath208 { |{0,\ldots,0}\rangle}_s{|{0}\rangle}_f \nonumber \\ & & \hspace*{-0.05em } = \sum_k s_k u(-\infty , t_1)\big [ i_s\otimes b^*(\gamma_k ) \big ] u^*(-\infty , t_1 ) { |{0,\ldots,0}\rangle}_s{|{0}\rangle}_f , \nonumber\end{aligned}\ ] ] where @xmath125 denotes the unitary time evolution . we here set the basis functions @xmath68 to the rising exponential functions @xmath209 given in eq . , meaning that the input pulse shape is chosen as @xmath210 then , noting that @xmath211 , we obtain @xmath212 \otimes { |{0}\rangle}_f,\ ] ] where @xmath213 with @xmath178 appearing only in the @xmath214th entry . therefore , through the interaction , at time @xmath118 the system completely acquires the input code state with coefficient @xmath215 ; the resulting system state is highly entangled among the nodes ( see fig . 2 ( b ) ) . the optimal input pulse shape is given by the rising exponential function of the form , as expected in section 4.2 . but the point here is that we now know that the parameter vector @xmath170 in eq . exactly corresponds to the superposition coefficients @xmath215 . together with the structure of the memory subsystem , this fact tells us how we should design the input pulse shape @xmath50 ; this will be more precisely discussed in the next subsection . * case ii : coherent state . * next , let us consider the case where the input is the coherent field state . again the system is in the ground state at @xmath126 . then , through the interaction the whole state becomes @xmath216 therefore , by setting the basis functions @xmath68 to the rising exponential i.e. @xmath217 and again noting that @xmath218 , we obtain @xmath219 that is , the system state is changed to the product of coherent states @xmath220 . hence , similar to the single photon case , the perfect state transfer is possible by sending the information @xmath221 over the rising exponential pulse field . the idea is the same for the coherent input case . that is , the state @xmath233 can be perfectly transferred and stored in the memory subsystem . the perfect memory procedure for the single photon input state in a 5-nodes passive linear network . the system can be tuned so that the ( 3 , 4 , 5 ) nodes become decoherence free ; hence these nodes constitute the memory subsystem . the ( 1 , 2 ) nodes does the buffer subsystem . ( a ) the single photon code state with @xmath227 is sent through the input optical field with pulse shape @xmath124 . ( b ) at time @xmath118 , the perfect state transfer has been completed . the system is then modulated and the memory subsystem is decoupled . ( c ) the transferred state is preserved during the period @xmath234 $ ] . ( d ) at @xmath235 the memory subsystem is again coupled to the buffer subsystem and thus the optical field . ( e ) the perfect copy appears in the output optical field with pulse shape @xmath236 . ] suppose that the state has been perfectly stored during the period @xmath234 $ ] ; hence the reading stage starts at time @xmath235 with the initial state @xmath237 for the single photon input case or @xmath238 for the coherent input case ; see fig . 2 ( d ) . note that , as described in section 5.2 , only some elements of @xmath215 or @xmath221 , which represents the classical information of the stored state , are not zero . to retrieve this initial state , we switch the system matrices so that the memory subsystem again couples to the buffer subsystem and thus the optical field ; in particular , we take the same system matrices @xmath34 and @xmath36 ( and thus @xmath21 ) as in the writing stage . thus note that @xmath21 is hurwitz . to describe the reading stage , first , we particularly focus on the following quantity : @xmath239dt \nonumber \\ & & \hspace*{-3em } = \big [ \int_{t_2}^\infty e^{a^\dagger ( t - t_2 ) } c^\dagger c e^{a ( t - t_2 ) } dt\big ] a(t_2 ) - \int_{t_2}^\infty e^{a^\dagger ( t - t_2 ) } c^\dagger c e^{a t } \big [ \int_{t_2}^t e^{-a s } c^\dagger b(s)ds \big ] dt \nonumber \\ & & \hspace*{1em } + \int_{t_2}^\infty e^{a^\dagger ( t - t_2 ) } c^\dagger b(t ) dt . \nonumber\end{aligned}\ ] ] the first term is @xmath240 . for the second and the third terms , by defining @xmath241 , which leads to @xmath242 , we find that they become @xmath243 dt = -e^{at_2}k(t_2)=0 . \nonumber\end{aligned}\ ] ] as a result , we have @xmath244 where @xmath245 as in the previous case , @xmath246 are orthonormal ; @xmath247 . note that @xmath236 is a generalization of a _ decaying _ exponential function . moreover , eq . leads to @xmath248^\top . \nonumber\end{aligned}\ ] ] * case i : single photon state . * the initial state is now @xmath237 . then , through the interaction , this state changes to : @xmath249 \otimes { |{0}\rangle}_f = u(t_2 , \infty ) \big [ \sum_k s_k a_k^*(t_2 ) \big ] { |{0,\ldots,0}\rangle}_s{|{0}\rangle}_f \nonumber \\ & & \hspace*{-0.8em } = u(t_2 , \infty ) \big [ \sum_k s_k a_k^*(t_2 ) \big ] u^*(t_2 , \infty ) { |{0,\ldots,0}\rangle}_s{|{0}\rangle}_f \nonumber \\ & & \hspace*{-0.8em } = \sum_k s_k \big [ i_s\otimes b^*(\tilde{\nu}_k ) \big ] { |{0,\ldots,0}\rangle}_s{|{0}\rangle}_f = { |{0,\ldots,0}\rangle}_s \otimes \sum_k s_k { |{1_{\tilde{\nu}_k}}\rangle}_f . \nonumber\end{aligned}\ ] ] therefore , certainly the field state recovers the input state , which is now carried by the output field with pulse shape . the system state returns to the ground state ; see fig . 2 ( d , e ) . * case ii : coherent state . * the initial state is @xmath238 . then , through the interaction , this state changes to : @xmath250 where @xmath251 is a coherent field state with pulse shape @xmath252 thus , similar to the single photon input case , the stored coherent states @xmath253 leaks into the output field with pulse shape , and we can retrieve the full information about @xmath221 contained in the coherent field state @xmath251 . here we derive the time evolution equations of the statistics in the writing stage . these equations are useful for numerical simulation , as demonstrated in the next section . * case i : single photon state . * in the case of single photon state , we evaluate the following matrix of operators : @xmath254 [ a_1 , \ldots , a_n].\ ] ] the photon is distributed in the system according to the statistics represented by the correlation matrix @xmath255 . the time - evolution equation of @xmath256 is , together with the vector @xmath257^\top$ ] , given by @xmath258 the solution of eq . is readily obtained as @xmath259 where @xmath260 is used . in general , the lyapunov differential equation @xmath261 , with @xmath262 time varying , has the solution of the form @xmath263 if @xmath21 is hurwitz , in the limit of @xmath126 , this becomes @xmath264 using this result and the expression , we have @xmath265 if we send the single photon code state over the input field with rising exponential pulse shape @xmath266 , then we have @xmath267 due to @xmath268 . this means that the input single photon state is distributed among the network so that the @xmath214th node has the mean photon number @xmath269 at time @xmath102 . * case ii : coherent state . * in this case the statistics is more convenient , because a coherent state is completely charactered only by its mean and variance . in particular , the dynamics of the mean , @xmath270 , was already obtained in eq . , with @xmath77 particularly given by @xmath271 in eq . . hence it is immediate to obtain the solution in @xmath173 : @xmath272 hence , by taking the limit @xmath126 , we have @xmath273 ; i.e. @xmath274 . also , we evaluate the covariance matrix @xmath275 with @xmath276 , which takes zero if and only if the state is a coherent state . similar to the single photon case , we find that @xmath277 obeys @xmath278 , which readily yields @xmath279 as @xmath126 . as a result , the @xmath214th node becomes the coherent state @xmath220 at time @xmath118 . we note that the mean of the output field is @xmath280 for all @xmath173 ; thus the zero - dynamics principle is certainly satisfied . this section is devoted to study a passive linear network composed of atomic ensembles , which contains a tunable df component . a numerical simulation will demonstrate how the input field state is transferred to the memory subsystem and how the input pulse shape to be engineered for perfect memory looks like . the passive linear network composed of three large atomic ensembles and a ring cavity . @xmath281 denotes the cavity mode , and @xmath282 is the annihilation operator approximating the collective lowering operator of the @xmath214th atomic ensemble . ] we can prove that , when @xmath300 , the matrix @xmath21 is hurwitz ; i.e. the real part of all the eigenvalues of @xmath21 is negative . a convenient way to see this fact is to use the property that the controllability matrix @xmath301 $ ] is of full rank iff @xmath21 is hurwitz @xcite . thus the system does not contain a df component when @xmath300 . on the other hand , if we turn off the magnetic field and set @xmath302 , then a df subsystem appears , as shown below . let us take the following unitary matrix : @xmath303.\ ] ] this transforms the system equation to @xmath304 where @xmath305 , \nonumber \\ & & \hspace*{0em } a'=u{^{\dagger}}au = \left [ \begin{array}{cccc } -\kappa/2 & \sqrt{3}g & 0 & 0 \\ -\sqrt{3}g & 0 & -\sqrt{2}i\delta/2 & \sqrt{6}i\delta/6 \\ 0 & -\sqrt{2}i\delta/2 & -i\delta/2 & -\sqrt{3}i\delta/6 \\ 0 & \sqrt{6}i\delta/6 & -\sqrt{3}i\delta/6 & i\delta/2 \\ \end{array } \right ] , \nonumber \\ & & \hspace*{0em } c ' = cu = [ \sqrt{\kappa},~0,~0,~0].\end{aligned}\ ] ] therefore , when @xmath302 , the system takes a form of eq . . that is , @xmath306^\top$ ] is not affected by the incoming field @xmath5 and it does not appear in the output field @xmath6 ; hence @xmath306^\top$ ] is the memory subsystem that can be switched to a df or non - df subsystem , just by controlling the external magnetic field . this two - mode subsystem works as a perfect memory that preserves any state of the form @xmath307 in the case of single photon state or @xmath308 in the case of coherent state . note that @xmath309 and @xmath310 depend on the atomic modes @xmath311 and not on the cavity mode @xmath281 , implying that the state is indeed stored in the atomic ensembles . also it should be remarked that @xmath309 and @xmath310 take the form of continuous - variable _ syndromes _ used for quantum error correction @xcite . here we describe the concrete procedure of the writing , storage , and reading processes , in the case of single photon input ; see fig . [ fig example ] . * a single photon field state is prepared in the form @xmath312 , where @xmath313 and @xmath314 are the third and fourth elements of the vector of rising exponential functions @xmath315 with @xmath316 and @xmath317 given in eq . . note in this stage the magnetic field is on ; @xmath300 . * the field couples to the system until @xmath173 . the perfect state transfer is achieved in the end , at @xmath118 , by sending the input state over the optical field with pulse shape @xmath318 described above . the whole state changes to @xmath319 . * we turn off the magnetic field and set @xmath302 ; then the memory subsystem with modes @xmath320 becomes decoherence free and its state @xmath321 is preserved during an arbitrary time interval @xmath234 $ ] . * at a later time @xmath322 , we turn on the magnetic field ( i.e. set @xmath300 ) to retrieve the stored state . then the memory subsystem again couples to the optical field , and the perfect copy @xmath323 appears in the output field with the pulse shape specified by @xmath324 . the memory procedure for the passive linear network composed of three atomic ensembles trapped in a single - mode cavity . the number @xmath325 indicates the subsystem with mode @xmath326 . ( a ) the single photon state is sent through the input optical field with pulse shape @xmath318 , where in this stage the magnetic field is turned on ( @xmath300 ) . ( b ) at time @xmath118 the system acquires the state @xmath307 ; that is , the input state is perfectly transferred into the 3rd and 4th nodes . ( c ) then the magnetic field is turned off ( @xmath302 ) so that the memory subsystem with modes @xmath327 is decoupled from the buffer subsystem with modes @xmath328 and the input - output optical field ; hence it becomes decoherence free and the transferred state is perfectly preserved . ( d ) at @xmath235 we again set @xmath300 . the memory subsystem again couples to the buffer subsystem and the optical field . ( e ) the perfect copy appears in the output field with the pulse shape @xmath329 . ] recall that the optimal input pulse shape is determined by the properties ( zeros ) of the memory system and this corresponds to the impedance matching mentioned in section 3.2 . now we should note that an additional matching condition is not imposed on the interaction between the cavity mode and the atomic ensembles , although perfect state transfer from the former to the latter is certainly achieved . this result seems to be inconsistent with the fact obtained in @xcite , showing that perfect state transfer from a cavity to an inhomogeneously broadened ( ib ) atomic ensemble requires a strict impedance matching between them . but there is a clear difference between our case and those studies ; in the case dealing with the ib ensemble , due to the matching condition , an input field state with _ arbitrary _ ( yet within a finite band - width ) temporal shape is allowed to be completely absorbed into the ensemble ( see e.g. @xcite for the recent experimental results ) , while in our case the optimal pulse shape has to be strictly specified . exploring a combined schematic of these two memory procedures , which would allow weaker pulse shaping and weaker impedance matching , should be an interesting future work . ( a ) time evolutions of the absolute value of @xmath313 ( red ) and @xmath314 ( green ) . ( b ) time evolution of the mean photon number at the @xmath330th nodes , @xmath331 . the blue , black , red , and green lines represent the time evolutions of @xmath332 , and @xmath333 , respectively . ] here we demonstrate a numerical simulation of the writing stage of the above memory procedure . the parameters are set to @xmath334 , @xmath335 , and @xmath336 ; note again in this stage @xmath300 and there is no df subsystem . the input is a single photon field state with coefficients @xmath337 , which is carried by the optical field with pulse shape @xmath313 and @xmath338 as mentioned above . the initial time is @xmath339 and the stopping time is @xmath133 . first , fig . [ network simulation ] ( a ) shows the absolute value of @xmath313 and @xmath338 . these are the pulse shapes we need to correctly engineer for the desirable perfect state transfer . a notable point is that they are not anymore of a rising exponential shape such as eq . ; particularly they take the value zero at the stopping time @xmath133 . a similar non - rising exponential pulse function was also found in @xcite , achieving perfect state transfer in an integrated quantum memory system . it looks that we can realize this kind of pulse shape by combining some gaussian wave packets , which might be a desirable feature from the engineering viewpoint . next , fig . [ network simulation ] ( b ) shows the time - evolutions of the mean photon number of each node , i.e. @xmath340 , which can be computed by numerically solving eqs . and . as expected from the theory , the memory subsystem with modes @xmath327 perfectly acquires the photon with mean photon number @xmath341 at @xmath133 . we should note that the transportation of the photon from the input field to the memory subsystem occurs rapidly only in the last few period ; in fact , almost all the energy contained in the input pulses @xmath313 and @xmath314 is confined in this short period . hence , we need to be very careful to stop the writing process at the accurate time @xmath133 , because the desired state @xmath342 is fragile in the following sense . for instance if we turn off the magnetic field a bit earlier than @xmath133 , say @xmath343 , then the whole system s state generated is roughly @xmath344 ( unnormalized ) ; thus the state of the memory subsystem becomes a mixed state ( unnormalized ) @xmath345 due to the decoherence added to the buffer subsystem with modes @xmath328 during the storage period . hence , an important future work is to find a suitable set of parameters @xmath346 so that the time - evolutions of the mean photon numbers of the memory subsystem become as flat as possible at the stopping time @xmath102 . in this paper , for a general passive linear system , we have provided a designing method of input pulse shape that perfectly transports a single photon or coherent field state to a memory subsystem , which can be switched to a df subsystem . the method is general and simple , so it can be directly applied to a large - scale network ; in fact , in the example studied in section 7 , we found that the explicit form of @xmath313 and @xmath314 are readily obtained . the results are based on the zero - dynamics principle . although in this paper this principle was used only for synthesizing the input pulse shape , it is indeed a wide concept that works in a more general situation . for example , the zero - dynamics principle can be applied to the case where , instead of pulse shaping of the input field , some time - varying controllable parameters of the system should be engineered due to practical limitation ; also the system can be nonlinear ; further , we could deal with an inhomogeneously broadened atomic ensemble memory that allows an arbitrary temporal shape for perfect state transfer , which was discussed in section 7.2 . in any case , following the zero - dynamics principle , we should design the system so that the output is zero or more generally the output is minimized . moreover , the zero - dynamics corresponds to the time - evolution of a state free from any energy loss , thus it represents a coherent , yet non - unitary , gate operation on the system state for quantum information processing ; that is , designing a desired manipulation of a state in an open system is no more than designing a desired zero - dynamics . all these problems will be addressed in future works . the authors acknowledges helpful discussions with m. r. hush and a. r. r. carvalho . this research was supported by the australian research council centre of excellence for quantum computation and communication technology ( project number ce110001027 ) , and afosr grant fa2386 - 12 - 1 - 4075 . also ny s work was supported by jsps grant - in - aid number 40513289 . in this paper , we thoroughly study a passive linear system , but there are many systems containing an _ active _ component . in general , for such an active system the energy balance identity does not hold , hence the zero - dynamics principle does not anymore mean the perfect energy transfer . hence , it should be worth doing a case study to see if an active system could allow perfect state transfer . let us consider the following active system : @xmath347 = -\frac{1}{2}\left [ \begin{array}{cc } \kappa & -\epsilon \\ -\epsilon & \kappa \\ \end{array } \right ] \left [ \begin{array}{c } a \\ a^ * \\ \end{array } \right ] -\sqrt{\kappa } \left [ \begin{array}{c } b \\ b^ * \\ \end{array } \right].\ ] ] in optics , this represents the dynamics of an optical parametric oscillator , where @xmath348 denotes the squeezing strength @xcite . note that the system becomes passive if @xmath349 . the above equation can be explicitly solved : @xmath350 \nonumber \\ & & \hspace*{-3em } \mbox { } - \sqrt{\kappa } \int_{t_0}^{t_1 } e^{-\kappa(t_1-s)/2 } \big [ \sinh(\epsilon(t_1-s)/2 ) b(s)ds + \cosh(\epsilon(t_1-t_0)/2 ) b^*(s)ds \big ] . \nonumber\end{aligned}\ ] ] unlike the passive case , the field annihilation operator @xmath351 appears in the equation . then under the same setting taken in section 3 where the input field state is given by a superposition of the vacuum and @xmath352 with the pulse shape function @xmath353 given below , we obtain ( @xmath126 and @xmath133 ) @xmath354 \otimes { |{0}\rangle}_f - \frac{\beta \epsilon}{\sqrt{2\kappa^2-\epsilon^2 } } u(t_0 , t_1 ) b(\xi_2 ) u^*(t_0 , t_1 ) { |{0}\rangle}_s{|{0}\rangle}_f , \nonumber\end{aligned}\ ] ] where @xmath355 this equation implies that , when @xmath356 , the perfect state transfer is impossible due to the third term , which clearly stems from the active element of the system . to carry out efficient state transfer , we need some approximation ; in the above case , if @xmath112 is much bigger than @xmath348 , then the system state becomes approximately the desired one to be stored . another example is found in @xcite , where the system is an atomic ensemble containing an active component , but by introducing a fast oscillating magnetic field it is approximated by a passive one , which was further shown to be a perfect memory . the basic idea of dark state principle is as follows . for a system coupled to a probe field , we continuously monitor the system by a photo detector measuring the output field ; then if the detector counts no photon , this means that the system is in a _ dark state _ and has a time evolution without loss of energy . here we apply this dark state principle to the writing problem discussed in section 3 and derive the same result ; that is , in this sense , the zero - dynamics principle and the dark state principle are equivalent , though there is a big difference in practice as shown below . first let us consider the case where we want to send a coherent field state to the system . in general , if we use a photon counter to estimate the system observables , our state ( knowledge ) conditioned on the measurement results is updated by the following stochastic master equation @xcite ( the scattering operator is now set to be the identity ) : @xmath357\alpha + [ l , \rho]\alpha^*\big)dt + \big [ \frac{1}{\cal n } ( l\rho l^ * + \alpha^*l\rho + \alpha \rho l^ * + |\alpha|^2\rho ) - \rho \big ] ( dy - { \cal n}dt),\ ] ] where @xmath358 + l\rho l^ * - l^ * l\rho/2 - \rho l^ * l/2,~~ { \cal n } = \tr\big [ \rho\big(l^*l + \alpha^*l + \alpha l^ * + |\alpha|^2\big ) \big].\ ] ] @xmath359 is the pulse shape of the input coherent light field and @xmath360 is the measurement result ( 0 or 1 ) obtained during the small time interval @xmath361 . also @xmath362 and @xmath363 are the system operators . now since the ensemble averaging over the measurement results leads to a standard master equation , we have @xmath364 . then , the counting probability of the measurement result 1 " during @xmath361 is given by @xmath365 . hence if @xmath366 , the system is in a dark state and loses no energy into the output field ; the state satisfying this condition is called the dark state . in our case where the system is the single - mode passive linear system with @xmath367 and @xmath368 , the dark state can be specified to a coherent state @xmath369 because we now know that the system s state is always a coherent state . the condition @xmath370 then becomes @xmath371 , which yields @xmath372 . now , under the dark state condition the time evolution of the conditional state is identical to that of the averaged one , which consequently leads to @xmath373 . these two equations yield @xmath374 , thus the input pulse shape must be a rising exponential function @xmath375 . next let us consider the case where the input is a single photon field state . as in the above case , the dark state principle is represented in terms of the conditional state subjected to the single - photon stochastic master equation ; see eq . ( 43 ) in @xcite . in this case the probability to obtain the measurement result 1 " during @xmath361 is given by @xmath376 where @xmath50 is the temporal pulse shape of the single photon field . @xmath377 are the operators characterizing the conditional state . under the dark state condition @xmath370 , they obey @xmath378\xi + [ l , \rho^{10}]\xi^*,~~ \dot{\rho}^{10 } = { \cal l}\rho^{10 } + [ \rho^{00 } , l^*]\xi,~~ \dot{\rho}^{00 } = { \cal l}\rho^{00},\ ] ] and @xmath379 , which are identical to the single - photon master equation @xcite . the initial conditions are @xmath380 and @xmath381 . in our case @xmath367 and @xmath368 , these differential equations can be explicitly solved , yielding @xmath382 , @xmath383 , and @xmath384 , where @xmath27 and @xmath385 satisfy @xmath386 and @xmath387 . substituting these solutions @xmath377 for the dark state condition @xmath370 , we have @xmath388 . combining these three equations , we end up with the relation @xmath389 and thus see that the input pulse shape has to be a rising exponential function @xmath390 . summarizing , we have recovered the same result obtained in section 3 , showing that the dark state principle is equivalent to the zero - dynamics principle . both principles require no energy leaking from the system into the output field , but their approaches are different ; the dark state principle is represented in the schrdinger picture , while the heisenberg picture is used to describe the zero - dynamics principle . as a result , in the former case we need to solve the master equation , which is sometimes a hard task as demonstrated above especially in the single photon field case . on the other hand , we have seen in section 4.2 that the zero - dynamics principle allows us to derive the rising exponential function very easily , even in the general setup ; also the transfer - function - based treatment of the principle is notable and possibly very useful from the viewpoint of the applicability of the linear response theory to quantum memory . of course these special advantages appear particularly in the linear case , and for more general nonlinear memory systems we should be careful in choosing the approach . 10 t. chaneliere , d. n. matsukevich , s. d. jenkins , s .- y . lan , t. a. b. kennedy , and a. kuzmich , storage and retrieval of single photons transmitted between remote quantum memories , nature * 438 * , 833/836 , 2005 . d. a. lidar and k. b. whaley , decoherence - free subspaces and subsystems , irreversible quantum dynamics , edited by f. benatti and r. floreanini , lecture notes in physics * 622 * , 83 , springer berlin / heidelberg , 2003 . a. v. gorshkov , a. andre , m. d. lukin , and a. s. sorensen , photon storage in lambda - type optically dense atomic media , i. cavity model , phys . a * 76 * , 033804 , 2007 ; ii . free - space model , phys . a * 76 * , 033805 , 2007 ; iii . effects of inhomogeneous broadening , phys . a * 76 * , 033806 , 2007 .

we discuss finite difference techniques for hyperbolic equations in non - trivial domains , as those that arise when simulating black hole spacetimes . in particular , we construct dissipative and difference operators that satisfy the _ summation by parts _ property in domains with excised multiple cubic regions . this property can be used to derive semi - discrete energy estimates for the associated initial - boundary value problem which in turn can be used to prove numerical stability .

gr-qc0308007

an important problem in astrophysics is to model in a detailed way the collision of two black holes @xcite . this requires numerically integrating einstein s field equations and extracting from the simulations the relevant physical information . unfortunately , it is difficult to obtain numerical solutions of these equations in generic three - dimensional settings , especially for long term simulations . obstacles to this goal are encountered in both the analytical and numerical arenas . in the _ analytical _ one , the formulation of a well posed initial - boundary value problem is not completely understood . this includes the definition of proper initial and boundary conditions and the equations determining the future evolution of the fields . in the _ numerical arena _ one seeks to define numerical techniques that allow for long term accurate evolutions . this requires the construction of appropriate discrete operators to implement the initial - boundary value problem . to date , despite considerable advances in both fronts , the challenge of simulating generic three - dimensional black hole systems remains unattained . this article intends to provide some initial steps for setting up numerical techniques suitable to address the numerical stability of equations like the ones in question , by extending and devising finite difference techniques to tackle first order symmetric hyperbolic problems in non - trivial domains , with numerical stability being guaranteed in the linear case . furthermore , via a local argument , one can assert that these methods should be useful in evolving smooth solutions of quasi - linear symmetric hyperbolic equations as well , as is the case of the full , non - linear einstein vacuum equations when appropriately written @xcite . although the main motivation of this work is to present techniques for the simulation of einstein s equations on domains with excised regions , the techniques here presented are readily applicable to any symmetric hyperbolic problem in such domains . applications of these techniques in a variety of scenarios will be presented elsewhere @xcite . this work is organized in the following way . in sections [ sect - numstab ] and [ sect - ob ] we review some of the issues involved in obtaining stable numerical schemes through the energy method , to set the stage for the specialized discussions that follow . in section [ sect - excision ] , the main new results of the article are presented : we derive three dimensional difference operators satisfying summation by parts for non - trivial domains which enable one to obtain energy estimates . we further introduce dissipative operators which do not spoil these estimates . to complete a stability proof , one needs to impose boundary conditions without affecting the semi - discrete energy estimates obtained thanks to the use of the constructed dissipative and difference operators . one way of doing so is by imposing boundary conditions through an orthogonal projection , as done by olsson @xcite , a technique which can even be applied to non - smooth domains . conclusions are drawn in section [ sect - conclusions ] . the appendices include technicalities , discussions about courant limits and further details about the implementation of boundary conditions . in order to construct stable finite difference schemes for initial - boundary value problems ( ibvps ) associated with partial differential evolution equations we will use the method of lines @xcite . this means that we first discretize the spatial derivatives appearing in the partial differential equations so as to obtain a large system of ordinary differential equations for the grid functions . this system is usually called _ semi - discrete system_. in the following , we assume that the solutions of the partial differential equations satisfy an _ energy estimate _ which bounds some norm of the solution at some fixed time @xmath0 in terms of a constant @xmath1 which is independent of the the initial and boundary data times the same norm of the initial data and a bound on the boundary data . as we will discuss in more details below , one can derive a similar energy estimate for the semi - discrete system if the partial derivatives and the boundary conditions are discretized in an appropriate way . provided that the constant @xmath2 involved in this semi - discrete estimate can be chosen to be resolution - independent for small enough resolution , this immediately implies numerical stability for the semi - discrete problem . an important ingredient in the derivation of such estimates is the _ summation by parts _ ( sbp ) property which is the discrete analogue of _ integration by parts_. there are many possible discretizations of the partial derivatives and the boundary conditions which are consistent with the ibvp and which yield a stable semi - discrete scheme , and in general , the constant @xmath2 will be larger than the corresponding constant @xmath3 of the continuum problem . by carefully choosing the discretization one can achieve optimal semi - discrete energy bounds , in the sense that @xmath4 which means that the norm of the solution to the semi - discrete problem satisfies the same estimate as the norm of the solution to the analytic solution . alternatively or complementary , one might want to add artificial dissipation in order to control the high frequency modes of the solution which are always poorly represented with finite difference approximations . finally , by discretizing the time derivatives one obtains the fully discrete system which is numerically implemented . if the semi - discrete system is stable one can show that the fully discrete system is stable as well , provided an appropriate time integrator is chosen . this will be briefly discussed later on . to summarize , numerical stability in this approach reduces to * formulate a well posed ibvp for the problem one wants to solve at the continuum . * construct difference operators for the domain of interest that satisfy the sbp property . * optionally or even complementary : * * construct dissipative operators that do not spoil the energy estimate . * * rearrange the semi - discrete equations to achieve optimal energy estimates . * impose boundary conditions without spoiling the semi - discrete stability . * achieve fully discrete stability by appropriately choosing the time integrator . the next subsections give an overview of these points . section [ sect - ob ] discusses some issues that appear in the ibvp case , while section [ sect - excision ] presents results for the case in which there are inner boundaries and excised regions in the computational domain . there are many definitions of well posedness though roughly , and without going into too many technical details , an ibvp is said to be well posed if 1 . a local in time solution with certain smoothness exists , 2 . the solution is unique , and , 3 . the solution depends continuously on the initial and boundary data of the problem . there are different approaches to obtaining well posed formulations of a given problem . a common one is the so called energy method @xcite where one seeks an energy norm which has the form @xmath5 where @xmath6 , the solution of the ibvp at some given time @xmath0 , lies in some hilbert space with scalar product @xmath7 . in many physical situations , this hilbert space can be motivated by the existence of a conserved energy . in general however , the expression @xmath8 does not need to coincide with any physical energy . for first order symmetric hyperbolic linear systems , the hilbert space can be taken to be the space of square integrable functions over some domain @xmath9 with scalar product @xmath10 where the symmetrizer @xmath11 is a symmetric positive definite matrix - valued function on @xmath9 which is bounded from above and away from zero , and @xmath12 denotes the standard scalar product between the vector - valued functions @xmath13 and @xmath14 . in this case , and for suitable boundary conditions , one has an a priori estimate of the form @xmath15 where @xmath16 is independent of the initial and boundary data , and @xmath17 bounds the boundary data . this means that the solution at time @xmath0 can be bounded by the energy at @xmath18 and a bound on the energy pumped in through the boundary of the domain . in particular , for each fixed @xmath19 , small variations in the data result in small variations of the solution . as an example , consider the initial - boundary value problem @xmath20 with smooth initial and boundary data , @xmath21 and @xmath22 , respectively , that satisfy appropriate compatibility conditions at @xmath23 . we also assume appropriate fall - off conditions at @xmath24 . taking a time derivative of the energy ( [ energynorm ] ) with @xmath25 and using the evolution equation ( [ eq : ev ] ) we obtain @xmath26 where we have used integration by parts in the last step . using the boundary condition ( [ eq : bc ] ) and integrating one obtains the energy estimate ( [ energygrow ] ) with @xmath27 and @xmath28 . the energy method is not only a valuable tool in studying the system at the analytical level , but it can also be used to produce stable numerical discretizations by considering a discrete analogue . essentially , this involves defining discrete operators with which one reproduces , at the discrete level , the steps taken at the continuum when obtaining the energy estimate . in what follows we briefly review the steps involved . before we proceed , we mention that the existence of an a priori energy estimate , where the energy norm has the form of an integral over a local density expression , is a sufficient but not necessary condition for well posedness @xcite . a different approach for analyzing stability is the laplace method @xcite which gives necessary conditions for well posedness . however , the application of this method to obtain sufficient conditions is rather cumbersome and problem dependent . on the other hand , the energy method is considerably simpler and so it is preferred when applicable . this is the approach we shall follow in this article . it can be applied to a very large set of physical problems . as we have seen , a tool that is used in the derivation of energy estimates is integration by parts . in one dimension ( 1d ) it reads @xmath29 where @xmath30 . let @xmath31 be a positive integer , @xmath32 and consider the grid defined by @xmath33 , @xmath34 . the function @xmath35 is approximated by a grid function @xmath36 . one of the ingredients needed to obtain an energy estimate for the semi - discrete problem is to construct difference operators d ( approximating the first order partial differential operator @xmath37 ) which satisfy a sbp rule @xcite @xmath38 } + ( dv , u)_{\sigma } ^{[0,n]}= \left . v_j \cdot u_j\right|_{j=0}^n \ ; , \label{eq : sumbyparts}\ ] ] with respect to some scalar product @xmath39 } = \delta x\sum_{i , j=0}^n \tilde{\sigma}_{ij } u_i \cdot v_j \ , . \label{scalar_product2}\ ] ] here , the weights @xmath40 must be chosen such that the norm @xmath41 , which can be seen as the discrete version of that defined in eq . ( [ energynorm ] ) , is positive definite . achievement of sbp in general requires a careful choice of _ both _ the scalar product and difference operator ; for a given choice of scalar product there might not exist any difference operators satisfying the sbp property . in the following , we skip the label @xmath42 when it refers to the trivial scalar product @xmath43 . also , when there is no need to specify the domain , we skip the superscript @xmath44 $ ] . having a 1d operator that satisfies sbp with respect to a _ diagonal scalar product _ @xmath45 , one can construct a 3d operator by simply applying the 1d difference operator in each direction . the resulting 3d operator satisfies sbp with respect to the scalar product @xmath46 with coefficients ( making a slight abuse of notation ) @xmath47 @xcite . this is _ not _ necessarily true if @xmath48 is not diagonal . for this reason , we will only consider diagonal scalar products in this article . in the absence of boundaries , sbp reduces to @xmath49 provided suitable fall off conditions are imposed . in this case simple 1d operators that satisfy sbp with respect to the trivial scalar product are the standard centered difference operators @xmath50 ; for example , @xmath51 etc . defining a 3d difference operator by just applying this one - dimensional one in each direction will satisfy sbp with respect to the trivial 3d scalar product . as a simple application , consider the initial value problem for the symmetric hyperbolic system , @xmath52 where @xmath53 is a vector valued function , the initial data @xmath21 vanishes for @xmath54 sufficiently large , and where the matrices @xmath55 are symmetric . assume for simplicity that the @xmath55 are constant and that @xmath56 . at the analytic level , the energy norm defined by eq . ( [ energynorm ] ) with @xmath57 satisfies @xmath58 . if we replace @xmath59 by the centered differencing operator @xmath60 in the @xmath61th direction , the discrete energy @xmath62 with respect to the trivial discrete scalar product satisfies the same estimate : @xmath63 , and the scheme is stable by construction . even though one can achieve numerical stability through sbp , it is sometimes convenient to add artificial dissipation to the problem . one possible reason for doing so is the presence of high frequency modes which even if they go away with increasing resolution grow in time at fixed resolution . although this would not be the case in our present discussion , it is well known that the addition of dissipation can aid in stabilizing schemes that would otherwise be numerically unstable . an example of this would be the case of system that is stable at the semidiscrete level for example , a symmetric system with discretized with operators satisfying sbp but becomes unstable because the region of absolute stability of the time integrator does not contain a neighborhood of the origin along the imaginary axis . as an example , consider the initial value problem ( [ eq : hypeq],[eq : id ] ) . the standard way @xcite to introduce dissipation in the problem is to modify the right - hand side ( rhs ) of the equations @xmath64 where @xmath60 is a differential operator satisfying sbp and where @xmath65 is a differential operator that vanishes in the limit of infinite resolution , such that the consistency of the scheme is not altered , and such that @xmath66 with some inner product @xmath67 for which a discrete energy estimate holds ( for example , the one with respect to which sbp holds ) . the dissipative property ( [ dissipation ] ) ensures that the discrete energy estimate is not spoiled and might be useful to stabilize an otherwise unstable scheme . furthermore , @xmath65 can be chosen such that it controls spurious high frequency modes in the solution . as mentioned above , in the absence of boundaries , standard centered derivatives satisfy sbp with respect to the trivial scalar product . similarly , the kreiss oliger dissipation @xcite @xmath68 where @xmath69 denote the one - sided difference operators , satisfies the dissipative property ( dp ) with respect to that scalar product . furthermore , if the accuracy of the scheme without artificial dissipation is @xmath70 , choosing @xmath71 does not affect the accuracy of the scheme . notice also that with a slight abuse of notation we have used @xmath72 to denote the dissipative parameter ; this should not cause confusion with respect to @xmath73 used in the context of weights of scalar products , since the latter have subindices . so far , we have discussed how to obtain energy estimates for the semi - discrete problem by constructing finite difference operators that satisfy sbp . these estimates imply numerical stability , by providing a discrete analogue of eq . ( [ energygrow ] ) . although the resulting scheme is stable the discrete estimate in principle might not agree with that found at the continuum ; that is , the constant @xmath74 appearing in the discrete analogue of eq . ( [ energygrow ] ) might be larger than @xmath16 . if one can show that @xmath75 we say that the scheme is _ strictly stable_. ] . the solution to non - strictly stable schemes can have unwanted features such as artificial growth in time of the errors . in the limit @xmath76 these features would disappear ; but one would like to avoid or minimize them even at fixed resolution . for first order symmetric hyperbolic linear systems , like the one described by eqs . ( [ eq : hypeq],[eq : id ] ) , one can achieve strict stability @xcite by rewriting the partial differential equation in skew - symmetric form : @xmath77 one can show that the resulting scheme is strictly stable with respect to the energy @xmath62 defined by the trivial discrete scalar product . in contrast to this , the simple discretization @xmath78 , while yielding a stable scheme , does not necessarily yield a _ strictly _ stable scheme if the matrices @xmath55 are not constant . strict stability is particularly useful if the formulation at the continuum admits a sharp estimate . in this case , the construction of a strictly stable scheme can be exploited to rule out artificial growth of the solutions @xcite . consider , for example , a system with time - independent coefficients , @xmath79 let @xmath11 be a symmetrizer , i.e. a symmetric positive definite matrix - valued function @xmath80 which is bounded from above and away from zero and which is such that the matrices @xmath81 are symmetric . if @xmath11 can be chosen such that @xmath82 one can show that the energy given by eq . ( [ energynorm ] ) is conserved , and that by rewriting the semi - discrete equations as @xmath83 the semi - discrete energy does not grow either . therefore the numerical solution will not have spurious growth even at a fixed resolution . in principle , there is no reason why such a symmetrizer should exist . however , on physical grounds , there should be one whenever there is a well defined local energy density . the usefulness of having a conserved energy at the semi - discrete level is discussed in detail in ref . @xcite . proceeding through the steps above described ( [ wp]-[sbp ] ) , and optionally ( [ disip]-[stricstability ] ) , one obtains an energy estimate for the _ semi - discrete problem _ which implies numerical stability for this semi - discrete system . however , one is of course interested in the stability of the _ fully discrete problem_. one particular simple but powerful approach to achieve this goal is to follow the strategy based on the method of lines and employ a time - integrator guaranteed to give rise to a stable scheme . a useful feature of this approach is that one can derive conditions for the time integrator that are sufficient for fully discrete stability and independent of the details of the spatial discretization @xcite . two rather straightforward options are given by third and fourth order runge - kutta schemes @xcite . in appendix [ app - cfl ] we present an elementary discussion , for a wave equation in a domain without boundaries , that gives an idea of possible values of the courant factor . for more complicated equations and domains see ref . @xcite . so far our discussion has ignored the presence of boundaries . however these are unavoidable in most problems of interest . in the next sections we discuss how to modify the previously described techniques such that one recovers an energy estimate in the presence of boundaries . for the sake of clarity in the presentation , we first concentrate on simple domains which do not posses inner boundaries and then consider the case of domains with holes . we here discuss appropriate finite - difference schemes for simple 3d domains . consider the following ibvp in the cubic domain @xmath84\times [ 0,1]\times [ 0,1]$ ] , with maximally dissipative boundary conditions ( @xmath85 denoting its ( non - smooth ) boundary ) , @xmath86 where the matrices @xmath55 are symmetric and @xmath87 represent the in- and outgoing characteristic variables with respect to the unit outward normal to the boundary . it is assumed that the coupling matrix @xmath88 does not depend on time and is small enough so as to imply an energy estimate @xcite . it is also assumed that the initial and boundary data are smooth enough and satisfy compatibility conditions at the intersection between the initial surface and boundaries . if one discretizes the rhs of equation ( [ eq : symhypsystem ] ) according to @xmath89 where the discrete derivative operator @xmath60 satisfies sbp , one will obtain an energy estimate , modulo boundary contributions that appear after sbp . as already mentioned , a simple strategy to construct 3d operators satisfying sbp is to use in each direction a 1d operator satisfying sbp with respect to a diagonal scalar product , and this is what we do in what follows . existence proofs of high order difference operators and scalar products satisfying sbp in 1d with boundaries can be found in @xcite . explicit expressions for these operators can be found in @xcite ( some of these operators have non - diagonal associated scalar products , and therefore their use beyond one dimensional cases does not guarantee numerical stability ) . in the interior the derivatives are approximated by one of the centered finite difference operators mentioned in the previous section , and the operator is modified near the boundary points . the simplest example is @xmath90 where the operators @xmath91 are defined in eqs . ( [ eq : secondsd ] ) and ( [ eq : onesided ] ) . the operator ( [ eq : strand21 ] ) is second order accurate in the interior and first order at the boundaries . it satisfies sbp , eq . ( [ eq : sumbyparts ] ) , with respect to the diagonal scalar product @xmath92 , @xmath93 for @xmath94 . for high order difference operators , the operator and the weights in the scalar product may needed to be modified at points that are near the boundary points as well . if the boundary is not smooth and possesses corners and edges the prescription remains the same , provided the vertices and edges are _ convex_. since in the diagonal case the 3d scalar product is just the product of the 1d one , the weights @xmath95 are @xmath96 in the interior points , and in the outer boundary : @xmath97 at the boundary faces , @xmath98 at the edges and @xmath99 at the vertices . the presence of _ concave edges and vertices _ requires modifications , which need to be treated carefully . these will arise , for instance , when considering a computational domain with an interior excised region . we discuss this case in section iv . having introduced the simple difference operator ( [ eq : strand21 ] ) that satisfies sbp in a cubic domain with respect to the 1d scalar product @xmath100 , we now construct some operators that satisfy the dissipative property with respect to the same scalar product . we start with the dissipation operator ( [ ko ] ) for @xmath101 but redefine @xmath72 and rewrite the operator as @xmath102 in the absence of boundaries , it satisfies the dp for all @xmath103 . notice that unless @xmath104 , the dissipation parameter @xmath72 is not dimensionless . as we will see later , this redefinition is convenient when boundaries are present . the goal is to define @xmath65 through eq . ( [ dis2 ] ) for @xmath105 , and to extend its definition to @xmath106 such that the dp holds . through a straightforward expansion one gets @xmath107 } = \frac{\delta x}{2 } u_0 q_du_0 + \sigma ( \delta x)^s ( u , d_+d_-u)^{[1,n-1 ] } + \frac{\delta x}{2 } u_n q_du_n \\ & & = -\sigma ( \delta x)^s \left ( \| d_- u \|^{[1,n ] } \right)^2 + \frac{\delta x}{2}u_0\left[q_du_0 - 2\sigma ( \delta x)^{s-1}d_+u_0\right ] \nonumber\\ & & + \frac{\delta x}{2}u_n\left[q_du_n + 2\sigma ( \delta x)^{s-1}d_-u_n\right ] \ ; , \end{aligned}\ ] ] where the second equality comes from eq . ( [ prop2 ] ) of appendix [ app - bp ] . certainly , there are many possible definitions of @xmath65 at the boundary points that imply the dp , eq . ( [ dissipation ] ) . the simplest one is @xmath108 note that this dissipative term vanishes in the limit @xmath109 only if @xmath110 . on the other hand , in those cases , the amplification factor of the high frequency modes depends on the resolution already in the absence of boundaries . that is , consistency requires @xmath110 , and in those cases the amplification factor converges to the non - dissipative one as @xmath109 . this means that the dissipative operator just constructed can not be expected to cure difference schemes which are unstable . it can only be applied to systems where the amplification factor does not become greater than one in the limit of high resolution ; that is , to schemes which are already stable in the absence of dissipation . we now improve on this . in the absence of boundaries , the standard way of correcting this is by considering dissipative operators of the form ( [ ko ] ) with @xmath111 , which satisfy the dp and do not change the accuracy of a scheme which uses @xmath70th order accurate difference operators . therefore , we now look for corrections to the operator ( [ ko ] ) at and near the boundary points . for simplicity , let us consider the case of a dissipative operator for an otherwise second order scheme ; that is , assume we are using difference operators of order two in the interior , and first order at boundary points . then , for @xmath112 we define @xmath65 through equation ( [ ko ] ) with @xmath113 , that is , we set @xmath114 where we have redefined @xmath72 , as before . the modification at and near the boundary points is motivated by a calculation that is similar to the one presented above : @xmath115 } & = & \frac{1}{2}\delta x u_0 q_du_0 + \delta x\ ; u_1 q_d u_1 - \sigma ( \delta x)^s \left(u , d_+^2d_-^2u \right)_{[2,n-2 ] } \nonumber\\ & & + \ ; \delta x u_{n-1 } q_du_{n-1 } + \frac{1}{2 } \delta x \ ; u_n q_d u_n \\ & = & u_0\left(\frac{\delta x}{2}q_du_0 + \sigma ( \delta x)^{s-1}d_+^2u_0 \right ) \nonumber\\ & & + \ ; u_1\left[\delta xq_du_1 + \sigma ( \delta x)^{s-1 } \left(d_+^2 \ , \ , - 2d_+d_-\right)u_1\right ] \nonumber\\ & & - \sigma ( \delta x)^s\left ( \|d_+d_-u\|^{[1,n-1 ] } \right)^2 \nonumber\\ & & + \ ; u_{n-1}\left [ \delta xq_du_{n-1 } + \sigma ( \delta x)^{s-1 } \left(d_-^2 \ , \ , -2d_+d_-\right)u_{n-1 } \right ] \nonumber \\ & & + \ ; u_n\left(\frac{\delta x}{2}q_du_n + \sigma ( \delta x)^{s-1}d_-^2 u_n \right)\ ; , \end{aligned}\ ] ] where in the last equality we have used properties listed in appendix [ app - bp ] . as before , there is more than one way of satisfying the dp , the simplest one being : @xmath116 for @xmath117 the high frequency modes are now damped in a resolution independent way while the dissipative term vanishes in the limit @xmath109 . in this case , the dissipative operator constitutes a third order correction in the interior and a first order one at the boundary points and the two grid points next to it . adding this operator to the rhs of the equations does not affect the consistency of the scheme , but reduces in one the order of accuracy at the grid points @xmath118 and @xmath119 ( recall that the difference scheme was already first order at the boundary points , so the accuracy there is not changed ) . finally , the maximally dissipative boundary conditions ( [ eq : maxdiss ] ) are implemented by multiplying the rhs of ( [ eq : symhypsystem_discrete ] ) from left by an appropriate projection operator @xcite @xmath120 . for homogeneous boundary conditions , @xmath120 is the projection on the space of grid functions that satisfy the boundary conditions ( [ eq : maxdiss ] ) with @xmath121 which is orthogonal with respect to the scalar product for which sbp holds . the orthogonality of @xmath120 makes sure that the energy estimate is not spoiled , and hence numerical stability still follows . furthermore , if @xmath120 is time - independent , a solution to the resulting semi - discrete system automatically satisfies the boundary conditions . inhomogeneous boundary conditions are discussed in appendix [ app - bc ] . in the previous section we restricted our discussions to computational domains without inner boundaries . now we extend those results to non - trivial computational domains . to simplify the discussion we restrict ourselves to the case where a single cubic box is cut out of the computational domain . however , it is straightforward to see that the difference and dissipative operators constructed in this section can also be applied to a domain with multiple cubic boxes excised from the computational domain , such that sbp and the dissipative property ( dp ) are satisfied . we will only analyze the case in which the second order centered difference operator @xmath122 , defined in eq . ( [ eq : secondsd ] ) , is used in the interior . the possibility of constructing higher order accurate operators is discussed in appendix [ app - fodiss ] . the modification of @xmath122 and its associated scalar product at the outer boundary points has already been discussed in the previous section . plane that passes through the interior of the excised cube ( left panel ) , or cuts the top or bottom of it ( right panel , also shown in fig.[fig : leftright ] ) . the weights @xmath95 that appear in the scalar product are 1 at interior grid points ( circles ) , @xmath123 at boundary faces ( squares ) , @xmath124 at convex edges ( hexagons ) , @xmath125 at convex vertices , @xmath126 at concave edges ( triangles ) and @xmath127 at concave vertices ( pentagons ) . [ fig : grid],title="fig:",height=302 ] plane that passes through the interior of the excised cube ( left panel ) , or cuts the top or bottom of it ( right panel , also shown in fig.[fig : leftright ] ) . the weights @xmath95 that appear in the scalar product are 1 at interior grid points ( circles ) , @xmath123 at boundary faces ( squares ) , @xmath124 at convex edges ( hexagons ) , @xmath125 at convex vertices , @xmath126 at concave edges ( triangles ) and @xmath127 at concave vertices ( pentagons ) . [ fig : grid],title="fig:",height=302 ] in order to define the scalar product and the difference operators at the excision boundary such that sbp holds , we restrict ourselves , without loss of generality , to an arbitrary line which is parallel to the @xmath128 axis . for points on such a line we assume that the difference operator in the @xmath128 direction depends only on neighboring points on this line ( as we will see next , this assumption suffices for our purposes ) . in order to label the grid points on such a line we only need the index @xmath61 and assume that there is an inner boundary at , say , @xmath129 . we will use centered differences at the neighbors @xmath130 and @xmath131 ( this will also turn out to be sufficient ) but leave the scalar product at these points undefined for the moment . that is , @xmath95 at points @xmath132 takes some values @xmath133 , @xmath134 , @xmath135 , respectively . note that since the difference operators at the outer boundaries have already been introduced in section [ sect - ob ] , for the sake of clarity and without loss of generality , we will now ignore this outer boundary and consider the case where only an inner boundary is present . therefore , in what follows , the subindices range from @xmath136 to @xmath137 . by splitting the scalar product in the different sub - domains one gets ( omitting the sum over @xmath138 and @xmath139 ) @xmath140 } + ( v , d_0u)^{(-\infty,-1 ] } \right ) + \nonumber\\ & & \beta\delta x \left ( u_0 dv_0 + v_0 du_0 \right ) + \gamma \left ( ( u , d_0v)^{[1,\infty ) } + ( v , d_0u)^{[1,\infty ) } \right ) \label{sbp } \\ & = & \frac{\alpha } { 2}(v_0 u_{-1 } + u_0v_{-1 } ) + \beta \delta x \left(u_0 dv_{0 } + v_0 du_0\right ) - \frac{\gamma } { 2 } ( u_0 v_1 + v_0u_1)\;. \nonumber\end{aligned}\ ] ] we now assume that the modified difference operator at @xmath129 has a stencil of width three ; that is , we assume that it has the form @xmath141 with @xmath70 , @xmath142 , @xmath143 to be determined later . taylor - expanding eq . ( [ gen_der ] ) in @xmath144 one obtains that in order for @xmath145 to be consistent with @xmath146 the conditions @xmath147 must be satisfied . inserting eq . ( [ gen_der ] ) and these conditions in eq . ( [ sbp ] ) one sees that in order for sbp to hold , the mixed terms of the form @xmath148 must vanish . this yields @xmath149 equations ( [ consistency ] ) and ( [ mixed ] ) completely fix , at @xmath129 , the derivative operator ( i.e. the coefficients @xmath70 , @xmath142 , @xmath143 ) , and the scalar product ( i.e. , @xmath134 ) : @xmath150 notice that , unless @xmath151 ( which in our case would correspond to @xmath129 actually not being part of a boundary , as we will now see ) , the modified difference operator ( [ nder ] ) is only a first order accurate approximation of @xmath146 , as is the case at the outer boundary points . we now explicitly write down the scalar product and difference operator just constructed for the different possible inner boundary points : * _ edges : _ + by definition of an edge , one neighbor lies in the interior ( and has scalar product weight equal to @xmath118 ) and the other lies at a face ( and has scalar product weight equal to @xmath123 ) . therefore , eq . ( [ beta ] ) implies that @xmath152 . thus , the expressions for the derivative operator ( depending on the face s side ) are , + [ cols="<,^,^",options="header " , ] the dissipative operators computed in section [ artdiss ] need to be modified at the excision boundary . we again assume that the grid point under consideration at the inner boundary is at index @xmath153 with @xmath154 , and that at the neighbors , @xmath155 , @xmath156 . we expand the corresponding scalar product : @xmath157 } + ( u , q_du)_{\sigma}^{(0,0 ) } + ( u , q_du)_{\sigma } ^{[1,\infty ) } \nonumber\\ & = & \alpha ( u , q_du)^{(-\infty,-1 ] } + \beta ( u , q_du)^{(0,0 ) } + \gamma ( u , q_du)^{[1,\infty ) } \nonumber\\ & = & \alpha \sigma ( \delta x)^s ( u , d_+d_-u)^{(-\infty,-1 ] } \nonumber\\ & & + \ ; \beta \delta x\ ; u_0 q_du_0 + \gamma \sigma ( \delta x)^s ( u , d_+d_-u)^{[1,\infty ) } \nonumber\\ & = & -\sigma ( \delta x)^s\left [ \alpha \left(\|d_-u\|^{(-\infty,0 ] } \right)^2 + \gamma \left ( \|d_-u\|^{[1,\infty ) } \right)^2 \right ] \nonumber\\ & & + \ ; u_0\left [ \beta q_d \delta x + \sigma ( \delta x)^s\left(\alpha d_- - \gamma d_+ \right ) \right ] u_0 \ ; , \nonumber\end{aligned}\ ] ] where the last equality is due to eq . ( [ prop2 ] ) . therefore one ( non - unique , but simple ) possibility is to choose : @xmath158 in which case @xmath159 } \right)^2 + \gamma \left ( \|d_-u\|^{[1,\infty ) } \right)^2 \right ] \;.\ ] ] making the expressions for the dissipation explicit : * _ edges : _ + @xmath160 * _ vertices : _ + @xmath161 the modification in this case follows along similar lines : @xmath162 } + ( u , q_du)_{\sigma } ^{[-1,-1 ] } + + ( u , q_du)_{\sigma}^{[0,0]}\nonumber\\ & & + \ ; ( u , q_du)_{\sigma}^{[1,1 ] } + ( u , q_du)_{\sigma } ^{[1,\infty ) } \nonumber\\ & = & \alpha ( u , q_du)^{(-\infty,-2 ] } + \alpha ( u , q_du)^{[-1,-1 ] } + \beta ( u , q_du)^{[0,0 ] } \nonumber\\ & + & \gamma ( u , q_du)^{[1,1 ] } + \gamma ( u , q_du)^{[2,\infty ) } \nonumber\\ & = & -\alpha \sigma ( \delta x)^s \left ( u,(d_+d_-)^2u \right)^{(\infty,-2 ] } + \alpha \delta xu_{-1}q_du_{-1}\nonumber\\ & & + \ ; \beta \delta x\ ; u_0 q_du_0 + \gamma \delta xu_1q_du_1 - \gamma \sigma ( \delta x)^s ( u , d_+^2d_-^2u)^{[2,\infty ) } \\ & = & -\sigma ( \delta x)^s\left[\alpha \left(\|d_+d_-u\|^{(-\infty,-1 ] } \right)^2 + \gamma \left ( \|d_+d_-u\|^{-\infty,-1 ] } \right)^2\right ] \nonumber\\ & & + \ ; u_{-1}\left[\sigma ( \delta x)^{s-1}\alpha ( 2d_+d_--d_-^2)+\alpha \delta x q_d \right]u_{-1}\nonumber\\ & & + \ ; u_0\left [ \beta \delta xq_d + \sigma ( \delta x)^{s-1}(\alpha d_-^2+\gamma d_+^2 ) \right]u_0\nonumber\\ & & + \ ; u_1\left[\gamma \delta x q_du_1 + \sigma ( \delta x)^{s-1}\gamma ( d_+^2 - 2d_+d_- ) \right]\ ; , \nonumber\end{aligned}\ ] ] where the last equality is due to eq . ( [ prop3 ] ) . therefore one possible choice is @xmath163 the case @xmath117 should be the preferred one , since then the amplification factor does not depend on resolution . making the expressions for the dissipation explicit : * _ edges : _ + @xmath164 * _ vertices : _ + @xmath165 we have presented a set of 3d operators satisfying summation by parts which can be used in non - trivial domains . the use of these operators to numerically implement first order symmetric hyperbolic systems guarantee stability of the semi - discrete system . additionally , the use of appropriate time integrators and a consistent treatment of the boundary provide a systematic way to achieve a stable implementation . furthermore , in order to rule out artificial growth , we have defined dissipative operators which do not spoil the energy growth estimates . the usefulness of these techniques will be highlighted in their application to different problems elsewhere . in particular , these applications include the evolution of scalar and electromagnetic fields propagating in black hole backgrounds in @xmath166 @xcite ; scalar fields in @xmath167 with a moving black hole @xcite ; bubble spacetimes which require evolving the 5d einstein equations in the presence of symmetries @xcite , the construction of schemes capable to deal with the axis of symmetry in @xmath167 , axisymmetric scenarios @xcite and 3d numerical simulations of single black holes in full nonlinear gr @xcite . we would like to thank h.o . kreiss , d. neilsen , p. olsson , j. pullin , and e. tadmor for very useful discussions , comments and suggestions throughout this work . this work was supported in part by grants nsf - phy-9800973 , nsf - phy-0244335 , nsf - int-0204937 , phy99 - 07949 , nasa - nag5 - 13430 , by funds of the horace hearne jr . institute for theoretical physics , and by agencia crdoba ciencia , conicet , and secyt unc . ll has been partially supported by the sloan foundation . in this appendix the definition and some properties of the first order @xmath168 and second order accurate @xmath169 finite difference operators are given . the proofs of the following statements can be found in ref . @xcite , for instance . one can show that with respect to the scalar product and norm @xmath170 } \equiv \sum_{j = r}^s u_jv_j \delta x , \qquad \left(\|u\|^{[r , s ] } \right)^2 = ( u , u)^{[r , s]}\;,\ ] ] the following properties hold : @xmath171 } & = & - ( d_-u , v)^{[r+1,s+1 ] } + u_j v_j |_r^{s+1}\\\ & = & -(d_+u , v)^{[r , s ] } - h(d_+u , d_+v)_{[r , s ] } + u_jv_j|_r^{s+1}\ ; , \\ ( u , d_-v)^{[r , s ] } & = & - ( d_+u , v)_{[r-1,s-1 ] } + u_j v_j |_{r-1}^s \\ & = & -(d_-u , v)_{[r , s ] } + h(d_-u , d_-v)_{[r , s ] } + u_jv_j|_{r-1}^s\ ; , \\ ( u , d_0v)^{[r , s ] } & = & -(d_0u , v)_{[r , s ] } + \frac{1}{2}(u_jv_{j+1 } + u_{j+1}v_j)|_{r-1}^s\;.\end{aligned}\ ] ] the following equalities are needed in the derivation of the modified derivative and dissipative operators near the boundary @xmath172 } & = & \frac{1}{2}(u_ju_{j+1})|_{r-1}^s\ ; , \\ ( u , d_+d_-u)^{[r , s ] } & = & -\left ( \|d_-u\|^{[r , s+1 ] } \right)^2 + u_{s+1}d_-u_{s+1 } - u_{r-1}d_+ u_{r-1 } \ ; , \label{prop1}\\ ( u,(d_+d_-)^2u)^{[r , s ] } & = & \left ( \|d_+d_-u\|^{[r-1,s+1 ] } \right)^2 + \frac{u_{s+1}}{\delta x}\left(2d_+d_- - d_-^2\right)u_{s+1 } \nonumber\\ & & - \frac{u_{s+2}}{\delta x}d_-^2 u_{s+2}+\frac{u_{r-1}}{\delta x}\left(2d_+d_- - d_+^2\right)u_{r-1 } - \frac{u_{r-2}}{\delta x}d_+^2 u_{r-2 } \ ; . \label{prop3 } \end{aligned}\ ] ] a sketch of the calculation that leads to the last equality is : @xmath173 } & = & \left ( \|d_+d_-u\|^{[r , s ] } \right)^2 \nonumber\\ & & + \left . \left ( u_j d_-d_+d_- u_j- d_-u_j d_+d_-u_j\right)\right|_r^{s+1 } \label{prop2}\\ & = & ( \|d_+d_-u\|^{[r , s ] } ) ^2 + \frac{1}{h}\left . ( u_{j-1}d_+^2u_{j-1 } - u_jd_-^2u_j)\right|_r^{s+1 } \nonumber\\ & = & \left ( \|d_+d_-u\|^{[r-1,s+1 ] } \right)^2 + \frac{u_{s+1}}{\delta x}\left(2d_+d_- - d_-^2\right)u_{s+1 } \nonumber\\ & & - \frac{u_{s+2}}{\delta x}d_-^2 u_{s+2}+\frac{u_{r-1}}{\delta x}\left(2d_+d_- - d_+^2\right)u_{r-1 } - \frac{u_{r-2}}{\delta x}d_+^2 u_{r-2}\;. \nonumber\end{aligned}\ ] ] here we analyze the question of whether one can construct high order difference operators satisfying sbp in a domain with excised cubic regions . more precisely , we seek difference operators that are fourth order in the interior and , as in the rest of the article , we assume that the 3d operator is inherited by a 1d one with diagonal scalar product , that the operator approximating , say , @xmath146 , depends only on points to the left or the right . unfortunately the calculation below shows that such an operator can not be constructed . however , a difference operator that satisfies sbp and is fourth order accurate almost everywhere in the interior could be obtained by means of a different approach where one decomposes the domain into cubes @xcite . the finite difference operator @xmath174 is a fourth order approximation of @xmath175 . strand @xcite showed that there exists a unique second order accurate modification of this operator near the boundaries , given by @xmath176 that satisfies sbp with respect to a diagonal scalar product , with @xmath177 next we discuss the question of whether one can modify this operator near inner boundaries , for the case of excised cubic regions , in a way such that sbp holds . let us assume that a 2d domain has a concave corner at the grid point @xmath178 , as is the case of an edge at the inner boundary of our domain . the scalar product near this point will have the following structure @xmath179 the weights @xmath180 for @xmath181 are unknown . we know that they must be positive and symmetric @xmath182 . we also need to compute a second order accurate difference operator . consider one of the rows with @xmath183 . their contribution to the scalar product @xmath184 must be zero , as there are no boundary terms at the continuum . we have @xmath185 } + \delta x \sum_{i=0}^3 \sigma_{ij } u_i ( du)_i + ( u , d^{(4)}u)^{[4,+\infty ) } \ ; , \label{rowj}\ ] ] where @xmath186 is the fourth order accurate difference operator defined in eq . ( [ d4 ] ) . the contribution from the first and last term of eq . ( [ rowj ] ) can be calculated using the properties listed in appendix [ app - bp ] . in general , one has that @xmath187 } = \frac{1}{12 } \left.\left ( 8 u_{j+1 } u_j - u_{j+2 } u_j - u_{j+1}u_{j-1}\right)\right|_{r-1}^s\;.\ ] ] we want to solve @xmath188 for the coefficients of the modified difference operator @xmath189 and the weights of the scalar product @xmath180 for @xmath190 . the operator @xmath189 can be at most a @xmath191 point stencil , i.e. @xmath192 a larger stencil would give rise to terms in ( [ rowj ] ) that would not cancel . by inspection of eq . ( [ rowjbexp ] ) carefully , one sees that there are many coefficients of the modified difference operators that must vanish . this reduces the number of parameters from @xmath193 to @xmath194 . ( [ rowjbexp ] ) becomes @xmath195\\ & & + \sigma_{1j } u_1 \left [ a_1 u_3 + \left ( \frac{1}{2 } - 2a_1\right ) u_2 + \left(-\frac{1}{2}+2a_1\right ) u_0 - a_1 u_{-1 } \right]\\ & & + \sigma_{2j } u_2 \left [ a_2 u_4 + \left ( \frac{1}{2 } - 2a_2\right ) u_3 + \left(-\frac{1}{2}+2a_2\right ) u_1 - a_2 u_{0 } \right]\\ & & + \sigma_{3j } u_3 \left [ a_3 u_5 + \left(\frac{1}{2 } -\frac{15}{8 } a_3 + \frac{1}{8 } b_3\right ) u_4 + \left ( -\frac{1}{2 } + \frac{5}{4 } a_3 - \frac{3}{4 } b_3\right ) \right.\nonumber\\ & & \left.+ b_3 u_1 -\frac{3}{8}(a_3 + b_3)u_0 \right]\\ & & - \frac{1}{12 } ( 8u_4u_3 - u_5u_3 -u_4u_2 ) \;. \end{aligned}\ ] ] unless @xmath93 there is no solution ; to see this it is sufficient to look at the coefficients of @xmath196 , @xmath197 , @xmath198 , @xmath199 , @xmath200 , @xmath201 and @xmath202 . on the other hand , @xmath203 is in contradiction with the structure of the scalar product under consideration , see ( [ scheme ] ) . when discretizing a hyperbolic system of partial differential equations with an explicit scheme the courant - friedrich - lewy ( cfl ) condition has to be satisfied in order to have numerical stability . below we obtain necessary and sufficient conditions for the numerical stability of a 3d wave equation , using a standard von neumann analysis . in particular , we want to determine what is the maximum value of @xmath204 that one can use if artificial dissipation is added to the rhs . one can use this information as guide for choosing the courant factor in more general situations . we start by considering the scalar advective equation . consider the advective equation @xmath205 , with @xmath206 a real constant . if the spatial derivative is discretized using a second order centered differencing operator and the resulting semi - discrete system is integrated using a runge kutta time integrator of order @xmath207 , the solution at a time step @xmath208 can be expressed in terms of the solution at the previous time step @xmath209 as @xmath210 this difference scheme ( [ eq : scheme_advective ] ) is stable if and only if it satisfies the von neumann condition . the courant limit is @xmath211 for the @xmath212 case ( third order runge - kutta ) and @xmath213 for the @xmath214 case ( fourth order runge - kutta ) . these limits change when artificial dissipation is added to the rhs . the fully discrete system becomes @xmath215 the amplification factor depends on the parameters @xmath216 and @xmath217 . if , for a given value of @xmath218 , we numerically compute the value of @xmath219 beyond which the amplification factor becomes in magnitude greater than one for some frequency , we obtain the plot in figure [ fig : courantlimits_1d ] . . [ fig : courantlimits_1d],height=302 ] consider now the 3d wave equation @xmath220 written in first order form . as before , we use the second order centered difference operator to approximate the spatial derivatives and @xmath207-th order runge - kutta . the finite difference scheme is @xmath221 where @xmath222 , @xmath223 we assume that @xmath224 and set @xmath225 . in fourier space the difference scheme becomes @xmath226 and the amplification matrix is given by @xmath227 where @xmath228 , @xmath229 , @xmath230 and @xmath231 . the maximum values for the courant factor are plotted in figure [ fig : courantlimits_3d ] . . [ fig : courantlimits_3d],height=302 ] here we discuss how to discretize inhomogeneous boundary conditions of the type ( [ eq : maxdiss ] ) . let @xmath120 be the orthogonal projector onto the space of grid functions that satisfy the homogeneous version of the boundary conditions ( [ eq : maxdiss ] ) . by _ we mean that @xmath120 is hermitian with respect to the scalar product for which sbp holds . eq . ( [ eq : symhypsystem ] ) is discretized in the following way : @xmath232 where @xmath233 denotes the identity matrix on the space of grid functions . numerical stability for the resulting semi - discrete scheme can be shown under additional assumptions on the boundary data ( see @xcite ) . * discretize the partial derivatives using a difference operators that satisfies sbp . * at each @xmath234 * * calculate the outward pointing unit normal @xmath209 and let @xmath235 denote the boundary matrix . note that the normal at edges and vertices needs special consideration . in the next subsection it is discussed how this normal is to be defined in these cases . * * multiply the rhs by @xmath236 , where @xmath237 is the ( orthogonal ) matrix that diagonalizes @xmath238 , i.e. @xmath239 , where @xmath240 is a diagonal positive ( negative ) definite matrix . let @xmath241 ; * * apply @xmath120 . since it is hermitian with respect to the scalar product for which sbp holds , this projector is non - trivial in the sense that , in general , it overwrites the ingoing _ and outgoing _ modes . consider , for simplicity , the case in which there is one ingoing and one outgoing mode ( or more generally , the case in which the coupling matrix between ingoing and outgoing modes is diagonal ) . then , we define @xmath242 where @xmath243 is the coupling that appears in the boundary condition @xmath244 . note that only when @xmath245 does the outgoing mode remain unchanged , otherwise it is overwritten . in the continuum limit , this overwriting is , of course , consistent with the partial differential equation . * go back to the primitive variables @xmath14 by multiplying by @xmath237 . it is useful to have stability results that allow for cubic domains , since they are used quite often . the corners in 2d or edges and vertices in 3d deserve special attention . in particular , the normal to the boundary is not defined there . the stability of the whole scheme is affected by the treatment of these points . in ref . @xcite it is discussed how to control ( i.e. how to achieve numerical stability ) the boundary terms that appear after sbp at non - smooth parts of the boundary . here we will summarize some aspects of this treatment . for simplicity , let us assume that the domain is a 2d square @xmath84\times [ 0,1]$ ] and that the partial differential equation is a symmetric hyperbolic , constant - coefficients one with no principal terms , @xmath246 where @xmath247 and @xmath248 are constant symmetric matrices . if we define the energy norm to be @xmath249 , the time derivative of the energy norm will be given the boundary terms @xmath250 in particular , in these boundary terms , there is no contribution due to corners , since they constitute a set of measure zero . the semi - discrete energy estimate obtained by discretizing the rhs with a difference operator satisfying sbp is simply the discrete version of ( [ eq : contestimate ] ) : @xmath251 one needs to prescribe boundary conditions to the semi - discrete system at the corners . there , the unit vector @xmath209 is not defined . in @xcite it is shown how to define @xmath209 so that a bound to the semi - discrete energy estimate can be given . essentially , the idea is that by looking at the contribution coming from the corner @xmath252 in ( [ eq : semidiscrestimate ] ) , which is given by , @xmath253 ( note that the `` cross - terms '' @xmath254 and @xmath255 are _ not _ a typo but do result from a non - trivial contribution to the discrete energy ) , we see that we need to control the positive speed characteristic variables in the effective direction @xmath256 , where @xmath257 . for uniform grids with @xmath258 this effective unit vector at the corners of @xmath9 is @xmath259 this is equivalent to providing boundary conditions as if the normal was at 45 degrees with respect to the faces of the cube . which data to give might be completely or partially determined by compatibility conditions . as discussed before , the way in which the boundary conditions are imposed in @xcite , including at corners , is through a non - trivial orthogonal projector . one might wonder whether or not giving boundary conditions at corners along the normal to one of the two faces that define the corner yields an energy estimate . that is , if controlling characteristic fields in one of the directions ( say , the term @xmath260 ) implies that the other one is also under control because of some compatibility conditions . the general answer is _ no _ , but it might work in some particular cases . let us illustrate this with one example . consider the 2d symmetric hyperbolic system @xmath261 in the domain @xmath262 , @xmath263 . the characteristic fields in the @xmath128 direction are @xmath13 ( eigenvalue @xmath118 ) and @xmath14 ( eigenvalue @xmath264 ) . in the @xmath265 direction they are @xmath266 ( eigenvalue @xmath118 ) and @xmath267 ( eigenvalue @xmath264 ) . as boundary conditions in the @xmath268 and @xmath269 boundaries use @xmath270 and @xmath271 , respectively , with @xmath272 . defining the energy @xmath273 taking a time derivative , using the evolution equations and integrating by parts gives ( note that in the discrete analogue this is where the `` cross terms '' comes from ) @xmath274 at the discrete level one is essentially left with the discrete version of this expression provided operators that satisfy sbp are used . in particular , the corner s contribution is @xmath275 where the dots represent the contributions from the boundary from points other than the corner . one has to control the complete corner term as it will not be canceled or changed by any contribution from other boundary points . suppose @xmath276 . the projector should set , for example , @xmath277 to zero . suppose one does not do this but , instead , controls only the ingoing mode in the @xmath128 direction ( that is , @xmath14 ) by setting it to zero . this does give an energy estimate , since @xmath278 which has the appropriate sign ( that is , the corner term can be bounded from above by zero ) . suppose on the other hand that one does not set it to zero but , instead , couples it to the outgoing mode , @xmath279 , with @xmath280 . then @xmath281 and the energy estimate follows or not , depending on the sign of @xmath143 . b. bruegmann , w. tichy , and n. jansen , phys . lett . * 92 * , 211101 ( 2004 ) . m. anderson and r. matzner , gr - qc/0307055 ; u. sperhake , k. smith , b. kelly , p. laguna and d. shoemaker , phys . d * 69 * , 024012 ( 2004 ) ; d. shoemaker , k. smith , u. sperhake , p. laguna , e. schnetter and d. fiske , class . quantum grav . * 20 * , 3729 ( 2003 ) ; l. lindblom and m. scheel , phys . rev . * d * 67 , 124005 ( 2003 ) ; t. baumgarte and s. shapiro , astrophys . j. * 585 * , 930 ( 2003 ) ; p. laguna and d. shoemaker , class . quantum grav . * 19 * , 3679 ( 2002 ) ; h. j. yo , t. w. baumgarte and s. l. shapiro , phys . d * 66 * , 084026 ( 2002 ) ; m. alcubierre , b. brugmann , d. pollney , e. seidel , and r. takahashi , phys . rev . d * 64 * , 061501 , ( 2001 ) ; m. alcubierre and b. brugmann , phys . rev . d * 63 * , 104006 ( 2001 ) . h. yo , t. baumgarte and s. shapiro , phys . rev . d*64 * , 124011 ( 2001 ) ; m. a. scheel , l. e. kidder , l. lindblom , h. p. pfeiffer and s. a. teukolsky , phys . d * 66 * , 124005 ( 2002 ) ; kidder , scheel , and teukolsky , phys rev . d * 64 * , 064017 ( 2001 ) . see , for instance , h. o. kreiss , j. lorenz , `` initial - boundary value problems and then navier - stokes equations '' , academic press , new york ( 1989 ) or ref . @xcite . + for applications in general relativity see : j. m. stewart , class . quantum grav . * 15 * , n9 , 1581 ( 1998 ) ; g. calabrese and o. sarbach , j. math . 44 * , 3888 ( 2003 ) . e. tadmor , `` from semi - discrete to fully discrete : stability of the runge - kutta schemes by the energy method . ii '' , in `` collected lectures on the preservation of stability under discretization '' , lecture notes from colorado state university conference , fort collins , co , 2001 ( d. estep and s. tavener , eds . ) proceedings in applied mathematics 109 , siam 2002 , 25 - 49 . ( downloadable from http://www.math.ucla.edu/ tadmor / pub / subj.html )

observations of the binary pulsar psr j1141 using the parkes radio telescope over 9.3 years show clear time - variations in pulse width , shape and polarization . we interpret these variations in terms of relativistic precession of the pulsar spin axis about the total angular momentum vector of the system changing our view of the emission beam . over the nine years , the pulse width at the 50% level has changed by more than a factor of three , reaching a maximum value of nearly in early 2007 . large variations have also been observed in the 1400-mhz mean flux density ; this reached a peak of mjy in mid-2002 but over the past several years has been relatively steady at mjy . the pulse polarization has been monitored since 2004 april using digital filterbank systems and also shows large and systematic variations in both linear and circular polarization . position angle variations , both across the pulse profile and over the data span , are complex , with major differences between the central and outer parts of the pulse profile . we interpret the outer parts as representing the underlying magnetic field and fit the rotating - vector model to these regions . modelling of the observed position angle variations by relativistic precession of the pulsar spin axis shows that the spin - orbit misalignment angle is about and that the precessional phase has passed through during the course of our observations . at the start of our observations , the line - of - sight impact parameter was about in magnitude and it reached a minimum very close to around early 2007 , consistent with the observed pulse width variations . we have therefore mapped approximately one half of the emission beam , at least out to a radius of about , showing that it is very asymmetric with respect to the magnetic axis . the derived precessional parameters imply that the pre - supernova star had a mass of about 2 m and that the supernova recoil kick velocity was relatively small , between 100 and 250 km s , depending on the assumed systemic velocity . with the reversal in the rate of change of the impact parameter , we predict that over the next decade we will see a reversed `` replay '' of the variations observed in the past decade .

1001.1483

binary pulsars with short orbital periods and massive companions exhibit a range of relativistic effects , the most prominent of which are relativistic periastron precession , gravitational time dilation , transverse doppler effects , orbit decay due to loss of energy to gravitational waves and the shapiro delay resulting from the passage of the pulsar signal through the gravitational field of the companion . these effects may be described by the so - called `` post - keplerian '' parameters @xcite and are most readily observed in double - neutron - star systems such as the hulse - taylor binary pulsar , psr b1913 + 16 @xcite and the double pulsar , psr j0737@xmath03039a / b @xcite . another relativistic effect which is potentially observable in such systems is the precession of the pulsar spin axis resulting from coupling between the spin and orbital angular momenta @xcite . the angular rate of precession , @xmath10 , is given by : @xmath11 where @xmath12 is the gravitational constant , @xmath13 is the velocity of light , @xmath14 is the binary orbital period , @xmath15 is the orbital eccentricity and @xmath16 and @xmath17 are the pulsar and companion masses respectively . for psr b1913 + 16 , the expected rate from equation ( [ eq : prec_freq ] ) is @xmath18 yr@xmath9 , corresponding to a precessional period of about 300 years , whereas for psr j0737@xmath03039a / b the rate is about a factor of four larger and the precessional period is about 75 years . since the emission from pulsars is believed to be beamed , probably along the open field lines associated with the magnetic poles on the neutron star , precession of the pulsar spin axis will result in changes in the beam aspect as viewed from the earth . one would expect this to lead to changes in the observed pulse profile and changes attributed to this effect have indeed been observed in psr b1913 + 16 @xcite and psr b1534 + 12 @xcite . for psr b1913 + 16 , a significant fraction of the expected beam diameter has been traversed in the time since its discovery and this has been used to map the two - dimensional structure of the emission beam @xcite . because of its relatively short precessional period , similar effects were expected to be observable for the double pulsar , but surprisingly , they have not so far been observed in the a pulsar @xcite . a possible reason for this is that a s spin - orbit misalignment angle ( the angle between the spin and orbital angular momenta ) is small ( see also * ? ? ? * ; * ? ? ? however , large variations in the pulse profile and orbital visibility of pulsar b have been observed @xcite and these are certainly due to some combination of relativistic periastron precession and precession of the b spin axis . there is now direct evidence for the relativistic spin precession of pulsar b. the emission from pulsar a is eclipsed for about 30 seconds when it passes behind pulsar b @xcite with the eclipse profile showing modulation related to the spin period of pulsar b @xcite . modelling of the long - term variations in the eclipse profile are consistent with the relativistic precession of b s spin axis with a spin - orbit misalignment angle of about @xmath19 @xcite . psr j1141@xmath06545 is a 394-ms pulsar in an eccentric 4.7-h binary orbit with a relatively massive companion ( @xmath20 m@xmath21 ) discovered in the parkes multibeam pulsar survey @xcite . with these characteristics , relativistic effects are expected to be detectable and indeed , measurement of the relativistic advance of periastron was reported in the discovery paper . the expected rate of relativistic precession of the pulsar spin axis is 136 yr@xmath9 , corresponding to a precessional period of about 265 yr . unusually for such a binary system , the pulsar characteristic age is relatively low , about @xmath22 yr , suggesting that the initially more massive primary star evolved to form a white dwarf , the present companion star , and in the process transferred mass to the secondary which subsequently exploded leaving behind a neutron star , the present pulsar @xcite . continued timing observations @xcite reinforced these ideas and gave detections of relativistic time dilation and orbital decay which were consistent with the predictions of general relativity . a clear orbital modulation in the timescale for interstellar scintillation was detected by @xcite , giving estimates of the orbital inclination angle @xmath23 ( or @xmath24 ) @xmath25 and the system transverse space velocity , @xmath26 km s@xmath9 . from measurements of the relativistic shapiro delay , @xcite derive a value for the inclination angle of @xmath27 , consistent with the scintillation measurement . observations at frequencies around 1.4 ghz by @xcite showed that the mean pulse profile for psr j1141@xmath06545 changed significantly over the five years from 1999 july to 2004 may ( mjd range 51381 53134 ) , with an approximately linear increase in @xmath28 , the pulse width at 10% of the pulse peak . furthermore , based on two measurments separated by about 0.7 yr , there appeared to be steepening of the gradient of polarization position angle ( pa ) @xmath29 , where @xmath30 is the pa and @xmath31 is the pulse phase ( measured in the same units as the pa ) near the profile center , at a rate of @xmath32 yr@xmath9 . variations of pa across pulse profiles are often well described by the `` rotating - vector model '' ( rvm ) in which the emission is assumed to be polarized parallel ( or perpendicular ) to the projected direction of magnetic fields in the vicinity of a magnetic pole on the rotating neutron star @xcite . in the rvm , @xmath30 varies as @xmath33 @xmath34 is the pa at @xmath35 , the center of symmetry of the pa variation and the projected direction of the rotation axis for a dipole field , @xmath36 is the inclination of the observer s line of sight relative to the rotation axis of the star , @xmath37 is the inclination of the magnetic axis relative to the rotation axis and @xmath38 is the impact parameter of the observer s line of sight , that is , the minimum angle between the magnetic axis and the observer s line of sight which occurs at @xmath39 . , [ sec : beam ] and [ sec : kick ] , @xmath30 is defined in a right - handed coordinate system , increasing in the clockwise direction from east ( looking toward the source ) following @xcite ( hereafter dt92 ) and @xcite . this is opposite to the iau convention in which pa increases in the counterclockwise direction from north toward east looking toward the source . the polarization conventions are discussed in more detail in section [ sec : obs ] . ] from equation [ eq : rvm ] , @xmath40 where @xmath41 is the maximum value of @xmath29 at @xmath35 . therefore , the increase in pa gradient observed by @xcite for psr j1141@xmath06545 was interpreted as a decrease in @xmath42 , implying that our line of sight was moving closer to the beam axis . this and the changing pulse width were interpreted in terms of relativistic precession of the pulsar spin axis with a spin - orbit misalignment angle greater than @xmath43 and probably less than @xmath44 . in this paper we report on observations of psr j1141@xmath06545 made over 9.3 years from 1999 august to 2008 november with the parkes radio telescope at frequencies close to 1.4 ghz . since 2004 april , we have observed with systems recording full polarization data but before that , only total intensity data were obtained . the observations and analysis procedures are described in section [ sec : obs ] and results are presented in section [ sec : results ] . in section [ sec : model ] we describe the interpretation of the observed variations in terms of relativistic precession of the pulsar spin axis and the implied shape of the emission beam is described in section [ sec : beam ] . the implications of our precessional model for the formation of the system are described in section [ sec : kick ] . in section [ sec : summary ] we summarize results and give our conclusions . we have observed psr j1141@xmath06545 using the parkes 64-m radio telescope between 1999 august and 2008 november ( mjds 51411 to 54785 ) using ( at different times ) two receiving systems , either the center beam of the parkes 20-cm multibeam ( mb ) receiver @xcite or the `` h - oh '' receiver . the mb receiver has a bandwidth of about 300 mhz centered at about 1.4 ghz and a system equivalent flux density ( @xmath45 ) of approximately 29 jy . the h - oh receiver has a wider bandwidth , approximately 600 mhz from 1.2 to 1.8 ghz , but a somewhat higher @xmath45 of about 35 jy . both receivers have orthogonal linearly - polarized feeds and a linearly - polarized , broad - band and pulsed calibration signal which can be injected into the feed at @xmath46 to the two signal probes . four different backend systems were used , all of which recorded data across a total bandwidth of 256 mhz : an analogue filterbank ( afb ) having 512 frequency channels on each polarization and recording total - intensity data using a one - bit digitiser system ( see * ? ? ? * ) , a wideband correlator ( wbc ) which recorded all four stokes parameters with 1024 frequency channels , and two digital filterbank systems ( pdfb1 and pdfb3 ) which also recorded full polarization data with 512 channels across the band . for the wbc and the pdfbs , each observation was preceded by a short ( 2-min ) observation of the pulsed calibration signal . observations of hydra a , assumed to have a flux density of 43.1 jy at 1.4 ghz , were used to set the flux - density scale . off - line processing made use of the psrchive pulsar data analysis system @xcite . a log of the observations is given in table [ tb : obs ] . afb observations , which commenced soon after the discovery of the pulsar , were recorded on tape and processed off - line to form mean total intensity profiles . frequency channels containing known interference were given zero weight before data were summed across the band . for the purposes of this paper , observations made over intervals ranging from a few hours to a few weeks have been grouped as shown in table [ tb : obs ] . pulse widths and flux densities were determined for each observation and mean values and their uncertainties for the group determined from these . the flux - density scale for the afb data was established by comparison with contemporaneous wbc observations . wbc and pdfb observations were typically of 1-h duration and were folded on - line to form 1-min sub - integrations . frequency channels containing strong interference , affected by resonances in the feed or with low system gain ( normally 5% of the band at the band edges ) and sub - integrations affected by strong impulsive interference were given zero weight . using the pulsed calibration signal as a reference , the variations in instrumental gain and phase across the band were removed , stokes parameters formed and the data placed on a flux - density scale . data were summed to form 15-min sub - integrations and mean profile widths and flux densities computed as for the afb data . all profiles were recorded with 1024 bins across the pulse period . observed pas given in this paper follow the astronomical convention , with pa measured from celestial north and increasing toward east ( counter - clockwise looking toward the source ) . the ieee definition of circular polarization is adopted ( i.e. , at a given point in space and looking at the source , the e - vector for a right - circular wave rotates in a counter - clockwise direction ) . both of these definitions are in accord with iau recommendations ( trans . iau , 15b , 166 , 1973 ) . stokes @xmath47 is defined in the sense left - circular minus right - circular to conform with established pulsar polarization conventions which pre - date the iau recommendations . file header parameters defined the polarization setup for each receiver / back - end combination and these were used by psrchive to correct the stokes parameters so that they conformed to these definitions . for further details on polarization conventions and their implementation , see @xcite . the strong millisecond pulsar , psr j0437@xmath04715 , was observed almost every session and used to check the sign of the calibrated stokes parameters and the absolute value of position angle by comparison with the results given by @xcite which are known to be in accordance with these definitions . the linearly polarized intensity @xmath48 is a positive definite quantity which is biased by noise . plotted values have been corrected for this bias using the relations given by @xcite . before summing in frequency , the observed pulse profiles must be corrected for faraday rotation across the band . the nominal rotation measure ( rm ) for psr j1141@xmath06545 is @xmath49 rad m@xmath50 @xcite . rms were measured for each epoch having polarization data by summing the upper and lower halves of the bandpass separately using the nominal rm , taking a weighted mean of the pa differences between the upper and lower bands across the profile , recomputing the rm and then iterating until convergence . we show below that the observed pa variations across the pulse profile are complex , both in time and in frequency , with quite different behavior in the central and outer parts of the profile . consequently , rms were determined separately for the central and outer parts ; we believe that the outer parts represent the true interstellar rm . the mb receiver suffers from significant coupling between the nominally orthogonal feed probes . observed polarization variations resulting from this coupling are a function of parallactic angle and the coupling parameters can be measured by analysis of a series of observations of a polarized source covering a wide range of parallactic angles @xcite . psr j0437@xmath04715 was observed every few months , typically for 10 minutes each hour during a 10-hour transit , for each of the wbc and pdfb configurations used with the mb receiver ( table [ tb : obs ] ) . these data were analysed using the psrchive program pcm to determine the feed ellipticities and their relative orientation as a function of frequency across the band . observations of psr j1141@xmath06545 were then calibrated to remove the effects of the cross - coupling . cross - coupling in the h - oh receiver is an order of magnitude less and this calibration step was not required . figure [ fg:0437 ] shows ( truncated ) pdfb1 polarization profiles for psr j0437@xmath04715 taken with the mb and h - oh receivers using the observational and processing methods described above . the results are essentially identical within the noise uncertainties , confirming that our calibration procedures are robust.4715 is only + 1.5 rad m@xmath50 @xcite and so the pa difference expected between 1369 mhz and 1433 mhz is only about @xmath51 , not visible in figure [ fg:0437 ] . ] while not the primary objective of this work , it was necessary to monitor period variations of the pulsar in order make accurate predictions of the topocentric period for use during the observations and to sum observations in off - line processing . initially , the timing model of @xcite was used for this purpose . however , while processing the 2007 july 18 observations , it was realised that the pulsar had suffered a sizable glitch . glitch parameters resulting from a fit of a model including post - glitch exponential decay ( see , e.g. * ? ? ? * ) to pulse time - of - arrival ( toa ) data from 2006 april to 2008 november are given in table [ tb : glt ] . toas were obtained from average pulse profiles for data segments of 15-min duration and the program tempo2 @xcite was used to determine the timing parameters . there was only a small relaxation of the pulse frequency toward the extrapolated pre - glitch solution following the glitch , i.e. , the @xmath52 parameter was small . significant timing noise remains after the fit and the uncertainties have been multiplied by five ( @xmath53 ) to allow for this . table [ tb : psrpar ] gives pulsar timing parameters from a fit to post - glitch data from 2007 december to 2008 november ; the @xmath54 d following the glitch were omitted to avoid biasing the result by the post - glitch relaxation . over the fitted data span there is significant timing noise and the fit uncertainties have been multiplied by a factor of three . only the pulse frequency @xmath55 and @xmath56 were fitted for ; the position and the binary parameters were held fixed at the values given by @xcite . parameters are quoted in tdb - compatible units , and the `` dd '' binary model @xcite and the jet propulsion laboratory solar - system ephemeris de405 @xcite were used . figure [ fg : w50s14 ] shows the observed time variations of the 50% mean pulse width ( @xmath57 ) and mean flux density at 1400 mhz over the 9.3-year data span . such dramatic long - term variations in pulse width ( more than a factor of three ) are unprecedented . the most straight - foward explanation is that they result from precession of the pulsar spin axis changing our view of the pulsar beam . the variations are complex with a @xmath58-d interval around mjd 52800 ( 2003 june ) where the width clearly decreases with time before increasing again . recent data show that a maximum width was reached around mjd 54150 ( 2007 february ) and now the width is decreasing . there is good agreement between the widths obtained with the different back - end systems , although the afb widths tend to be a few percent less than those measured with the wbc or pdfb1 at the same or similar time owing to differences in the instrumental impulse response . the time variation in @xmath28 is similar , but with a smaller relative change over the data span . the mean pulsed flux density has also changed dramatically over the data span with a broad peak around mjd 52500 ( mid-2002 ) . there is a significant day - to - day variation in the measured flux densities . this is indicated by the large error bars on the afb averages ( which typically cover a week or more ) and by the scatter in the wbc / pdfb measurements ( which typically are made on a single day ) , both much larger than the uncertainty in an individual measurement . these short - term variations can be attributed to refractive scintillation for which the expected timescale is about seven days @xcite . however , the broad rise and decay of the flux density over the nine - year data span is unlikely to be a scintillation effect and we attribute this to the changing aspect of a complex beam pattern sweeping across the earth as the pulsar spin axis precesses . average total - intensity profiles as a function of time are shown in figure [ fg : prf ] illustrating the profile evolution at early times the profile was dominated by a strong trailing peak with a broad ramp of emission leading up to it , whereas at late times , the profile is more symmetric with an approximately gaussian shape . figure [ fg : grey ] shows the profile evolution in the form of greyscale plots with both linear ( left ) and logarithmic ( right ) intensity scales . the logarithmic plot clearly illustrates a striking property of the profile evolution , that the width at a very low level ( the lowest contour is at 1% of the peak ) is remarkably constant , suggesting that the low - level flux - density contours represent the overall beam extent . as will be further described in section [ sec : model ] , the outer parts of the profile are relatively stable , both in shape and flux density . the profiles in figures [ fg : prf ] and [ fg : grey ] have therefore been aligned using the midpoint at a constant flux density of 2% of the maximum as a reference point . this is an important assumption as it affects not only the interpretation of the various profile components , but also the parameters derived from any timing analysis . with this assumption , the peak of the profile has moved in a somewhat step - wise fashion from near its trailing edge to near its center . this evolution will be interpreted in terms of the growth and decay of components representing bright regions of the beam in section [ sec : model ] below . figure [ fg : poln ] shows the mean pulse polarization profiles for psr j1141@xmath06545 at two epochs separated by about three years . plotted pas refer to the band - center frequency as marked on the plots.6545 given by @xcite have the opposite sign of @xmath47 and the opposite sense of position - angle swing to the iau and pulsar conventions used for observational results in this paper . we also note that the same comments apply to the polarization profiles for the double pulsar , psr j0737@xmath03039a / b , presented by @xcite and @xcite . as a consequence , the rotation measures given in these latter papers have the wrong sign . the current best estimate of the rm for psr j0737@xmath03039a / b with the corrected sign is @xmath59 rad m@xmath50 @xcite . ] dramatic changes are observed over the three - year span in the ( total intensity ) pulse profile and in the polarization parameters . once the sign differences are allowed for and ignoring the absolute value of pa , the 2004 profile in figure [ fg : poln ] agrees well with the profile given by @xcite taken about two months earlier . in particular , the pa changes rapidly near the profile center , indicating that our line of sight traverses the beam relatively close to the magnetic axis . however there are clear and signficant departures from the pa variation expected for the simple rvm . the evolution of the profiles of linearly polarized intensity @xmath60 and of stokes @xmath47 across the pulse over the 4.5-year span of the polarization data are shown in figure [ fg : lvpol ] . both show systematic changes as a function of time . in the case of the @xmath60 profiles , components at different pulse phases appear to get stronger or weaker as a function of time , whereas the variation of stokes @xmath47 is more consistent across the whole profile . it is striking that the variation of @xmath47 across the pulse changes smoothly from a positive to negative sign change at early epochs to a negative to positive change at late epochs . especially at later times , there are clear dips in the linearly polarized intensity at pulse phases near @xmath61 . as will be discussed further below , these are attributed to overlapping of approximately orthogonally polarized components . figure [ fg : pa ] shows the evolution of the pa variation across the pulse . the observed pa variations are clearly not well described by the rvm and there are clear systematic trends in the pa variations as a function of time . while there are some differences between adjacent pa profiles which may in part result from residual calibration errors or low - level radio - frequency interference , in general the trends in polarization properties with time are smooth and consistent despite the use of different receivers and backend systems at different times , again showing that our calibration procedures are effective . most striking is the clear difference in pa evolution between the central and the outer parts of the profile . figure [ fg : pa_evol ] shows the pa variations over the 4.5-year data span averaged over the inner part ( phase @xmath62 to 0.0068 ) and the outer parts ( @xmath63 to @xmath64 and 0.0136 to 0.0253 ) of the profile . variations of the mean pa for the inner part of the pulse profile are very significant with an intial decrease followed by a rapid increase which slows at later epochs . for the outer parts of the profile , the mean pa at early times shows some fluctuations but overall is consistent with a slow decrease , flattening at later times . especially at early times , the polarization in the outer parts of the profile is quite weak and hence the fluctuations in pa can be attributed to remaining systematic errors and/or contributions from the wings of the stronger emission in the center of the profile . these pa changes could be due to variations in rm or to aspect changes resulting from the precessional motion . in figure [ fg : rm ] we show the rms separately for the central and outer parts of the profile computed using the iterative method described in section [ sec : obs ] . for the outer parts of the profile there is no significant time variation . the weighted mean rm value is @xmath65 rad m@xmath50 with a reduced @xmath66 of 3.1 ; the quoted uncertainty is the weighted rms deviation multiplied by @xmath67 . in contrast , for the central part , there is a clear systematic variation of measured rm with time . we believe that this is not a true variation in the interstellar rm it would be much larger than rm variations in other pulsars that are attributed to a changing path through the interstellar medium , e.g. , for the vela pulsar @xcite . furthermore , it is inconsistent with the changes in the central pa shown in figure [ fg : pa_evol ] . rather , we believe the observed changes in both pa and apparent rm for the central part of the psr j1141@xmath06545 profile are due to variations in the relative amplitudes of observed profile components with different pas and spectral indices resulting from the changing aspect of the line of sight relative to the beam axis as the pulsar spin axis precesses . this is somewhat analogous to the apparent rm variations ( as a function of pulse phase ) seen in psr b2016 + 28 by @xcite which are attributed to over - lapping non - orthogonal components . we adopt pas in the outer zones as representing the underlying magnetic field structure . consequently , these pa variations should be consistent with the rvm . the two sides are consistently at approximately the same pa , as expected in the rvm . also , in some cases , e.g. , the 2007 august 4 profile shown in figure [ fg : poln ] , there is evidence for the pa variations expected for the rvm at the inner edges of the outer pas . since there is no signficant rm variation for these parts , we interpret the observed slow decrease in outer - zone pas shown in the right panel of figure [ fg : pa_evol ] as intrinsic and resulting from precession of the pulsar spin axis . by modelling the observed pa variations in terms of precessional motion of the pulsar s spin axis , we can infer additial properties of the system , for example , the spin - orbit misalignment angle and the precessional phase . as a first step in the modelling , we fit the observed total - intensity profiles ( figure [ fg : prf ] ) with gaussian components . a total of six components are fitted , although for later epochs only four are required . we have used fixed central phases for the components ; only the amplitude and width of each component is allowed to vary as a function of epoch . the component phases , given in table [ tb : comp_phs ] , were chosen using an iterative process by fitting the first and last observed profiles . fitted components and their sum are shown in figure [ fg : prof ] for a selected set of profiles which cover most of the observed data span , showing that an excellent fit to the observed profiles is obtained . this retroactively justifies our decision to fix the component phases . figure [ fg : comps ] gives the widths and mean flux densities of the fitted components . the mean flux densities are scaled so that their sum equals the smoothed flux - density variation shown in figure [ fg : w50s14 ] . for components 3 and 5 , the fitted amplitudes dropped to zero after mjds 52100 and 52500 , respectively . it is striking that the amplitude of each component varies smoothly from one epoch to the next and that the widths of the components are relatively stable . this suggests that these components represent real physical zones of emission which are fixed in longitude on the star . we note that most of the observed flux density variation comes from the central and trailing components . although the leading component 1 is relatively weak at early epochs and dominates the profile at late epochs , its absolute amplitude remains approximately constant across the whole data span . as discussed above , the observed polarization pa and @xmath60 variations ( figs [ fg : pa ] and [ fg : lvpol ] ) strongly suggest that these components may be grouped into two overlapping and quasi - independent emission zones , one dominant in the wings of the observed pulse and the other dominant in the central regions . we note that similar non - rvm variations ( not accountable for by overlapping orthogonal polarization modes ) are seen in other short - period pulsars , for example , psr b1913 + 16 @xcite and psr b1534 + 12 @xcite . the outer components ( 1 and 6 ) may be identified with the outer zones in the pa and @xmath60 plots . as discussed above , the pas in these zones together are well fitted by a single rvm . note that , as discussed in sections [ sec : intro ] and [ sec : obs ] , for modelling purposes , we have adopted a right - hand coordinate system in which the sign of pa is reversed compared to the astronomical convention used in section [ sec : results ] . we also choose @xmath68 and hold this fixed for all epochs . most previous discussions of relativistic spin precession in this and other pulsars ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have concentrated on measurement and interpretation of the rate of change of impact parameter . for psr j1141@xmath06545 , the impact parameter is not very well determined because of the presence of the evidently independent emission in the central part of the profile . however , since we have measured absolute position angles , we have another , potentially more sensitive , observable : the central pa of the rvm , @xmath34 ( cf . * ? ? ? * ; * ? ? ? * ) . in the rvm , this is the direction of the spin axis of the star projected on the sky plane ( see figure [ fg : angles ] ) . this of course varies as the pulsar precesses : @xmath69 where @xmath70 is the longitude of the ascending node and @xmath71 , the precessional longitude , is the angle of the pulsar spin axis relative to the ascending node projected on the sky plane . in this equation , @xmath34 is the observed pa corrected to infinite frequency using the measured rm of @xmath72 rad m@xmath50 . the polar angle @xmath73 varies with the precession according to : @xmath74 @xmath75 where @xmath76 is the spin - orbit misalignment angle , @xmath77 is the precessional angle measured in the orbit plane from @xmath78 and @xmath79 is the value of @xmath77 at @xmath80 ( dt92 ) and @xmath81 from equation ( [ eq : psi0 ] ) , since @xmath70 is effectively constant , the precessional variation in @xmath34 is determined by the variation of @xmath71 . from equation ( [ eq : lambda1 ] ) , we can determine @xmath82 as a function of time and the fixed angles @xmath76 , @xmath23 and @xmath79 . we take @xmath83 from @xcite , take @xmath84 at mjd 53000.0 , near the center of the data span and the beginning of our polarization monitoring , and take the gr prediction for the spin precession rate @xmath10 , approximately @xmath85 yr@xmath9 . then , using equation [ eq : lambda2 ] ( where there is no sign ambiguity in @xmath86 since @xmath87 ) , we obtain @xmath88 . from the same spherical triangle used to derive equation ( [ eq : lambda2 ] ) , we also have @xmath89 giving us @xmath90 and hence @xmath71 as a function of time ( and the fixed but unknown angles @xmath70 , @xmath76 and @xmath79 ) without further ambiguity ( cf . * ? ? ? consequently , the set of three parameters , @xmath76 , @xmath79 , @xmath91 , ( where @xmath91 absorbs the contribution to @xmath34 by @xmath70 and any other intrinsic constant offset to the pa values ) describes the variation of @xmath34 as a function of time . at the same time , the pa swing at each epoch is ideally described by the rvm , which uses two further parameters : the magnetic inclination @xmath37 and the impact parameter @xmath38 . ( we have held the central longitude of the rvm , @xmath35 , at zero as discussed above . ) however , since @xmath92 the angle @xmath38 can be computed from eqn . ( [ eq : lambda1 ] ) for a given @xmath37 . as a result , only four parameters should describe the behavior of the pa variations at all epochs . as described above , the central range of the pa swing shows clear deviations from an rvm . however , in a blind search for the best fit solution , it nevertheless adds valuable information as it clearly indicates ( by its slope ) that @xmath93 ( in the dt92 convention ) . it can also be noted that the slope steepens and then flattens again towards the end of our dataset ( see figure [ fg : pa ] ) . in order to use the geometric information provided by the central part but to still tie the results to the outer wings , we decided to decrease the relative weight of the central part in the fitting process ( by multiplying its errors by a factor of five ) . we allowed a pa offset between the central part relative to the rvm determined by the outer wings , and this was fitted for independently for each epoch . we take longitudes in the range of @xmath94 ( 0.00833 in phase ) to define the central region , whereas pas in the transition zone @xmath95 ( 0.00833 to 0.0139 in phase ) were ignored in the fit . as a first step of the fitting procedure , we constrain the allowable values of @xmath76 and @xmath79 using just the outer - zone pas . for this purpose , we compute the mean pa of the outer wings and fit its behavior as a function of time ( see figure [ fg : psi0 ] ) . we use a grid search in the @xmath76 @xmath79 plane , where at each grid point we compute the @xmath66 for the fit of the model values ( based on equations [ eq : psi0 ] [ eq : cosdelta ] ) to the computed values of @xmath34 ( figure [ fg : psi0 ] ) . figure [ fg : chi2pa0 ] is a plot of @xmath66 in the @xmath76 @xmath79 plane , showing two clear and well defined solutions . for this plot , we have taken @xmath96 ; the plot for @xmath24 is mirror - symmetric . the solution with @xmath97 corresponds to @xmath98 while the other solution implies @xmath99 . we can use the information provided by the pa swing to identify the appropriate solution in the following way . from the pa swing ( in the dt92 convention ) , the slope is negative and hence @xmath93 . also , the fact that the pa becomes flat at the outer wings suggests an outer line of sight ( that is , the line of sight is on the equatorial side of the magnetic axis when @xmath100 , see , e.g. , lorimer & kramer 2005 ) rather than an inner one . the combination of both is only possible in a solution with @xmath101 , corresponding to solution where @xmath102 . this uniquely identifies the region around @xmath103 and @xmath104 as the correct solution . the best solution from this fitting procedure corresponds to @xmath73 being very close to zero . we also know that @xmath38 is close to zero from the rapid swing of pa in the central region and the flat pas in the outer zones which are only a few degrees from the pulse center . therefore , from equation ( [ eq : lambda3 ] ) , we have that @xmath105 . however , the total - intensity profiles and pa swing show that the magnetic axis can not be exactly aligned ( or counter - aligned ) with the rotation axis . the fact that the sign of the central pa swing remains the same at all epochs also shows that @xmath38 does not change sign as a result of the precessional motion . this suggests that the fit to the pa data as a whole will lie somewhat to the right of the best - fit point shown in figure [ fg : chi2pa0 ] . this idea is confirmed by a blind fit to all pa data using only the four parameters @xmath76 , @xmath79 , @xmath106 and @xmath37 . the search is again performed on a @xmath76 @xmath79 grid , where at each grid point we execute a simplex algorithm to minimize for the other two parameters . the results ( collapsed in @xmath106 and @xmath37 space ) are shown in figure [ fg : chi2final ] . we show in grayscale the previous results of the pa - offset fit and overlay contour levels for the global rvm fit to the pa data . the model variation of @xmath34 , shown in figure [ fg : psi0 ] , has a reduced @xmath66 of 7.4 with 920 degrees of freedom . assumed and derived parameters for the system are given in table [ tb : model ] . figure [ fg : delta_pdf ] shows the joint probability density function for the spin - orbit misalignment angle @xmath76 derived from the fits to the pa variations and the global rvm fit ( figure [ fg : chi2final ] ) . given the geometry of the system we can obtain the model time variation of impact parameter @xmath38 using equations [ eq : lambda1 ] [ eq : phi ] and [ eq : lambda3 ] . the result , given in figure [ fg : beta ] , shows that the observer s line of sight has approached the magnetic axis direction through most of our observed data span , reaching its closest approach around mjd 54000 , and is now receding again . this reversal in the variation of @xmath38 is supported by the fact that the pulse width reached a maximum at about mjd 54200 and is now decreasing ( figure [ fg : w50s14 ] ) . note that we have obtained these results by fitting only the polarization data , i.e. , we did not make use of the total intensity information . although the over - all variation of pulse width is consistent with the precessional model , the detailed variations are more complex . this shows that the assumption of a uniform circular beam is not appropriate . given the variation of @xmath38 over the data span ( figure [ fg : beta ] ) and the time variation of component shapes and amplitudes ( figure [ fg : comps ] ) , we can compute the two - dimensional intensity profile of the emission beam over the traversed region . for each epoch we compute the traverse of our line of sight across the polar region and accumulate the pulse intensity in a two dimensional grid centered on the magnetic axis . the resulting beam pattern is shown in figure [ fg : beam ] . because of the reversal in the time - derivative of @xmath38 when it was close to zero , we only see one half of the polar - cap region . despite this we can clearly see that the beam is quite asymmetric with no evidence for a core - cone or ring structure that is symmetric about the magnetic pole . the partially filled beam can be described as `` patchy '' , albeit with just one major patch in the region scanned so far . although this is the first two - dimensional map of an emission beam to clearly show such patchy structure , there is good evidence that the emission beam from most pulsars is best described in this way @xcite . although the observed pulse width is about average ( the median pulse width for all pulsars is about @xmath107 ) , the small value of @xmath108 implies that the intrinsic beamwidth is small . figure [ fg : beam ] shows that the emitting region fits within a circle of radius about @xmath6 centered on the magnetic axis . excluding millisecond pulsars , observed pulse widths are generally consistent with the relation @xmath109 @xcite . for psr j1141@xmath06545 the predicted value of @xmath110 is about @xmath111 , much larger than the emitting zone traversed so far . psr j1141@xmath06545 was not detected in the parkes 70 cm survey @xcite although , even at the present relatively low flux - density levels , a detection with signal / noise ratio of the order of 50 would have been expected . observations within half a beamwidth of the pulsar position were made 1993 july 14 ( mjd 49182 ) . figure [ fg : beta ] shows that the impact parameter @xmath38 at that time was about @xmath112 . this non - detection therefore suggests that the beam half - width in latitude is @xmath113 , although this must be qualified because of the patchy beam structure . there are good arguments ( e.g. , * ? ? ? * ; * ? ? ? * ) that beams in young pulsars are elongated in the latitude direction and hence more fan - like than circular . an elongated beam has also been suggested for psr b1913 + 16 @xcite . observations over the next decade or two will establish whether or not this is the case for psr j1141@xmath06545 as @xmath38 returns to large ( negative ) values . there is good and increasing evidence ( e.g. * ? ? ? * ; * ? ? ? * ) that the velocity and spin vectors of young pulsars are nearly aligned . this suggests that the kick from the supernova ( sn ) explosion , the dominant factor in determining the pulsar space motion , may also determine the spin direction ( see also * ? ? ? since psr j1141@xmath06545 is a young pulsar , its spin direction is therefore likely to be aligned with the sn kick . while precession will have caused the azimuthal angle of the pulsar spin relative to the ( post - sn ) orbital angular momentum to wrap many times over the lifetime of the pulsar , the polar angle ( @xmath76 in this paper ) should still reflect the direction between the kick and the post - sn orbit normal . we have explored if this expectation can set any interesting constraints on the kick imparted to the forming neutron star , using the binary orbit and kick description in @xcite and @xcite . in terms of their variables , we find : @xmath114^{1/2}}.\ ] ] we follow the procedure set out in @xcite and the bayesian version in @xcite , which follows the principles of @xcite but which puts priors on the system velocity , the angle @xmath115 between the pre- and post - sn orbit normals in @xcite and @xcite and corresponds to the spin - orbit misalignment angle for a recycled pulsar whose companion has undergone a sn explosion . ] and the orientation @xmath116 of the binary system on the sky . each sampling of the priors determines the coefficients of a quadratic equation in @xmath117 , the mass of the pre - sn star ( equation 8 of * ? ? ? * ) , which may or may not have a solution that falls within acceptable ranges for the pre - sn mass and the size of the ( assumed circular ) pre - sn orbit . as this is a young system , we did not evolve the system s motion back in time through the galaxy to identify plausible birth sites @xcite , nor did we evolve the orbital size and eccentricity to account for gravitational radiation losses @xcite , but simply assumed the current velocity and orbital parameters reflect the birth properties . note that the results are therefore independent of the assumed distance ; we assumed a distance of 3.7 kpc @xcite . we put uniform priors on @xmath115 ( sampling from @xmath7 to @xmath5 and testing both positive and negative values in each trial ) and @xmath116 ( @xmath7 to @xmath118 ) . we sampled the companion mass over ( @xmath119 ) m@xmath120 and the total system mass over ( @xmath121 ) m@xmath120 , based on the orbital parameters reported in @xcite . we subtracted these to get the pulsar mass for the given trial , and used the mass function computed from the orbit presented in @xcite , 0.176550265 m@xmath120 , to derive @xmath122 . we considered only the cases where @xmath123 in order to match the modeling in the previous section . while there is a scintillation velocity measurement for this system @xcite which could in principle provide some constraints on the angle between the proper motion and @xmath116 , similar measurements have been shown to be quite unreliable in the double pulsar , probably because of the effects of anisotropies in the interstellar medium @xcite and hence we do not make use of this information here . instead we investigate maxwellian velocity distributions with 1-dimensional dispersions of 50 km s@xmath9 and 100 km s@xmath9 ( these dispersions are of order the transverse velocity derived from the scintillation measurements ) , combined with a second maxwellian with 1-dimensional dispersion of 12 km s@xmath9 representing the pre - sn peculiar velocity . we then derive bayesian probability distributions for the pre - sn and kick parameters as described in @xcite , considering pre - sn masses in the range 1.4 10.0 m@xmath120 to allow for the binding energy of the neutron star @xcite . for each acceptable solution of the quadratic equation , the relevant set of parameters is assigned a uniform likelihood ( as in * ? ? ? * ) for a `` natural '' weighting . we then make use of the probability distribution for the post - sn pulsar spin - orbit misalignment angle @xmath76 obtained from the pa fitting ( figure [ fg : delta_pdf ] ) to give a constrained weighting on the range of acceptable pre - sn parameters and kick velocity . the derived probability density functions for the pre - sn stellar mass , the kick velocity and the spin - orbit misalignment angle after constraining with the allowed misalignment - angle distribution ( figure [ fg : delta_pdf ] ) are shown in figure [ fg : pdfc ] . table [ tb : pdf ] gives the median values of the parameters and the limits at 68% and 95% confidence . we note that the natural weighting tends to disfavour spin - orbit misalignment angles around @xmath124 , so the constrained distribution is biased toward slightly larger angles compared to figure [ fg : delta_pdf ] . not unexpectedly , the assumed distribution of post - sn system velocities has a significant effect on the likely kick velocity , with larger values requiring a larger kick . however , the most probable kick velocity is relatively small in both cases , about 100 km s@xmath9 for a dispersion of 50 km s@xmath9 and about 180 km s@xmath9 for a dispersion of 100 km s@xmath9 . median velocities ( table [ tb : pdf ] ) are somewhat higher since the distributions have a high - velocity tail . never - the - less , they are still small compared with mean or median of the inferred three - dimensional pulsar velocity distribution @xcite . relatively low progenitor masses are strongly favoured with median values of about 2 m@xmath8 . the pulsar mass implied by the timing fits for companion mass and total mass @xcite is just 1.27 m@xmath8 , lower than average @xcite , but very similar to the mass of psr j0737@xmath03039b , the second - born and slow pulsar in the double pulsar system . this system also has a relatively low implied kick velocity from the b pulsar formation ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) but see also @xcite . these low masses and kick velocities may imply formation of the neutron star in an electron - capture collapse of the core of an onemg white dwarf @xcite but other evolutionary histories are also possible ( e.g. , * ? ? ? the inferred spin - orbit misalignment angle @xmath76 in the psr j1141@xmath06545 system is large , with a most probable value of between @xmath125 and @xmath4 . that @xmath76 is not small ( in contrast to pulsar a of the double pulsar system , * ? ? ? * ; * ? ? ? * ) , is consistent with the major changes that we see in the pulse profile as a result of the precession and in fact results from the fitting of the precessional model to these changes . the dramatic long - term changes in pulse shape and amplitude observed for psr j1141@xmath06545 ( figures [ fg : w50s14 ] and [ fg : prf ] ) are unprecedented in pulsar astronomy . pulse profile changes are observed in mode - changing pulsars ( e.g. * ? ? ? * ) and in radio emission from magnetars ( e.g. * ? ? ? * ) but these are of a quite different character to the variations seen in psr j1141@xmath06545 and are certainly caused by fluctuations in the emitted pulsar beam . slow and systematic variations similar to those observed in psr j1141@xmath06545 are seen in the pulse profiles of psr b1913 + 16 and psr b1534 + 12 , but these are of much smaller amplitude . all three of these pulsars are in close binary orbits with massive companions . consequently , the observed profile variations most probably result from precession of the pulsar spin axis resulting from spin - orbit coupling changing our view of the pulsar beam . precession of the pulsar s spin axis has resulted in large and systematic changes in both the amplitude and the shape of the observed pulse profile as our line of sight scanned across the polar region . there are clearly large variations in emissivity across the nominal beam area , consistent with the idea of a partially filled or patchy beam . the large variations in the observed mean pulse profile raise the issue of profile alignment when comparing different epochs . figure [ fg : grey ] shows that the pulse width is relatively stable at the lowest contour levels . we have chosen to align the profiles according to the mid - point of a constant flux - density level near their extreme wings . this choice is motivated by the fact that the outer components are relatively stable , both in flux density and shape ( figure [ fg : comps ] ) . with this choice , alignment of the polarization features over the data span is also very stable , giving added support to our method of defining the absolute pulse phase relative to the star . we note that this choice also defines a reference phase for pulse timing . it is interesting to note that , for psr b1913 + 16 also , the pulse width defined by the lower beam contours is more constant than , say , the pulse width at the 50% level @xcite . major changes are also observed in the pulse polarization parameters over the 4.5 years in which the polarization has been monitored . there has been a steady evolution of the circular polarization with the sign of the sense reversal near the profile center changing from positive - to - negative to negative - to - positive over the data span . variations in the linear polarization are complex with quite different behavior in the central and outer zones of the pulse profile ( figure [ fg : pa ] ) . our choice of taking the outer - zone pas to represent the underlying magnetic - field structure is supported by a number of factors . first , the outer - zone pas are well fitted by the rotating vector model ( rvm ) whereas this is less true of the central - zone pas . more importantly , there is a large and variable offset between the central pas and the outer - zone pas . the time variation of the mean central - zone pa is rapid and inconsistent with precessional motion . furthermore there is an apparent time variation of rm for the central zone which is not present for the outer zones . we attribute the complex behavior of the central - zone polarization to the overlapping of independent components which have different pas and different spectral indices . it may be possible to model the polarization variations based on the variations of the total - intensity components shown in figure [ fg : comps ] but this is beyond the scope of the present work . we note that similar but much less dramatic non - rvm variations are observed in the central parts of the profiles for psr b1913 + 16 and psr b1534 + 12 . fitting of a precessional model to the observed pa variations leads to the conclusion that the spin - orbit misalignment angle @xmath76 is very large in this system , with the pulsar spin axis nearly orthogonal to the orbital angular momentum vector . this is consistent with the large profile shape and polarization changes as the spin axis precesses . unfortunately , because of the complex and poorly understood emission physics , it is not possible to turn the argument around and use these observations as a test of the precessional predictions of general relativity . it is interesting to note though that a large spin - orbit misalignment angle is quite possible despite the low progenitor mass and modest kick velocity . this result is not in conflict with the conclusion of @xcite , that the spin - orbit misalignment angle is probably less than @xmath44 , since that was on the proviso that the pulsar was not at a precessional phase when the angle between the observer s line of sight and the spin axis ( @xmath73 in our terminology ) was changing rapidly . in our solution , @xmath73 is close to zero and is changing rapidly . our observational data span encompassed the time when the precessional longitude passed through @xmath5 . this means that the impact parameter @xmath38 reached an extremum during our data span . in fact , @xmath42 reached a minimum value very close to zero at about mjd 54000 ( figure [ fg : beta ] ) . this solution is consistent with the observed maximum in the pulse width at around the same time and implies that we will only ever traverse one side of the emission beam . the slope of the @xmath38 time - variation has now reversed and so we are now retracing our earlier path across the beam . we therefore predict that over the coming decade we will see a time - reversed `` replay '' of the recent profile amplitude and shape evolution . we thank our colleagues of the parkes multibeam pulsar survey team for assistance with the observations reported in this paper . we also thank steve thorsett and rachel dewey for their contributions to the modelling of the effects of the pulsar kick velocity ( section [ sec : kick ] ) . pulsar research at ubc is supported by an nserc discovery grant and a cfi new opportunities grant to ihs and m.berciu . cam held an nserc usra and ihs an nserc ufa for part of this work . ihs further acknowledges support from the atnf distinguished visitors program and the swinburne university visiting distinguished researcher scheme . the parkes radio telescope is part of the australia telescope which is funded by the commonwealth government for operation as a national facility managed by csiro . , m. , possenti , a. , manchester , r. n. , kramer , m. , mclaughlin , m. a. , lorimer , d. r. , stairs , i. h. , joshi , b. c. , lyne , a. g. , camilo , f. , damico , n. , freire , p. c. c. , sarkissian , j. m. , hotan , a. w. , & hobbs , g. b. 2005 , apj , 624 , l113 , v. , valsecchi , f. , & willems , b. 2008 , in 40 years of pulsars : millisecond pulsars , magnetars and more , ed . c. bassa , z. wang , a. cumming , & v. m. kaspi , vol . 983 ( new york : american institute of physics ) , 433441 , m. , stairs , i. h. , manchester , r. n. , mclaughlin , m. a. , lyne , a. g. , ferdman , r. d. , burgay , m. , lorimer , d. r. , possenti , a. , damico , n. , sarkissian , j. m. , hobbs , g. b. , reynolds , j. e. , freire , p. c. c. , & camilo , f. 2006 , science , 314 , 97 lyne , a. g. , burgay , m. , kramer , m. , possenti , a. , manchester , r. n. , camilo , f. , mclaughlin , m. a. , lorimer , d. r. , damico , n. , joshi , b. c. , reynolds , j. , & freire , p. c. c. 2004 , science , 303 , 1153 , r. n. , kramer , m. , possenti , a. , lyne , a. g. , burgay , m. , stairs , i. h. , hotan , a. w. , mclaughlin , m. a. , lorimer , d. r. , hobbs , g. b. , sarkissian , j. m. , damico , n. , camilo , f. , joshi , b. c. , & freire , p. c. c. 2005 , apj , 621 , l49 , m. a. , lyne , a. g. , lorimer , d. r. , possenti , a. , manchester , r. n. , camilo , f. , stairs , i. h. , kramer , m. , burgay , m. , damico , n. , freire , p. c. c. , joshi , b. c. , & bhat , n. d. r. 2004 , apj , 616 , l131 cccccccc 1999/08/21 & 51411.7 & 5140751413 & mb & afb & 1390 & 11 & 1.72 + 1999/10/02 & 51452.9 & 5145151454 & mb & afb & 1390 & 8 & 1.23 + 1999/12/01 & 51513.5 & 5149851529 & mb & afb & 1390 & 4 & 0.66 + 2000/02/04 & 51577.7 & 5155451632 & mb & afb & 1390 & 12 & 2.10 + 2000/06/27 & 51721.6 & 5171051754 & mb & afb & 1390 & 4 & 0.99 + 2000/11/23 & 51851.6 & 5184151940 & mb & afb & 1390 & 5 & 0.80 + 2001/04/24 & 52022.9 & 5196952115 & mb & afb & 1390 & 8 & 0.78 + 2001/10/20 & 52201.9 & 5213252251 & mb & afb & 1390 & 6 & 0.60 + 2002/03/28 & 52361.0 & 5230552428 & mb & afb & 1390 & 6 & 0.60 + 2002/07/30 & 52485.0 & 5245952507 & mb & afb & 1390 & 8 & 1.14 + 2002/11/21 & 52598.9 & 5257152624 & mb & afb & 1390 & 6 & 0.60 + 2003/03/17 & 52715.4 & 5266052770 & mb & afb & 1390 & 4 & 0.40 + 2003/11/19 & 52961.6 & 5291453004 & mb / h - oh & afb & 1390 & 6 & 0.52 + 2004/04/07 & 53102.5 & 5310253102 & h - oh & wbc & 1375 & 1 & 1.00 + 2004/04/23 & 53118.0 & 5310353146 & h - oh & afb & 1390 & 3 & 0.30 + 2004/07/04 & 53190.4 & 5318353194 & h - oh & afb & 1390 & 5 & 3.20 + 2004/07/07 & 53193.3 & 5319353194 & h - oh & wbc & 1375 & 8 & 8.00 + 2004/08/31 & 53248.2 & 5322353282 & h - oh / mb & afb & 1390 & 3 & 0.30 + 2004/10/31 & 53309.6 & 5330953311 & mb & wbc & 1433 & 12 & 11.76 + 2004/11/01 & 53310.2 & 5330653311 & mb & afb & 1390 & 8 & 3.85 + 2005/01/02 & 53372.1 & 5337153372 & mb & wbc & 1433 & 3 & 3.00 + 2005/01/02 & 53372.5 & 5337153372 & mb & afb & 1390 & 6 & 5.70 + 2005/04/23 & 53482.9 & 5348253484 & mb & afb & 1390 & 9 & 8.55 + 2005/06/05 & 53526.1 & 5352653526 & mb & wbc & 1433 & 2 & 2.00 + 2005/08/04 & 53586.1 & 5358653586 & mb & wbc & 1433 & 12 & 7.70 + 2005/08/13 & 53594.6 & 5352253620 & mb & afb & 1390 & 4 & 2.16 + 2005/09/05 & 53618.7 & 5361753620 & mb & pdfb1 & 1433 & 3 & 2.05 + 2005/12/27 & 53730.8 & 5371453741 & mb & afb & 1390 & 3 & 1.55 + 2006/01/05 & 53740.8 & 5373553745 & mb & pdfb1 & 1433 & 2 & 1.95 + 2006/02/04 & 53770.5 & 5377053770 & mb & pdfb1 & 1433 & 2 & 1.52 + 2006/03/03 & 53797.7 & 5379753797 & mb & pdfb1 & 1369 & 2 & 1.67 + 2006/04/09 & 53834.5 & 5383453834 & mb & pdfb1 & 1369 & 2 & 1.67 + 2006/05/10 & 53865.5 & 5386553865 & mb & pdfb1 & 1369 & 2 & 1.67 + 2006/05/28 & 53883.5 & 5388353883 & mb & pdfb1 & 1369 & 2 & 1.67 + 2006/07/10 & 53926.7 & 5392153932 & mb & pdfb1 & 1369 & 2 & 1.37 + 2006/08/14 & 53961.3 & 5396153961 & mb & pdfb1 & 1369 & 2 & 1.67 + 2006/09/27 & 54005.5 & 5399554015 & mb & pdfb1 & 1369 & 4 & 2.83 + 2007/01/19 & 54119.8 & 5411954120 & h - oh & pdfb1 & 1433 & 4 & 3.00 + 2007/02/09 & 54140.7 & 5413754147 & h - oh & pdfb1 & 1433 & 3 & 1.63 + 2007/05/03 & 54223.5 & 5422354223 & h - oh & pdfb1 & 1433 & 4 & 3.33 + 2007/06/06 & 54257.4 & 5425754257 & mb & pdfb1 & 1369 & 2 & 1.67 + 2007/07/18 & 54299.2 & 5429954299 & mb & pdfb1 & 1369 & 2 & 2.13 + 2007/08/04 & 54316.0 & 5431554316 & mb & pdfb1 & 1369 & 2 & 2.13 + 2007/11/05 & 54409.1 & 5440954409 & mb & pdfb1 & 1369 & 2 & 2.13 + 2008/01/24 & 54489.6 & 5448954489 & mb & pdfb1 & 1369 & 1 & 1.06 + 2008/02/24 & 54509.6 & 5450954509 & mb & pdfb1 & 1369 & 1 & 1.06 + 2008/08/03 & 54681.2 & 5468154681 & mb & pdfb3 & 1369 & 2 & 2.11 + 2008/11/14 & 54784.8 & 5478454784 & mb & pdfb3 & 1369 & 2 & 2.13 + ll glitch epoch ( mjd ) & @xmath126 + @xmath127 & @xmath128 + @xmath129 & @xmath130 + @xmath52 & @xmath131 + @xmath132 ( d ) & @xmath133 + data span ( mjd ) & 53834 54785 + number of toas & 254 + rms residual ( @xmath134s ) & 220 ll r.a . ( j2000 ) & 11@xmath135 41@xmath136 @xmath137 + dec . ( j2000 ) & @xmath138 @xmath139 @xmath140 + pulse frequency ( @xmath55 ) ( hz ) & 2.538723048486(4 ) + pulse frequency time - deriv . ( @xmath56 ) ( s@xmath50 ) & @xmath141 + epoch ( mjd ) & 54637.00 + dispersion measure ( @xmath142 pc ) & 116.080 + binary period ( d ) & 0.1976509593 + binary period time - derivative & @xmath143 + orbit semi - major axis ( s ) & 1.858922 + eccentricity & 0.171884 + periastron time ( mjd ) & 51369.8545515 + longitude of periastron ( @xmath144 ) & 42.4561 + longitude time - derivative ( @xmath144 yr@xmath9 ) & 5.3096 + grav . redshift / time dilation ( @xmath145 ) ( s ) & 0.000773 + data span ( mjd ) & 54435 54785 + rms timing residual ( @xmath134s ) & 56 lc assumed parameters : & + orbit inclination @xmath23 & @xmath147 + precession rate @xmath10 & @xmath85 yr@xmath9 + rvm reference phase @xmath35 & 0.0 + reference time for precession & mjd 53000.0 + derived parameters : & + spin - orbit misalignment angle @xmath76 & @xmath148 ( @xmath149,@xmath150 ) + precession angle @xmath79 & @xmath151 ( @xmath152,@xmath153 ) + offset in pa @xmath91 & @xmath154 + magnetic inclination @xmath37 & @xmath155 ( @xmath156,@xmath157 ) + lcccc progenitor mass ( m@xmath8 ) & 50 & 1.80 & 1.54 2.10 & 1.42 2.48 + & 100 & 2.24 & 1.58 3.06 & 1.42 4.00 + kick velocity ( km s@xmath9 ) & 50 & 116 & 86 164 & 56 240 + & 100 & 258 & 150 434 & 94 452 + spin - orbit misalignment angle ( @xmath144 ) & 50 & 116 & 101 131 & 92 149 + & 100 & 106 & 94 120 & 87 135 +

an _ obstacle representation _ of a graph is a set of points in the plane representing the vertices of , together with a set of polygonal obstacles such that two vertices of are connected by an edge in if and only if the line segment between the corresponding points avoids all the obstacles . the _ obstacle number of _ is the minimum number of obstacles in an obstacle representation of . we provide the first non - trivial general upper bound on the obstacle number of graphs by showing that every-vertex graph satisfies . this refutes a conjecture of mukkamala , pach , and plvlgyi . for-vertex graphs with bounded chromatic number , we improve this bound to . both bounds apply even when the obstacles are required to be convex . we also prove a lower bound on the number of-vertex graphs with obstacle number at most for and an asymptotically matching lower bound for the complexity of a collection of faces in an arrangement of line segments with endpoints .

1610.04741

in a _ geometric drawing of a graph @xmath0 _ , the vertices of @xmath0 are represented by distinct points in the plane and each edge @xmath10 of @xmath0 is represented by the line segment between the pair of points that represent the vertices of @xmath10 . as usual , we identify the vertices and their images , as well as the edges and the line segments representing them . let @xmath11 be a finite set of points in the plane in _ general position _ , that is , there are no three collinear points in @xmath11 . the _ complete geometric graph _ @xmath12 is the geometric drawing of the complete graph @xmath13 with vertices represented by the points of @xmath11 . an _ obstacle _ is a polygon in the plane . an _ obstacle representation _ of a graph @xmath0 is a geometric drawing @xmath14 of @xmath0 together with a set @xmath15 of obstacles such that two vertices of @xmath0 are connected by an edge @xmath10 if and only if the line segment representing @xmath10 in @xmath14 is disjoint from all obstacles in @xmath15 . the _ obstacle number _ @xmath1 of @xmath0 is the minimum number of obstacles in an obstacle representation of @xmath0 . the _ convex obstacle number _ @xmath16 of a graph @xmath0 is the minimum number of obstacles in an obstacle representation of @xmath0 in which all the obstacles are required to be convex . clearly , we have @xmath17 for every graph @xmath0 . for a positive integer @xmath2 , let @xmath18 be the maximum obstacle number of a graph on @xmath2 vertices . in this paper , we provide the first nontrivial upper bound on @xmath18 ( theorem [ thm - generalupperbound ] ) . we also show a lower bound for the number of graphs with small obstacle number ( theorem [ thm - count - lb ] ) and a matching lower bound for the complexity of a collection of faces in an arrangement of line segments that share endpoints ( theorem [ thm - arrangements ] ) . all proofs of our results are based on so - called _ @xmath19-dilated bipartite drawings of @xmath20 _ , which we introduce in section [ sec : bipartitedrawings ] . in the following , we make no serious effort to optimize the constants . all logarithms in this paper are base 2 . the obstacle number of a graph was introduced by alpert , koch , and laison @xcite who showed , among several other results , that for every positive integer @xmath6 there is a graph @xmath0 with @xmath21 . using extremal graph theoretic tools , pach and sariz @xcite proved that the number of labeled @xmath2-vertex graphs with obstacle number at most @xmath6 is at most @xmath22 for every fixed integer @xmath6 . this implies that there are bipartite graphs with arbitrarily large obstacle number . mukkamala , pach , and sariz @xcite established more precise bounds by showing that the number of labeled @xmath2-vertex graphs with obstacle number at most @xmath6 is at most @xmath23 for every fixed positive integer @xmath6 . it follows that @xmath24 . later , mukkamala , pach , and plvlgyi @xcite improved the lower bound to @xmath25 . currently , the strongest lower bound on the obstacle number is due to dujmovi and morin @xcite who showed @xmath26 . surprisingly , not much has been done for the general upper bound on the obstacle number . we are only aware of the trivial bound @xmath27 for every graph @xmath0 on @xmath2 vertices . this follows easily , as we can consider the complete geometric graph @xmath12 for some point set @xmath11 of size @xmath2 and place a small obstacle @xmath28 on every _ non - edge _ @xmath10 of @xmath0 such that @xmath28 intersects only @xmath10 in @xmath12 . a non - edge of a graph @xmath29 is an element of @xmath30 . concerning special graph classes , fulek , saeedi , and sariz @xcite showed that the convex obstacle number is at most five for every outerplanar graph , and at most four for every bipartite permutation graph . chaplick et al . @xcite showed that @xmath31 whenever @xmath32 and that @xmath33 . they also found an @xmath34-vertex graph @xmath0 with @xmath35 such that in every its representation with a single obstacle , the obstacle lies in one of the bounded cells of the drawing of @xmath0 . berman et al . @xcite constructed a @xmath36-vertex planar graph @xmath0 with @xmath37 . alpert , koch , and laison @xcite asked whether the obstacle number of every graph on @xmath2 vertices can be bounded from above by a linear function of @xmath2 . we show that this is true for bipartite graphs and split graphs , even for the convex obstacle number . split graph _ is a graph for which the vertex set can be partitioned into an independent subset and a clique . [ thm - bipartiteupperbound ] if @xmath29 is a bipartite graph or a split graph , then we have @xmath38 where @xmath39 denotes the complement of @xmath0 . on the other hand , a modification of the proof of the lower bound by mukkamala , pach , and plvlgyi @xcite implies that there are bipartite graphs @xmath0 on @xmath2 vertices with @xmath40 for every positive integer @xmath2 . in contrast to the above question of alpert , koch , and laison on the existence of a linear upper bound , mukkamala , pach , and plvlgyi @xcite conjectured that the maximum obstacle number of @xmath2-vertex graphs is around @xmath41 . we refute this conjecture by showing the first non - trivial general upper bound on the obstacle number of graphs . in fact , we prove a stronger result that provides a general upper bound for the convex obstacle number . [ thm - generalupperbound ] for every positive integer @xmath2 and every graph @xmath0 on @xmath2 vertices , the convex obstacle number of @xmath0 satisfies @xmath42 the question whether the upper bound on @xmath18 can be improved to @xmath4 remains open . we can , however , prove @xmath43 provided that the chromatic number ( or even the _ subchromatic number _ ) of @xmath0 is bounded from above by a constant . let @xmath29 be a graph and let @xmath44 be a function that assigns a _ color _ @xmath45 to every vertex @xmath46 of @xmath0 . we call @xmath47 a _ subcoloring of @xmath0 _ if for every @xmath48 the color class @xmath49 induces a disjoint union of cliques in @xmath0 subchromatic number of @xmath0 _ , denoted by @xmath50 , is the least number of colors needed in any subcoloring of @xmath0 . we prove the following result that asymptotically implies theorems [ thm - bipartiteupperbound ] and [ thm - generalupperbound ] . [ thm - chromaticupperbound ] for every positive integer @xmath2 and every graph @xmath0 on @xmath2 vertices , the convex obstacle number of @xmath0 satisfies @xmath51 note that we have @xmath52 , thus theorem [ thm - chromaticupperbound ] gives the bound @xmath43 if the chromatic number of @xmath0 or the chromatic number of its complement @xmath53 is bounded from above by a constant . theorem [ thm - chromaticupperbound ] immediately implies an analogous statement for _ cochromatic number of @xmath0 _ , since the cochromatic number of a graph @xmath0 , first defined by lesniak and straight @xcite , is at most @xmath50 for every graph @xmath0 . for positive integers @xmath6 and @xmath2 , let @xmath54 be the number of labeled @xmath2-vertex graphs with obstacle number at most @xmath6 . the lower bounds on the obstacle number by mukkamala , pach , and plvlgyi @xcite and by dujmovi and morin @xcite are both based on the upper bound @xmath55 . in fact , any improvement on the upper bound for @xmath54 will translate into an improved lower bound on the obstacle number @xcite . dujmovi and morin @xcite conjectured @xmath56 where @xmath57 . we show the following lower bound on @xmath54 . [ thm - count - lb ] for every pair of integers @xmath2 and @xmath6 satisfying @xmath58 , we have @xmath59 an _ arrangement @xmath60 of line segments _ is a finite collection of line segments in the plane . the line segments of @xmath60 partition the plane into _ vertices _ , _ edges _ , and _ cells_. a vertex is a common point of two or more line segments . removing the vertices from the line segments creates a collection of subsegments which are called edges . the cells are the connected components of the complement of the line segments . a _ face _ of @xmath60 is a closure of a cell . note that every geometric drawing of a graph is an arrangement of line segments and vice versa . the edges of the graph correspond to the line segments of the arrangement and the vertices of the graph correspond to the endpoints of the line segments . a line segment @xmath61 of @xmath60 is _ incident _ to a face @xmath62 of @xmath60 if @xmath61 and @xmath62 share an edge of @xmath60 . the _ complexity of a face _ @xmath62 is the number of the line segments of @xmath60 that are incident to @xmath62 . if @xmath63 is a set of faces of @xmath60 , then the _ complexity of @xmath63 _ is the sum of the complexities of @xmath62 taken over all @xmath64 . an _ arrangement of lines _ is a finite collection of lines in the plane with faces and their complexity defined analogously . edelsbrunner and welzl @xcite constructed an arrangement of @xmath65 lines having a set of @xmath66 faces with complexity @xmath67 for every @xmath65 and @xmath68 . wiernik and sharir @xcite constructed an arrangement of @xmath65 line segments with a single face of complexity @xmath69 . these two constructions can be combined to provide the lower bound @xmath70 for the complexity of @xmath66 faces in an arrangement of @xmath65 line segments , where @xmath68 . the best upper bound for the complexity of @xmath66 faces in an arrangement of @xmath65 line segments is @xmath71 by aronov et al . @xcite . arkin et al . @xcite studied arrangements whose line segments share endpoints . that is , they considered the maximum complexity of a face when we bound the number of endpoints of the line segments instead of the number of the line segments . they showed that the complexity of a single face in an arrangement of line segments with @xmath2 endpoints is at most @xmath72 . an @xmath73 lower bound was then proved by matouek and valtr @xcite . arkin et al . @xcite asked what is the maximum complexity of a set of @xmath66 faces in an arrangement of line segments with @xmath2 endpoints . since every arrangement of line segments with @xmath2 endpoints contains at most @xmath74 line segments , the upper bound @xmath75 can be deduced from the bound by aronov et al . @xcite . together with the upper bound @xmath76 , which follows from the bound by arkin et al . @xcite on the complexity of a single face , we obtain an upper bound @xmath77 on the complexity of @xmath66 faces in an arrangement of line segments with @xmath2 endpoints . we give the following lower bound . [ thm - arrangements ] for every sufficiently large integer @xmath2 , there is an arrangement @xmath60 of line segments with @xmath2 endpoints such that for every @xmath66 satisfying @xmath78 there is a set of at most @xmath66 faces of @xmath60 with complexity @xmath79 . whenever @xmath80 , this lower bound matches the upper bound up to a multiplicative constant . for @xmath81 , the lower bound differs from the upper bound by at most an @xmath82 multiplicative factor . see also table [ tab : arrgbounds ] . .summary of the best known upper and lower bounds on the complexity of @xmath66 faces in an arrangement of segments with @xmath2 endpoints . [ cols="<,<,<,<",options="header " , ] for a point @xmath83 , let @xmath84 and @xmath85 denote the @xmath86- and the @xmath87-coordinate of @xmath88 , respectively . an _ intersection point _ in a geometric drawing @xmath14 of a graph @xmath0 is a common point of two edges of @xmath0 that share no vertex . let @xmath65 and @xmath2 be positive integers . we say that a geometric drawing of @xmath20 is _ bipartite _ if the vertices of the same color class of @xmath20 lie on a common vertical line and not all vertices of @xmath20 lie on the same vertical line . for the rest of this section , we let @xmath14 be a bipartite drawing of @xmath20 and use @xmath89 and @xmath90 with @xmath91 and @xmath92 to denote the point sets representing the color classes of @xmath20 in @xmath14 . we let @xmath93 and @xmath94 be the vertical lines that contain the points of @xmath11 and @xmath95 , respectively . the _ width _ @xmath96 of @xmath14 is @xmath97 . in the following , we assume that @xmath93 is to the left of @xmath94 and that @xmath98 and @xmath99 . we set @xmath100 for @xmath101 and @xmath102 for @xmath103 . we call @xmath104 the _ left step _ of @xmath14 and @xmath105 the _ right step _ of @xmath14 . an example can be found in part ( a ) of figure [ fig - bipdraw ] . that is not regular . ( b ) a regular drawing of @xmath106 . ] we say that @xmath14 is _ regular _ if we have @xmath107 and @xmath108 ; see part ( b ) of figure [ fig - bipdraw ] for an example . note that every regular drawing of @xmath20 is uniquely determined by its width , left step , and right step . a _ regularization _ of a ( possibly non - regular ) bipartite drawing @xmath14 is the regular bipartite drawing of @xmath20 with the vertices @xmath109 and @xmath110 for @xmath111 and @xmath112 . for @xmath113 , the _ @xmath114th level _ of @xmath14 is the set of edges @xmath115 with @xmath116 ; see figure [ fig - levels ] . note that the levels of @xmath14 partition the edge set of @xmath20 and that the @xmath114th level of @xmath14 contains @xmath117 edges . if @xmath14 is regular , then , for every @xmath118 , the edges of the @xmath114th level of @xmath14 share a unique intersection point that lies on the vertical line @xmath119 . into levels . the edges in the same level are denoted by black line segments . ] for an integer @xmath120 , an ordered @xmath121-tuple @xmath122 of edges of @xmath14 is _ uniformly crossing _ if we have @xmath123 and @xmath124 . in particular , a set of edges forming a level of @xmath14 , ordered by their decreasing slopes , is uniformly crossing . note that if @xmath122 is uniformly crossing , then the edges @xmath125 of the regularization of @xmath14 share a common intersection point , which we call the _ meeting point of @xmath122_. note that this point does not necessarily lie in @xmath14 ; see figure [ fig - meetingpoint ] . in the other direction , if @xmath14 is regular and @xmath126 is a maximal set of edges of @xmath14 that share a common intersection point and are ordered by their decreasing slopes , then @xmath126 is uniformly crossing . in a bipartite drawing @xmath14 of @xmath106 . ( b ) the meeting point of @xmath127 in the regularization of @xmath14 . ] let @xmath128 be a real number . we say that @xmath14 is _ @xmath19-dilated _ if we have @xmath129 and @xmath130 ; see part ( a ) of figure [ fig - cap ] . in a geometric drawing @xmath131 of a ( not necessarily bipartite ) graph , let @xmath126 be an ordered @xmath121-tuple of edges of @xmath131 with finite slopes and such that @xmath132 and @xmath133 share an intersection point @xmath134 for @xmath135 . we say that @xmath126 _ forms a cap _ , if @xmath136 and the slopes of @xmath137 are strictly decreasing . for every @xmath138 , the segment @xmath132 is the graph of a linear function of @xmath139 defined on an interval of real numbers . the _ lower envelope _ of @xmath126 is the graph of the function defined on the union of these intervals as the pointwise minimum of the functions @xmath140 and is undefined elsewhere . that is , the lower envelope is a union of finitely many piecewise linear curves , called _ components_. the _ cap _ formed by @xmath126 is the component of the lower envelope that contains @xmath141 . the points @xmath134 are _ vertices of @xmath142 _ and @xmath143 are _ edges of @xmath142 _ ; see part ( b ) of figure [ fig - cap ] . a cap @xmath142 is _ good in @xmath131 _ , if the edges of @xmath142 are incident to the same bounded face of @xmath131 or if @xmath142 has only one edge . if @xmath131 is bipartite and the edges of one of its levels form a cap @xmath142 , then we call @xmath142 a _ level - cap of @xmath131_. -dilated drawing of @xmath106 . ( b ) an example of a cap with vertices denoted by empty circles and with edges denoted black . ] the following lemma is crucial in the proofs of all our main results . let @xmath14 be a bipartite drawing of @xmath20 . [ lem - cap ] 1 . [ item - lemcap1 ] if @xmath14 satisfies @xmath144 and @xmath145 , then , for every @xmath146 , every uniformly crossing @xmath121-tuple of edges of @xmath14 forms a cap . [ item - lemcapnew ] for all @xmath147 and @xmath148 , there is an @xmath149 such that for every positive @xmath150 the following statement holds . if @xmath14 is an @xmath19-dilated bipartite drawing of @xmath20 with width @xmath96 , left step @xmath104 , and right step @xmath105 , then the intersection point between any two edges @xmath115 and @xmath151 of @xmath14 lies in distance less than @xmath152 from the intersection point @xmath153 . [ item - lemcap2]for all @xmath154 and @xmath148 , there is an @xmath155 such that for every positive @xmath150 the following statement holds . if @xmath14 is an @xmath19-dilated bipartite drawing of @xmath20 with width @xmath96 , left step @xmath104 , and right step @xmath105 , then for every @xmath146 every uniformly crossing @xmath121-tuple of edges of @xmath14 forms a good cap in @xmath14 . for part [ item - lemcap1 ] , let @xmath126 be a uniformly crossing @xmath121-tuple of edges of @xmath14 with @xmath156 for every @xmath157 . consider edges @xmath158 , @xmath159 , @xmath160 and let @xmath161 and @xmath162 be the points @xmath163 and @xmath164 , respectively . the points @xmath161 and @xmath162 exist , as @xmath165 and @xmath166 . consider the midpoint @xmath88 of @xmath167 and the midpoint @xmath168 of @xmath169 . since @xmath126 is uniformly crossing and @xmath144 and @xmath170 , we have @xmath171 and @xmath172 ; see part ( a ) of figure [ fig - lemma ] . the edges @xmath173 , @xmath158 , and @xmath160 share a common point that lies above @xmath159 . since @xmath161 and @xmath162 lie on @xmath159 , we obtain @xmath174 . the slopes of @xmath158 , @xmath159 , @xmath160 are strictly decreasing , thus @xmath126 forms a cap . of lemma [ lem - cap ] . ( b ) a situation in the proof of part [ item - lemcap2 ] of lemma [ lem - cap ] . ] part [ item - lemcapnew ] follows from the fact that for fixed @xmath96 , @xmath104 , @xmath105 , all @xmath175-dilated drawings of @xmath20 with width @xmath96 , left step @xmath104 , and right step @xmath105 converge to their common regularization as @xmath176 tends to zero . we now show part [ item - lemcap2 ] . we let @xmath177 be the half of the minimum distance between two intersection points of the regular drawing @xmath131 of @xmath20 with width @xmath96 , left step @xmath104 , and right step @xmath105 . we take @xmath178 from part [ item - lemcapnew ] and we let @xmath14 be an @xmath19-dilated drawing of @xmath20 with width @xmath96 , left step @xmath104 , and right step @xmath105 . according to [ item - lemcap1 ] , every uniformly crossing @xmath121-tuple @xmath126 of edges of @xmath14 forms a cap . let @xmath61 be the meeting point of @xmath179 in the regular drawing @xmath131 . let @xmath180 be the open disc with the center @xmath61 and the radius @xmath152 . let @xmath181 be the set of edges that intersect some other edge inside @xmath180 . by the choice of @xmath152 , @xmath181 is the set of edges that correspond to the edges of the regular drawing @xmath131 that contain @xmath61 . in particular , @xmath182 . by part [ item - lemcap1 ] , the edges of @xmath181 form a cap . all the vertices of this cap lie inside @xmath180 . since no edge outside @xmath181 crosses any edge of @xmath181 inside @xmath180 , there is a face @xmath183 in @xmath14 incident to all edges of @xmath181 . since @xmath184 , @xmath181 contains at least two edges and thus @xmath183 is bounded ; see part ( b ) of figure [ fig - lemma ] . let @xmath185 be a bipartite graph and @xmath53 be its complement . using lemma [ lem - cap ] , we can easily show @xmath186 . let @xmath128 be chosen as in part [ item - lemcap2 ] of lemma [ lem - cap ] for @xmath20 and @xmath187 . consider an @xmath19-dilated drawing @xmath14 of @xmath20 with @xmath187 , @xmath98 , and @xmath188 . since edges of every level of @xmath14 are uniformly crossing , part [ item - lemcap2 ] of lemma [ lem - cap ] implies that the edges of the @xmath114th level of @xmath14 form a good level - cap @xmath189 in @xmath14 for every @xmath113 . that is , there is a bounded face @xmath190 of @xmath14 such that each edge of @xmath189 is incident to @xmath190 or @xmath189 contains only one edge . for every integer @xmath114 satisfying @xmath113 , we construct a single convex obstacle @xmath191 . if @xmath189 contains only one edge @xmath10 , the obstacle @xmath191 is an arbitrary point of @xmath10 or an empty set . otherwise every edge @xmath192 of the @xmath114th level of @xmath14 shares a line segment @xmath193 of positive length with @xmath190 . the obstacle @xmath191 is defined as the convex hull of the midpoints of the line segments @xmath193 where @xmath192 is not an edge of @xmath53 ; see figure [ fig - obstacle ] . that blocks three edges of @xmath20 . ] the levels partition the edge set of @xmath20 , therefore we block every non - edge of @xmath53 . since every bounded face of @xmath14 is convex , we have @xmath194 . therefore no edge of @xmath53 is blocked and we obtain an obstacle representation of @xmath53 . in total , we produce at most @xmath195 obstacles . to show @xmath196 , we proceed analogously as above , except the vertices of @xmath14 are suitably perturbed before obstacles @xmath191 are defined , which allows to add two ( long and skinny ) convex obstacles @xmath197 and @xmath198 blocking all the edges @xmath199 and @xmath200 , respectively . the addition of the obstacles @xmath197 and @xmath198 may be compensated by using a single convex obstacle to block non - edges in the first and the second level and in the @xmath201nd and the @xmath202st level . if @xmath0 is a split graph , then we add only one of the obstacles @xmath197 and @xmath198 depending on whether @xmath11 or @xmath95 , respectively , represents an independent set in @xmath0 . we show that the convex obstacle number of every graph @xmath0 on @xmath2 vertices is at most @xmath203 . the high - level overview of the proof is as follows . we partition the edges of @xmath0 to edge sets of @xmath4 induced bipartite subgraphs of @xmath0 by iteratively partitioning the vertex set of @xmath0 into two ( almost ) equal parts and considering the corresponding induced bipartite subgraphs of @xmath0 . for every @xmath204 , the number of such bipartite subgraphs of size about @xmath205 is @xmath206 . then we construct an obstacle representation of @xmath0 whose restriction to every such bipartite subgraph resembles the obstacle representation from the proof of theorem [ thm - bipartiteupperbound ] . since the obstacle representation of every bipartite subgraph of size about @xmath205 uses about @xmath205 obstacles , we have @xmath207 obstacles in total . let @xmath65 be a positive integer and let @xmath208 be a finite set of @xmath65 points on a vertical line . the _ bottom half of @xmath208 _ is the set of the first @xmath209 points of @xmath208 in the ordering of @xmath208 by increasing @xmath87-coordinates . the set of @xmath210 remaining points of @xmath208 is called the _ top half of @xmath208_. let @xmath128 be a number smaller than @xmath211 from part [ item - lemcap2 ] of lemma [ lem - cap ] for @xmath212 and @xmath187 and smaller than @xmath211 from part [ item - lemcapnew ] of lemma [ lem - cap ] for @xmath212 and @xmath187 and @xmath213 . let @xmath14 be an @xmath19-dilated bipartite drawing of @xmath212 with width , left step , and right step equal to 1 and with @xmath214 for every @xmath215 . we let @xmath216 and @xmath90 be the color classes of @xmath14 ordered by increasing @xmath87-coordinates such that @xmath98 and @xmath188 . by part [ item - lemcap2 ] of lemma [ lem - cap ] , edges of each level of @xmath14 form a good cap in @xmath14 . for the rest of the proof , the @xmath87-coordinates of all points remain fixed . let @xmath217 be a real number to be determined later . first , we let @xmath218 be the drawing obtained from @xmath14 by removing the top half of @xmath11 and the bottom half of @xmath95 . we use @xmath219 and @xmath220 to denote the left and the right color class of @xmath218 , respectively . we map the vertices of @xmath0 to the vertices of @xmath218 arbitrarily . let @xmath221 be the set of the level - caps of @xmath218 . since every level - cap in @xmath14 is good in @xmath14 , every cap in @xmath221 is good in @xmath218 . the drawing @xmath218 is the first step towards making an obstacle representation of @xmath0 . in fact , we can now block a large portion of non - edges of @xmath0 by placing obstacles in @xmath218 as in the proof of theorem [ thm - bipartiteupperbound ] . then we take care of the edges between vertices in the left color class @xmath219 of @xmath222 ( edges between vertices in the right color class @xmath220 of @xmath222 are dealt with analogously ) . we slightly shift the top half of @xmath219 horizontally to the right . only some of the edges of a copy of @xmath223 between the top and the bottom half of @xmath219 belong to @xmath0 . hence we place convex obstacles along the level - caps of this copy , again , same as in the bipartite case . to take care of the edges between vertices in the same color class of @xmath223 , and for each of the color classes we proceed similarly as above . we now describe this iterative process formally . having chosen point sets @xmath224 for some @xmath225 , in the @xmath226th step , we define @xmath227 as follows . for @xmath228 , let @xmath229 be the bottom half of @xmath230 and let @xmath231 be the top half of @xmath230 . let @xmath232 be a small real number . if @xmath114 is odd and @xmath233 , we move the points from @xmath231 to the right by @xmath234 . if @xmath114 is even and @xmath233 , we move the points from @xmath229 to the left by @xmath234 ; see figure [ fig - construction ] for an illustration . this modified drawing is @xmath235 and @xmath236 is transformed into @xmath237 . and the first two iterations in the construction of the drawing of @xmath238 in the proof of theorem [ thm - generalupperbound ] . the edges added in the current iteration are denoted black . note that the vertical strip between @xmath239 and @xmath240 in @xmath241 is the same as the vertical strip between @xmath219 and @xmath220 in @xmath218 . ] for @xmath228 , we add all edges between points from @xmath229 and @xmath231 to create a bipartite drawing @xmath242 of @xmath243 for some integers @xmath244 satisfying @xmath245 and @xmath246 . let @xmath247 be the vertical strip between @xmath229 and @xmath231 . that is , @xmath247 contains @xmath242 . we let @xmath248 be the union of @xmath237 with a set of level - caps of the drawings @xmath242 for @xmath228 . we also set @xmath249 . we choose @xmath234 small enough so that each good cap @xmath250 , is transformed to a good cap @xmath251 . such @xmath234 exists , as every geometric drawing of a graph is compact and the distance of two points is a continuous function . we also choose @xmath234 small enough so that the following holds for every edge @xmath10 of @xmath235 . let @xmath252 be the smallest index such that @xmath253 . for every vertical strip @xmath254 intersecting @xmath10 and with @xmath255 and @xmath256 , the portion of @xmath10 in the vertical strip @xmath254 is contained in the horizontal strip @xmath257 for some endpoint @xmath88 of @xmath10 . this can be done , as the endpoints of @xmath10 move by at most @xmath234 , the vertical strips @xmath254 for @xmath258 and @xmath256 do not change during the translations by @xmath234 and the vertical strips @xmath259 for @xmath260 are contained in the vertical strips @xmath261 with @xmath262 . see figure [ fig - shift ] for an illustration . while choosing @xmath234 . in the example , @xmath263 , @xmath114 is even , @xmath264 and @xmath265 . the portions of @xmath266 in @xmath267 and @xmath268 are contained in the horizontal strips of height @xmath269 around @xmath88 and @xmath270 , respectively . the portions of @xmath271 in @xmath272 , @xmath267 and @xmath268 are contained in the horizontal strips of height @xmath269 around @xmath88 . the portions of @xmath271 in @xmath273 , @xmath274 and @xmath275 are contained in the horizontal strips of height @xmath269 around @xmath276 . ] after @xmath277 steps , the drawings @xmath278 contain at most two vertices and the construction stops . we show that we can add at most @xmath279 convex obstacles to the drawing @xmath280 to obtain an obstacle representation of @xmath0 . for @xmath225 and @xmath281 , let @xmath282 be the affine mapping @xmath283 where @xmath284 is chosen such that the left color class of @xmath285 lies on @xmath286 . note that the drawing @xmath285 is contained in the drawing @xmath14 and thus edges of the levels of @xmath285 form good caps in @xmath285 ; see part ( a ) of figure [ fig - alfa ] . since @xmath287 does not change the edge - face incidences in @xmath288 , edges of the levels of @xmath288 form good caps in @xmath288 . let @xmath142 be a level - cap formed by edges of a level @xmath289 of @xmath285 and let @xmath290 be the bounded face of @xmath285 such that all edges of @xmath142 are incident to @xmath290 . edges of @xmath289 are also edges of a level @xmath291 of @xmath14 . if @xmath291 contains an edge @xmath292 for some @xmath293 , we let @xmath294 be the horizontal line containing @xmath292 . otherwise @xmath291 contains edges @xmath295 and @xmath296 for some @xmath297 and we let @xmath294 be the horizontal line @xmath298 . no vertex of the level - cap formed by the edges of @xmath291 lies strictly above @xmath294 and thus whenever two edges of @xmath291 intersect above @xmath294 , one of them has a positive slope and the other one negative . therefore no vertex of @xmath142 lies above @xmath294 . since the edge @xmath299 is not part of @xmath285 , no vertex of @xmath142 lies on @xmath294 . thus there is @xmath300 such that every edge of @xmath142 is incident to @xmath301 ; see part ( b ) of figure [ fig - alfa ] . observe that @xmath302 since all the vertices of the cap @xmath142 are at distance at most @xmath152 from @xmath294 . and @xmath303 ( denoted black ) in the @xmath19-dilated drawing @xmath14 of @xmath304 . ( b ) all edges of a cap @xmath142 are incident to a part of a face @xmath290 strictly below @xmath294 . empty circles represent vertices of the level - cap formed by the edges of the level @xmath291 . grey area is the face @xmath290 , which is incident to all edges of the level - cap @xmath142 of the level @xmath289 . ] we choose @xmath305 to be the minimum of @xmath306 over all level - caps @xmath142 of @xmath285 with @xmath225 and @xmath281 . since @xmath285 is a drawing contained in @xmath14 and determined by @xmath226 and @xmath114 , we see that @xmath307 depends only on @xmath19 . since @xmath287 does not change the @xmath87-coordinates , for every level - cap @xmath142 of @xmath288 , there is a bounded face @xmath290 of @xmath288 such that all edges of @xmath142 are incident to the part of @xmath290 that lies below @xmath294 in the vertical distance larger than @xmath307 . by induction on @xmath226 , @xmath308 , we show that every cap from @xmath248 is good in @xmath309 in the @xmath226th step of the construction . we already observed that this is true for @xmath310 . suppose for a contradiction that there is a cap @xmath311 that is not good in @xmath309 for @xmath312 . using the inductive hypothesis and the choice of @xmath234 , @xmath142 is not in @xmath237 . therefore there is a drawing @xmath242 for @xmath228 such that @xmath142 is a level - cap of @xmath242 . since @xmath142 is good in @xmath242 , all edges forming @xmath142 are incident to a single bounded face @xmath290 of @xmath288 . the drawings @xmath313 are contained in pairwise disjoint vertical strips , thus @xmath142 is good in @xmath314 . it follows from part [ item - lemcapnew ] of lemma [ lem - cap ] that all edges of @xmath142 are incident to @xmath290 in a @xmath152-neighborhood of @xmath294 for @xmath315 . moreover , all edges of @xmath142 are incident to the part of @xmath290 that is strictly below @xmath316 . therefore all edges of @xmath142 are incident to the part of @xmath290 that is in the horizontal strip @xmath317 . because @xmath142 is not good in @xmath309 , some edge @xmath10 of @xmath235 intersects @xmath318 in the vertical strip @xmath247 . by the choice of @xmath234 , the portion of @xmath10 inside @xmath247 is contained in the horizontal strip @xmath257 for an endpoint @xmath88 of @xmath10 . since @xmath319 , @xmath318 is disjoint with all the horizontal strips @xmath257 for all endpoints @xmath88 in @xmath235 and we obtain a contradiction . let @xmath320 be the part of the drawing @xmath280 transformed from @xmath242 by the steps @xmath321 . for every @xmath320 , we place the obstacles to its good level caps as in the first part of the proof of theorem [ thm - bipartiteupperbound ] . using the fact that bounded faces of every geometric drawing of @xmath322 are convex , it follows from the construction of @xmath323 that we obtain an obstacle representation of @xmath0 . the total number of obstacles that we use can be estimated recursively . every drawing @xmath324 is a bipartite drawing of @xmath243 for some integers @xmath244 satisfying @xmath245 and @xmath246 . for such @xmath324 we use @xmath325 obstacles . it follows from the construction that the maximum number of obstacles that we use for an @xmath2-vertex graph can be estimated from above by a function @xmath326 that is given by a recursive formula @xmath327 with @xmath328 . now it suffices to show @xmath329 . we proceed by induction on @xmath2 . the inequality is trivially satisfied for @xmath330 . for @xmath331 , we have @xmath332 we have @xmath333 for every real number @xmath334 and thus @xmath335 , since @xmath336 . therefore @xmath337 this finishes the proof of theorem [ thm - generalupperbound ] . an alternative construction , in which the underlying points set is a variant of the well - known _ horton sets _ @xcite , can be found in the conference version of this paper @xcite . let @xmath11 be a finite set of points of the plane , let @xmath0 be a graph with the vertex set @xmath338 , @xmath339 , and let @xmath340 be a one - to - one correspondence . an obstacle representation of @xmath0 is _ induced by @xmath341 _ if every vertex @xmath46 of @xmath0 is represented by the point @xmath342 . the proof of theorem [ thm - generalupperbound ] gives the following stronger claim . [ cor - generalupperbound ] for every positive integer @xmath2 , there is a set @xmath11 of @xmath2 points in the plane such that for every graph @xmath29 on @xmath2 vertices and for every one - to - one correspondence @xmath340 there is an obstacle representation of @xmath0 induced by @xmath341 with at most @xmath343 convex obstacles . we also note that the vertices of the drawing @xmath323 can be perturbed so that the vertices of the resulting drawing are in general position . when the perturbation is small enough , the obstacles can be modified to block exactly the same set of edges that they blocked before the perturbation . to prove thoerem [ thm - chromaticupperbound ] , we start with the same @xmath19-dilated drawing @xmath14 of @xmath212 and use @xmath11 and @xmath95 to denote the left and the right color class of @xmath14 , respectively . again , by part [ item - lemcap2 ] of lemma [ lem - cap ] , edges of each level of @xmath14 form a good cap in @xmath14 . for a finite set @xmath208 of points on a vertical line @xmath344 , an _ interval in @xmath208 _ is a subset @xmath345 of @xmath208 such that there is no point of @xmath346 lying between two points of @xmath345 on @xmath344 . if @xmath208 is partitioned into intervals @xmath347 , then the _ bottom part of @xmath208 with respect to @xmath347 _ is the set of points of @xmath208 that are contained in the intervals @xmath348 . the set of remaining points of @xmath208 is called the _ top part of @xmath208 with respect to @xmath347_. let @xmath47 be a subcoloring of @xmath0 with @xmath50 colors . we map the vertices of @xmath0 to the points of @xmath11 such that the color classes of @xmath47 form intervals in @xmath11 ordered by increasing @xmath87-coordinates and such that in each color class the vertex sets of the disjoint cliques also form intervals in @xmath11 ordered by increasing @xmath87-coordinates . we map the vertices of @xmath0 to @xmath95 in the same order by increasing @xmath87-coordinate . let @xmath219 be the bottom part of @xmath11 and let @xmath220 be the top part of @xmath95 with respect to the partitioning of @xmath11 and @xmath95 into the color classes of @xmath47 . we let @xmath218 be the drawing of the complete bipartite graph with the bottom part of @xmath11 as the left color class of @xmath218 and with the top part of @xmath95 as the right color class of @xmath218 . the construction of the obstacle representation of @xmath0 then proceeds analogously as in the proof of theorem [ thm - generalupperbound ] with only one difference . instead of partitioning the set @xmath349 into the bottom and the top half , we partition @xmath349 into the bottom and the top part with respect to the partitioning of @xmath349 into the color classes of @xmath47 . if all points of @xmath349 are contained in the same color class of @xmath47 , then we do not partition @xmath349 at all . for every drawing @xmath288 , we place the obstacles to block non - edges of @xmath0 in the same way as in the proof of theorem [ thm - generalupperbound ] . we place obstacles only to block non - edges of @xmath0 between vertices from different color classes of @xmath47 . we now estimate the number of obstacles used so far . the number of steps is at most @xmath350 , as the number of intervals in the partitioning of @xmath351 is at most @xmath352 . we place at most @xmath353 obstacles in every step , as the number of obstacles used in a drawing @xmath288 is less than the number of vertices in @xmath288 and the total number of vertices in the drawings @xmath288 is at most @xmath2 for a fixed @xmath226 . thus the total number of obstacles is at most @xmath354 . additionally , if there is a set @xmath351 with vertices contained in the same color class of @xmath47 , then we define a single - point obstacle @xmath355 for every pair @xmath356 of consecutive vertices of @xmath351 that are not contained in a common clique of the color class @xmath351 such that @xmath355 blocks no edge of @xmath0 . since the cliques in the color class @xmath351 form intervals , the obstacles @xmath355 exist . this gives us less than @xmath2 additional obstacles , as there is at most @xmath353 such pairs @xmath356 of consecutive vertices of @xmath0 . note that the construction from the proof of theorem [ thm - generalupperbound ] is a special case of this construction , in which the color classes of @xmath47 consist of a single point . the vertices of the final drawing in the obstacle representation of @xmath0 are not in general position . however , they can be perturbed to be in general position and the obstacles can be modified to keep blocking the same set of edges , in particular , the one - point obstacles can be replaced by short horizontal segments . we note that there are graphs for which the bound @xmath357 from theorem [ thm - chromaticupperbound ] is significantly better than @xmath358 where @xmath359 . for example , if @xmath0 is a disjoint union of @xmath360 cliques , each of size @xmath360 for some positive integer @xmath360 , then the first bound gives @xmath361 for @xmath362 , as @xmath363 . on the other hand , we have @xmath364 and thus the latter bound gives only @xmath365 . note that we can choose vertices and obstacles in the obstacle representation @xmath15 of a graph @xmath29 from theorem [ thm - chromaticupperbound ] in such a way that every non - edge of @xmath0 intersects exactly one obstacle and it intersects the obstacle in exactly one point . in particular , if we add a non - edge @xmath10 of a graph @xmath0 to @xmath0 , then by removing the point @xmath366 from the obstacle @xmath367 that intersects @xmath10 , we construct an obstacle representation of @xmath368 with at most one additional obstacle when compared to @xmath15 . motivated by this observation , we pose the following problem . if @xmath0 is a graph and @xmath10 is a non - edge of @xmath0 , how much larger can @xmath369 be when compared to @xmath1 ? the same question can be also asked for the convex obstacle number . note that @xmath370 and @xmath371 for every graph @xmath0 and every non - edge @xmath10 of @xmath0 . let @xmath6 and @xmath2 be given positive integers with @xmath372 . we show that the number @xmath54 of labeled @xmath2-vertex graphs of obstacle number at most @xmath6 is at least @xmath5 . for a point set @xmath373 in general position , let @xmath374 be the maximum integer for which there is a set @xmath63 of at most @xmath6 bounded faces of @xmath12 and a set of @xmath374 edges of @xmath12 that are incident to at least one face from @xmath63 . let @xmath375 be the maximum of @xmath374 over all sets @xmath11 of @xmath2 points in the plane in general position . we have @xmath376 . to prove the claim , let @xmath11 be a set of @xmath2 points in the plane in general position for which @xmath377 . let @xmath63 be the set of at most @xmath6 bounded faces of @xmath12 such that @xmath375 edges of @xmath12 are incident to at least one face from @xmath63 . for a face @xmath64 , let @xmath378 denote the set of edges of @xmath12 that are incident to @xmath62 . we use @xmath0 to denote the graph with the vertex set @xmath11 and with two vertices connected by an edge if and only if the corresponding edge of @xmath12 is incident to no face @xmath62 of @xmath63 . we show that every subgraph @xmath379 of @xmath12 containing @xmath0 satisfies @xmath380 . the claim then follows , as the number of such subgraphs @xmath379 is @xmath381 . let @xmath379 be a subgraph of @xmath12 such that @xmath382 . for every face @xmath64 , we define a convex obstacle @xmath383 as the convex hull of midpoints of line segments @xmath384 for every @xmath385 that represents a non - edge of @xmath379 . note that , since all bounded faces of @xmath12 are convex , the obstacle @xmath383 is contained in @xmath62 and thus @xmath383 blocks only non - edges of @xmath379 . since every non - edge of @xmath379 is contained in @xmath378 for some @xmath64 , we obtain an obstacle representation of @xmath379 with at most @xmath6 convex obstacles . this finishes the proof of the claim . since @xmath7 , the following and the previous claim give theorem [ thm - count - lb ] . for @xmath386 , we have @xmath387 . let @xmath128 be chosen as in part [ item - lemcap2 ] of lemma [ lem - cap ] for @xmath388 and @xmath187 . let @xmath14 be an @xmath19-dilated drawing of @xmath388 with @xmath187 , @xmath98 , and @xmath188 . by part [ item - lemcap2 ] of lemma [ lem - cap ] , the edges of the @xmath114th level of @xmath14 form a good cap @xmath189 in @xmath14 for every @xmath389 . we perturb the vertices of @xmath14 such that the vertex set of the resulting geometric drawing @xmath131 of @xmath388 is in general position . we let @xmath12 be the geometric drawing of @xmath322 obtained from @xmath131 by adding the missing edges . note that if the perturbation is sufficiently small , then every good cap @xmath189 in @xmath14 corresponds to a good cap @xmath390 in @xmath12 . let @xmath391 be the set of ( not necessarily distinct ) bounded faces of @xmath12 such that , for @xmath392 , all edges of the cap @xmath393 are incident to @xmath394 . that is , @xmath395 are faces incident to edges of @xmath6 middle caps @xmath390 ; see figure [ fig - lowerbound ] . since caps @xmath393 are good in @xmath12 and @xmath386 , the faces @xmath394 exist . for @xmath396 and @xmath397 . the faces @xmath395 of @xmath12 are denoted gray . ] every cap @xmath390 is formed by @xmath398 edges for every @xmath389 . therefore , for every @xmath392 , the face @xmath394 is incident to at least @xmath399 edges of @xmath12 . summing over @xmath392 , we obtain the following expression for the number of edges of @xmath12 incident to at least one face of @xmath63 : @xmath400 if @xmath6 is even , we have @xmath401 if @xmath6 is odd , we have @xmath402 this implies @xmath403 and proves the claim . for a sufficiently large constant @xmath142 and every sufficiently large integer @xmath2 , we find a bipartite drawing @xmath14 of @xmath212 such that for every integer @xmath66 satisfying @xmath404 there is a set of at most @xmath66 faces of @xmath14 with complexity at least @xmath8 . when @xmath405 , we obtain a set of at most @xmath406 faces with complexity at least @xmath407 . for @xmath408 , it suffices to take only the faces with the highest complexity from this set to obtain the lower bound @xmath409 . theorem [ thm - arrangements ] then follows , as @xmath14 can be treated as an arrangement of @xmath41 line segments with @xmath410 endpoints . let @xmath131 be the regular bipartite drawing of @xmath212 with width , left step , and right step equal to 1 , @xmath98 , and @xmath188 . for all coprime integers @xmath138 and @xmath114 satisfying @xmath411 , every intersection point of a uniformly crossing @xmath121-tuple of edges @xmath122 of @xmath131 with @xmath412 and @xmath413 is called a _ uniform @xmath414-crossing _ ; see part ( a ) of figure [ fig - arrang ] . a point that is a uniform @xmath414-crossing for some integers @xmath138 and @xmath114 is called a _ uniform crossing_. note that all uniform @xmath414-crossings lie on the vertical line @xmath415 and that no uniform @xmath414-crossing is a uniform @xmath416-crossing for any pair @xmath417 , as @xmath138 and @xmath114 are coprime . since the @xmath87-coordinate of every uniform @xmath414-crossing equals @xmath418 for some @xmath419 , the number of uniform @xmath414-crossings is at most @xmath420 . we now show that there are at least @xmath421 edges of @xmath131 that contain a uniform @xmath414-crossing . this follows from the condition @xmath422 , as for every edge @xmath151 of @xmath131 with @xmath423 and @xmath424 either @xmath425 or @xmath426 is an edge of @xmath131 and forms a uniform @xmath414-crossing with @xmath151 . we choose @xmath128 as in part [ item - lemcap2 ] of lemma [ lem - cap ] for @xmath212 and @xmath187 . let @xmath14 be an @xmath19-dilated drawing of @xmath212 with width , left step , and right step equal to 1 , with the left lowest point @xmath427 , and with the right lowest point @xmath428 . by part [ item - lemcap2 ] of lemma [ lem - cap ] , every uniformly crossing @xmath121-tuple of edges of @xmath14 forms a good cap in @xmath14 . in particular , every uniform crossing @xmath47 in @xmath131 is the meeting point of edges of @xmath14 that form a good cap @xmath429 . let @xmath430 be the bounded face of @xmath14 such that all edges of @xmath429 are incident to @xmath430 and all vertices of @xmath429 are vertices of @xmath430 ; see part ( b ) of figure [ fig - arrang ] . -crossings ( denoted light gray ) , @xmath431-crossings ( gray ) , and @xmath432-crossings ( dark gray ) in the regular drawing @xmath131 of @xmath433 . ( b ) faces corresponding to the @xmath434-crossings ( light gray ) , to the uniform @xmath431-crossings ( gray ) , and to the @xmath432-crossings ( dark gray ) in the @xmath19-dilated drawing @xmath14 . ] let @xmath47 be the meeting point of a maximal uniformly crossing @xmath121-tuple @xmath289 of edges from @xmath131 for some @xmath120 . similarly , let @xmath435 be the meeting point of a maximal uniformly crossing @xmath436-tuple @xmath291 of edges from @xmath131 for some @xmath437 . let @xmath430 and @xmath438 be the faces of @xmath14 that are incident to all edges of @xmath289 and @xmath291 , respectively , in @xmath14 . we show that the faces @xmath430 and @xmath438 of @xmath14 are distinct . let @xmath439 and @xmath440 be the faces that correspond to @xmath430 and @xmath438 in @xmath131 . it suffices to show that @xmath439 and @xmath440 are distinct , so suppose for a contradiction that @xmath441 . we can assume without loss of generality that @xmath47 and @xmath435 lie on a common edge @xmath442 of @xmath131 , since all vertices @xmath443 of @xmath62 with @xmath444 are meeting points of a maximal uniformly crossing tuple of edges from @xmath131 and they form an interval on @xmath62 . therefore we can choose @xmath47 and @xmath435 to be two such vertices of @xmath62 that share an edge of @xmath62 . we also assume that @xmath47 is to the left of @xmath435 on @xmath10 . then @xmath10 has the smallest slope among edges that contain @xmath47 . we let @xmath445 be the edge that contains @xmath47 and that has the second smallest slope among edges that contain @xmath47 . similarly , @xmath10 has the largest slope among edges that contain @xmath435 and we let @xmath446 be the edge that contains @xmath435 and that has the second largest slope among edges that contain @xmath435 ; see part ( a ) of figure [ fig - arrangfaces ] . note that @xmath447 and @xmath448 . without loss of generality , we assume that @xmath449 . let @xmath86 be the point on @xmath93 above @xmath450 with @xmath451 and let @xmath87 be the point on @xmath94 below @xmath452 with @xmath453 . observe that @xmath47 is the common point of @xmath10 and @xmath454 . the point @xmath86 is below @xmath455 , since otherwise we have @xmath456 and , since @xmath457 , @xmath458 . this implies @xmath459 , which contradicts our assumption . therefore @xmath86 lies between @xmath450 and @xmath455 on @xmath93 and , in particular , @xmath460 . the point @xmath87 lies below @xmath461 , since otherwise @xmath462 and @xmath457 implies that @xmath454 gives @xmath463 . however , if @xmath87 is below @xmath461 , then the edge @xmath464 gives @xmath465 and we obtain a contradiction ; see part ( b ) of figure [ fig - arrangfaces ] . let @xmath466 be a positive integer whose value we specify later . for coprime integers @xmath138 and @xmath114 satisfying @xmath467 , let @xmath468 be the set of faces @xmath430 where @xmath47 is a uniform @xmath414-crossing in @xmath131 . it follows from our observations that @xmath468 contains at most @xmath420 faces and that the complexity of @xmath468 is at least @xmath469 . we let @xmath470 where the union is taken over all coprime integers @xmath138 and @xmath114 satisfying @xmath467 . then @xmath63 contains at most @xmath471 faces where @xmath472 denotes the euler s totient function . the last inequality follows from @xmath473 for every positive integer @xmath474 . since the sets @xmath468 are pairwise disjoint , the complexity of @xmath63 is at least @xmath475 the totient summatory function satisfies @xmath476 @xcite . thus the complexity of @xmath63 is at least @xmath477 . let @xmath66 be a given integer that satisfies @xmath478 . we set @xmath479 . we may assume that @xmath183 is an integer , as it does not affect the asymptotics . for @xmath478 , we have @xmath480 . the set @xmath63 then contains at most @xmath66 faces and its complexity is at least @xmath481 which is @xmath8 for a sufficiently large absolute constant @xmath142 and @xmath404 . we let @xmath482 , @xmath483 and assume that @xmath484 . if we expand the edges of a regular bipartite drawing @xmath485 to lines , we obtain a set of @xmath2 lines and @xmath66 points with @xmath486 incidences , which is the maximum possible number , up to a multiplicative constant . this construction is , however , only a transformation of the construction of erds described for example in a paper of pach and tth @xcite . _ geometric duality _ is a mapping from points and lines in the plane to lines and points , respectively , where the lines are restricted to those with finite slope . a point @xmath487 is mapped to the line whose every point @xmath488 satisfies @xmath489 . notice that the duality preserves point - line incidences . let @xmath490 be the line passing through the points @xmath491 and @xmath492 . its dual is the point @xmath493 . thus , the set of lines @xmath490 with @xmath494 is mapped to the set of points with integer coordinates inside the quadrangle with vertices @xmath495 , @xmath427 , @xmath496 and @xmath497 . this point set can be transformed by a linear transformation of the plane into the set @xmath498 , which is the point set used in the construction of erds . alpert , h. , koch , c. , laison , j. d. : obstacle numbers of graphs . discrete comput . 44(1 ) , 223244 ( 2010 ) arkin , e. m. , halperin , d. , kedem , k. , mitchell , j. s. b. , naor , n. : arrangements of segments that share endpoints : single face results . discrete comput . 13(3 ) , 257270 ( 1995 ) aronov , b. , edelsbrunner , h. , guibas , l. j. , sharir , m. : the number of edges of many faces in a line segment arrangement . combinatorica 12(3 ) , 261274 ( 1992 ) balko , m. , cibulka , j. , valtr , p. : drawing graphs using a small number of obstacles . in : graph drawing and network visualization , volume 9411 of the series lecture notes in computer science , pp . 360372 . springer international publishing ( 2015 ) . dujmovi , v. , morin , p. : on obstacle numbers , electron . j. combin . 22(3 ) , p3.1 ( 2015 ) edelsbrunner , h. , welzl , e. : on the maximal number of edges of many faces in an arrangement . j. combin . theory ser . a 41(2 ) , 159166 ( 1986 ) fulek , r. , saeedi , n. , sariz , d. : convex obstacle numbers of outerplanar graphs and bipartite permutation graphs . in : thirty essays on geometric graph theory , pp . springer , new york ( 2013 ) lesniak , l. m. , straight , h. j. : the cochromatic number of a graph . ars combin . 3 , 3945 ( 1977 ) matouek , j. , valtr , p. : the complexity of lower envelope of segments with @xmath6 endpoints . in : intuitive geometry , pp . 407411 , bolyai society of math . studies 6 ( 1997 ) mukkamala , p. , pach , j. , plvlgyi , d. : lower bounds on the obstacle number of graphs . j. combin . 19(2 ) , p32 ( 2012 ) mukkamala , p. , pach , j. , sariz , d. : graphs with large obstacle numbers . in : graph theoretic concepts in computer science . volume 6410 of the series lecture notes in computer science , pp . springer , heidelberg ( 2010 ) pach , j. , sariz , d. : on the structure of graphs with low obstacle number . graphs combin . 27(3 ) , 465473 ( 2011 ) pach , j. , tth , g. : graphs drawn with few crossings per edge . combinatorica 17(3 ) , 427439 ( 1997 ) valtr , p. : convex independent sets and 7-holes in restricted planar point sets . discrete comput . 7(2 ) , 135152 ( 1992 ) wiernik , a. , sharir , m. : planar realizations of nonlinear davenport schinzel sequences by segments . discrete comput . 3(1 ) , 1547 ( 1988 )

we use the high angular resolution in the near - infrared of the wfc3 on hst to determine _ yhvz _ color - color selection criteria to identify and characterize galaxies in the hudf09 and ers ( goods - south ) fields . the wfc3 nir images reveal galaxies at these redshifts that were undetected in the rest - frame uv hudf / goods images , as well as true centers and regular disks in galaxies classified as highly irregular in rest - frame uv light . across the redshift range , regular disks are unveiled in the wfc3 images of% of both intermediate and high mass galaxies , i.e. , above . meanwhile , galaxies maintaining diffuse and/or irregular morphologies in the rest - frame optical light i.e . , not yet dynamically settled at these epochs are almost entirely restricted to masses below . in contrast at these diffuse and/or irregular structures overwhelmingly dominate the morphological mix in both the intermediate and high mass regimes , while no regular disks , and only a small fraction (% ) of smooth spheroids , are evident above . strikingly , by roughly 2 out of every 3 galaxies at the highest masses are spheroids . in our small sample , the fraction of star - forming galaxies at these mass scales decreases concurrently from% to% . if confirmed , this indicates that is the epoch of both the morphological transformation and quenching of star - formation which assemble the first substantial population of massive ellipticals . = 1

1007.2422

the redshift interval @xmath9 represents an important epoch for the study of galaxy formation and evolution , encompassing both the peak of cosmic star formation ( lilly et al.1996 ; madau et al . 1996 , 1998 ; hopkins & beacom 2006 ; bouwens et al.2007 ; tresse et al . 2007 ) and the peak of active galactic nuclei activity ( warren , hewett , & osmer 1994 ; fan et al . 2001 ; croom et al . 2004 ; barger et al . 2005 ) . one efficient method for identifying large samples of galaxies at these redshifts is via application of color - based selection criteria to optical - to - near - infrared ( nir ) imaging surveys . established selection criteria now exist for the identification of @xmath10 systems via the lyman break in observed uv - to - optical colors ( cf . lbgs : steidel et al . 1996 , 2003 ) , and @xmath11 systems via the balmer break or 4000@xmath12 break in observed optical - to - nir colors ( cf . eros : thompson et al . 1999 ; franx et al . 2003 ; mccarthy et al . 2004 ; van dokkum et al . 2006 ; bx / bms : steidel et al . 2004 ; adelberger et al . 2004 ; and _ _ bzk__s : daddi et al . 2004 , 2005 ) . importantly , with optical - to - nir color - selection criteria it is possible to sample uniformly both actively star - forming and passively - evolving systems at high redshift ( e.g. daddi et al . 2004 ) . the new wide field camera 3 ( wfc3 ) on the hubble space telescope ( hst ) has greatly enhanced the uv and nir imaging capabilities of this observatory , offering sharper imaging and greater throughput with a new complement of narrow- and broad - band filters . in particular , the unprecedented depth and sharpness of the nir wfc3 images further enhances our capacity to detect lbgs at the era of peak star formation in the universe as well as at much higher redshifts ; see , e.g. , the wfc3 detections of numerous galaxy candidates at @xmath13 - 8 ( bouwens et al . 2010a , b ; oesch etal . 2010 ; wilkins et al . 2010 ; mclure et al . 2010 ; bunker et al . 2010 ; finkelstein et al . 2010 ) , and even @xmath14 ( bouwens et al . 2011 ) . in this paper we use the wfc3 data acquired so far in the context of the hubble ultra deep field 2009 ( hudf09 ) and early release science ( ers ) programs to ( i ) extract @xmath15-band based galaxy catalogs down to , respectively , ab=27 and 25 mag ; ( ii ) derive optimal color - selection criteria ( _ yhvz _ ) for the identification and characterization of galaxies at @xmath9 , based on the wfc3 nir filter set ( complemented by optical acs data ) , and ( iii ) use the constructed wfc3 @xmath15-band source catalogs to investigate , through qualitative comparisons with archival acs optical data in the hudf09 and ers fields , differences and similarities in the rest - frame uv and optical morphologies of @xmath16 galaxies . the efficiencies of the _ yhvz _ selection criteria presented here are established via comparisons with the spectroscopic and photometric redshifts of galaxies in both datasets . alternative color - selection criteria , _ vih _ and _ jhvi _ , relevant to the filter set of candels / wide , are also explored using the ers dataset . the outline of the paper is as follows . in section [ data ] we describe the imaging data now available from the hudf09 and ers observational programs , and briefly review the other datasets from which additional uv - to - ir magnitudes and spectroscopic redshifts are obtained . in section [ method ] we detail the construction of our master source catalogs based on the hudf09 and ers @xmath15 imaging , and the matching of these catalogs to multi - wavelength photometry from previous studies . in section [ sec4 ] we describe the estimation of photometric redshifts , specific star formation rates , and stellar masses via spectral energy distribution ( sed ) fitting . in section [ results ] we present _ yhvz _ color - selection criteria for the identification of galaxies at @xmath17 using the wfc3 and acs filters . in addition , we carefully quantify selection efficiencies and contamination rates for these criteria according to the spectroscopic , and ( up to ) 15 band photometric , redshifts of galaxies in our master source catalogs . furthermore , the characterization of passively - evolving galaxies at @xmath9 via a simple extension of our selection criteria ( _ passive yhvz _ ) is explored in section [ pyhvz ] , and the alternative color - selections _ jhvi _ and _ vih _ , using filters from the candels / wide program ( excluding @xmath18 which is available only within the goods - n field ; koekemoer et al . 2011 ) , are presented in section [ mct ] . in section [ sourcecatalogs ] we publish our master source catalogs containing positions , photometric apertures , and photometric redshifts for all robust detections to @xmath19 mag in the hudf09 and @xmath20 mag in the ers . in section [ morph ] we compare the rest - frame uv and optical morphologies of galaxies in the hudf09 and ers fields at @xmath21 , and examine the mass scales over which key structural sub - types are revealed by the wfc3 . we summarize in section [ conclusions ] . note that we adopt a cosmological model with @xmath22 , @xmath23 , and @xmath24 , and that all magnitudes are quoted in the ab system throughout . the hudf09 program ( bouwens et al . 2010 ) now in progress will observe a @xmath24.7 arcmin@xmath25 central region of the hudf ( beckwith et al.2006 ) with the new wide field camera 3 ( wfc3 ) on the _ hubble space telescope _ ( hst ) for a total of 96 orbits . the exposure time is divided between three nir filters : @xmath18 ( f105w ) , @xmath26 ( f125w ) , and @xmath15 ( f160w ) . the first epoch data used here total 60 orbits ( 16 in @xmath18 , 16 in @xmath26 , and 28 in @xmath15 ) , and reach a uniform depth in all filters of @xmath228.8 mag in a @xmath27 arcsec radius aperture ( despite the necessary rejection of two orbits of @xmath18 imaging due to severe persistence issues ) . our reduction of the wfc3 imaging data using the ` multidrizzle ` package ( koekemoer et al . 2002 ) produces final science grade images with psf fwhm of @xmath20.15 arcsec and a pixel size of 0.06 arcsec ( see bouwens et al . rebinning of the hudf acs @xmath28 ( f435w ) , @xmath29 ( f606w ) , @xmath30 ( f775w ) , and @xmath31 ( f850lp ) images to the scale of our wfc3 frames was also performed to facilitate aperture - matched photometry . the early release science ( ers ) program has observed a northern section of the goods - south field ( giavalisco et al . 2004 ) in the wfc3/ir channel in 10 pointings for a total of 60 orbits ( cf . windhorst et al . 2010 ) . the exposure time was divided evenly between three nir filters : @xmath32 ( f098 m ) , @xmath26 ( f125w ) , and @xmath15 ( f160w ) , reaching a depth of @xmath227.5 mag ( in a 0.2 arcsec radius aperture ) over an area of @xmath240 arcmin@xmath25 . our reduction of this data ( bouwens et al . 2010 ) was performed using the same procedures as for the hudf09 wfc3 imaging , resulting in final science grade images with psf fwhm @xmath20.15 arcsec and a pixel size of 0.06 arcsec . the goods acs @xmath28 ( f435w ) , @xmath29 ( f606w ) , @xmath30 ( f775w ) , and @xmath31 ( f850lp ) images were again rebinned to the same scale as the reduced wfc3 frames . a wealth of complementary data , including photometry at wavelengths beyond the range of the hudf / hudf09 and goods / ers hst imaging has been gathered for objects in the hudf and goods - south fields by a number of different teams . much of this data where publicly available has been compiled in the goods - music catalog ( grazian et al . 2006 ; santini et al . 2009 ) , which we employ in this study for objects with unambiguous matches in our primary source catalogs ( see section [ catalog ] below ) . the goods - music catalog contains psf- and aperture - matched photometry ( derived via the ` convphot ` package ; de santis et al . 2007 ) in the following additional filters : eso 2.2-wfi and vlt - vimos @xmath33 , eso vlt / isaac @xmath34 , @xmath35 , and @xmath36 , spitzer irac 3.6 , 4.5 , 5.8 , and 8.0 @xmath37 m , and mips 24 @xmath37 m . a compilation of spectroscopic redshifts is also included in the goods - music catalog , mainly sourced from the goods ( vanzella et al . 2005 , 2006 , 2008 ) , k20 ( mignoli et al . 2005 ) , and vvds ( le fvre et al . 2004 ) projects . in addition to the goods - music catalog , we also employ the published spectroscopic redshifts and spectral classifications for passive galaxies at @xmath38 in the hudf and goods - south fields from daddi et al . ( 2005 ) and cimatti et al . ( 2008 ) ( hereafter _ pbzk _ ) . the former sample consists of 7 high redshift sources in the hudf with _ passive bzk _ colors , confirmed via analysis of low resolution grapes spectra , of which 3 lie in the region of wfc3/ir imaging from the hudf09 program . the latter sample consists of 13 galaxies from the gmass catalog of spitzer 4.5 @xmath37 m detected sources in the goods - south field with a prominent mg@xmath39 feature in their eso vlt / fors2 spectra , of which 10 lie in the region of wfc3/ir imaging from the ers program . in total , 2 and 8 of these _ pbzk _ galaxies in the daddi et al . and cimatti et al . samples , respectively , lie within the target range of our _ yhvz _ color - selection criteria ( @xmath9 ) . a further 3 and 6 gmos spectroscopic redshifts from the @xmath40 roche et al . ( 2006 ) ero sample were employed for objects in the hudf09 and ers , respectively . although only one of the corresponding objects was missing a ` confident ' spectroscopic redshift in the goods - music catalog , the roche et al . redshifts were preferred due to the higher confidence and precision of these measurements . importantly , for one object ( ers0325 ) assigned @xmath41 in the goods - music catalog , the value of @xmath42 reported by the roche et al . team is much more consistent with its observed uv - to - ir sed . finally , for the purposes of star - galaxy separation we also refer to the pears - s star catalog of pirzkal et al . master source catalogs based on all robust @xmath15 detections to @xmath43 mag and @xmath20 mag in the hudf09 and ers imaging , respectively , were constructed as follows . first , ` sextractor ` ( bertin & arnouts 1996 ) was run on each of the reduced @xmath15 science images with four different choices of deblend parameters representing _ hot _ ( ` nthresh ` @xmath44 64 ) , _ medium _ ( ` nthresh ` @xmath44 16 ) , _ cold _ ( ` nthresh ` @xmath44 8) , and _ very cold _ ( ` nthresh ` @xmath44 2 ) extractions . in each case , the rms frames output from ` multidrizzle ` were scaled to reflect the true signal - to - noise of the reduced science images before their application as weight maps , and the following key control parameters were adopted : ` detect_thresh ` @xmath44 @xmath45 , ` detect_minarea ` @xmath44 12 , and ` min_cont ` @xmath44 0.0001 . based on these @xmath15 extractions we then re - ran ` sextractor ` in dual image mode to recover aperture - matched fluxes and kron magnitudes from the remaining optical - to - nir hst images ( @xmath28 to @xmath26 ) . careful visual inspection of the output kron ellipses of sources identified in each extraction was then performed to remove spurious detections , and to identify the most appropriate aperture for flux measurement of each real source . when choosing the best aperture we considered both the positions of possible matches in the goods - music catalog and the appearance of each source in all seven acs plus wfc3 ( @xmath28 through @xmath15 ) filters . in ambiguous cases we favoured the deblending of objects into distinct concentrations of @xmath15 flux ( the filter expected to best trace the underlying stellar mass distribution , modulo some uncertainty regarding the role of tp - agb stars ) . for calibration of the nir magnitudes we employed the latest ( ab system ) wfc3/ir zero - points available from the stsci webpage , and for the optical magnitudes we adopted the acs zero - points determined by the goods team . a total of 1052 and 3078 sources were thereby recovered to @xmath46 mag and @xmath20 mag in the hudf09 and ers imaging , respectively . to define the master source catalogs employed in this analysis we limit our samples to those systems with high s / n coverage in all 7 available acs plus wfc3 filters and lying within the region of sky covered by the goods - music catalog , a total of 993 and 2630 objects in the hudf09 and ers , respectively . for all objects in our master source catalogs we searched for counterparts in the goods - music catalog ( santini et al . 2009 ) with matching positions and photometric apertures . all object pairs with centroid differences of @xmath470.24 arcsec ( i.e. , 4 pixels in our drizzled wfc3 images ) and @xmath28 , @xmath29 , @xmath30 , and @xmath31 magnitudes differing by less than 0.5 mag between the two catalogs were classified automatically as successfully matched426 sources in the hudf09 and 1954 in the ers fields . the remaining object pairs not satisfying these criteria were visually inspected , and a further 44 and 294 matches were thereby recovered for the hudf09 and ers fields , respectively . for most galaxies at @xmath48 mag no counterpart could be identified due to the brighter goods - music selection limits ( @xmath49 mag or @xmath50 mag ) . in total , 523 and 382 objects in our master source catalogs from the hudf09 and ers , respectively , were ultimately deemed to be missing valid counterparts in the goods - music catalog . star - galaxy separation in our hudf09 and ers samples was performed as follows . first , all sources matched to counterparts in the goods - music catalog with spectroscopic redshift quality flag @xmath472 and @xmath51 ( 2 objects in the hudf09 , and 29 in the ers ) and all objects with centroids matched within @xmath52 arcsec to members of the pears - s star catalog of pirzkal et al . ( 2009 ) ( an additional 3 and 16 sources in the hudf09 and ers , respectively ) were identified as spectroscopically - confirmed stars . in addition , all ultra - compact outliers from the observed @xmath15 apparent magnitude@xmath53kron radius distributions at @xmath54 ( with @xmath55 0.2 for the hudf09 and 0.075 for the ers ) not already identified as spectroscopically - confirmed stars were visually - inspected and classified as candidate stars ( 2 in the hudf09 , and 35 in the ers ) unless resolved structure was clearly evident . figure [ fig1 ] shows these candidate stars on the @xmath15 apparent magnitude@xmath53kron radius diagnostic diagram . we exclude these sources from our analysis of galaxy color - selection efficiencies in section [ results ] . we have computed photometric redshifts for all objects in our hudf09 and ers source catalogs via template fitting to the observed spectral energy distributions in the hst acs and wfc3 filters ( @xmath28 to @xmath15 ) , supplemented by the additional photometry in the goods - music catalog ( for a total of 15 passbands ) where a valid counterpart could be identified . the actual template fitting was performed using the _ zurich extragalactic bayesian redshift analyzer _ ( zebra ; feldmann et al.2006 ) . to recover optimal photometric redshifts zebra modifies an empirical template library ( polletta et al . 2007 ; plus model seds for very blue objects from bruzual & charlot 2003 ) to account for systematic mismatches against the observed spectral energy distributions . we quantify empirically our mean photometric redshift uncertainties by comparison against matched objects with confident ( quality flag @xmath472 ) spectroscopic redshifts in the goods - music , daddi et al . , cimatti et al . , and roche et al . catalogs ( 86 in the hudf09 , and 446 in the ers ) , as illustrated in the panels in the left most columns of fig . we thereby estimate typical uncertainties ( @xmath56 ) of @xmath57 and @xmath58 for the hudf09 and ers master source catalogs , respectively . the larger uncertainties in the hudf09 photometric redshifts reflect the relatively small number of confident spectroscopic measurements available in this field for calibration of our sed template library . we also note low rates of catastrophic failures for the zebra photometric redshift pipeline , namely that @xmath475% of galaxies in the hudf09 , and @xmath479% of galaxies in the ers , are in disagreement with their spectroscopic redshifts by @xmath59 . at @xmath60 , where the greatest numbers of spectroscopic redshifts are available ( 78 in the hudf09 , and 355 in the ers ) , the mean uncertainties are reduced to @xmath61 , and @xmath62 , respectively . nevertheless , at @xmath63 our photometric redshift accuracy remains satisfactory with uncertainties of only @xmath64 , and @xmath65 , respectively although we note that in the case of the hudf09 the reliability of this comparison is limited by the small number ( 8) of spectroscopic redshifts in this interval . hence , as a further stage of quality control on our zebra photometric redshifts we also evaluate here the performance of the output confidence intervals derived from the posterior probability distribution in redshift space for each galaxy . specifically , we examine the ratio between @xmath66 ( i.e. , the absolute difference between the maximum likelihood photometric redshift and the true spectroscopic redshift ) and the width of the one sigma confidence interval for each galaxy ( cf . ilbert et al . if our posterior probability distributions are well estimated then we would expect the value of this quotient to be less than one for roughly 68% of each sample . in fact , our investigation reveals a slight underestimation of uncertainties by zebra such that a rescaling of these confidence interval widths by 1.25 and 1.35 for the hudf09 and ers samples , respectively , is required in order to achieve the 68% target . having calibrated our photometric redshift uncertainties we fold these via monte carlo simulation into our subsequent error calculations for the selection efficiencies of the color - color criteria presented in section [ results ] . in fig . [ fig2 ] we summarize graphically the key properties of our hudf09 and ers galaxy redshift catalogs . in order from the lefthand side of the page the four columns of this figure each contain two panels demonstrating for the hudf09 and ers ( top and bottom rows , respectively ) : ( i ) the close agreement between our photometric redshift estimates and the benchmark spectroscopic values where available in each sample ; ( ii ) the necessity of our ( slight ) re - scaling of the ( zebra ) estimated uncertainties to ensure consistency in the cumulative distribution functions for ( absolute ) empirical redshift error divided by one sigma confidence interval width , i.e. , @xmath67/(1@xmath68 error ) ; ( iii ) relative frequency histograms contrasting the redshift distributions of galaxies with spectroscopic redshifts against those for galaxies without ( further subdivided according to presence / absence of additional goods - music photometry)the latter distinction thereby highlighting the unique power of wfc3/ir imaging for exploring the @xmath69 discovery space ; and ( iv ) the overall redshift distributions for all galaxies in our magnitude limited samples both of which peak around @xmath70 , as is typical for untargeted , deep imaging surveys with optical / near - ir selection ( e.g. rudnick et al . 2001 ; ilbert et al . 2009 ) . finally , we note that our hudf09 @xmath19 mag master source catalog contains both @xmath13 candidates at this brightness from oesch et al . ( 2010 ) ( udfz-42566566 and udfz-38807073 ) plus the probable supernova ( udfz-34537360 ) , all of which were assigned photometric redshifts @xmath71 by zebra . however , the remainder of the recently reported @xmath13 and @xmath72 candidates in the hudf09 and ers fields ( bouwens et al . 2010a , b ; oesch et al . 2010 ; wilkins et al . 2010 ) are fainter than the apparent magnitude limits employed in this study . specific star formation rates ( ssfrs ) and stellar masses for all galaxies in our hudf09 and ers samples were estimated using an updated version of zebra , zebra+ ( oesch et al . 2010 , in prep . ) . zebra+ computes various galaxy physical properties based on template fitting to model seds with the ( previously - computed ) photometric redshift , or measured spectroscopic redshift , held fixed . for this analysis we employed a library of sed templates based on the bruzual & charlot ( 2003 ) stellar population synthesis code with a chabrier imf ( chabrier 2003 ) , drawn from a grid of exponential decay star formation rate models well - sampled over a wide range of ages ( 0.01@xmath5312 gyr ) , decay rates ( @xmath73 @xmath44 0.05@xmath539 gyr ) , and metallicities ( 0.05@xmath532 @xmath74 ) . the impact of dust reddening is accounted for during template fitting via a calzetti extinction law ( calzetti 2001 ) with the e@xmath75 value a free parameter ( see sargent et al . 2010 for a demonstration of the impact of dust reddening on galaxies at intermediate redshift ) . of course , there is a substantial degree of degeneracy in the choice of stellar population template and dust extinction model inherent in the fitting of broad - band seds , which is expected from previous studies to dominate the error budget ( contributing @xmath76 dex ) in samples for which broad - band flux measurements are available in a significant number of filters adequately sampling each galaxy sed ( e.g. bolzonella et al . 2010 ; taylor et al . thus , for galaxies in our sample matched to counterparts in the goods - music survey we expect to achieve roughly this level of accuracy , although for galaxies with hst - only photometry our uncertainties will be necessarily higher . details for zebra+ , including a further investigation into the relevant uncertainties , are given in oesch et al . ( 2011 , in prep . ; see also oesch et al . 2009 ) . we note that of the 3 and 10 spectroscopically - confirmed passive galaxies of daddi et al . and cimatti et al . , respectively , which enter our hudf09 and ers samples , all but one have zebra+ estimated ssfrs representative of passively - evolving stellar populations ( i.e. , ssfr @xmath77 yr@xmath78 ) . the one discrepant object ( ers2953 ) was assigned an intermediate ssfr of @xmath79 yr@xmath78 with an extreme e@xmath75 fit of 0.6 mag . in fig . [ fig3 ] we present the distributions of galaxies in our hudf09 and ers master source catalogs in both absolute ( rest - frame @xmath80-band ) magnitude and stellar mass as function of redshift . two selection limits are indicated on each plot : ( i ) the sed-_dependent _ completeness limit corresponding to the lower bound on luminosity / mass of the faintest / least massive systems detectable at a given redshift above our @xmath15 magnitude threshold all members of our hudf09 and ers samples thus lie above , or on , this line ; and ( ii ) the sed-_independent _ completeness limit corresponding to the respective lower bound under the constraint of full detectability for galaxies of _ all _ spectral types . note that the observed distributions of galaxies are not tightly bound by the former until @xmath81 as the largest k - correction from observed @xmath15 to rest - frame @xmath80 below this redshift occurs for the reddest spectral type ( which is not matched to any of the faintest / least massive systems in these surveys ; see , for comparison , cameron & driver 2007 , 2009 ) . the sed-_independent _ limits presented here set the thresholds ( @xmath82@xmath4 for the hudf09 and @xmath83@xmath4 for the ers ) for selection of the mass - limited galaxy samples used to investigate the morphological mix at @xmath38 in section [ morph ] . through visual inspection of synthetic evolutionary tracks for model stellar populations in our zebra+ template library we have identified @xmath85 vs. @xmath86 as the most favorable parameter space for selection and characterization of @xmath9 galaxies given the set of wfc3 and acs filters employed in the hudf09/hudf and ers / goods imaging programs . example tracks for instantaneous burst and constant star formation rate models ( with metallicities of @xmath87 @xmath74 and formation redshifts of both @xmath88 and @xmath89 ) are presented in fig . [ fig4 ] to illustrate the expected evolution of active and passive systems in this plane . these tracks indicate that galaxies at @xmath9 should be significantly redder in @xmath85 , at fixed @xmath86 , than galaxies at lower or higher redshifts . moreover , according to the calzetti ( 2001 ) reddening law , the effect of dust extinction on the observed galaxy colors in this diagram is simply a diagonal shift ( as indicated in fig . [ fig4 ] ) with the potential to move lower redshift objects into the region of high redshift color - color space only for passively - evolving systems at @xmath90 . when the corresponding color - color distributions of real galaxies at @xmath91 in the hudf09 and ers are contrasted against those at lower and higher redshifts ( see fig . [ fig4 ] again ) it is clear that galaxies at these epochs are indeed well separated , occupying distinct color - color sequences with minimal overlap . [ table1 ] -0.5 cm [ cols="<,<,^,^,^,^,^",options="header " , ] @xmath92 parameters in columns 4 - 21 derived from ` sextractor ` output ; see bertin & arnouts ( 1996 ) for detailed definitions . + @xmath93 regions of `` high signal - to - noise '' imaging defined as the area on each frame with local median rms less than twice the global median rms . the complete source catalogs employed in this study , containing all robust detections to @xmath19 mag in the hudf09 and @xmath20 mag in the ers , are publicly - available in electronic form from the following url : ` http://www.astro.phys.ethz.ch/cameronetal2011/ ` . these catalogs have been carefully inspected both to remove spurious detections and to ensure selection of the optimum photometric aperture for flux measurement of each object . a series of flags is used to indicate which objects appear within the area covered by the goods - music catalog , which are classified as well - matched to a counterpart in the goods - music catalog , which appear in a region of high signal - to - noise imaging in all 7 acs plus wfc3 filters , which have a goods - music catalog spectroscopic redshift significantly in conflict with their observed colors , which are spectroscopically - confirmed stars or unresolved stellar candidates , and which are in our _ passive yhvz _ selection . the column numbers and definitions for these catalogs are listed in table [ catalogs ] below . as a first application of our @xmath15-selected hudf09 and ers source catalogues presented in section [ sourcecatalogs ] we establish the most basic , and thus fundamental , facts concerning the morphologies of @xmath0 galaxies in the rest - frame optical window compared with the morphologies of the same galaxies in rest - frame uv light . in recent years , ground - based or hst / nicmos nir imaging sampling rest - frame optical wavelengths at @xmath96 has uncovered a galaxy population at these epochs rich in diversity of structure : with smooth disk / bulge - disk galaxies , and rather normal ellipticals coexisting with massive interacting / disturbed systems , as well as ultra - compact , elliptical - like galaxies , large , turbulent , clumpy disks , and compact , dispersion - dominated disks ( corbin et al . 2000 ; daddi et al . 2005 ; papovich et al . 2005 ; franx et al . 2008 ; stockton et al . 2008 ; genzel et al . 2008 ; conselice et al . 2008 ; buitrago et al . 2008 ; frster - schreiber et al . 2009 ; elmegreen et al . 2009 ) . interestingly , many galaxies at @xmath63 exhibit genuinely disturbed / interacting morphologies , i.e. , not only the young stellar populations visible in uv light , but , rather , most of the stellar mass in these galaxies , is arranged in an irregular manner . with the significantly higher angular resolution and uniformly high imaging sensitivity of wfc3/ir , the new hudf09 and ers @xmath15 images offer the most detailed view to - date of the high - redshift galaxy population at rest - frame optical wavelengths . through visual classification of the galaxies in our @xmath9 color - selected hudf09 and ers samples , we thus investigate here , qualitatively , the diversity of galaxy structure as revealed by the wfc3/ir @xmath15 ( rest - frame optical ) images relative to the acs @xmath31 ( rest - frame uv ) images . the first indication of the paramount importance of the rest - frame wfc3 images in understanding the @xmath0 galaxy population comes from the comparison with the goods - music catalog discussed in section [ catalog ] . in most cases where no counterpart to one of our @xmath15 sources was found in the goods - music catalog , the centroid offset that prevented automatic catalog matching was due to morphological differences between the rest - frame uv @xmath31 image ( used as a basis for most goods - music detections ) and the new rest - frame optical @xmath15 image provided by the wfc3 . in a number of instances , the greater sensitivity of the wfc3 @xmath15 image to the smoother , underlying stellar mass distribution provided a confident aperture definition and centroid conflicting significantly with that adopted in the goods - music analysis , and no match was possible . it is evident that the rest - frame uv images on which the goods - music catalog was based often identify , as the galaxy center , a knot of star formation substantially displaced from the true galaxy center . this is , on the other hand , well identified in the rest - frame optical wfc3 images . also , the combination of passive - sed color - selection ( _ passive yhvz _ ) and rest - frame optical morphologies ( see section [ pyhvz ] ) reveals that passive galaxies at such early epochs are indeed mostly spheroids , as shown in previous studies ( e.g. van dokkum et al . 2004 , 2010 ; daddi et al . 2005 ; trujillo et al . 2006 ; cimatti et al . 2008 ; franx et al . 2008 ) , but , in addition , a substantial fraction ( @xmath215 - 20% ) show disk morphologies . we speculate that this population of ` quenched ' disks at such early epochs might be a smoking gun for disk galaxies which have been cut out of their feeding streams of cold gas , as seen in numerical simulations and expected on theoretical grounds ( e.g. kere et al . 2005 ; dekel et al . mergers of such disks would be ` dry ' while nonetheless involving galaxies with a dynamically - cold disk component . foreseen large @xmath15-band surveys with the hst will be instrumental in understanding what fraction of dry mergers at @xmath38 - 3 involve disk galaxies an important constraint for galaxy evolution models . furthermore , by comparison of the acs and wfc3 morphologies of all galaxies in our master source catalogs within the @xmath0 redshift window we identify four main galaxy sub - types . the first category ( class 1 ) are galaxies which , in the @xmath15-band wfc3 images , reveal a regular disk morphology ( with or without a bulge)in contrast with their irregular rest - frame uv morphologies . these class 1 galaxies are the only ones which show a morphological difference between their rest - frame uv and rest - frame optical images . the remaining three classes do separate galaxies with very different morphologies , but , strikingly , these morphologies are unchanged in rest - frame uv and rest - frame optical light . these classes are , respectively : galaxies which , both in the rest - frame uv and rest - frame optical nir wfc3 images maintain a regular , spheroid - like morphology ( class 2 ) ; galaxies which retain , in the nir wfc3 images , a highly irregular / disturbed morphology , similar to that observed in the acs rest - frame uv images ( class 3 ) ; and , finally , the population of small - sized , smooth and rather symmetric ` blobs ' of light in both the @xmath15 and @xmath31 images for which it is unclear whether these systems are proto - disks or spheroids ( class 4 ) . examples of each class are given in figures [ fig8 ] , [ fig9 ] , [ fig10 ] , and [ fig11 ] . in class 1 objects , the rest - frame optical morphology is substantially smoother , i.e. , less clumpy , and also more centrally concentrated , than in the rest - frame uv . indeed , in such systems , the wfc3 @xmath15 images expose a well - defined center in contrast with the clumpy / asymmetric appearance of these galaxies in the rest - frame uv light . two such galaxies in our sample ( udf0093 , ers2047 ) exhibit clear bulge - dominated morphologies , and a number of others ( e.g. udf0590 , ers3032 ) indicate the presence of small , central , disk light excesses , most likely small(er ) bulges , consistent with counterparts identified by papovich et al . ( 2005 ) and stockton et al . ( 2008 ) . class 2 objects show the smooth and compact appearance indicated by previous studies to be typical for many spheroids at these epochs ( e.g. daddi et al . 2004 , 2005 ; cimatti et al . 2008 ; franx et al . 2008 ; damjanov et al . already szomoru et al . ( 2010 ) have used the hudf09 wfc3/ir images to establish , on firm grounds , the compactness of one ` quenched ' spheroid at @xmath97 ( udf0884 ) , extracted from the daddi et al . sample . while many galaxies in this morphological class are identified as _ passive yhvz _ systems in our color - color selection criteria ( section [ pyhvz ] ) , we stress that a non - negligible fraction of them have instead star - forming seds , indicating that such smooth , compact morphologies , while more typical of the passively - evolving galaxy population at these redshifts , can also , to a smaller extent , be achieved by galaxies that are still actively forming stars at those epochs ( e.g. ers0556 , ers1202 ) . class 3 objects exhibit a wide variety of morphologies , from systems with a compact nucleus plus a highly asymmetric disk - like structure ( e.g. udf0124 ) to systems with multiple bright nuclei ( e.g. udf0435 , ers2010 , ers1203 ) . interestingly , the disk - like structures in these systems are often warped ( e.g. udf0124 , udf0870 ) , a phenomenon that is observed to occur in high - resolution , hydrodynamical numerical simulations in cdm cosmology , as disk galaxies stream down the hot gas filaments of the cosmic web ( steger et al 2010 , in prep . ) . in a number of cases ( e.g. udf0435 , udf0870 ) there exist clear differences between the @xmath31 and @xmath15 brightnesses of specific nuclei within these galaxies , reflecting a complex , spatially - dependent i.e . , ` clumpy'star formation activity , and thus history , in these systems . since these systems will most likely evolve into relatively massive galaxies by redshift zero , a substantial stellar population mixing , through mergers or internal dynamical instabilities , must occur over cosmic time to achieve the massive galaxy population with rather smooth stellar population gradients that we observe today . one class 3 irregular galaxy in our sample , a lop - sided disk ( udf0870 ) at @xmath98 , might exhibit a large - scale bar . while a decline in the bar fraction with increasing redshift has been demonstrated by several studies out to redshifts of @xmath99 ( e.g. abraham et al . 1999 ; sheth et al . 2008 ; cameron et al . 2010 ) , it is only with the newly available rest - frame optical , deep and high - resolution images of high-@xmath100 galaxies that a firm statement on the bar fraction at @xmath101 will become feasible . the discovery of one such system in the hudf09 images indicates that large - scale , coherent dynamical instabilities such as bars can and do exist at such early epochs ( see also the @xmath102 ` bar - like structures ' classified in the rest - frame uv by ravindranath et al . 2006 ) , although they may be not as frequent as at redshift @xmath70 , given that disks at such early times appear to be , on average , substantially hotter and more turbulent than at later times ( e.g. genzel et al . 2006 , 2008 , 2010 ) . class 4 objects are relatively small - sized systems exhibiting smooth , symmetric morphologies , but with less centrally - concentrated light distributions than the class 2 objects . as such , these class 4 objects could either be diffuse counterparts of the class 2 spheroids , or else small proto - disks in the earliest stages of growth . future studies of the quantitative structural parameters and , ideally , kinematical properties of this object class are required to establish the place of these systems within the cosmic history of galaxy structural assembly . in fig . [ fig12 ] we present the morphological mix of these galaxy types in each of the @xmath21.2 gyr intervals , @xmath103 and @xmath104 , based upon counts in mass - limited samples of 30 galaxies at @xmath82@xmath4 in the hudf09 and 30 galaxies at @xmath83@xmath4 in the ers ( see fig . [ fig3 ] for the relationship between our mass completeness limits and the distribution of galaxies in our hudf09 and ers source catalogs as a function of redshift ) . interestingly , as we confirm via artificial noise degradation analysis of the hudf09 images in the appendix to this paper , despite a @xmath214@xmath105 shorter exposure time the depth of the ers imaging is sufficient to perform morphological classification of the ( relatively crude ) nature described here at comparable accuracy to that in the hudf09 . the combined ( i.e. , @xmath106@xmath3@xmath4 ) fractions of objects in the four different classes are , respectively , @xmath107% , @xmath108% , @xmath109% and @xmath109% at @xmath103 , and @xmath110% , @xmath111% , @xmath112% and @xmath112% at @xmath104 ( with 1@xmath68 binomial uncertainties estimated using a bayesian approach ; cf . cameron 2010 ) . note that the evolution of these fractions likely implies substantial morphological transformations across the redshift @xmath38 epoch of peak star formation activity . we highlight , in particular , the emergence of a substantial population of regular disk morphologies , and a substantial increase in the fraction of spheroidal morphologies at epochs @xmath113 . importantly , the four morphological classes show different redshift and stellar mass distributions . at @xmath104 class 3 objects ( with highly irregular morphologies also in rest - frame optical light ) and class 4 objects ( with small - sized , ` blob'-like morphologies ) are overwhelmingly dominant at both intermediate and high mass scales , while class 2 spheroids are largely restricted to above @xmath5 @xmath4 . at @xmath103 class 3 and class 4 objects are now largely restricted to intermediate masses , while the class 2 spheroid population now dominates at high masses . at @xmath103 the class 1 systems emerge and contribute @xmath225% of the morphological mix at both mass scales . at and around the epoch of peak cosmic star formation rate in the universe , therefore , the statistical census of the brightest galaxies indicates that the most massive of them are assembled in regular , hubble - type elliptical and disk ( plus bulge ) rest - frame optical morphologies , while less massive systems often reveal , in their rest - frame optical images , clear signs of dynamical youth . perhaps the most striking morphological difference between these two redshift intervals is the transition across the @xmath38 epoch for galaxies at the highest stellar masses , @xmath106@xmath114 , from a diversity of structural types to a dominant population of spheroids . although sed - fit based ssfrs suffer from large uncertainties , and our samples of galaxies with masses above @xmath114 at these redshifts are only small ( 16 at @xmath6 and 15 at @xmath7 ) , we note that the assembly of the massive spheroids appears to be accompanied by a rapid ` quenching ' of star formation . adopting a redshift - dependent division between star - forming and passively - evolving systems based on the ssfr necessary for a galaxy to double its stellar mass between its observed redshift and the present day , and employing our sed - fit ssfrs computed with zebra+ ( see section [ ssfrs ] ) , we find the fraction of star - forming galaxies at @xmath106@xmath114 falls from @xmath115% at @xmath6 to @xmath116 at @xmath7 . using our own color - based criteria for identification of a `` passively - evolving '' sed , i.e. , _ passive yhvz _ , the fraction of quenched galaxies above @xmath3 @xmath4 increases from @xmath117% at @xmath6 to @xmath118% at @xmath7 . this is consistent with the observed decline in the star - forming fraction estimated by feulner et al . ( 2005 ) for a selection of galaxies in the fors - deep and goods - south fields . if confirmed by larger samples , these results would indicate that @xmath8 is the epoch when both morphological transformations first produce a significant population of massive spheroids and physical processes quench their star formation activity and move them to the red sequence , as expected in recent models ( peng et al . the wealth of wfc3 imaging of the high redshift universe achieved or planned , coupled with archival acs optical imaging , is self - sufficient , without additional ground - based photometry , to robustly identify galaxies in the @xmath0 redshift window . in this paper we have presented , to this aim , new , robust and efficient color - selection criteria established primarily on the @xmath85 vs.@xmath86 color - color diagram ( and similar ones , slightly less efficient but still valuable , which use combinations of the hst @xmath26 , @xmath15 , @xmath29 , and @xmath119 filters , or simply the @xmath29 , @xmath119 , and @xmath15 filters ) . we have also shown that an additional color selection on the @xmath85 vs.@xmath86 color - color diagram further enables the disentanglement of passively - evolving , ` quenched ' galaxies from actively star - forming galaxies at these redshifts . as a first use of the new galaxy catalog that we have produced , starting from the @xmath15 wfc3 images , we have performed a qualitative comparison of @xmath0 galaxy morphologies between rest - frame uv and rest - frame optical light . regular disk galaxies with or without a bulge are present at these epochs the disk components being unseen or irregular in rest - frame uv images , and being clearly revealed in the rest - frame optical images . together with the population of smooth spheroids , these regular disk galaxies have substantial stellar masses , @xmath106@xmath120 ( or , for the spheroids , even above this threshold ) . at masses @xmath106@xmath114 , the wfc3 nir images reveal a transition of morphologies from a variety of structures at @xmath121 to a marked dominance of elliptical - like stellar distributions at @xmath113 . across the @xmath8 boundary the fraction of star - forming galaxies at @xmath106@xmath114 decreases from @xmath260% to @xmath25% . consistent with analysis of the @xmath122 population ( peng et al . 2010 ) , and observed here in rest - frame optical morphologies and colors , @xmath38 appears therefore to be the epoch at which assembly and quenching of the massive spheroid population begins in earnest . the wfc3 nir images also reveal , however , that , less massive galaxies at the same epochs often keep a highly irregular morphology , also in the rest - frame optical light , as they had in the rest - frame uv light . this indicates that their previously reported irregular morphologies in rest - frame uv are not driven by detection of star - forming populations on top of settled , underlying older stellar populations , but rather an indication of a genuine juvenile dynamical state for much of the @xmath123@xmath4 galaxy population at these redshifts . + abraham , r. g. , et al . 1999 , mnras , 308 , 2 , 569 - 576 adelberger , k. l. , steidel c. c. , shapley , a. e. , hunt , m. p. , erb , d. k. , reddy , n. a. , & pettini , m. 2004 , apj , 607 , 1 , 226 - 240 barger , a. j. , & cowie , l. l. 2005 , apj , 635 , 1 , 115 - 122 beckwith , s. v. w. , et al . 2006 , aj , 132 , 5 , 1729 - 1755 bertin , e. , & arnouts , s. 1996 , a&a , 117 , 393 - 404 bolzonella , m. , et al . 2010 , a&a , 524 , a76 bouwens , r. j. , illingworth , g. d. , franx , m. , & ford , h. 2007 , apj , 670 , 2 , 928 - 958 bouwens , r. j. , et al . 2010a , apj , 709 , 2 , l133-l137 bouwens , r. j. , et al . 2010b , preprint(arxiv:1006.4360 ) bouwens , r. j. , et al . 2011 , nature , 469 , 7331 , 504 - 507 bruzual , g. , & charlot , s. 2003 , mnras , 344 , 4 , 1000 - 1028 bunker , a. , et al . 2010 , mnras , 409 , 2 , 855 - 866 calzetti , d. 2001 , pasp , 113 , 790 , 1449 - 1485 cameron , e. , & driver , s. p. , 2007 , mnras , 377 , 1 , 523 - 534 cameron , e. , & driver , s. p. , 2009 , a&a , 493 , 2 , 489 - 500 cameron , e. , driver , s. p. , graham , a. w. , & liske , j. 2009 , apj , 699 , 1 , 105 - 117 cameron , e. , et al . 2010 , mnras , 409 , 1 , 346 - 354 cameron , e. 2010 , preprint(arxiv:1012.0566 ) cassata , p. , et al . 2010 , apjl , 714 , 1 , l79-l83 chabrier , g. 2003 , pasp , 115 , 809 , 763 - 795 cimatti , a. , et al . 2008 , a&a , 482 , 1 , 21 - 42 corbin , m. r. , et al . 2000 , aj , 119 , 3 , 1062 - 1077 croom , s. m. , smith , r. j. , boyle , b. j. , shanks , t. , miller , l. , outram , p. j. , & loaring , n. s. 2004 , mnras , 349 , 4 , 1397 - 1418 daddi , e. , cimatti , a. , renzini , a. , fontana , a. , mignoli , m. , pozzetti , l. , tozzi , p. , & zamorani , g. 2004 , apj , 617 , 2 , 746 - 764 daddi , e. , et al . 2005 , apj , 626 , 680 - 697 damjanov , i. 2009 , apj , 695 , 1 , 101 - 115 dekel , a. , & birnboim , y. 2008 , mnras , 383 , 1 , 119 - 138 de santis , c. , grazian , a. , fontana , a. , & santini , p. 2007 , new astronomy , 12 , 4 , 271 - 288 elmegreen , d. m. , et al . 2009 , apj , 701 , 1 , 306 - 329 fan , x. , et al . 2001 , aj , 122 , 6 , 2833 - 2849 feldmann , r. , et al . 2006 , mnras , 372 , 2 , 565 - 577 feldmann , r. , et al . 2010 , apj , 709 , 1 , 218 - 240 finkelstein , s. l. , papovich , c. , giavalisco , m. , reddy , n. a. , ferguson , h. c. , koekemoer , a. m. , & dickinson , m. 2010 , apj , 719 , 2 , 1250 - 1273 frster - schreiber , n. m. , et al . 2009 , apj , 706 , 2 , 1364 - 1428 frster - schreiber , n. m. , et al . 2011 , preprint(arxiv:1104.0248 ) franx , m. , et al . 2003 , apj , 587 , 2 , l79-l82 franx , m. , van dokkum , p. g. , frster schreiber , n. m. , wuyts , s. , labb , i. , & toft , s. 2008 , apj , 688 , 770 - 788 genzel , r. , et al . 2008 , apj , 687 , 1 , 59 - 77 giavalisco , m. , et al . 2004 , 600 , 2 , l93-l98 grazian , a. , et al . 2006 , a&a , 449 , 3 , 951 - 968 grogin , n. a. , et al . 2011 , preprint(arxiv:1105.3753 ) hopkins , a. m. , & beacom , j. f. 2006 , apj , 651 , 1 , 142 - 154 ilbert , o. , et al . 2009 , apj , 690 , 2 , 1236 - 1249 kere , d. , katz , n. , weinberg , d. h. , & romeel , d. 2005 , mnras , 363 , 1 , 2 - 28 koekemoer , a. m. , fruchter , a. s. , hook , r. n. , & hack , w. 2002 , the 2002 hst calibration workshop , space telescope science institute , baltimore , maryland , 337 koekemoer , a. m. , et al . 2011 , preprint(arxiv:1105.3754 ) le fvre , o. , et al . 2004 , a&a , 428 , 1043 - 1049 lilly , s. j. , le fvre , o. , hammer , f. , & crampton , d. 1996 , apj , 460 , l1 madau , p. , ferguson , h. c. , dickinson , m. e. , giavalisco , m. , steidel , c. c. , & fruchter , a. 1996 , mnras , 283 , 4 , 138 - 1404 madau , p. , pozetti , l. , & dickinson , m. 1998 , apj , 498 , 106 mccarthy , p. j. , et al . 2004 , apj , 614 , 1 , l9-l12 mclure , r. j. , dunlop , j. s. , cirasuolo , m. , koekemoer , a. m. , sabbi , e. , stark , d. p. , targett , t. a. , & ellis , r. s. 2010 , mnras , 403 , 2 , 960 - 983 mcgrath , e. , stockton , a. , canalizo , g. , iye , m. , & maihara , t. 2008 , apj , 682 , 303 - 318 mignoli , m. , et al . 2005 , a&a , 437 , 3 , 883 - 897 oesch , p. a. , et al . 2007 , apj , 671 , 2 , 1212 - 1226 oesch , p. a. , et al . 2009 , apj , 690 , 2 , 1350 - 1357 oesch , p. a. , et al . 2010 , apj , 709 , 1 , l16-l20 papovich , c. , et al . 2005 , apj , 631 , 1 , 101 - 120 peng , y. , et al . 2010 , apj , 721 , 1 , 193 - 221 pirzkal , n. , et al . 2009 , apj , 695 , 2 , 1591 - 1603 polletta , m. , et al . 2007 , apj , 663 , 1 , 81 - 102 ravindranath , s. , et al . 2006 , apj , 652 , 963 - 980 rudnick , g. , et al . , 2001 , aj , 122 , 2205 - 2221 roche , n. d. , et al . 2006 , mnras , 370 , 74 - 90 santini , p. , et al . 2009 , 504 , 3 , 751 - 767 sargent , m. t. , et al . 2010 , apj , 714 , 1 , l113-l117 scarlata , c. , et al . 2007 , apjs , 172 , 1 , 406 - 433 sheth , k. , et al . 2008 , apj , 675 , 2 , 1141 - 1155 steidel , c. c. , giavalisco , m. , pettini , m. , dickinson , m. , & adelberger , k. l. 1996 , apj , 462 , l17 steidel c. c. , adelberger , k. l. , shapley , a. e. , pettini , m. , dickinson , m. , & giavalisco , m. 2003 , apj , 592 , 2 , 728 - 754 steidel , c. c. , shapley , a. e. , pettini , m. , adelberger , k. l. , erb , d. k. , reddy , n. a. , & hunt , m. p. 2004 , apj , 604 , 2 , 534 - 550 stockton , a. , mcgrath , e. , canalizo , g. , iye , m. , & maihara , t. 2008 , apj , 672 , 146 - 152 szomoru , d. , et al . 2010 , apjl , 714 , l244-l248 taylor , e. n. , franx , m. , brinchmann , j. , van der wel , a. , van dokkum , p. g. 2010 , apj , 722 , 1 , 1 - 19 thompson , d. , et al . 1999 , apj , 523 , 1 , 100 - 106 toft , s. , franx , m. , van dokkum , p. , frster schreiber , n. m. , labbe , i. , wuyts , s. , & marchesini , d. 2009 , apj , 705 , 225 tresse , l. , et al . 2007 , a&a , 472 , 2 , 403 - 419 trujillo , i. , et al . 2006 , mnras , 373 , 1 , l36-l40 warren , s. j. , hewett , p. c. , & osmer , p. s. 1994 , apj , 421 , 412 - 433 windhorst , r. a. , et al . 2010 , preprint(arxiv:1005.2776 ) wilkins , s. m. , et al . 2010 , mnras , 403 , 2 , 938 - 944 van dokkum , p. g. , et al . 2004 , apj , 611 , 703 - 724 van dokkum , p. g. , et al . 2006 , apj , 638 , 2 , l59-l62 van dokkum , p. g. , et al . 2010 , apjl , 677 , l5-l8 vanzella , e. , et al . 2005 , a&a , 434 , 1 , 53 - 65 vanzella , e. , et al . 2006 , a&a , 454 , 2 , 423 - 435 vanzella , e. , et al . 2008 , a&a , 478 , 1 , 83 - 92 based on observations made with the nasa / esa _ hubble space telescope _ , obtained from the data archive at the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . p.o.acknowledges support from the swiss national foundation ( snf ) . in this brief appendix we confirm through statistical image degradation that the limited depth of the ers imaging with wfc3/ir is indeed sufficient to allow for confident morphological classification of @xmath8 galaxies in the rest - frame optical , at least to the relatively crude level of detail employed in the analysis of section [ morph ] . adopting the ultra - deep hudf09 @xmath15 images of our @xmath9 master sample as the benchmark for morphological classification at these redshifts we thus aim to demonstrate a consistency of visual appearance following artificial degradation to the noise level of the ers . in particular , the net exposure time of the master udf @xmath15 image is a factor of @xmath214 times greater than that of the ers @xmath15 image ; indeed such an extraordinary depth at these wavelengths was crucial to the principal aim of the hudf09 program namely , the search for @xmath124 - 10 candidates appearing as @xmath18/@xmath26 drop - outs ( bouwens et al . 2010a , 2011 ) . consequently , the contributions of both background sky noise and shot noise in the ers are @xmath2@xmath125 times higher than in the hudf09 . the final image noise in our master frames is , moreover , weakly correlated between adjacent pixels as a result of the reduction process . hence , rather than simply adding an independent noise component to each pixel to simulate the degradation in sky noise we attempt to at least partially approximate the impact of pixel - to - pixel correlation by degrading our udf images instead through the stacking of blank regions of real background sky . for the addition of shot noise , however , we simply suppose that all flux detected above the 3@xmath68 sky noise level in the original udf images corresponds to real object flux , and we directly resample this flux ( with poisson uncertainty ) in the @xmath126 statistical package at the ers rate . in figure [ fig13 ] we compare the original ( @xmath15 ) hudf09 postage stamp images of three archetypal class 1 ( disk , or bulge / disk ) systems from fig . [ fig8 ] , and three archetypal class 3 ( irregular ) systems from fig . [ fig10 ] , against our new versions of these images degraded to the noise level of the ers . it is clear from inspection of this figure that the impact of the additional noise is to reduce the relative prominence of certain faint and/or fine - structure features in these @xmath8 galaxies , but that the broad morphological properties of each system ( e.g. , the presence / absence of multiple nuclei , a central bulge , an outer disk , or a warped disk ) remain readily discernable , and thus each galaxy remains clearly recognisable as a member of its assigned class . of course , studies attempting to achieve finer morphological classifications than those conducted here , or else attempting to perform quantitative ` clump - science ' ( e.g. , frster - schreiber et al . 2011 ) , may well prove more sensitive to the imaging depth than our relatively crude analysis of section [ morph ] . further , one should not attempt to infer from the results of this appendix that certain quantitative morphological analysis procedures ( e.g. , bulge - disk decomposition ) will also prove insensitive to the difference in the hudf09 and ers imaging depths such a statement clearly requires a more detailed investigation tailored to each morphological analysis tool under study . as a further robustness test of the morphological evolution described in section [ morph ] we have investigated the possibility that the decreasing fraction of massive spheroids with increasing redshift ( past @xmath8 ) is an artifact of the impact of cosmological surface brightness dimming ( and band - pass shifting ) on our observations . in particular , we have verified that our visual classifications of all massive elliptical galaxies in our @xmath103 sub - sample remain consistent after artificial redshifting of their @xmath15 images to @xmath127 . the results of this redshifting for six archetypal @xmath128 ellipticals from our sample are displayed in fig . [ fig14 ] . the extremely centrally - concentrated light profiles of these systems ensure that even under substantial cosmological surface brightness dimming they remain quite distinct in appearance from the remaining three classes .

we show that a smooth unknotted curve in satisfies an isoperimetric inequality that bounds the area of an embedded disk spanning the curve in terms of two parameters : the length of the curve and the thickness ( maximal radius of an embedded tubular neighborhood ) of the curve . for fixed length , the expression giving the upper bound on the area grows exponentially in . in the direction of lower bounds , we give a sequence of length one curves with for which the area of any spanning disk is bounded from below by a function that grows exponentially with . in particular , given any constant , there is a smooth , unknotted length one curve for which the area of a smallest embedded spanning disk is greater than .

math0306313

the classical isoperimetric inequality gives an upper bound for the area of the disk bounded by a curve in @xmath7 . for a simple closed curve of length @xmath1 in the plane that encloses area @xmath6 , it states that @xmath8 with equality if and only if the curve is a circle . there are many extensions of the isoperimetric theorem to higher dimensions ; some of these are discussed in almgren @xcite , burago and zalgaller @xcite , and osserman @xcite . there are two natural extensions of the isoperimetric inequality to a simple closed @xmath9-curve @xmath10 in @xmath0 of length @xmath1 . \(1 ) there exists an immersed smooth disk in @xmath0 , with @xmath10 as boundary , having area @xmath6 with @xmath11 this disk may self - intersect and may intersect the curve @xmath10 , as indeed it must if the curve is knotted . if the curve is not a circle , then there exists such an immersed disk for which strict inequality holds . \(2 ) there exists an orientable smooth surface embedded in @xmath0 , with no restriction on its genus , with @xmath10 as boundary , having area @xmath6 , with @xmath12 if the curve is not a circle , then there exists such a surface for which strict inequality holds . result ( 1 ) traces back to a result of andre weil in 1926 concerning isoperimetric inequalities for surfaces of non - positive gaussian curvature . see also beckenbach and rado @xcite . result ( 2 ) traces back to a 1930 argument in blaschke @xcite , see osserman @xcite . for completeness we indicate proofs of these well - known results in an appendix ( [ classical ] ) . we derive ( 2 ) directly from ( 1 ) . the hypotheses in ( 1 ) and ( 2 ) can be weakened to require only that @xmath10 be rectifiable ; however we assume @xmath9-curves for ease of comparison with later results . this paper treats the situation when the two conditions above are imposed simultaneously . that is , we suppose the curve @xmath10 is unknotted , and therefore is the boundary of an embedded disk , and ask for upper bounds on the area of some embedded spanning disk . note that an embedded spanning disk attaining the minimal area under these conditions need not exist , because the embeddedness condition may not be preserved under limits . however there is still an infimum for the area of all embedded spanning disks , which we refer to as the minimal area of a spanning disk . we will show that under these conditions there is no upper bound on the minimal area given in terms of the length @xmath1 of @xmath10 alone . to obtain an upper bound on the area of some spanning disk we must therefore impose more geometrical control on @xmath10 than just bounding its length . for this we consider an additional geometric quantity , the thickness of @xmath10 . the _ thickness _ @xmath13 of a @xmath9-curve @xmath10 in @xmath0 is the supremum of @xmath2 such that a radius @xmath2 normal tubular neighborhood of the curve is embedded . this notion can be used to define a knot invariant . the supremum of the ratio @xmath14 over all @xmath9-curves @xmath10 having a given knot type defines a knot invariant called the _ thickness _ of a knot , studied in buck and simon @xcite and litherland et al . @xcite , where it is related to various knot energies . for a @xmath9-closed curve @xmath10 the quantity @xmath15 is called the _ ropelength _ of @xmath10 , and the infimum of the ropelength over all @xmath9-representatives of a knot type , called the minimal ropelength , is the inverse of the thickness . various properties of ropelength , including the existence of a minimizer in the class @xmath16 , have been investigated recently @xcite . our first two results concern area bounds in terms of the length @xmath1 and thickness @xmath2 of an unknotted closed curve @xmath10 . in [ upper.bounds2 ] we show that for an unknotted curve there is an upper bound for the area of a spanning disk depending on these quantities . + [ upperbound0 ] for any embedded closed unknotted @xmath9-curve @xmath10 in @xmath0 having length @xmath1 and thickness @xmath2 , there exists a smooth embedded disk of area @xmath6 having @xmath10 as boundary with @xmath17 where @xmath18 is a constant independent of @xmath10 , @xmath1 and @xmath2 . in the other direction , we show that an upper bound of this type must be at least exponential in @xmath19 , and that there is no upper bound on the area depending solely on @xmath1 . + [ noupperbound1 ] there is a constant @xmath20 and a sequence of unknotted , @xmath21-curves @xmath22 embedded in @xmath0 , each having length @xmath23 and thickness @xmath24 , with the property that for large enough @xmath25 , any embedded disk spanning @xmath26 has area @xmath27 we include the constant @xmath28 for comparision to theorem [ upperbound0 ] . these exponential lower bounds differ from the exponential upper bound in theorem [ noupperbound1 ] in having an exponent linear in @xmath29 rather than quadratic . such examples necessarily have @xmath30 for a constant @xmath31 as @xmath32 , by comparing the lower bound @xmath33 with the upper bound in theorem [ upperbound0 ] . we next show that a bound on the length together with an upper bound on the curvature of an unknotted curve is also not sufficient to give an upper bound for the area of some spanning disk . without loss of generality we may normalize by setting the length of the curve to one . + [ noupperbound2 ] there is an infinite sequence @xmath34 of unknotted @xmath35-curves of length @xmath23 embedded in @xmath0 , each with ( pointwise ) curvature everywhere bounded above by a constant @xmath36 , independent of @xmath25 , such that any embedded disk spanning @xmath37 has area @xmath38 the proofs of theorems [ noupperbound1 ] and [ noupperbound2 ] are both given in [ lower.bounds ] . they are proved via a detour into piecewise - linear topology , and an investigation of similar questions for piecewise - linear curves , i.e. embedded polygons in @xmath0 . for piecewise linear curves , one can consider combinatorial analogues of isoperimetric inequalities , in which the `` length '' is replaced by the number of edges @xmath25 in a polygon , and the `` area '' is the number of triangles in an embedded triangulated surface having the polygon as boundary , as in @xcite . we establish the following pl analogue of theorem [ upperbound0 ] above . + [ plupperbound00 ] there is a constant @xmath39 such that given any polygonal unknotted curve @xmath40 in @xmath0 containing @xmath25 line segments , there is a triangulated pl disk embedded in @xmath0 having @xmath40 as boundary that contains at most @xmath41 triangles . this result is established as part of theorem [ plupperbound ] in 2 . to deduce theorem [ upperbound0 ] from it , we show that we can obtain such a triangulated surface in which the area of each triangle is bounded above by a constant . hass , snoeyink and thurston @xcite established the following pl analogue of theorems [ noupperbound1 ] and [ noupperbound2 ] . + [ plowerbound00 ] there is a constant @xmath42 and an infinite sequence of unknotted polygonal curves @xmath43 having between @xmath25 and @xmath44 segments , such that any triangulated pl disk embedded in @xmath0 having @xmath45 as boundary contains at least @xmath46 triangles . this result is not quite sufficient to deduce theorem [ noupperbound1 ] . it is necessary to strengthen the construction in @xcite , and we carry this out in theorem [ properties ] in 4 . using this strengthened construction , in 5 we show that a large number of triangles occurring in the triangulated pl disk have their areas bounded below by a fixed positive constant , and that such the resulting lower bound carries over to smooth embedded disks with the same or nearby boundaries . in the direction of pl combinatorial bounds , hass and lagarias @xcite showed that there is a ( qualitative ) combinatorial analogue of the isoperimetric inequality ( [ isop ] ) above in @xmath0 . the plan of the paper is as follows . in [ upper.bounds ] we derive the pl upper bound theorem [ plupperbound00 ] . from it we obtain upper bounds on the area of a piecewise - smooth embedded disk spanning a polygonal unknot @xmath40 with @xmath25 edges . in [ upper.bounds2 ] we deduce theorem [ upperbound0 ] from these results . in [ pl ] we review some results of @xcite , and construct a family of unknotted polygonal curves @xmath43 that do not have pl spanning disks having few triangles , and have additional useful properties . the families of curves @xmath47 and \{@xmath48 used to prove theorems [ noupperbound1 ] and [ noupperbound2 ] are constructed by a smoothing process from these curves . in [ lower.bounds ] we relate the pl complexity to area , and obtain lower bounds for the area of any spanning disk of @xmath37 , resp . @xmath26 , that increase to infinity with @xmath25 , and deduce theorems [ noupperbound1 ] and [ noupperbound2 ] . we note that the proofs of the lower bound estimates theorems [ noupperbound1 ] and [ noupperbound2 ] in [ pl ] and [ lower.bounds ] can be read independently of the proofs of the upper bound results theorem [ upperbound0 ] and [ plupperbound00 ] in [ upper.bounds ] and [ upper.bounds2 ] . [ upper.bounds ] let @xmath40 be a polygonal , unknotted curve of length @xmath1 in @xmath0 having @xmath25 edges . in this section we deduce upper bounds on the number of triangles in some embedded pl spanning disk for @xmath40 , and for the minimal area of such a disk . [ plupperbound ] there exists a constant @xmath49 such that whenever @xmath40 is a closed , unknotted , polygonal curve in @xmath0 having @xmath25 edges and length @xmath1 , the following hold : \(1 ) there exists an embedded pl disk spanning @xmath40 whose number of triangles is at most @xmath50 . \(2 ) such an embedded pl disk can be chosen to lie inside a ball of radius @xmath51 , and its area then satisfies @xmath52 theorem [ plupperbound ] is proved by a modification of ideas used in hass and lagarias @xcite . that paper obtains upper bounds on the number of elementary moves required to move from an unknotted polygonal curve in the one - skeleton of a triangulated 3-manifold having @xmath53 tetrahedra , to a single triangle , by a sequence of moves which deform the curve across three surfaces , an annulus , a torus and a disk . here we need to produce a spanning disk . the construction of @xcite can be modified to produce such a disk , with the bound on the number of triangles in the disk having the same functional form @xmath54 as the bounds obtained in ( * ? ? ? * theorem 1.2 ) , but with a different constant @xmath55 . a second necessary change is that we must construct a triangulation of a 3-ball containing the polygon @xmath40 in its one - skeleton , and we do not have the freedom to move this curve ; our construction obtains a quadratic upper bound for the number of tetrahedra in the constructed 3-ball triangulation . [ unknot.plmanifold ] there exists a constant @xmath56 such that whenever @xmath57 is a triangulated pl @xmath58-manifold with boundary , containing @xmath53 tetrahedra , and @xmath59 is a closed polygonal curve contained in the @xmath60-skeleton of the interior of @xmath57 that is unknotted in @xmath57 , then there is a pl triangulated disk in @xmath57 that has a subdivision of @xmath59 as boundary , and that contains at most @xmath61 triangles . * proof : * theorem 1.2 of @xcite states that there is a constant @xmath62 , independent of @xmath57 , such that if @xmath59 is an unknot embedded in the 1-skeleton of interior@xmath63 , then @xmath59 can be isotoped to a single triangle using at most @xmath64 elementary moves . the proof of this result deforms the original knot @xmath59 to a triangle with three sequences of elementary moves , made inside three triangulated surfaces . a peripheral torus is constructed by taking two barycentric subdivisions and taking the torus that is the boundary of a regular neighborhood @xmath65 of @xmath59 . an embedded disk is used to construct elementary moves taking an essential curve @xmath66 on @xmath67 to a single triangle in @xmath57 . on @xmath67 there is a curve @xmath68 with prescribed number of segments that is a longitude , and is isotopic to @xmath59 in @xmath65 by an isotopy with an explicitly computed bound on the number of elementary moves . an embedded annulus is used to go from @xmath59 to the longitude @xmath68 . the 2-torus @xmath67 is then used to define a sequence of elementary moves from @xmath68 to the boundary of a normal disk @xmath66 . these moves define a homotopy between two embedded curves on @xmath67 . the disk and the annulus used to construct the elementary moves are each embedded in @xmath57 . however the elementary moves that connect @xmath68 to @xmath66 in @xmath67 may go back and forth over the torus @xmath67 many times , and the annulus that is swept out on @xmath67 may not be embedded . this problem is rectified by a modification of the construction given in @xcite that perturbs this swept out annulus to an embedding . in theorem 1.2 of @xcite it is shown that : \(1 ) a normal disk with a longitude @xmath66 as boundary contains at most @xmath69 triangles . \(2 ) there is an embedded @xmath70 annulus @xmath71 in the solid torus @xmath65 whose two boundary components are another longitude @xmath68 and @xmath59 , and which consists of at most @xmath72 triangles . \(3 ) the number of elementary moves needed to deform @xmath68 to @xmath66 on the torus @xmath67 is at most @xmath73 . let @xmath25 denote the number of elementary moves taking @xmath68 to @xmath66 , so that @xmath74 and let @xmath75 denote the pl surface @xmath67 . we retriangulate @xmath57 so as to insert a product @xmath76 $ ] in place of the triangulated torus @xmath75 , with @xmath77 and @xmath78 triangulated identically to @xmath75 . the interior of @xmath76 $ ] is triangulated as follows : the torus @xmath79 is triangulated identically to @xmath77 , @xmath80 . we triangulate the product @xmath81 $ ] , @xmath82 in two steps . begin by dividing this region into triangular prisms , one for each triangle on @xmath79 . then divide each prism into 14 tetrahedra , by first dividing each rectangular face of the prism into four triangles with a common vertex at the center of the rectangle , and then coning the 14 triangles on the boundary of the prism to a new vertex at the center of the prism . we now construct an embedded annulus in @xmath76 $ ] whose boundary on @xmath77 is @xmath66 and whose boundary on @xmath78 is @xmath68 . the intersection of the annulus with @xmath83 consists of the image of @xmath68 in @xmath75 after @xmath84 elementary moves @xmath85 . on @xmath79 the annulus consists of the image of @xmath68 in @xmath75 after @xmath84 elementary moves plus the single triangle swept out by the @xmath86th elementary move if the @xmath86th move is of type 2 . if the @xmath86th move is of type 1 then it consists of image of @xmath68 in @xmath75 after @xmath84 elementary moves plus or minus a single vertex corresponding to the @xmath86th move . the image of @xmath68 on @xmath79 after @xmath86 elementary moves for any @xmath86 consists of less than @xmath87 edges . each vertical annular piece lying in @xmath83 can be triangulated with four triangles for each vertical rectangle above an edge in @xmath79 , and each horizontal piece lying in a torus @xmath79 contains at most one triangle . there are at most @xmath73 such annuli in @xmath76 $ ] , one for each elementary move . so the total number of triangles required to triangulate the annulus in @xmath88 $ ] is at most @xmath89 . we obtain a total of at most @xmath90 triangles in the annulus connecting @xmath66 on @xmath77 to @xmath68 on @xmath78 . adding to this the contributions of the disk and annulus in ( 1 ) and ( 2 ) above results in a spanning disk containing less than @xmath91 triangles . we take @xmath92 . @xmath93 to apply this result , we construct a triangulated convex polyhedron @xmath94 in @xmath0 which contains the polygonal curve @xmath40 in its one - skeleton ; @xmath94 will be the pl @xmath58-manifold in the result above . [ triangulation ] let @xmath40 be an unknotted polygonal curve in @xmath0 having @xmath25 vertices and length @xmath1 . then there exists a triangulated convex polyhedron @xmath94 ( with interior vertices ) in @xmath0 , which contains @xmath40 strictly in its interior , in its @xmath60-skeleton , and which has the following properties . \(1 ) the manifold @xmath94 is contained in a sphere of radius @xmath51 in @xmath0 . \(2 ) the number of tetrahedra in @xmath94 is less than @xmath95 . * proof : * the proof is similar in spirit to the construction of a triangulated @xmath58-manifold given in ( * ? ? ? * theorem 8.1 ) . however the argument in @xcite must be modified , because here the polygonal curve @xmath40 is fixed , and its vertices must be in the triangulation , whereas in @xcite only the projection is specified , and there is freedom to move around the vertices to simplify the triangulation . after rescaling @xmath40 by the homothety @xmath96 , we may assume its length is @xmath60 . in this case @xmath40 lies in the interior of a ball of radius @xmath97 in @xmath0 , which can without loss of generality be assumed to be centered at the origin . by rotating the coordinate system if necessary , we can suppose that the projection of @xmath40 in the @xmath98-direction onto the @xmath99-plane is a regular projection , and then , as in lemma 7.1 of @xcite , this projection has less than @xmath100 crossings . the graph associated to the projection has straight - line edges , and its vertices consist of the projections of vertices of the polygon @xmath10 together with all crossing points . as a preliminary step , we construct an augmented version of this projection graph in the @xmath101 plane , which will be a triangulated planar graph that we call @xmath102 . let @xmath103 denote the convex hull of the vertices of the projection graph , and add extra edges as necessary to fill out the boundary of the convex hull . note that all crossing vertices in the projection lie strictly in the interior of the convex hull . add new vertices to the graph by adding on each edge of this graph two new interior `` special '' vertices , spaced one - third of the way along the edge from its two endpoints . the resulting graph has at most @xmath104 vertices . this graph is planar with convex faces , and we triangulate it as follows . we first enclose each crossing vertex in four triangles , by adding the four edges connecting pairs of interior `` special '' vertices nearest to it . we then triangulate all remaining ( convex ) polygonal faces of the graph in an arbitrary fashion , adding no new vertices . as a final step we add a border to this graph consisting of three points arranged in an equilateral triangle in the @xmath105 plane at distance @xmath58 from the origin , together with the three edges of this triangle . note that each edge gets no closer to the origin than @xmath106 , and so remains outside the convex hull @xmath103 of the projection graph , which is contained in the ball of radius @xmath97 around the origin . finally we triangulate the outer region thus added by connecting each vertex on the boundary of @xmath103 to appropriate vertices of the equilateral triangle . the resulting triangulated graph @xmath102 has at most @xmath107 vertices , and so there are exactly @xmath108 triangles in its triangulation . its convex hull is contained in the circle of radius @xmath58 centered at the origin . to begin construction of the triangulated polyhedron , we start with two vertically translated copies of this augmented graph , in the planes @xmath109 and @xmath110 . the convex hull of this set is a triangular prism , which will be the polyhedron @xmath94 . we proceed to triangulate @xmath94 , i.e. to dissect it into tetrahedra sharing common faces , such that @xmath10 is in the one - skeleton . we will use the triangulation of the upper and lower faces on @xmath111 and @xmath112 arising from the triangulation of the augmented graph . we proceed to construct a simplicial complex @xmath113 inside @xmath94 which will include @xmath40 in its 1-skeleton , and which vertically projects to @xmath102 on the plane @xmath101 ; we will include @xmath113 in the triangulation and not include @xmath102 . the simplicial complex @xmath113 is mostly two - dimensional , consisting of triangles lying over each triangle of @xmath102 not including a crossing vertex , but thickened to a three - dimensional tetrahedron over each triangle of @xmath102 that does include a crossing vertex . to begin , we add the polygon @xmath40 , and add as new vertices along each edge of @xmath40 the points that vertically project to crossing vertices of the graph ; there are then two vertices on @xmath40 that project to each crossing vertex of the graph . to these upper and lower crossing vertices of @xmath40 add a vertical edge connecting them , which is called the vertical edge corresponding to the crossing . then add extra vertices on this configuration , two along each edge , that project to the `` special '' vertices of the graph @xmath102 . we add the three equilateral vertices of @xmath102 lying in the plane @xmath101 , plus the edges connecting them to the vertices in @xmath113 that project to the corresponding vertices in @xmath102 that they connect to . each vertex in @xmath102 that is not a crossing vertex has a unique `` lifted '' vertex in @xmath113 that vertically projects to it , while crossing vertices have two `` lifted '' vertices , an upper and a lower one . each triangle in the triangulation @xmath102 not containing a crossing vertex then lifts uniquely to a triangle in @xmath113 having the corresponding `` lifted '' vertices . corresponding to each crossing vertex of @xmath102 we create four tetrahedra in the simplicial complex @xmath113 , which share the vertical edge corresponding to the crossing , and completely surround it . each of the four tetrahedra has as vertices the upper and lower crossing vertex of @xmath40 plus two `` special '' vertices whose projection gives a triangle in the triangulation @xmath102 . we have now specified the simplicial complex @xmath113 . it has the property that each top - dimensional simplex in it projects vertically to a unique triangle in @xmath102 and each such triangle is covered once . furthermore @xmath113 has an `` upper surface '' and a `` lower surface '' obtained by extracting from each tetrahedron its triangular upper and lower face having the same vertical projection . we now add vertical edges connecting each vertex in the `` upper surface '' of @xmath113 to the corresponding vertex of the graph @xmath102 in the plane @xmath114 . we can match triangular faces in this copy of @xmath102 with the triangular face in the `` upper surface '' and their convex hull is a convex triangular prism , with boundary edges the vertical edges we just added . each such triangular prism may be subdivided into fourteen tetrahedra , by first subdividing each of its three quadrilateral vertical faces into four triangles by adding a vertex at its barycenter , and then coning all the triangular faces to the barycenter of the triangular prism . triangulations of adjacent prisms are automatically compatible ( adjoining triangular faces match ) ; this triangulates the `` upper '' portion of @xmath115 the `` lower '' portion is cut up into compatible tetrahedra in an exactly similar fashion , adding vertical edges . in this triangulation of @xmath94 , we have @xmath116 tetrahedra corresponding to each triangle in @xmath102 that does not include any crossing vertex , and @xmath117 tetrahedra for those faces that do . thus this triangulation uses at most @xmath118 tetrahedra . we have obtained a linear triangulation of a convex region @xmath94 contained in the ball of radius @xmath119 such that @xmath40 lies in the one - skeleton of @xmath94 , strictly in the interior of @xmath94 . upon rescaling by the homothety @xmath120 we obtain the desired triangulation , with @xmath94 inside a sphere of radius @xmath51 . @xmath93 + * remark . * we do not know whether the bound @xmath121 in the triangulation above is the optimal order of magnitude . avis and elgindy @xcite give examples of sets of @xmath25 points in @xmath0 , for arbitrarily large @xmath25 , such that any triangulation of the convex hull including these points as vertices required at least @xmath122 tetrahedra . however , they also showed that for @xmath25 points in general position ( no four in a plane ) there exists a triangulation of the convex hull using @xmath123 tetrahedra . we do not know whether such an improvement to @xmath123 is possible in the case above for general position vertices , because we impose the stronger requirement that @xmath40 be in the @xmath60-skeleton of the triangulation . see @xcite for a possible approach . + * proof of theorem [ plupperbound ] : * ( 1 ) let @xmath94 be the triangulated ball constructed in lemma [ triangulation ] , that has at most @xmath124 tetrahedra . by hypothesis @xmath40 is unknotted in @xmath0 , and it follows that it is unknotted in the ball @xmath94 , since an unknotting disk can be moved into @xmath94 by an isotopy . by theorem [ unknot.plmanifold ] there is an embedded pl spanning disk in @xmath94 having @xmath40 as boundary and containing at most @xmath61 triangles . this gives an upper bound of @xmath50 triangles , with @xmath125 , on noting that @xmath126 . this yields ( 1 ) . \(2 ) the embedded pl spanning disk constructed in lemma [ triangulation ] is contained in a ball @xmath94 of radius @xmath51 . each triangle in it therefore has area at most @xmath127 . thus the total area is bounded by @xmath128 which yields ( 2 ) . @xmath93 + * remark : * we have spanned the polygonal curve @xmath40 with a pl disk . such disks can be approximated on their interior by smooth disks of no greater area , so theorem [ plupperbound ] implies the same upper bound for the area of a smooth spanning disk for @xmath40 . [ upper.bounds2 ] in this section we consider closed unknotted @xmath9-curves @xmath10 embedded in @xmath0 . we obtain upper bounds for the minimal area of an embedded smooth disk having the curve as boundary as a function of the length @xmath1 and the thickness @xmath2 of the curve . in [ lower.bounds ] we will show that there is no upper bound for the minimal area of an embedded disk having the curve as boundary that is purely a function of the length and the maximal curvature of such a curve . let @xmath10 be a rectifiable closed curve in @xmath0 of length @xmath1 , and let @xmath129 be a parameterization of the curve by arc length , starting from a fixed point @xmath130 on the curve . recall that an _ @xmath2-neighborhood _ @xmath131 for a curve @xmath10 consists of all points in @xmath0 within euclidean distance @xmath2 of some point on the curve . an @xmath2-neighborhood @xmath132 is _ tubular _ if it is foliated by disjoint two - dimensional closed euclidean disks @xmath133 of radius @xmath2 isometrically embedded in @xmath0 , whose center is the point @xmath134 for @xmath135 , and each disk is normal to the curve @xmath10 at the point @xmath136 . if so , we call the disks @xmath133 _ fibers _ of the @xmath2-tubular neighborhood . every closed @xmath9-curve embedded in @xmath0 has an @xmath2-tubular neighborhood for some positive @xmath2 , see ( * ? ? ? * theorem 5.1 , pp . 109 - 110 ) . [ unknot.smoothupper ] suppose that @xmath10 is an unknotted @xmath9-curve in @xmath0 of length @xmath1 that has an @xmath2-tubular neighborhood . then there is an embedded smooth disk having boundary @xmath10 whose area @xmath6 satisfies @xmath137 with @xmath138 is a constant independent of @xmath10 . the idea of the proof is to approximate @xmath10 by an inscribed polygonal closed curve @xmath139 which remains entirely inside a thinner normal tubular neighborhood and has @xmath140 segments . by theorem [ plupperbound ] there exists a pl spanning disk for @xmath139 having area at most @xmath141 . this is at most @xmath142 , for a different constant @xmath143 . we deform this pl spanning disk to a spanning disk for @xmath10 using a homeomorphism @xmath144 of @xmath0 that moves @xmath139 to @xmath10 , is the identity outside the @xmath2-tubular neighborhood of @xmath10 , and is a piecewise differentiable lipschitz map with lipschitz constant everywhere bounded by 5 . it follows that the area of the resulting spanning disk for @xmath10 is larger by at most a factor of @xmath145 over that of the pl spanning disk for @xmath139 . the lipschitz map we construct is the identity outside a neighborhood of @xmath10 ; this neighborhood is a sort of necklace of beads around @xmath10 , in which each bead is a right circular cone , with cone point on the curve @xmath10 , rather then a round ball . we first establish some preliminary results concerning the deformations of a curve to an inscribed polygon . [ cone ] let @xmath146 \to { { \mathbb{r}}}^3 $ ] be a smooth , properly embedded unit speed @xmath9-curve with curvature bounded above by @xmath5 , with @xmath147 . the tangent vectors @xmath148 of @xmath149 lie inside a right circular cone with cone point at @xmath150 , axis on the line in direction @xmath151 , and with cone angle @xmath152 subtended to its axis . the entire curve @xmath153 lies within this closed cone . * proof : * let @xmath148 denote the unit tangent at @xmath154 and @xmath155 the magnitude of the curvature vector . then @xmath156 , and @xmath157 it follows from the geometry of the isosceles triangle formed by @xmath158 and @xmath159 that all tangent vectors to @xmath160 lie in a cone of angle @xmath152 centered around the line through @xmath161 . in particular , @xmath162 over this whole range , since @xmath147 . the curve @xmath149 itself also lies inside the cone with axis in the direction @xmath161 and based at @xmath150 . for if the curve ever left this cone then it would also leave a cone with a slightly larger cone angle , and would exit the boundary of this slightly wider cone transversely . the tangent vector of the @xmath149 at such a point would not lie within the original cone . we next compare the distance of two points on a curve with specified tubular neighborhood as measured along the curve and as measured in @xmath0 . [ relative distance ] let @xmath163 and let @xmath164 \to { { \mathbb{r}}}^3 $ ] be a smooth , properly embedded unit speed @xmath9-curve with curvature everywhere bounded by @xmath5 . let @xmath165 denote the line segment in @xmath0 connecting @xmath150 and @xmath166 . then @xmath149 is contained in the right circular cone with cone point @xmath150 , axis @xmath165 , and that subtends the angle @xmath167 to its axis . moreover @xmath168 * proof : * lemma [ cone ] shows that two tangent vectors on @xmath153 differ by an angle at most @xmath167 , so that their inner product satisfies @xmath169 , using @xmath170 in the last inequality . let @xmath75 denote the unit tangent vector pointing from @xmath150 to @xmath166 , so that @xmath171 then for any @xmath172 , the inner product @xmath173 it follows that the entire curve lies in the right circular cone with vertex at @xmath150 , axis the line segment @xmath165 pointing in direction @xmath75 , subtending angle @xmath174 , and with its endpoint @xmath166 being the center of the base of the cone . now since @xmath175 we obtain @xmath176 this yields @xmath177 as asserted . @xmath93 in what follows we let @xmath10 denote a @xmath9-curve having a normal tubular neighborhood of radius @xmath2 . the existence of the tubular neighborhood for @xmath10 implies that the absolute value of the curvature at each point of @xmath10 is at most @xmath5 . ( the points where the normal exponential map develops singularities are called the focal set . they occur at distance @xmath2 for a curve at a point of ( absolute ) curvature @xmath5 , see @xcite . ) therefore lemmas [ cone ] and [ relative distance ] are applicable . we construct a particular polygonal curve @xmath139 approximating the @xmath9-curve @xmath10 as follows . let @xmath139 in @xmath0 have vertices on the curve @xmath10 taking @xmath178 equally spaced points @xmath179 along @xmath10 ( measured by arclength of @xmath10 ) and inscribing line segments connecting successive points . let @xmath180 be the arc of @xmath10 between @xmath181 and @xmath182 and let @xmath183 $ ] be the line segment of @xmath139 between these points in @xmath0 , with last segment @xmath184.$ ] all arcs @xmath180 are of equal length @xmath185 and we define the quantity @xmath186 by @xmath187 , which implies that @xmath188 we will verify below that @xmath139 is embedded in @xmath0 . let @xmath189 denote the right circular cone with cone point @xmath181 , axis the line segment @xmath190 , subtending the angle @xmath191 , note that @xmath189 has base diameter no larger than @xmath192 since @xmath193 , and the center of its base is @xmath182 . [ cones.disjoint ] let @xmath10 be a closed @xmath9-curve with tubular neighborhood of radius @xmath2 . in the construction above , the arc @xmath180 lies inside the closed cone @xmath189 . distinct cones @xmath189 and @xmath194 with @xmath195 are disjoint unless @xmath196 or @xmath197 , and in this case the two cones intersect in the single point @xmath182 , resp . @xmath105 if @xmath198 . the polygonal curve @xmath139 is embedded in @xmath0 . * proof : * lemma [ relative distance ] applies to the arc @xmath180 to show that it is contained inside the cone @xmath189 . the cone @xmath189 subtends an angle to its axis of @xmath199 and the base of this cone has diameter at most @xmath200 if two cones @xmath189 and @xmath194 intersect , then they necessarily contain points @xmath201 and @xmath202 separated by distance at most @xmath203 . let @xmath165 be the line segment between @xmath204 and @xmath205 . we claim that since @xmath10 has an @xmath2-tubular neighborhood , the shorter of the two arcs of @xmath10 connecting @xmath204 and @xmath205 has length at most @xmath206 . to see this , consider the arc of @xmath10 starting from @xmath207 to a variable point @xmath208 where @xmath209 is an arclength parametrization of @xmath10 that increases from @xmath210 as @xmath211 moves along the curve in one direction . the distance @xmath212 is initially increasing . let @xmath213 be the first maximum point reached . at that point the tangent vector @xmath214 is orthogonal to the line @xmath215 $ ] , and part of this line lies in the normal disk of radius @xmath2 around @xmath216 so we must have @xmath217 . similarly allowing @xmath209 to decrease from @xmath218 we find that @xmath219 increases monotonically to the first point @xmath220 where @xmath221 reaches a maximum , and this maximum is at least @xmath2 by the same reasoning . now the remaining part of the curve falls outside the normal tubular neighborhood of radius @xmath2 determined by the arc @xmath222 . this normal tubular neighborhood includes all points within distance @xmath223 of @xmath204 , for if @xmath224 is a point with @xmath225 then the closest point @xmath226 on the arc @xmath222 to @xmath224 has @xmath227 from it , and attains a local minimum for the distance function from @xmath224 . it is at distance at least @xmath228 from each of the endpoints @xmath229 , since otherwise @xmath230 a contradiction . the local minimality implies that @xmath231 is perpendicular to the tangent vector @xmath232 , so lies in the disk of radius @xmath2 at @xmath226 normal to @xmath232 . we conclude that all points on the curve @xmath10 within distance @xmath233 of @xmath204 must lie in that part of the curve where @xmath212 is monotone increasing as @xmath234 increases . the remark above shows that this holds at least for @xmath235 . since @xmath236 , the point @xmath205 must fall on this part of the curve . now we can apply lemma [ relative distance ] to observe that on taking @xmath237 the segment @xmath238 $ ] already satisfies @xmath239 we conclude that the shorter of the two arcs of @xmath10 connecting @xmath204 and @xmath205 has length at most @xmath206 , as asserted . since the length of each segment @xmath180 is at least @xmath240 , it follows that @xmath189 and @xmath194 are adjacent cones , i.e. @xmath241 or @xmath197 . adjacent cones however intersect in a single point , the center of the base of one and the cone point of the other . finally , since each segment @xmath190 of @xmath242 lies in a different cone , and overlaps are only possible at endpoints of the @xmath243 , the closed polygonal curve @xmath139 is embedded in @xmath0 . @xmath93 the next two lemmas will be used to construct a lipschitz homeomorphism inside the tubular neighborhood that carries @xmath139 to @xmath149 . we start with a two - dimensional map . given a two - dimensional euclidean disk @xmath244 of radius @xmath245 centered at @xmath246 and a point @xmath247 in the disk , with @xmath248 , let @xmath249 be the map that sends the origin to @xmath250 and that maps each line segment from the origin to a boundary point of @xmath251 linearly to the line segment from @xmath250 to the same boundary point . it is explicitly given by @xmath252 [ sigma.lipschitz ] the map @xmath253 is a lipschitz homeomorphism with lipschitz constant at most @xmath254 . its inverse map @xmath255 is lipschitz with lipschitz constant at most @xmath256 if @xmath257 and @xmath258 lie in the disk @xmath251 , then for any point @xmath259 @xmath260 * proof : * the map @xmath261 is clearly a homeomorphism , which leaves the boundary of @xmath251 fixed . it remains to estimate a lipschitz bound . for @xmath259 we have @xmath262 we calculate that @xmath263 similarly @xmath264 and below by @xmath265 . it follows that the inverse map @xmath255 has lipschitz constant bounded above by @xmath266 finally , we have latexmath:[\[||\sigma_{q_1,d}(w ) - \sigma_{q_2,d}(w)|| = @xmath93 + next we give a parametrized generalization of the map of lemma [ sigma.lipschitz ] which is a three - dimensional map . we define in @xmath268-space the right circular cone @xmath269 of height @xmath270 and base of radius @xmath271 , centered on the @xmath53-axis , and suppose that @xmath272 we consider a curve @xmath273 parametrized by the @xmath53-variable which is strictly monotone increasing in @xmath53 , with @xmath274 and @xmath275 . and we assume that the whole curve is contained in the right circular cone @xmath276 . concerning the smoothness of this curve we assume only that @xmath277 is lipschitz - continuous with lipschitz constant @xmath278 , i.e. that @xmath279 holds for @xmath280 now , associated to this curve @xmath149 , we define a homeomorphism @xmath281 of @xmath0 supported in the cone @xmath276 , which leaves its boundary fixed . it takes horizontal disks with fixed @xmath98-coordinate into themselves moving each of them by the map @xmath282 of lemma [ sigma.lipschitz ] , and moves the point @xmath283 on the axis of the cone to @xmath284 . this homeomorphism is defined inside the cone @xmath285 by @xmath286 and is the identity map on the boundary of the cone . [ tube.curve ] suppose @xmath287 . the map @xmath288 associated to the curve @xmath289 is a homeomorphism of @xmath276 , leaving its boundary fixed . this map is a lipschitz map with lipschitz constant @xmath290 satisfying @xmath291 * proof : * let @xmath292 be two points in @xmath276 . if they lie at the same height @xmath293 , then since @xmath294 lies inside @xmath295 , lemma [ sigma.lipschitz ] yields the lipschitz bound @xmath296 . now if @xmath297 we may arrange that @xmath298 , by exchanging @xmath299 and @xmath224 if necessary . set @xmath300 and our object is to show that @xmath301 we study the effect of radial projection using the cone vertex @xmath302 . given a point @xmath303 in the plane @xmath304 we let @xmath305 be its radial projection to the plane @xmath306 . then for all such points @xmath216 in the cone @xmath307 we have @xmath308 the upper bound being tight for @xmath216 on the edge of the cone . since @xmath309 and @xmath224 both lie in the cone in the same plane @xmath310 , we have @xmath311 to bound the term @xmath312 we observe that in terms of the function of lemma [ sigma.lipschitz ] applied in the plane @xmath306 of the cone @xmath276 we have @xmath313 while @xmath314 . then @xmath315 where the first line above follows using lemma [ sigma.lipschitz ] . substituting this in the previous inequality yields @xmath316 on the other hand we have @xmath317 and @xmath318 we average these two inequalities in a suitable ratio to obtain @xmath319 which gives the lipschitz bound for @xmath144 . @xmath93 + * proof of theorem [ unknot.smoothupper ] : * the curve @xmath10 has a normal tubular neighborhood of radius @xmath320 as described above we inscribe in @xmath10 a closed polygon @xmath139 with @xmath25 edges , with @xmath321 , labelling the points @xmath322 , the @xmath323-th edge @xmath324 $ ] of @xmath139 , and @xmath180 the corresponding arc of the curve between @xmath325 and @xmath326 , as described above . we also have the set of @xmath25 right circular cones @xmath327 , where @xmath189 has the segment @xmath190 as its axis . now lemma [ cones.disjoint ] applies to show that @xmath139 is embedded in @xmath0 and that these cones are disjoint except at the points @xmath181 . we define a homeomorphism @xmath328 , which is the identity outside all the cones @xmath189 and which inside each cone @xmath189 is given by the homeomorphism @xmath329 of @xmath189 given by lemma [ tube.curve ] , taking the arc @xmath289 to be the arc @xmath180 , reparameterized using the variable @xmath53 that measures distance along @xmath190 , rather than the arc length variable @xmath218 . with this reparameterization the map @xmath277 is a lipschitz map . to bound the lipschitz constant we use results from the proof of lemma [ cones.disjoint ] . it shows that the tangent vector @xmath75 in the direction of @xmath190 satisfies @xmath330 for all tangent vectors on the arc @xmath180 , where @xmath331 which yields @xmath332 . this implies that the lipschitz constant @xmath333 we conclude using lemma [ tube.curve ] that @xmath144 is a lipschitz map and obtain a bound for its lipschitz constant inside the individual cones @xmath189 . we take @xmath334 and @xmath335 which yields the bound @xmath336 since @xmath144 is the identity map outside the cones , this lipschitz bound carries over to @xmath328 , using the triangle inequality . now consider the triangulated pl surface @xmath337 having @xmath139 as boundary given by theorem [ plupperbound ] , having area @xmath338 bounded by @xmath339 where @xmath340 is the length of @xmath341 using the bound @xmath342 and that @xmath343 and @xmath344 this gives @xmath345 for an enlarged constant @xmath346 . we can assume , possibly after a small perturbation that induces an arbitrarily small increase in the area of @xmath337 , that @xmath337 intersects the boundary of the @xmath233-tubular neighborhood of @xmath10 and the boundary of each ball @xmath347 transversely . this follows from the fact that the space of smooth disks is dense in the space of piecewise smooth disks in @xmath0 , and the area functional is continuous on the space of piecewise smooth disks with the smooth topology . see for example kosinski ( * ? ? ? * theorem 2.5 ) . + applying the homeomorphism @xmath144 then gives a piecewise-@xmath9 embedded surface @xmath348 that has @xmath10 as boundary and whose area is increased from the area @xmath6 of @xmath337 by at most a factor of 25 , using the bound of 5 on the lipschitz constant of the map @xmath144 . ( a lipschitz map @xmath349 with lipschitz constant @xmath350 can stretch the volume of an @xmath351-dimensional submanifold by at most a factor @xmath352 , see federer@xcite , morgan ( * ? ? ? * prop . the fact that @xmath353 is piecewise @xmath9 follows using the fact that the map @xmath144 is piecewise @xmath9 , with its smoothness inherited from the smoothness of the curve @xmath136 . thus , enlarging the constant @xmath354 to a suitable @xmath138 we have @xmath355 finally , we can perturb the interior of @xmath353 to obtain a smooth surface while again changing its area by an arbitrarily small amount , and without moving the boundary @xmath10 , see ( * ? ? ? * theorem 2.5 ) . this yields the theorem , where we give up another factor of @xmath356 on the right side of the bound above to absorb the effect of the perturbation . in this section we construct a series of polygonal unknotted curves @xmath45 in @xmath0 whose minimal spanning disks must have exponential complexity , according to various measures . this is an extension of the results of @xcite . [ properties ] there exists a sequence of unknotted polygonal curves @xmath357 in @xmath0 , for @xmath358 , having the following properties . \(1 ) @xmath45 contains at most @xmath359 edges . \(2 ) any embedded , piecewise - smooth disk spanning @xmath45 intersects the @xmath360-axis in at least @xmath361 points . \(3 ) any embedded pl triangulated disk @xmath362 bounded by @xmath45 contains at least @xmath361 triangles . \(4 ) each @xmath45 has length one and there exists a constant @xmath363 independent of @xmath25 such that each @xmath45 is disjoint from the solid cylinder @xmath364 of radius @xmath365 around the @xmath360-axis . three curves in this sequence are shown in figure [ kn ] . , @xmath68 and @xmath366 . ] , @xmath68 and @xmath366 . ] , @xmath68 and @xmath366 . ] [ [ remark . ] ] remark . + + + + + + + the construction of the curves @xmath45 in theorem [ properties ] differs from that in @xcite , in order to make condition ( 4 ) hold . the main focus of @xcite was to construct examples of curves with @xmath123 edges satisfying property ( 3 ) , while in the application here it is properties ( 2 ) and ( 4 ) that are needed . can be embedded in @xmath0 using 12 segments so that the top and bottom ends of each arc are vertical segments on the @xmath367-plane . the leftmost strand is made from five segments , and rises above and below the @xmath367-plane . the other three strands lie completely in the @xmath367-plane . ] * proof : * the assertions ( 1)-(3 ) of the theorem are invariant under homothety @xmath368 , so restricting to curves @xmath45 of length one results in no loss of generality . the construction of @xmath45 begins with the 4-braid @xmath369 shown in figure [ alpha ] . we construct @xmath149 to consist of vertical segments near its endpoints and to have ( total ) length one . we stack @xmath25 copies of @xmath149 to get a braid @xmath370 , and then rescale by a homothety to get a braid @xmath371 whose length is @xmath372 . similarly we stack @xmath25 copies of @xmath373 and scale to get a braid @xmath374 of length @xmath372 . the curve @xmath45 is constructed by starting with a copy of each of @xmath375 and adding eight arcs to obtain a closed curve in @xmath0 . the eight arcs are shown in figure [ caps ] . used in forming the top , middle and bottom pieces of @xmath45 . as @xmath25 increases , the horizontal separation between the endpoints of the arcs decreases to 0 , but their total length remains equal to 1/3 . all arcs lie in the @xmath367-plane and are disjoint from a disk in this plane of radius @xmath365 centered at the origin . ] the added eight arcs all lie in the @xmath367-plane , as shown in figure [ caps ] . they consist of two arcs @xmath376 connecting the first and second and the third and fourth endpoints at the top of @xmath377 , four arcs @xmath378 connecting the bottom four endpoints of @xmath377 to the top four endpoints of @xmath374 , and two arcs @xmath379 connecting the first and fourth and the second and third endpoints of the bottom of @xmath374 . forty line segments are used to construct @xmath380 . the resulting closed curves @xmath45 , shown in figure [ kn ] , are each unknotted , since the braid @xmath381 is trivial , and each @xmath45 uses at most @xmath382 segments , with a contribution of @xmath383 coming from each of @xmath384 , @xmath374 . for each @xmath25 , we construct the eight added arcs so that they have total length @xmath372 , and so that they are disjoint from a solid cylinder of fixed radius @xmath365 . with the total length of all the segments set to one , a calculation shows that we can take @xmath365 to be @xmath385 . see figure [ cylinder ] . is disjoint from a cylinder of radius @xmath365 around the @xmath360-axis . ] as @xmath32 the separation between the top four points of @xmath384 converges to zero , and the endpoints of @xmath386 and @xmath387 which are identified to them also converge . endpoints of @xmath388 converge similarly . each curve @xmath45 is a length one curve disjoint from the cylinder @xmath364 of radius @xmath365 around the @xmath360-axis . now consider an arbitrary piecewise smooth disk @xmath362 with boundary @xmath45 . the intersections of @xmath362 with the level sets @xmath389 can be analyzed using morse theory and the theory of train tracks and surface diffeomorphisms . this analysis is carried out in theorem 1 of @xcite , where it is shown that the @xmath360-axis must intersect @xmath362 in at least @xmath361 points . all cases of theorem [ properties ] have now been established . [ lower.bounds ] in this section we show that there is no upper bound on the minimal area of any embedded disk that depends only on the length @xmath1 of the curve , as stated in theorem [ noupperbound2 ] . this is based on the sequence of curves @xmath45 constructed in theorem [ properties ] . [ perturb ] let @xmath43 be the sequence of length one curves constructed in theorem [ properties ] . there is a constant @xmath391 such that any embedded disk @xmath362 having boundary @xmath45 has area @xmath6 satisfying @xmath392 the same lower bound holds for the area of any embedded disk having boundary a curve @xmath393 that is isotopic to @xmath45 by an isotopy that fixes the cylinder @xmath364 of radius @xmath365 around the @xmath360-axis . * proof : * suppose a line @xmath1 in @xmath364 parallel to its axis intersects @xmath362 in less than @xmath361 points . there is an isotopy of @xmath0 fixing @xmath45 and carrying @xmath1 to the @xmath360-axis . the image of @xmath362 under this isotopy would be a disk @xmath394 spanning @xmath45 and intersecting the @xmath360-axis in less than @xmath361 points , contradicting theorem 1 of @xcite . so any such line in the cylinder @xmath364 intersects @xmath362 in at least @xmath361 points . the area of a perpendicular cross - section of @xmath364 is @xmath395 , so the projection of @xmath362 to the @xmath367-plane has area at least @xmath396 , as does @xmath362 itself . we choose @xmath397 now suppose that @xmath398 is a curve isotopic to @xmath45 in the complement of @xmath364 and @xmath399 is a disk spanning @xmath398 . then there is an isotopy of @xmath0 carrying @xmath399 to a disk spanning @xmath45 that restricts to the identity on @xmath364 . the area of @xmath399 in @xmath364 is at least @xmath400 by the previous argument . @xmath401 to obtain @xmath26 we will construct a smooth curve of length one that approximates @xmath45 , and in particular remains disjoint from @xmath364 and has thickness greater than @xmath402 , as shown in figure [ deltan ] . + * proof of theorem [ noupperbound1 ] : * it suffices to find curves @xmath26 having length at most one with the other claimed properties , as rescaling by a homothety to make these curves have length precisely one increases both the thickness and the area of any spanning disk . we will approximate each @xmath45 by a smooth curve @xmath403 , isotopic to @xmath45 in the complement of the cylinder @xmath364 . finally we will take @xmath404 + 1}$ ] for an appropriate constant @xmath405 . we first smooth the braids @xmath149 and @xmath373 by replacing neighborhoods of their vertices by smooth arcs , getting a smooth braid @xmath406 of thickness @xmath407 and curvature bounded above by a constant @xmath36 . we do this smoothing so that @xmath406 still consists of vertical arcs near its endpoints . then there is a constant @xmath408 with the property that the thickness of @xmath406 is greater than @xmath409 , and each vertical segment near an endpoint of @xmath406 has length greater than @xmath409 . form @xmath410 and @xmath411 as before by stacking copies of @xmath406 or @xmath412 . since copies of @xmath406 are identified along points that have neighborhoods coinciding with vertical segments , the resulting curve is @xmath35 . moreover the thickness of @xmath410 and @xmath411 is still greater than @xmath409 , since distinct copies of @xmath406 have @xmath409-tubular neighborhoods with disjoint interiors . scaling by a homothety gives @xmath413 , with length @xmath372 , and similarly @xmath414 . the thickness of each of @xmath415 and @xmath414 is greater than @xmath416 . we can also smooth the curves @xmath417 in neighborhoods of their internal vertices , to obtain nearby arcs @xmath418 that are smooth with arcs near the endpoints remaining as straight vertical segments , with the length of each straight segment no less than a constant @xmath419 ( which does not depend on @xmath25 ) . the approximating arcs are chosen sufficiently close to maintain their disjointness from @xmath364 , and with their total length remaining less than @xmath372 . then combining @xmath420 with @xmath418 gives a closed curve @xmath403 of length one that is isotopic to @xmath45 in the complement of @xmath364 , and the thickness of @xmath403 is given by @xmath421 , where @xmath422 is the smaller of @xmath423 . is a smooth curve that has length one and thickness at least @xmath424 . @xmath425 and @xmath426 are shown . ] is a smooth curve that has length one and thickness at least @xmath424 . @xmath425 and @xmath426 are shown . ] we now establish a lower bound on the area @xmath6 of any embedded spanning disk for @xmath403 . lemma [ perturb ] implies that there is a @xmath18 such that for large enough @xmath25 , any embedded disk with boundary @xmath403 has area @xmath6 greater than @xmath427 . note that @xmath428 since @xmath403 has length one and thickness less than @xmath422 . thus we can rescale @xmath25 to eliminate the constant @xmath422 appearing in the thickness bound by taking a subsequence @xmath429 + 1}$ ] , and setting @xmath430 we then have minimal area @xmath431 . @xmath390 + in the next result we show how to deform @xmath45 to construct a similar sequence of smooth curves @xmath37 having uniformly bounded curvature . recall that the curvature function for a @xmath9-curve embedded in @xmath0 is computed by taking an arclength parameterization @xmath432 \to { { \mathbb{r}}}^3 $ ] , so that @xmath433 , and the curvature function is then @xmath434 + * proof of theorem [ noupperbound2 ] : * the construction of the family @xmath435 is similar to that of @xmath436 . we again use smoothings @xmath418 of the arcs @xmath417 to curves that lie nearby them , and assert that the approximating curves can be chosen to have curvature at each point bounded above by a fixed constant @xmath59 . to see this we use the fact that a sequence of smooth curves converging to a smooth limit curve have curvatures converging pointwise to the curvature of the limit curve . for any curve with monotonically increasing @xmath98-component , a deformation of the curve by a diffeomorphism @xmath437 carries it to a curve that , for @xmath438 small , lies very close to to the @xmath98-axis . for @xmath438 sufficiently small the curvature of the image curve under @xmath439 is uniformly close to zero . we choose @xmath440 to be a sequence converging to 0 sufficiently fast so that the curvature of @xmath441 , where @xmath410 is as before , and similarly we define @xmath414 . we also smooth the curves @xmath417 in neighborhoods of their internal vertices , to obtain nearby arcs @xmath418 that are smooth with arcs near the endpoints remaining straight vertical segments . while these vertical segments all converge to the @xmath98-axis as @xmath442 , the curvature of each of @xmath418 remains uniformly bounded . one way to see this explicitly is to choose smooth limiting curves @xmath443 of bounded curvature that are not embedded , but rather have arcs near their endpoints that agree with the @xmath98-axis . any sequence of smooth curves @xmath418 that converge to these curves smoothly will have uniformly bounded curvature . then the curvature of the curves @xmath37 actually converges to 0 at points corresponding to the braids @xmath410 , @xmath414 , and is uniformly bounded at other points . the curvature is zero along the straight segments where @xmath418 join @xmath410 and @xmath414 . since the required area for a spanning disk goes to infinity as @xmath32 , we can rechoose @xmath37 to be a suitable subsequence to have the required property @xmath444 this concludes the proof of theorem [ noupperbound2 ] . [ classical ] this appendix establishes two versions of the isoperimetric theorem in @xmath0 as stated as ( 1 ) and ( 2 ) in 1 . result ( 1 ) follows from the solution of the plateau problem , due to douglas and rado , @xcite , @xcite , and on a result of carleman on isoperimetric inequalities for minimal disks @xcite . see also reid @xcite . we deduce ( 2 ) from ( 1 ) , using standard cut and paste methods of 3-manifold topology , which can be used to convert an immersed surface to an embedded surface , possibly with an increase in genus , but maintaining orientability . such a cut and paste can be achieved without increasing area by a `` rounding of creases '' . alternate approaches to proving ( 2 ) are possible based on work of blaschke @xcite , which was the original approach to ( 2 ) . the discussion in osserman @xcite indicates that blaschke s argument is heuristic , but can be made valid for area - minimizing immersed surfaces having the curve as boundary . + [ iso - immersed ] let @xmath10 be a simple closed @xmath9-curve in @xmath0 of length @xmath1 . then there exists an immersed disk having area @xmath6 in @xmath0 , with @xmath10 as boundary , such that @xmath446 if the curve is not a circle , then there exists such an immersed disk for which strict inequality holds . * proof of theorem [ iso - immersed ] : * span @xmath10 by a least area disk @xmath447 , whose existence is guaranteed for rectifiable @xmath10 by the solution of the plateau problem @xcite , @xcite . the regularity results of osserman @xcite and gulliver @xcite show that @xmath447 is a smooth immersion in its interior . ( the interior of @xmath447 may transversely intersect @xmath10 , as indeed it must if @xmath10 is knotted . ) the argument in reid @xcite ( see also @xcite ) shows that @xmath448 , with equality if and only if @xmath10 is a circle . @xmath93 + [ iso - embedded ] let @xmath10 be an embedded closed @xmath9-curve in @xmath0 of length @xmath1 . then there exists an embedded orientable smooth spanning surface for @xmath10 in @xmath0 having area @xmath6 , with @xmath449 if the curve is not a circle , then there exists such a surface for which strict inequality holds . * proof of theorem [ iso - embedded ] : * apply theorem [ iso - immersed ] to get an immersed disk @xmath447 spanning @xmath10 satisfying the isoperimetric inequality ( [ isop ] ) . if @xmath447 is not embedded then perturb it slightly so that its self - intersections are in general position , i.e. transverse with finitely many triple points . since smooth general position maps are dense among @xmath9 maps , this can be done with an arbitrarily small increase in area . the self - intersections then consist of a finite number of immersed double curves . there is a uniquely defined surface @xmath450 obtained by oriented cut - and - paste along these curves . an arbitrarily small perturbation of @xmath450 moves it to an embedded surface . the resulting surface is orientable with the same boundary as @xmath447 , but may have higher genus . if @xmath447 has area strictly less than @xmath451 then the area of @xmath450 can also be taken to be less than @xmath451 . if @xmath447 has area equal to @xmath451 then it spans a circle and is already embedded . in all other cases strict inequality applies for both @xmath447 and @xmath450 . @xmath93 + * remark : * almgren @xcite established isoperimetric inequalities for varifolds and currents in arbitrary codimension , with optimal constants . in some range of dimension and codimension , these give upper bounds for the area of embedded spanning surfaces and spanning submanifolds , but with unspecified topological type . j. hass and j. c. lagarias , the minimal number of triangles needed to span a polygon embedded in @xmath452 , pp . 509526 in : j. goodman - r . pollack festscrift volume ( b. aronov , j. pach , m. sharir , eds . ) , springer - verlag : new york 2003 . department of mathematics + university of california + davis , ca 95616 + [email protected] + at&t labs - research + florham park , nj 07932 - 0971 + [email protected] + department of mathematics + university of california + davis , ca 95616 + [email protected]

we present a hadronic model of activity for galactic -loud binaries , in which the multi - tev neutrino flux from the source can be much higher and/or harder than the detected tev flux . this is related to the fact that most neutrinos are produced in pp interactions close to the bright massive star , in a region optically thick for the tev s. considering the specific example of , we derive upper bounds for neutrino fluxes from various proton injection spectra compatible with the observed multi - wavelength spectrum . at this upper level of neutrino emission , we demonstrate that icecube will not only detect this source at c.l . after one year of operation , but , after 3 years of exposure , will also collect a sample marginally sufficient to constrain the spectral characteristics of the neutrino signal , directly related to the underlying source acceleration mechanisms .

0812.0306

the recently discovered `` -loud '' binaries ( grlb ) form a new sub - class of galactic binary star systems which emit gev - tev s @xcite . these are high mass x - ray binaries ( hmxrb ) composed of a compact object ( a black hole or a neutron star ) orbiting a massive star . the detection of s with energies up to 10 tev from these systems shows that certain hmxrbs host powerful particle accelerators producing electrons and/or protons with energies above 10 tev . at the moment it is not clear which physical process leads to the very high energy ( vhe ) particle acceleration and whether the vhe particle acceleration is a generic feature of the hmxrbs or the result of specific physical conditions in a restricted hmxrb sub - class . it is possible that particle acceleration is taking place in a large number of x - ray binaries , but the s can be detected only from the objects with preferred orientations w.r.t . the line of sight ( e.g. with jets pointing toward the observer ) @xcite . alternatively , it may be that the conventional accretion - powered binaries are not capable of accelerating particles and that the activity of a binary is powered by a different mechanism ( see e.g. @xcite ) . the theoretical models of activity of hmxrbs @xcite all assume emission from the interaction of a relativistic outflow from the compact object ( jet from a black hole , or wide angle wind from a pulsar ) with the wind and radiation emitted by the companion massive star . the basic properties of the outflows , such as the composition ( e@xmath1e@xmath2 pairs or electron - nuclei plasma ) , anisotropy ( a collimated jet or a wide angle outflow ) etc . are as yet poorly constrained by the data . in one of the three persistent ( periodic ) -loud binaries , psr b1259 - 63 , the relativistic outflow is known to be produced by a young pulsar . in principle , a similar mechanism could power the activity of other sources ( except for cyg x-1 ) , although direct proof of the presence of the young pulsar in these systems is not possible : the radio emission from the compact object is free - free absorbed . recent puzzling detection of a short soft flare from one of the -loud binaries , lsi + 61 303 @xcite , may indicate the presence of a neutron star in the system . the outflows from lsi + 61 303 and ls 5039 , extending to the distances @xmath3 cm ( far beyond the binary orbit ) are revealed by the radio observations @xcite . the nature of the outflows ( a collimated jet or a wide - angle outflow ) is still debated . with the exception of cyg x-1 , all the known grlbs have similar spectral energy distributions ( sed ) , peaking in the mev - gev energy band . the physical mechanism producing the mev - gev bump in the spectra is , however , not clear . the emission from grlbs is supposed to come from internal and/or external shocks formed in the relativistic outflow either as a result of the development of intrinsic instabilities or through interactions with the stellar wind of the massive star . the mev - gev emission can be the synchrotron emission from electrons with energies much above tev @xcite , produced locally at the shock , or else , be produced via inverse compton scattering of the uv thermal emission from the massive star by electrons of the energies @xmath4 mev @xcite . the available multi - wavelength data do not constrain the composition of the relativistic wind from the compact object . on the one hand , the multi - tev or 10 mev electrons , responsible for the production of the mev - gev bump in the sed , could be injected into the shock region from the e@xmath1e@xmath2 pairs loaded wind . on the other , these electrons could be secondary particles produced in e.g. proton - proton collisions , if the relativistic wind is proton - loaded . the only direct way to test if relativistic protons are present in the emission region would be the detection of multi - tev neutrinos . the possibility of detection of neutrinos from grlbs by a km@xmath5-class neutrino telescopes was considered in several references @xcite . if the sources are assumed to be transparent to the tev s , the estimation of the neutrino flux is straightworward , given a known source flux and spectrum . the assumption of source transparency was adopted e.g. in the estimates of the number of detected neutrinos as in ref . however , the tev flux from the grlbs can be significantly attenuated by the pair production on the uv photon field of the massive star @xcite . in addition , if the s and neutrinos are produced close to the compact object , the flux can be further suppressed by pair production on the soft photons emitted by the accretion flow @xcite . the derivation of the estimate of the neutrino flux and spectral characteristics based on the observed tev emission is inconclusive due to the uncertain attenuation of the flux in the tev band ( see @xcite for s specific discussion of ) . in the absence of a direct relation between the characteristics of the observed tev and the neutrino emission from a grlb , the only way to constrain possible neutrino signals from the source is via the detailed modelling of the broad - band spectrum of the source within the hadronic model of activity . the idea is that the pp interactions , which result in the production of neutrinos , also result in the production of e@xmath1e@xmath2 pairs and the subsequent release of their energy via synchrotron , inverse compton and bremsstrahlung emission . the electromagnetic emission from the secondary e@xmath1e@xmath2 pairs is readily detectable . the total power released in the pp interactions determines the overall luminosity of the emission from the secondary pairs . the known electromagnetic luminosity and broad - band spectral characteristics of the source can be used to constrain the power released in pp interactions as well as the spectrum of the primary high energy protons . in the following we develop the hadronic model of activity of grlbs and derive the constraints on the spectrum and overall luminosity of neutrino emission from the analysis of the broad - band spectral characteristics of grlbs . although the following discussion is generically applicable for the grlbs as a class , we concentrate on the particular example of the system , because it is the only known persistent grlb in the northern hemisphere , available for observations with the icecube neutrino telescope @xcite . the existing observational data are found to be consistent with a possible very strong neutrino emission from ( with a flux at the level of @xmath6 erg/@xmath7s ) , close to the best reported amanda upper limit , @xmath8 ) @xcite ( a more recent reference @xcite combining more data reported a slightly degraded upper limit for over approximately the same range of energy , while the sensitivity had been improved by roughly a factor two ; this however translates a non significant but large background fluctuation at the source location ) . moreover , the assumption of an almost arbitrarily hard neutrino spectrum ( e.g. a powerlaw @xmath9 with index @xmath10 ) is not ruled out by the available multi - wavelength observational data . this means that , in principle , the soure could be readily detectable with icecube . the plan of the paper is as follows : in section [ sec : model ] we describe the general features of the hadronic model of activity of grlbs . in particular , we stress that when the massive companion star is a be type star , the presence of a dense equatorial decretion disk around the massive star can boost the pp interaction rate . at the same time , the emission from the pp interactions in the disk would be strongly suppressed , because of the large density of the soft photon background in the direct vicinity of the star . in section [ sec : numeric ] we perform the detailed numerical modeling of the broad - band spectra of grlbs in the hadronic model and show , on the particular example of , that the model provides a suitable explanation of the typical shape of the grlb seds . satisfactory fits to the observed seds can be achieved under quite arbitrary assumptions about the shape of the initial high energy proton spectrum . in particular , the initial spectra as different as an @xmath11 power law with a high energy cut - off and a monochromatic proton spectrum peaked at @xmath12pev energy could both explain the observed x - ray - to- spectrum of the source . finally , in section [ sec : icecube ] we work out the predictions for the detection of neutrinos from with icecube . we show that in an optimistic scenario , when the anisotropy of the neutrino emission of the source does not result in a suppression of the neutrino flux in the direction of the earth , should be detectable within a year of exposure . nevertheless , the spectral characteristics of the emission can only be marginally delineated after three years of exposure , given the wash out of the original neutrino spectrum from the measurement of the muon spectrum . in the hadronic model , the primary source of the system s high energy activity are high energy protons . the presence of the high energy protons in relativistic outflows from compact objects ( stellar mass and supermassive black holes , neutron stars ) is usually difficult to detect , because of their very low energy loss rates . for example , in the case of relativistic winds ejected by young pulsars , it is possible that most of the wind power is carried by the high energy protons or ions , but the presence of protons / ions in the wind can be established only indirectly , because most of the radiation detected from a nebula powered by the pulsar wind is emitted by electrons accelerated at the shock whose properties are determined by the parameters of the proton / ion component of the pulsar wind @xcite . emission initiated by interactions of the pulsar wind protons with the ambient medium can be detected only if the density of the medium is high enough @xcite . grlbs provide a unique opportunity to `` trace '' the presence of protons / ions in the relativistic outflows generated by compact objects . the dense matter and radiation environment created by the companion massive star provides abundant target material for the protons in the relativistic outflow . interactions of high energy protons with the ambient matter and radiation fields , created by the presence of a bright massive companion star , lead to the production and subsequent decays of pions . this results in the emission of neutrinos and s from the source and to the deposition of e@xmath1e@xmath2 pairs throughout the proton interaction region . radiative cooling of the secondary e@xmath1e@xmath2 pairs leads to the broad - band synchrotron and inverse compton emission from the source . if the primary high energy protons are produced close to the compact object , the maximal energies of particles accelerated in the region of the size @xmath13 in magnetic field @xmath14 , can be estimated as @xmath15\left[\frac{r_{\rm c}}{10^{6}\mbox { cm}}\right]\mbox { ev}\ ] ] the parameter @xmath16 characterizes the acceleration process efficiency . for example , in the case when the compact object is a rotation - powered neutron star , @xmath17 g , while @xmath18^{-2}$ ] , with @xmath19 being respectively the neutron star radius , angular velocity and rotation period . depending on the particular acceleration mechanism , the proton energy spectrum below this maximal energy can range from the conventional power law shape with spectral index close to @xmath20 ( acceleration in non - relativistic shocks ) , to almost monochromatic spectra ( if the acceleration proceeds in large scale electric fields in the magnetic reconnection regions of accretion flow , or in the vacuum gaps in black hole or pulsar magnetosphere ) . otherwise , protons can be accelerated via shock acceleration in an extended region of the size @xmath21 with magnetic field @xmath22 in a jet or in a shocked relativistic wind emitted by the compact object . in this case one expects a conventional @xmath11 type spectrum with a cut - off at the energy at which the larmor radius of the high energy particles becomes comparable to the size of the system @xmath23\left[r_{\rm ext}/10^{12}\mbox { cm}\right]\mbox { ev}$ ] . efficient interaction of high energy protons with the radiation field produced by the bright massive star in the system ( e.g. a be star with temperature @xmath24 k in the case of and psr b1259 - 63 ) occurs above @xmath25m_p\simeq 2\times 10^{16}\left[\epsilon_*/10\mbox { ev}\right]$ ] ev , where @xmath26 is the typical photon energy of the stellar radiation . in contrast , high energy protons efficiently interact with the protons from the dense stellar wind already at much lower energies . in the case of massive stars of type be , the pp interaction rate can be highly enhanced if the high energy protons are able to penetrate the dense equatorial disk known to surround them . an obstacle for the penetration of the high energy protons accelerated e.g. close to the compact object into the disk could be the presence of magnetic fields , which would deviate the proton trajectories away from the disk . however , the larmor radius of the highest energy protons , @xmath27\left[b/1\mbox { g}\right]^{-1}$ ] cm , could be comparable to the size of the system . thus , if the magnetic field in the region of contact between the stellar wind and the relativistic outflow is not larger than several gauss , the highest energy protons can freely penetrate the dense stellar wind region . note that the same effect may also result in the production of very hard injection spectra of the high energy protons in the pp interaction region ( penetration of the lower energy protons into the dense stellar wind suppressed by the magnetic fields ) . the estimation of the pp interaction rate in the disk is straighforward , provided a known disk density profile . for example , in the case of the be star in the system , the density profile of the equatorial disc has been measured , using the ir free - free and free - bound radiation ( see e.g. refs . @xcite ) . assuming a disc half - opening angle of @xmath28 and a stellar radius of @xmath29 , the disk density profile can be modelled as a power law of the distance @xmath30 @xmath31 where @xmath32 @xmath33 and @xmath34 . given @xmath35 @xmath7 , the pp interaction cross - section at @xmath36 ev , one finds the characteristic time of pp interactions in the disk @xmath37^\gamma\mbox { s}\ ] ] to estimate the efficiency of the pp interactions , the pp interaction time should be compared to the escape time from the system . since the details of the escape regime are not known , one can only compare ( [ tpp ] ) with possible relevant time scales . the shortest possible time scale is set up by the light - crossing time of the system , @xmath38\mbox { s } \label{t_lc}\ ] ] this time scale is relevant e.g. if the larmor radius of the high energy protons is comparable to the size of the disk , so that these protons can not be trapped in the disk interior . if the high energy protons escape on this shortest time scale , the pp interactions can only be efficient in the direct vicinity of the be star , at @xmath39 cm . the escape of the high energy particles can be slowed down in the presence of a sufficiently strong tangled magnetic field @xmath40 . in this case , the characteristic escape time is given by the diffusion time which is , in the bohm diffusion approximation @xmath41 ^ 2 \left[\frac{b}{1\,\mbox{g}}\right]\left[\frac{10^{14}\,\mbox{ev}}{e_p}\right ] \mbox{s } , \label{t_diff}\ ] ] even if the high energy particles are trapped by the tangled magnetic fields in the stellar wind , they do not stay in the innermost dense stellar wind region for a long time : the high energy particles escape together with the stellar wind of velocity @xmath42 , so that the escape time is limited from above by @xmath43\left[\frac{v_{\rm w}}{10 ^ 8\,\mbox{cm s}^{-1}}\right]^{-1}\,\mbox{s . } \label{tesc}\ ] ] in the equatorial disk of a be star , the initial wind velocity close to the surface of the star is believed to be slow with @xmath44 km / s close to the surface of the star @xcite . the polar component of the wind is faster and less dense than the equatorial one . both equatorial and polar winds accelerate with distance , reaching asymptotically the velocity @xmath45@xmath46 km s@xmath47 . comparing ( [ tesc ] ) with ( [ tpp ] ) one finds that the pp interactions in the be star disk can efficiently transfer the power contained in the high energy protons into neutrinos and s. pp interactions result in the production of pions , which subsequently decay onto s , neutrinos and electrons / positrons . the neutrinos freely escape from the source . the electrons / positrons suffer synchrotron and inverse compton ( ic ) energy loss and , in this way release their energy through emission . if the larmor radius of the highest energy protons is comparable to the size of the disk , most of the pions are produced by the protons propagating toward the companion star . in this case the neutrino and emission from the pion decays is expected to be anisotropic , with most of the neutrino flux emitted toward the massive star , as it is shown in fig . [ fig : ellipse ] . conversely , the synchrotron and ic emission from the e@xmath1e@xmath2 pairs is , most probably , isotropized at energies at which the radiative cooling time of electrons becomes longer than the period of gyration in the magnetic field . differing anisotropy patterns for neutrino emission and for broad - band emission from the secondary e@xmath1e@xmath2 pairs should , in principle , lead to significant differences in the expected orbital lightcurves of neutrino and electromagnetic emission from the source . the ic cooling time of e@xmath1e@xmath2 pairs is shortest at the boundary between the thomson and klein - nishina limits for compton scattering , @xmath48 , when @xmath49 gev : @xmath50 \left[\frac{d}{r_*}\right]^2 \left[\frac{10^{10}\,\mbox{ev}}{e}\right]\,\mbox{s } , \label{t_ic}\end{aligned}\ ] ] at the same time , the period of rotation around the larmor circle is @xmath51\left[b/1\mbox { g}\right]^{-1}$ ] ms@xmath52 . the ic scattering on higher energy electrons proceeds in the klein - nishina regime . in this regime , @xmath53 , and the electron energy loss time grows with energy just slightly slower than the larmor radius @xcite @xmath54 \left[\frac{d}{r_*}\right]^2\mbox{s,}\end{aligned}\ ] ] where the density of the blackbody photons has been diluted by the @xmath55 factor . this means that if the dominant mechanism of radiative cooling is the ic energy loss , the emission from the secondary e@xmath1e@xmath2 pairs is isotropic . the synchrotron cooling time , @xmath56 ^ 2 \left[\frac{10^{14}\,\mbox{ev}}{e}\right]\,\mbox{s}\ ] ] can be much shorter than the period of rotation around the larmor circle only at the highest energies @xmath57^{-1/2}\mbox { ev}\ ] ] taking into account that the synchrotron emission from electrons of energy @xmath58 is emitted in the energy band @xmath59\left [ \frac{e}{10^{13}\,\mbox{ev}}\right]^{2}\mbox { mev},\ ] ] this emission is expected to be isotropic below several mev . comparing the synchrotron and ic cooling times at energies @xmath60 gev and @xmath61 gev , one finds that the lower energy electrons predominantly dissipate their energy via inverse compton in the energy band @xmath62 \left[\frac{e}{10^{10}\,\mbox{ev}}\right]^{2}\mbox{gev},\ ] ] while synchrotron radiation losses dominate for the highest energy electrons . a significant contribution to the cooling of electrons and positrons in the densest part of the pp interaction region may arise from bremsstrahlung emission , with cooling time @xmath63^{-1}$ ] s , which compares to the synchrotron / ic cooling time of @xmath12tev electrons . the flux of s from the pp interaction region is absorbed due to the pair production on the uv photon background in the vicinity of the be star . maximal optical depth with respect to the pair production is achieved at energies @xmath64^{-1}$ ] tev , where the pair production cross section reaches its maximum @xmath65 @xmath7 . the density of photons close to the surface of the be star is @xmath66 ^ 3\left[d / r_*\right]^{-2}$ ] @xmath33 , thus an estimate for the optical depth for s of this energy is given by @xmath67 ^ 3\left[\frac{d}{r_*}\right]^{-1},\ ] ] assuming the radius of the star @xmath68 cm . at higher energies , @xmath69 , the pair production cross - section and the optical depth decrease as @xmath70 . the attenuation of the flux due to the pair production becomes small only at energies above @xmath71 tev . below this energy , the spectrum of the emission can be significantly different from that of the neutrino emission , so that no solid prediction for the neutrino flux can be derived based on the observed flux and spectrum in the tev band . the power of the absorbed s is re - distributed to the secondary e@xmath1e@xmath2 pairs , which subsequently loose their energy through synchrotron and inverse compton emission . depending on the magnetic field strength in the pair production region , the bulk of the electromagnetic emission from the secondary pairs with energies ranging approximately between 10 gev and 10 tev can be either re - emitted back in the gev - tev energy band , if the inverse compton loss dominates , or in the x - ray band , in case the synchrotron losses dominate ( see eqs . ( [ esynch ] ) , ( [ eic ] ) ) . in order to demonstrate the possibility of dramatically different spectra of neutrino and emission from the system in the tev energy band , we have modeled the emission from pp interactions , assuming different proton injection spectra . our numerical code follows the evolution of the secondary particle spectra produced by the interaction of the high energy protons with the stellar wind protons in the course of their escape from the system . the equatorial wind of the massive star is supposed to have the radial density profile described by eq . ( [ n_disc ] ) . we assume that the high energy protons are initially injected at a distance @xmath72 and then escape together with the secondary particles produced in pp interactions toward larger distances . we explore the different options for the escape regime , as described in section [ sec : model ] . one possibility occurs when the high energy protons are traversing the pp interaction region without being trapped by the magnetic fields , the escape velocity is comparable to the speed of light . a similar `` fast escape '' situation is present when the magnetic field in the contact region between the stellar wind and the relativistic outflow is ordered ( the situation present e.g. in the scenario of interaction of relativistic pulsar wind with the be stellar wind ) . to model this `` fast escape '' regime , we assume that the primary high energy protons and the secondary e@xmath1e@xmath2 pairs move toward larger distances with speed @xmath73 . another possibility is that the high energy protons , which penetrate into the stellar wind , and/or the secondary e@xmath1e@xmath2 pairs , deposited throughout the stellar wind in result of pp interactions , are trapped by the tangled magnetic field in the stellar wind . in this case they would escape with velocity @xmath74 approximately equal to the stellar wind velocity . the slow down of escape mostly affects the low energy parts of the electron spectra , at energies at which the radiative cooling time is comparable or larger than the escape time from the system . we calculate the production spectra of neutrinos , s and e@xmath1e@xmath2 pairs using the approximations given in the ref . the synchrotron , ic and bremsstrahlung emission from the secondary e@xmath1e@xmath2 pairs , as well as the evolution of the spectra of the pairs due to the radiative cooling effects is modeled in a standard way @xcite , via a solution of the kinetic equations with the derivative over the distance @xmath75 substituted for the time derivative @xmath76 . to model the synchrotron emission , we assume a certain radial profile of the magnetic field , @xmath77 with @xmath78 . the parameters @xmath79 and @xmath80 , used for numerical calculations shown in figs . [ fig : without_pion ] , [ fig : gamma0 ] and [ fig : gamma1 ] is @xmath81 g , while for the calculation of fig . [ fig : gamma2 ] , @xmath82 g. taking into account that the trajectories of the secondary e@xmath1e@xmath2 pairs are isotropized by the magnetic fields and that we are interested in the orbit - averaged spectrum of the source , we take the angle - averaged cross - section of ic scattering for the calculation of the ic emission . the broad - band spectra of the three known persistent or periodic loud binaries , lsi + 61 303 , ls 5039 and psr b1259 - 63 have similar shape ( which does not resemble the shape of typical accretion - powered x - ray binaries @xcite ) . the -loud binaries spectral energy distribution are peaked in the mev - gev energy band . the observed x - ray and tev emission apparently form the low and high energy tails of the mev - gev `` bump '' in the spectrum . [ fig : without_pion ] shows an example of lsi + 61 303 sed . e@xmath2 pairs produced in pp interactions close to the surface of the be star , calculated assuming a nearly monochromatic proton injection spectrum with @xmath83 , @xmath84 ev . the upper panel shows the injection spectrum of e@xmath1e@xmath2 pairs ( dashed line ) and the spectrum formed as a result of cooling via synchrotron , ic and bremsstrahlung emission as well as coulomb energy loss . in the lower panel , thin red solid , dashed and dotted lines show respectively the synchrotron , the ic and the bremsstrahlung emission from the pairs . the black thick solid line shows the overall broad - band model spectrum . ] taking into account the possible anisotropy of the neutrino and neutral pion decay emission , we first show in fig . [ fig : without_pion ] the broad - band isotropic emission spectrum from the secondary e@xmath1e@xmath2 pairs . the primary proton injection spectrum is assumed to be hard , with @xmath85 , so that most of the protons have the energy close to the cut - off energy assumed to be @xmath86 ev . as discussed above , such an almost monochromatic spectrum of protons injected into the stellar wind can be produced when the high energy protons originate from a `` cold '' relativistic wind with bulk lorentz factor @xmath87 or when only the highest energy protons have large enough larmor radii to be able to penetrate deep into the stellar wind . within the hadronic model of activity , the sed mev - gev bump can be ascribed to the synchrotron emission from the e@xmath1e@xmath2 pairs produced in the pp and @xmath88 interactions . one should note that if the primary proton injection spectrum is hard , the shape of the mev - gev synchrotron bump in the spectrum depends slightly on the details of the proton injection spectrum , because the shape of the e@xmath1e@xmath2 pair spectrum ( shown in the upper panel of fig . [ fig : without_pion ] ) is determined by the radiative cooling effects , rather than by the details of the e@xmath1e@xmath2 injection spectrum ( shown by the dashed line in the upper panel of the figure ) . the hard e@xmath1e@xmath2 pair spectrum below @xmath89 ev is explained by the dominance of the coulomb loss in the densest innermost part of the stellar wind . the @xmath11 type spectrum between @xmath90 ev and @xmath91 ev is explained by the dominance of the synchrotron loss in this energy range . the hardening of the spectrum in the range @xmath91 ev@xmath92 ev is explained by the dominance of the ic energy loss proceeding in the klein - nishina regime . above @xmath93 ev , the ic cooling becomes less efficient than the synchrotron cooling , which leads to the softening of the spectrum . the electron energy at which the synchrotron loss takes over the klein - nishina energy loss can be estimated by comparing the synchrotron cooling time ( [ tsynch ] ) to the inverse compton cooling time ( [ tkn ] ) : @xmath94^{-1}\mbox { ev}\ ] ] ( where @xmath79 is the magnetic field in the innermost part of the stellar wind and we assume the radial profile @xmath95 ) . the synchrotron emission by electrons of such energies is produced in the energy band above @xmath96^{-1}\left[\frac{b_0}{1\mbox { g}}\right]^{-1}\mbox { mev , } \ ] ] resulting in a broad `` bump '' in the synchrotron spectrum above @xmath97 . the comparison of synchrotron , inverse compton and bremsstrahlung emission spectra from the `` cooling - shaped '' electron spectrum to the multi - wavelength data @xcite , done in the lower panel of fig . [ fig : without_pion ] , shows that the model in which all the e@xmath1e@xmath2 pairs are initially injected at very high energies ( above @xmath98 tev ) provides a good fit to the broad - band spectrum of the source . to produce this figure , we have taken into account the attenuation of the spectrum of the emission due to the @xmath88 pair production in the photon field of the be star , assuming that the s escape almost radially from the emission region . this corresponds to the situation of the compact object close to the inferior conjunction of the orbit , where the effects of the absorption of the flux are minimized . [ fig : gamma0 ] shows an opposite situation in which the spectrum strongly absorbed in the energy band 0.1 - 1 tev ( long - dashed thick solid line ) due to the pair production . this situation can occur when the emission comes from the densest innermost part of the stellar wind in the vicinity of the be star . the real value of the average @xmath99 can not be estimated unless the details of the 3-dimensional geometry of the emission region and the relative orientations of the extended emission region , of the be star and of the observer are known . taking into account this uncertainty , we choose for @xmath99 the minimal value which ensures the non violation of the flux upper bound at energies above @xmath100 tev from veritas observations of the source @xcite . [ fig : gamma0 ] also shows the emission spectrum from neutral pion decays ( blue thick solid line ) and the neutrino spectrum ( green thick solid line ) . the observation of the mev - gev band synchrotron emission from the secondary e@xmath1e@xmath2 pairs enables to constrain of the possible neutrino emission as explained previously . contrary to the emission from e@xmath1e@xmath2 pairs , neutrino and @xmath101 decay emission is anisotropic . in principle , the anisotropy could lead to a boosting of the neutrino and @xmath101 decay components . however , if the orbital plane is not aligned with the line of sight , the neutrinos and s from @xmath101 decay should be emitted into a cone with rather wide opening angle , to be observable ( the opening angle of the cone should be larger than the inclination angle of the binary orbit , see fig . [ fig : ellipse ] ) . this means that the flux is at most moderately enhanced by the anisotropy effect . at the same time , if the inclination of the orbit is larger than the opening angle of the neutrino and @xmath101 decay emission cone , the flux in these emission components could be negligible . , but considering the possibility of strong neutrino emission from the source . the notations for the spectrum are the same as in the lower panel of fig . [ fig : without_pion ] . the green thick solid line shows the neutrino spectrum . blue thick solid line shows the @xmath101 decay contribution to the spectrum . the black thick line shows the overall broad - band model spectrum . the dashed part of the line shows the part of the spectrum attenuated by @xmath88 pair production . ] [ fig : gamma1 ] and [ fig : gamma2 ] show the results of calculation of the broad - band emission spectrum from pp interactions obtained assuming softer spectra for the primary protons , with @xmath102 and @xmath103 , but keeping the same cut - off energy , @xmath86 ev . in these figures , we illustrate the effect of an additional uncertainty in the calculation of the broad - band spectrum of the source , related to the uncertainty of the cascade contribution to the observed source spectrum . the problem is that the shape of the electron spectrum below 10 tev is modified by the injection of @xmath104 pairs via the process of absorption of the s on the uv photon background . in case the synchrotron radiation dominates the energy loss of the `` tertiary '' electrons , the synchrotron emission from the cascade @xmath104 pairs will further increases the height of the mev - gev synchrotron bump in the spectrum . in case inverse compton dominates the energy loss of the cascade electrons , the tertiary pairs will emit mostly in the 1 - 10 gev energy band . the overall flux of the cascade contribution is determined by the average optical depth of the extended source with respect to the pair production . the thick short dashed lines in figs . [ fig : gamma1 ] and [ fig : gamma2 ] show the calculated modification of the broad - band spectrum of the source by the cascade . . the thick short dashed line shows the overall spectrum calculated taking into account the tertiary @xmath104 pairs from @xmath88 interactions . ] . the magnetic field is assumed to be @xmath82 g. ] in the previous section we have shown that the hadronic model of activity of -loud binaries could provide a good fit to the broad - band spectra of these sources . within this model , the observed mev - gev bump in the spectral energy distribution is explained by the synchrotron emission from the secondary e@xmath1e@xmath2 pairs produced either in pp interactions or in result of @xmath88 pair production process . a direct test of the hadronic model would be the detection of neutrino flux from the known -loud binaries . in principle , the expected neutrino flux from a given source can be estimated from the known flux . in the simplest case when the source is transparent to the s , the estimate is straightforward , because the normalization and spectrum of the flux produced by the neutral pion decays in the source directly gives an estimate of the spectrum and normalization of the neutrino flux . however , modeling of the broad - band spectra of emission from -loud binaries , described in the previous section , shows that this simple possibility , most probably , does not hold . first , the spectrum of emission in the tev band is affected by the @xmath88 pair production , so that neither the overall flux , nor the spectral index of the emission are related to the ones of the neutrino spectrum . next , a sizable contribution to the spectrum of emission in the tev band can be given by the ic emission from the secondary @xmath104 pairs produced in pp interactions . this further modifies the flux and the spectrum of tev emission . thus , the information on the properties of the tev emission from a -loud binary can not be used to estimate the neutrino flux from the source . surprisingly , the source flux in the mev - gev , rather than tev energy range can be used for the neutrino flux prediction . this possibility arises because within the hadronic model of activity , the mev - gev bump in the spectral energy distribution is produced by the emission from the secondary @xmath104 pairs from the pp interactions . contrary to the tev flux , the mev - gev flux from the -loud binary is not affected by the pair production and can be used to estimate the total energy output from the pp interactions in the source . the only uncertainty of such an estimate is that the synchrotron emission from the secondary @xmath104 pairs in the mev - gev energy band is , most probably , isotropic , while the neutrino and @xmath101 decay emissions are not . as explained in the previous section , the observed source flux in the mev - gev band gives , in fact , an upper limit on the possible neutrino flux from the source . although the total power of the neutrino emission can be estimated from the mev - gev luminosity of the -loud binary , the modelling of the broad - band emission spectrum of the source gives only mild constraints on the neutrino emission spectrum : acceptable models of the broad - band spectra can be found assuming initial proton injection spectra ranging from a @xmath11 power law to an almost monochromatic injection spectrum . below we estimate the maximal neutrino event rate in icecube for a particular example of lsi + 61 303 , taking into account the uncertainty of the spectrum of neutrino emission from the source . in the most optimistic case scenario , when the neutrino flux is at the level of the observed mev - gev flux from the source , we also investigate the possibility to measure the parameters of the neutrino spectrum ( and , therefore , of the primary proton spectrum ) , such as the spectral slope and/or the cut - off energy . the rate of neutrino - induced muon events from a point source at declination @xmath105 with differential neutrino spectrum @xmath106 @xmath107 where @xmath108 and @xmath109 are respectively the neutrino and the muon effective areas of the detector . the neutrino effective area , often provided in papers in contrast to the muon effective area , folds the detector efficiency and the neutrino propagation effects . therefore eq . ( [ eq : nevnu ] ) is convenient for a direct calculation of the number of detected neutrino - induced muon events in a specified neutrino energy range . however , the neutrino energy is not an experimental observable , contrary to the reconstructed muon energy . if we are interested in understanding whether the different neutrino spectra can be experimentally differentiated , _ i.e. _ we want to assess the spectral response of the detector for each model , we should use instead eq . ( [ eq : nevmu ] ) . this can not be done without further derivation of the muon effective area and knowledge of the details of the detector , namely the angular resolution ( of the order of 1@xmath110 @xcite for icecube in its final 80 string configuration ( ic80 ) and precisely accounted for in the following ) and the energy resolution of the order @xmath111 @xcite . the differential neutrino induced muon spectrum at the detector , for a given source at location @xmath105 , @xmath112 , is obtained after propagation of neutrinos up to the interaction point and further propagation of the muons to the detector . this can be sketched as follows : @xmath113 here @xmath114 is the muon energy at the interaction vertex , given by @xmath115 , where @xmath116 is the mean interaction inelasticity , taken from @xcite ( calculations below will adopt a simplifying hypothesis , considering an average inelasticity ) . @xmath117 is the probability density for a muon produced with energy @xmath114 to reach the detector with energy @xmath118 . for a given distance between the interaction vertex and the detector , @xmath118 is calculated from @xmath114 by integrating the muon energy loss rate @xmath119 , where the @xmath120 are the energy independent standard rock coefficients from ref . @xcite and @xmath121 is the grammage traversed by the muon on its way to the detector , @xmath122 where @xmath123 . the function @xmath124 is set to zero outside of the interval @xmath125 $ ] . @xmath126 is the grammage function for a muon of initial energy @xmath127 and final energy @xmath118 and @xmath128 is its range . finally , the function @xmath129 , which enters the final stage of calculation sketched in ( [ scheme ] ) characterizes the energy resolution of the detector and is discussed below . the differential muon flux at the detector , which enters equ . ( [ eq : nevmu ] ) can be calculated via the propagation of the muon flux from the interaction vertex as @xmath130 where @xmath131 is the neutrino interaction probability in the vicinity of the detector ( potentially producing a muon within the reach of the detector ) and @xmath132 is the earth transmission probability of a neutrino arriving from declination @xmath105 ( _ i.e. _ after crossing a grammage @xmath133 ) . @xmath134 is the avogadro number , @xmath135 and @xmath136 are respectively the charged current and the total muon neutrino cross sections , taken from @xcite . the separability feature of the kernel @xmath117 , together with the implicit assumption of the absence of muon propagation fluctuation allows us to analytically extract @xmath137 by substituting eq . ( [ fmu ] ) into ( [ eq : nevnu ] ) , ( [ eq : nevmu ] ) : @xmath138 the neutrino effective area @xmath139 without absorption is extracted from the neutrino effective area averaged over the whole northern hemisphere , @xmath140 from @xcite , using the relation @xmath141 the transmission probability @xmath142 which appears in the denominator ( numerically computed using the preliminary earth model @xcite ) strongly decreases with energy . therefore , to estimate the neutrino effective area at the declination of , the effect of earth absorption is accounted for according to @xmath143 it should be mentioned that @xmath144 is approximately , at the modest energy considered here , the effective area at zero declination , assuming an isotropic detector response and analyses uniform down to the horizon ( of course , analyses proceed non completely uniformly w.t.r . the declination to account for the difficult background rejection near the horizon . nevertheless , lately , as can be seen e.g. in @xcite , there is a remarkable uniformity of the sensitivity , down to very close to the horizon ) . the effective areas are shown for ic22 ( the icecube 22 string configuration ) and ic80 , in fig . [ fig : nu_spectranu_aeff ] . in the lower panel , the icecube muon effective area derived in @xcite shows a remarkable agreement above @xmath121 tev , while the behavior at lower energy is rather different : our calculation exhibits a soft exponential decrease below @xmath121 tev contrasting to the sharp cutoff at @xmath12100 gev . considering the upper panel , we notice the fast decrease of the neutrino effective area at lsi+61 303 declination above @xmath12300 tev . this is due to the fact that the neutrino trajectories towards icecube are slightly passing through the high density earth core . this leads to a significant suppression of the neutrino flux above this energy ( the observation of psr b1259 - 63 and ls 5039 with the km3net@xcite would not suffer this limitation and thus offer the possibility to probe spectral features up to the highest cutoff energies ) . with no absoption . blue dot - dashed line shows the effective area after averaging over the declinations @xmath105 for an isotropic neutrino flux @xmath145 @xcite . curves running in parallel are for ic22 and ic80 . black solid line shows the effective area for . bottom : full lines are the muon effective area calculated with ( [ amu ] ) for ic22 and ic80 . the dashed curve is the muon effective area derived in @xcite . ] with no absoption . blue dot - dashed line shows the effective area after averaging over the declinations @xmath105 for an isotropic neutrino flux @xmath145 @xcite . curves running in parallel are for ic22 and ic80 . black solid line shows the effective area for . bottom : full lines are the muon effective area calculated with ( [ amu ] ) for ic22 and ic80 . the dashed curve is the muon effective area derived in @xcite . ] using the analytical muon effective area ( [ amu ] ) , the diffuse atmospheric neutrino background rate and the point source signal event rate as a function of the chosen reconstructed muon threshold energy @xmath146 can be calculated , @xmath147 @xmath148 is the solid angle subtended within a radius @xmath149 for the diffuse atmospheric neutrino background and @xmath150 is the fraction of signal events with energy @xmath151 reconstructed within an angle @xmath149 from the source : @xmath152\right),\ ] ] where the point spread function ( psf ) @xmath153 decomposes into two distinct psf contributions : the energy - dependent neutrino to muon kinematics @xmath154 and the muon reconstruction resolution @xmath155 . for our analysis we extracted the parameters @xmath155 and @xmath156 from a fit to the ic22 and ic80 psf found in @xcite ( for ic80 , we obtain : @xmath157 . e.g. at 10 tev , 50% of the neutrinos coming from a specified source are reconstructed within 0.9@xmath110 for ic80 ) . in the following , the results will be presented using the optimized @xmath158 for ic80 . above , in eqs [ equ : nb ] and [ equ : ns ] , the energy argument for @xmath159 is set to @xmath160 . this is a conservative statement because the muon energy is smaller than the neutrino energy , we have @xmath161 . the weight factor used to calculate the number of events above some reconstructed energy : @xmath162 follows from an assumed gaussian energy resolution distribution function in @xmath163 , @xmath164 , with @xmath165 . in our calculations , the muon propagation was implemented through @xmath166 , with the energy - independent parameters @xmath167 and @xmath168 for standard rock ( adequate , given the source location and the proximity of icecube to the bedrock ) , in order to derive the muon energy distribution at the detector @xcite . therefore , fluctuations of muon energy due to the stochastic nature of high energy muon energy losses which implies muon energy distributions , were not accounted for @xcite . a more refined treatment , which requires numerical methods , proceed in integrating over the distance the probability @xmath169 @xcite for a muon generated at distance @xmath170 with energy @xmath127 to reach the detector with energy @xmath118 . our simplified treatment has nevertheless the advantage of transparently combining the latest published neutrino effective area with some phenomenological assumptions for the detector muon effective area , detector response , muon propagation and neutrino interaction , resulting in a simple analytical formula for the rates , thus avoiding more generic integro - differential forms . we choose the angular - dependent parametrization from volkova ( prompt + conventional neutrino fluxes ) for the atmospheric neutrino background model @xcite . this corresponds to 7.7 detected muons when using the neutrino effective area and per year with ic80 in the 1.3@xmath110 circular bin at declination . conversely , calculating this number with the analytical muon effective area [ amu ] yields 8.8 detected muons . this 15% difference may come from the various simplifications and is probably dominated by the delta function differential cross section approximation ( at low energies , the average range of the neutrino induced muons is larger ) . at declinations , the atmospheric neutrino zenith angle to be considered is @xmath171 ( @xmath172 is for vertical incidence ) . the results , in terms of the atmospheric background - subtracted muon spectra for the baseline neutrino flux models discussed in the previous sections , are presented in fig . [ fig : nu_spectra ] . the exposure time is taken to be 3 years of running the full icecube array . for comparison , we also show by the red solid thick line the @xmath0 level above the atmospheric background ( extracted from binomial statistics , where the trial factor is the total expected number of atmospheric neutrino @xmath173 in a 2@xmath149 width declination band reconstructed with energy @xmath174 and the probabiblity of success is @xmath175 in the declination band , @xmath176 . note however that an unbinned calculation slightly lower the signal requirements @xcite ) . we calculate the number of muon events as a function of the energy threshold for the three neutrino spectra discussed in the previous section . the blue dotted line show the detected muon spectrum for the model with the proton injection spectrum with spectral index @xmath103 , the solid black line corresponds to the proton injection spectrum with @xmath102 and cut - off at the same energy , while the magenta dashed line corresponds to @xmath83 . in all three cases the cut - off energy is assumed to be @xmath177 pev . one can see that the uncertainty of the proton injection spectra results in a factor of @xmath178 uncertainty in predictions of the number of neutrinos detectable in icecube . assuming that the neutrino flux saturates the upper limit imposed by the observed flux in the mev - gev energy band , one finds that the neutrino source should be detectable at @xmath0 level in the energy band @xmath179 tev in roughly one year , if the proton injection spectrum is not too hard . inspecting the muon spectra shown in the upper panel of fig . [ fig : nu_spectra ] , one sees that softed proton injection spectrum results in a slight excess of muon events at lower energies . if the overall normalization of the neutrino flux were known , the measurement of the muon event spectrum would allow to constrain the spectrum of the primary protons in the source . however , taking into account the uncertainty of the overall normalization of neutrino flux introduced by the uncertainty of the anisotropy pattern of neutrino emission , one finds that the statistics of the signal will in the best case only allow for marginal estimate of the neutrino spectrum parameters . this is illustrated in the lower panel of fig . [ fig : nu_spectra ] , where a comparison of the shapes of the muon spectra is shown . if one assumes the same total number of muon events , the difference in the spectra for the three models mostly remain within @xmath180 error bars over the entire energy range . the neutrino effective area in the direction of peaks at the energy @xmath181 tev ( see fig . [ fig : nu_spectranu_aeff ] . taking into account the typical inelasticity of pp interactions , one finds that neutrinos of this energy are injected by the primary protons with energies @xmath182 pev . this means that assuming the cut - off in the proton spectrum at the energy @xmath183 pev , one maximizes the number of neutrino events detectable with icecube . if the cut - off in the proton spectrum is much above 1 pev , the neutrino signal will be lost because of the opacity of the earth . otherwise , if the cut - off in the proton spectrum is much below 1 pev , the signal is lost because of the transparency of the detector for the neutrinos and because of the high atmospheric neutrino background . thus , the typical numbers of neutrino - induced events in the icecube , shown in fig . [ fig : nu_spectra ] should be considered as the upper bound on the possible neutrino signal from . , expected after the 3-year icecube exposure for the three model neutrino spectra of , discussed in the previous sections ( error bars of signal muon spectra are the sum in quadrature of statistical errors of signal + atmospheric neutrino background ) . the blue dotted / black solid / magenta dashed line respectively show the spectra corresponding to proton injection spectra with spectral indices @xmath184 , @xmath185 and @xmath186 . the red thick solid line shows the @xmath0 excess above the atmospheric neutrino background ( the `` discovery threshold '' ) . ] we have estimated the neutrino flux from grlbs expected within the hadronic model of activity of these sources . within such a model , the measured spectral characteristics of emission from the source in the tev energy band are not directly related to the spectral characteristics of the neutrino emission , because of absorption of the tev s on the thermal photon background produced by the massive star in the system . the uncertainty of the calculation of the attenuation of the tev flux introduces a large uncertainty to the estimate of the neutrino flux based on the measured tev flux . taking this uncertainty into account , we have adopted a different approach for the estimate of the neutrino flux from a grlb . namely , we have noted that the energy output of proton - proton interactions , and hence the neutrino flux , can be constrained by the broad - band spectrum of the source . the idea is that the @xmath104 pairs , produced in the decays of charged pions , release their energy in the form of the synchrotron , ic and bremsstrahlung emission , which contributes to the observed broad - band sed of the source . the requirement that the power of electromagnetic emission from the secondary @xmath104 pairs in different energy bands is not higher than the observed source luminosity imposes a constraint on the neutrino luminosity of the grlb . we have worked out the numerical model of the broad - band emission from a grlb , assuming that pp interactions take place in the vicinity of the bright massive star , in the innermost and densest part of the stellar wind . we have shown that the interactions of high energy protons ( with energies reaching @xmath12pev ) in the dense stellar wind can explain the observed broad - band emission , with the mev - gev `` bump '' of the sed being due to the synchrotron emission from the secondary @xmath104 pairs produced either in the decay of charged pions or resulting from the absorption of 10 gev - tev s on the thermal photons from the massive star ( see figs . 1 - 4 ) . although the observed bolometric luminosity of the source ( i.e. the height of the mev - gev bump of the sed ) constrains the overall neutrino luminosity , the shape of the neutrino spectrum and the overall normalization of the neutrino flux are only mildly constrained by the multi - wavelength data . the problem is that the properties of the broad - band emission from the secondary @xmath104 pairs are mostly determined by the effects of the radiative cooling and the escape of the pairs from the source , rather than by the initial injection spectrum of the pairs . this means that quite different injection spectra of the primary high energy protons ( and hence of the neutrinos ) can result in approximately the same broad - band seds , as is clear from figs . 1 - 4 . in addition , the anisotropy pattern of neutrino emission is expected to differ from that of the synchrotron emission from the @xmath104 pairs . this results in an uncertainty of the anisotropy - related suppression of the neutrino flux compared to the mev - gev synchrotron flux . taking into account these uncertainties of the neutrino emission spectrum , we have estimated the expected number of neutrinos which will be detected by icecube , assuming that the neutrino flux saturates the upper bound imposed by the observed flux in the mev - gev energy band . considering the particular example of , we have found that if the spectrum of high energy protons in the source extends to the pev energies , the source would be readily detectable within roughly one year of exposure with icecube . we have also explored the potential of the full icecube detector for the measurement of the spectral characteristics of the neutrino signal from . we find that in the case when the neutrino flux is at the level of the upper bound imposed by the observed mev - gev flux , an exposure time longer than 3 years will be required to reliably constrain the spectral index of the primary high energy proton spectrum via observations of neutrino signal in icecube . the mere fact of a possible spectral characterization of neutrino sources within the lifetime of icecube represents however a most exciting prospect for the future of mutimessenger astronomy with the possible emergence of a new field , the astrophysics of neutrinos . the authors are thankful to d.v . semikoz for useful suggestions . m. ribordy is supported by the swiss national research foundation ( grant pp002114800 ) . aharonian f. et al . , 2005 , science , 309 , 746 ; 2006 , a&a , 460 , 743 . albert j. et al . 2006 , science , 312 , 1771 ; acciari , v. a. et al . , 2008 , ap.j . , 679 , 1427 . aharonian a. et al . , 2005 , a&a , 442 , 1 . hinton j.a . et al . , 2008 , albert j. et al . , 2007 , ap.j . , 665 , l51 . mirabel i.f . , 2006 , science , 312 , 1759 . maraschi l. , treves a. 1981 , mnras , 194 , 1 . tavani m. , arons j. , 1997 , ap.j . , 477 , 439 . bosch - ramon v. et al . , 2006 , a&a , 447 , 263 . dubus g. , 2006 , a&a , 451 , 9 ; bednarek w. , 2006 , mnras , 368 , 579 ; khangulyan d. et al . 2008 , mnras , 383 , 467 . romero g. et al . 2007 , a&a , 474 , 15 . chernyakova m. , neronov a. & walter r. , 2006a , mnras , 372 , 1585 . chernyakova m. , neronov a. , et al . mnras , 2006b , 367 , 1201 ; de pasquale m. et al . , 2008 , gcn 8209 . dhawan v. , mioduszewski a. , rupen m. in proc . of vi microquasar workshop ( como ) . rib m. et al . , 2008 , a&a , 481 , 17 . aharonian f.a . , 2007 , talk at neutrino 06 , santa fe , new mexico , june 13 - 19 , 2006astro - ph/0702680 . aharonian f.a . , anchordoqui l. ; khangulyan d. ; montaruli t. , 2006 , j.phys . : conf . series , 39 , 408 . orellana m. ; bordas p. ; bosch - ramon v. ; romero g. e. ; paredes j. m. , 2007 , a&a,476 . torres d.f . ; halzen f. , 2007 , a.ph . , kappes a. ; hinton j. ; stegmann c. ; aharonian f. a. , 2007 , ap.j . , 656 , 870 . christiansen h.r . ; orellana m. ; romero g.e . , 2006 , d , 73 , 3012 . gallant y.a . , arons j. , 1994 , ap.j . , 435 , 230 . horns d. , aharonian f. , santangelo a. , hoffmann a.i.d . , masterson c. , 2006 , a&a , l51 . bednarek w. , 2003 , a&a , 407 , 1 . waters l. b. f. m. , van den heuvel e. p. j. , taylor a. r. , habets g. m. h. j. , persi p. , 1988 , a&a , 198 , 200 mart j. , paredes j. m. , 1995 , a&a , 298 , 151 porter j. m. , 1998 , a&a , 333 , l83 blumenthal g. r. , gould r. j. , 1970 , rev . phys . , 42 , 237 kelner s. r. ; aharonian f. a. ; bugayov v. v. , 2006 , phys . d , 74 , 4018 . chernyakova m. , neronov a. , walter r. , 2006 , mnras , 372 , 1585 . acciari , v. a. et al.,2008 , ap.j . , 679 , 1427 . km3net coll . , see http://www.km3net.org/publications.php achenberg a. , et al . , 2007 , phys.rev.d75:102001,2007 . t. i. collaboration , arxiv : astro - ph/0809.1646 . p. gondolo , g. ingelman , m. thunman , astropart . phys . * 5 * 309 , 1996 . p. lipari , astropart . phys . * 1 * 195 , 1993 . volkova , sov . . phys . * 31 * 784 , 1980 . g. anton , nucl . * 143 * 351 , 2005 . t. montaruli _ et al._in proc . of 10th int . conf . on topics in astropart . and underground phys . ( taup ) 2007 , sendai , japan , astro - ph/0712.3524 . a.m. dziewonski , d.l . anderson , phys . earth planet . 25 , 297 ( 1981 ) . a. d. martin , w. j. stirling , and r. g. roberts , phys . b 354 , 155 ( 1995 ) ; int . a 10 , 2885 ( 1995 ) ; phys . rev . d 51 , 4756 , ( 1995 ) . icecube coll . , juan - de - dios zornoza et al . , in proc . 30th icrc , mrida ( 2007 ) , arxiv:0711.0353 . m. ribordy , nucl . instrum . a * 574 * , 137 ( 2007 ) . p. lipari and t. stanev , phys . d 44 , 3543 ( 1991 ) . t. k. gaisser and t. stanev , phys . d * 30 * , 985 ( 1984 ) . m. c. gonzalez - garcia , f. halzen and m. maltoni , phys . d * 71 * ( 2005 ) 093010 . r. gandhi , c. quigg , m. h. reno and i. sarcevic , phys . d * 58 * ( 1998 ) 093009 . halzen , kappes , murchada j. braun , j. dumm , f. de palma , c. finley , a. karle and t. montaruli , astropart . * 29 * ( 2008 ) 299 . d. chirkin and w. rhode , arxiv : hep - ph/0407075 .

we formulate general conditions necessary for a linear - response detector to reach the quantum limit of measurement efficiency , where the measurement - induced dephasing rate takes on its minimum possible value . these conditions are applicable to both non - interacting and interacting systems . we assess the status of these requirements in an arbitrary non - interacting scattering based detector , identifying the symmetries of the scattering matrix needed to reach the quantum limit . we show that these conditions are necessary to prevent the existence of information in the detector which is not extracted in the measurement process .

cond-mat0211001

issues of quantum measurement in mesoscopic systems have recently garnered considerable interest , both because of their relevance to attempts at quantum computation @xcite and quantum - limited amplifiers @xcite . a general consequence of any quantum measurement is that it must induce decoherence in the system variable conjugate to that being measured . this basic fact naturally leads to the issue of measurement efficiency : what conditions must a particular detector satisfy so that it induces the _ absolute minimum _ amount of dephasing required by quantum mechanics ? this minimum dephasing rate is identical to the measurement rate @xmath0 , the rate at which information is extracted during the measurement process ; thus , the measurement efficiency ratio @xmath1 is defined by @xmath2 , where @xmath3 is the measurement - induced dephasing rate . besides being of great conceptual interest , near - ideal measurement schemes are necessary to detect signatures of coherent qubit oscillations in the output noise of a detector @xcite , and are essential if one wishes to construct a quantum limited amplifier ( i.e. an amplifier whose noise energy is the minimum allowed by quantum mechanics ) @xcite . while the question of measurement efficiency has received attention in the context of general measurement theory @xcite , it is only recently that it has been considered in the context of solid state detectors . averin @xcite has considered the status of the quantum limit in a number of solid state detectors , while recently pilgram and bttiker @xcite considered the quantum limit for a system in which a mesoscopic conductor acts as a detector . in this paper , we formulate general conditions which are needed for an arbitrary detector in the linear - response regime to reach the quantum limit of detection , where @xmath4 . these general conditions are valid for both interacting and non - interacting systems , and can be given a direct physical interpretation . we also discuss the quantum limit in terms of a simple concept from quantum information theory , the accessible information . to make these considerations more concrete , we apply them to a mesoscopic scattering detector similar to that considered in ref . , identifying precise conditions and symmetries needed to reach the quantum limit . we find that the required symmetries are most easily understood if one considers the scattering detector in terms of information ; these symmetries are not the same as those usually considered in mesoscopic systems . for example , we find that time reversal symmetry is not necessary for reaching the quantum limit . we also find that , surprisingly , an adiabatic point contact @xcite system remains a quantum limited detector even for voltages large enough that several channels contribute to transport and that the energy dependence of scattering is important ; previous studies @xcite have only shown that the quantum limit is achieved in the small voltage regime . our results for the mesoscopic scattering detector are complementary to those obtained in ref . . we start by considering a generic system consisting of a qubit ( i.e. a two - level system described as a spin @xmath5 ) coupled to an arbitrary detector . the system hamiltonian is @xmath6 , where @xmath7 , @xmath8 , and we leave @xmath9 unspecified . @xmath10 is the detector input " operator which couples to the qubit , while @xmath11 characterizes the strength of the qubit - detector coupling . mixing effects , where the detector causes transitions in the qubit , are neglected by taking @xmath12=0 $ ] ; such effects always cause a deviation from the quantum limit . we work in the weak - coupling regime ( @xmath13 ) , and can thus use linear response theory to describe the output of detector . taking @xmath14 to be the detector observable that is measured ( i.e. the output " operator ) , one has to lowest order in @xmath11 : @xmath15 where the zero - frequency linear - response coefficient ( or forward gain " ) @xmath16 is given by @xmath17 \rangle_{\rho_0 } \\ & = & \frac{2}{\hbar } \textrm{im } \int^\infty_{0 } d \tau \langle i(\tau ) q(0 ) \rangle_{\rho_0 } \label{lambdadefn}\end{aligned}\ ] ] here , @xmath18 is the initial density matrix of the detector , and @xmath19 is the initial density matrix of the qubit . we have assumed that the qubit splitting frequency @xmath20 is much smaller than the rate which characterizes the detector , which allows us to approximate the detector s response to the qubit as instantaneous . alternatively , one can restrict attention to the case where the qubit is in a @xmath21 eigenstate , and thus @xmath22 is time independent . the operators on the rhs in the above equation evolve in the heisenberg picture generated by @xmath23 . next , we connect the detector noise in the output operator @xmath14 and input operator @xmath10 to , respectively , the measurement rate @xmath0 and the dephasing rate @xmath3 . defining the fluctuating part of an operator @xmath11 as @xmath24 , the required zero - frequency noise correlators are given by : [ corrs ] @xmath25 here , we use the short hand @xmath26 , where @xmath27,@xmath28 are eigenstates of @xmath29 with energies @xmath30 . the probability @xmath31 is defined as @xmath32 ; we assume that @xmath18 is diagonal in the basis of eigenstates . taking the detector noise to be gaussian , the standard expressions for the dephasing rate @xmath3 and measurement rate @xmath0 are given by : @xcite @xmath33 we briefly review the origin of eqs . ( [ rates ] ) . the dephasing rate describes the measurement - induced decay of the off - diagonal elements of the qubit density matrix . it can be derived by looking at the decay at long times of the phase correlator @xmath34 , where @xmath35 ( @xmath36 ) is the spin raising ( lowering ) operator : @xmath37 \big\rangle \label{firstdepheq } \\ & \simeq & e^{-i \widetilde{\omega } t } \exp \left ( \frac{-2 a^2}{\hbar^2 } \int_0^t dt_1 \int_0^t dt_2 \langle \widetilde{q}(t_1 ) \widetilde{q}(t_2 ) \rangle \right ) \nonumber \\ & \ra & e^{-i \widetilde{\omega } t } e^{-\gamma_{\varphi } t}\end{aligned}\ ] ] here , @xmath38 . the measurement rate describes how long the measurement must be on before the signal associated with the two qubit states can be distinguished from the noise in @xmath14 . the quantity of interest is the time - integral of the detector output , @xmath39 . one needs that the distributions of @xmath40 corresponding to the two different qubit states ( i.e. @xmath41 and @xmath42 ) be statistically distinguishable . assuming gaussian distributions , distinguishability is defined as : @xmath43 where @xmath44 denotes the variance of the distribution , and the @xmath45 factor is included in order to make the final upper bound on @xmath46 unity . using eq . ( [ linresp ] ) for @xmath47 , and letting @xmath48 , the condition becomes : @xmath49 which directly yields the expression in eq . ( [ rates ] ) for @xmath0 . note that we have taken @xmath50 in the last step ; this is sufficient to obtain the leading order expression for @xmath0 . to relate @xmath3 and @xmath0 , we first note that the righthand sides of eqs . ( [ sispect])-([siq ] ) implicitly define an inner product ( i.e. , interpret the matrix elements @xmath51 and @xmath52 as defining vectors ) . the schwartz inequality then immediately yields : @xmath53 where we have introduced the reciprocal response coefficient ( or backwards gain " ) @xmath54 : @xmath55 @xmath54 would describes the response of @xmath56 to a perturbation which couples to the operator @xmath14 . note that as @xmath16 and @xmath54 are defined in terms of commutators , we may substitute @xmath57 , @xmath58 in their definitions . general stability considerations lead to the condition @xmath59 . using eqs . ( [ rates ] ) , we thus have : @xmath60 the best one can do is measure the qubit as quickly as one dephases it @xcite . note that this derivation only requires the validity of linear response and the weak - coupling approximations which give rise to eqs . ( [ rates ] ) ; very little is specified of the detector . similar derivations of the quantum limit are presented in refs . and . the inequality of eq . ( [ finalineq ] ) is in many ways intuitively reasonable . both dephasing and measurement involve entangling the state of the qubit with states in the detector . in principle , there may be degrees of freedom in the detector which become entangled with the qubit _ without _ providing any detectable information in a measurement of @xmath61 ; any such entanglement would lead to @xmath62 . more precisely , imagine that when the measurement is initially turned on , the system is in a product state : @xmath63 where @xmath64 is the initial state of the detector , and @xmath65,@xmath66 denote qubit @xmath21 eigenstates . at some later time @xmath67 , the state of the system may be written as : @xmath68 to say that we have measured the state of the system implies that the states @xmath69 and @xmath70 are distinguishable ; to say that the qubit has been dephased only implies that the detector states @xmath69 and @xmath71 are orthogonal . while distinguishability implies orthogonality , the opposite is not true ; thus , in general , @xmath72 . note that in this formulation , the dephasing rate will be related to the overlap between the two detector states : @xmath73 we have thus seen that on a heuristic level , reaching the quantum limit requires that the detector have no extraneous " degrees of freedom which couple to the qubit . equivalently , all information on the state of the qubit residing in the detector should be accessible in a measurement of @xmath61 . the virtue of the derivation presented in the last subsection is that these statements can be given a precise meaning . one sees that three conditions are necessary to reach the quantum limit : ( i ) the schwartz inequality of eq . ( [ schwartzi ] ) must be optimized , ( ii ) the cross - correlator @xmath74 must vanish , and ( iii ) the backwards gain @xmath54 must vanish . conditions ( i ) and ( ii ) can be succinctly re - expressed as a single condition , leading to the following necessary and sufficient requirements : @xmath75 @xmath76 here , @xmath77 is a real number which is independent of the detector eigenstates @xmath78 and @xmath28 . ( [ bigcondition ] ) and ( [ gaincondition ] ) are central results of this paper . the first of these equations expresses the fact that to reach the quantum limit , there must be a close similarity between the detector s input and output operators as far as the zero - frequency noise correlators are concerned , _ the operators @xmath14 and @xmath10 must be proportional to one another_. this required similarity between the detector input and output is a formal expression of the intuitive idea that a quantum limited detector has no extraneous " internal degrees of freedom . the second condition , eq . ( [ gaincondition ] ) , expresses the fact that a quantum - limited detector must have a strong intrinsic directionality which discriminates between the input and output . the output operator is influenced by behaviour at the input , but not vice - versa . this requirement is consistent with our tacit assumption that the quantity @xmath61 can be measured without problems . to measure @xmath14 , one needs to introduce a coupling in the hamiltonian to @xmath14 ; the vanishing of @xmath54 implies that this additional coupling will not contribute to @xmath79 , and thus can not further dephase the qubit ( c.f . ( [ firstdepheq ] ) ) . on a technical level , eq . ( [ bigcondition ] ) follows from the optimization of the schwartz inequality and the requirement that @xmath80 ( i.e. conditions ( i ) and ( ii ) above ) . the vanishing of @xmath54 ( eq . ( [ gaincondition ] ) ) can be interpreted in terms of causality . to see this , we first introduce the frequency - dependent cross - correlator @xmath81 : @xmath82 we may use this to write : @xmath83 \nonumber \\ & & - \frac{1}{\pi } { p } \int_{-\infty}^{\infty } de \frac { \textrm{re } \left [ s_{iq}(e ) \right ] } { e } \bigg ) \label{lambdadecomp}\end{aligned}\ ] ] if @xmath84 , it follows from the above that at @xmath85 , the imaginary part of @xmath81 coincides with the hilbert transform of the real part of @xmath81 : @xmath86 \bigg|_{e=0 } = \bigg(- \frac{1}{\pi } { p } \int_{-\infty}^{\infty } de ' \frac { \textrm{re } \left [ s_{iq}(e ' ) \right ] } { e ' - e } \bigg ) \bigg|_{e=0 } \label{weakcausal}\ ] ] if this held for all @xmath87 , it would follow from the titchmarsh theorem @xcite that @xmath88 is causal : it would vanish for @xmath89 . this would clearly be sufficient to satisfy eq . ( [ gaincondition ] ) . more generally , the vanishing of @xmath54 only requires the weaker condition of eq . ( [ weakcausal ] ) . we close this section by formalizing the connection between the quantum limit and information . a deviation from the quantum limit ( i.e. @xmath90 ) implies the existence in the detector of missing information " regarding the state of the qubit , information which is not revealed in a measurement of @xmath91 . the dephasing rate thus corresponds to what the measurement rate would be _ if _ we could make use of all the available information . this notion can be quantified by borrowing a concept from quantum information theory , the accessible information @xcite . to define this , note first that if we choose a specific detector quantity ( or set of quantities ) @xmath92 to measure ( described by , e.g. , a set of commuting observables ) , we can think of our system as a noisy classical communication channel . the two possible inputs to the channel are the qubit states @xmath93 and @xmath66 ; interaction with the detector for a time @xmath67 then leads to two corresponding detector states @xmath69 and @xmath94 ( c.f . eq . ( [ heuristicpsi])).@xcite finally , the outputs from the channel are the outcomes of the measurement of @xmath92 . the noise " here is a result of the intrinsic uncertainties of @xmath92 in the states @xmath95 and @xmath96 ; the output will thus be described by the conditional probability distributions @xmath97 determined by these states , where @xmath98 represents possible outcomes of the measurement . letting @xmath99/2 $ ] , the mutual information @xmath100 of this channel is @xcite : @xmath101 = h [ \bar{p}(y ) ] - \frac{1}{2 } \big ( h [ p(y| \ua ) ] + h [ p(y | \da ) ] \big ) \label{reqn}\ ] ] where @xmath102 $ ] is the shannon information entropy associated with the distribution @xmath103 : @xmath104 = -\sum_{y_i } p(y_i ) \log(p(y_i))\ ] ] note that we have chosen to equally weight our two inputs to the channel . assuming that this choice is optimal , shannon s noisy channel coding theorem implies that @xmath105 $ ] is the maximum rate at which messages can be reliably transmitted down the channel by modulating the state of the qubit and making measurements of @xmath92 . @xcite alternatively , @xmath105 $ ] may be considered as being related to a generalized measurement rate describing the chosen measurement @xmath92 . for example , if the distributions @xmath106 and @xmath107 are gaussian , one finds that at small times ( i.e. before the two distributions are well separated ) : @xmath101_{\rm gaussian } = \frac{1}{8 } \frac { \left ( \langle y(t ) \rangle_{\ua } - \langle y(t ) \rangle_{\da } \right)^2 } { \sigma_{\ua}(t)\sigma_{\da}(t ) } \ ] ] this corresponds to our definition of the measurement rate , c.f . ( [ mr1 ] ) and ( [ mr2 ] ) . we thus have a new way to interpret the measurement rate @xmath0 : given that one is monitoring @xmath61 , @xmath0 represents the maximum rate at which information can be sent to the detector by modulating the qubit . the quantum mechanical accessible information @xmath108 is now defined by maximizing the mutual information @xmath105 $ ] over all possible measurement schemes @xmath92 . remarkably , for the case considered here ( where the detector is described by a pure state ) it can be calculated exactly @xcite ; a simplified proof is presented in appendix a , where we also demonstrate that there are several possible optimal measurement schemes . letting @xmath109 , we have : @xmath110 \label{accessible}\end{aligned}\ ] ] this expression corresponds to having equally weighted our two input states , as we did in eq . ( [ reqn ] ) ; one can check that this choice maximizes @xmath111 . at small times ( @xmath112 ) , comparison against eq . ( [ phendephasing ] ) yields @xmath113 , and we have : @xmath114 as expected , the growth of the accessible information is determined by the dephasing rate . achieving @xmath115 thus implies that the rate that we actually obtain information , @xmath116 , coincides with the growth of the total accessible information . thus , there is no missing " information in the detector . we can also think of eqs . ( [ accessible ] ) and ( [ weakia ] ) as providing an alternate route for deriving the quantum limit inequality @xmath117 , i.e. : @xmath101 \simeq \gamma_{\rm meas } t \leq { \mathcal i } \simeq \gamma_{\varphi } t\ ] ] the utility of thinking about back action effects and the quantum limit in terms of information will become clear in the next section , where we discuss the mesoscopic scattering detector . note also that the relation between information and state disturbance has been studied in a slightly different context by fuchs et al . to make the preceding discussion more concrete , we now consider the status of the quantum limit in a slightly less general detector set - up , the mesoscopic scattering detector considered in ref . . we determine the conditions needed to reach the quantum limit of detection by directly applying the general conditions derived in the last section , namely the proportionality condition of eq . ( [ bigcondition ] ) , and the causality condition of eq . ( [ gaincondition ] ) . this is in contrast to ref . , which developed conditions needed for the quantum limit by directly calculating @xmath3 and @xmath0 . we explicitly show that a violation of eq . ( [ bigcondition ] ) implies the existence of unused information in the detector , information which is not extracted in the measurement process . the detector here is a two terminal scattering region ( see fig . 1 ) characterized by a scattering matrix @xmath118 . taking the contact to both the right and left reservoirs to have @xmath119 propagating transverse modes , @xmath118 will have dimension @xmath120 . the output operator of the detector @xmath14 is simply the current through the region ; the state of the qubit alters @xmath61 by modulating the potential in the scattering region . note that while we focus on the limit of a weak coupling between the qubit and detector , so that the linear response approach of the previous section is valid , we do not assume that the voltage is small enough that @xmath121 . @xcite the mesoscopic scattering detector describes the setup used in two recent which path " experiments @xcite . these experiments used a quantum point contact to detect the presence of an extra electron in a nearby quantum dot . as the dot was imbedded in an aharanov - bohm ring , the dephasing induced by the measurement could be studied directly . schematic of the mesoscopic scattering detector , in which the current through a phase coherent scattering region is used to detect the qubit . @xmath10 denotes the charge in the scattering region , while @xmath122 ( @xmath123 ) is the current in the right ( left ) contact.,width=6 ] we start by considering the simplest situation , also considered in ref . , where the state of the qubit provides a uniform potential change in the scattering region . in this case the input operator @xmath10 is the _ total _ charge in the scattering region . unlike ref . , we do not explicitly consider the effects of screening here . within an rpa scheme , consideration of such effects allows an explicit calculation of the qubit - detector coupling strength @xmath11 , but does not result in any other changes over a non - interacting approach . in the weak coupling regime , the particular value of @xmath11 does not affect the approach to the quantum limit . letting @xmath124 represent the creation operator for an incident wave in contact @xmath125 , transverse mode @xmath126 , and at energy @xmath87 , the detector current operator for contact @xmath127 takes the form @xcite : @xmath128 \label{idefn}\end{aligned}\ ] ] @xmath129^{\dagger } s_{\alpha \gamma}(e ' ) \right)_{nm}\ ] ] a positive current corresponds to a current incident on the scattering region ; note that throughout this section , we neglect electron spin for simplicity . the total charge @xmath10 in the scattering region may be defined in terms of the total current incident on the scattering region in the heisenberg picture , @xmath130 . one obtains : @xmath131 \label{qdefn}\end{aligned}\ ] ] @xmath132 . \label{nndefn}\end{aligned}\ ] ] in the limit where @xmath133 , @xmath134 reduces to the well - known wigner - smith delay time matrix : @xmath135\ ] ] finally , the assumption that the qubit couples to the total charge in the scattering region is equivalent to assuming that the potential it creates is smooth in the wkb sense . we can use the fact that the sensitivity of the scattering matrix @xmath118 to a global change of potential in the scattering region is the same as its sensitivity to energy . thus , the linear response coefficient @xmath16 has the form : @xmath136 \nonumber \\ & = & - \frac{e^2}{h } \int_{\mu_r}^{\mu_l } d \ve \sum_j \frac{d t_j ( \ve)}{d \ve } , \label{sclambda}\end{aligned}\ ] ] where the @xmath137 are the transmission eigenvalues of the system . without loss of generality , we have assumed that our detector is biased such that the chemical potential of the left reservoir is greater than that of the right reservoir : @xmath138 ; we also consider the limit of zero temperature . given these definitions , we can now turn to eqs . ( [ bigcondition ] ) and ( [ gaincondition ] ) and ask what is required of the scattering matrix @xmath118 in order to reach the quantum limit . we first focus on the case @xmath139 , where there is a single propagating mode in both contacts . the scattering matrix @xmath118 is thus @xmath140 , and may be written as : @xmath141 where @xmath142 . at zero temperature , the detector is described by a single many - body state @xmath27 in which all incident states in lead @xmath127 with @xmath143 are occupied , and all other incident states are unoccupied : @xmath144 first , we consider the causality condition of eq . ( [ gaincondition ] ) which requires that the backwards gain @xmath54 vanishes . as we know the initial state of the detector and have explicit expressions for @xmath14 and @xmath10 , we can directly evaluate the function @xmath81 appearing in eq . ( [ siqfreq ] ) in terms of @xmath118 . a direct calculation can be performed to show that : @xmath145 } { e } & = & \int_{-\infty}^{\infty } de \frac { \textrm{re } \left [ f(e ) \right ] } { e } \label{causalone } \\ \textrm{im } \left [ s_{iq}(0 ) \right ] & = & \textrm{im } \left [ f(0 ) \right ] \label{causaltwo } \end{aligned}\ ] ] where , letting @xmath146,the function @xmath147 is defined as : @xmath148 note that eqs . ( [ causalone ] ) and ( [ causaltwo ] ) are independent of whether @xmath14 is take to be @xmath123 , @xmath122 , or a linear combination of the two . now , causality dictates that the scattering matrix @xmath118 is analytic in the upper half complex plane , and thus so is the function @xmath147 . the real and imaginary parts of @xmath149 are thus related by a hilbert transform , and eqs . ( [ weakcausal ] ) , ( [ causalone ] ) and ( [ causaltwo ] ) imply that @xmath150 for the scattering detector _ irrespective of the choice of s_. thus , the causality properties of the scattering matrix @xmath118 ensure that one of the conditions necessary for reaching the quantum limit is always satisfied . note that substituting these expressions for @xmath81 in eq . ( [ lambdadecomp ] ) does indeed yield the expected form of @xmath16 ( eq . ( [ sclambda ] ) ) . it is also useful to note that gauge invariance can be used to directly establish@xcite @xmath150 . the essence of the argument is that a coupling to the current ( i.e. @xmath151 ) is equivalent to introducing a local vector potential . the gauge transformation which removes this term will only modify the transmission phases in the scattering matrix @xmath118 ( i.e. @xmath152 and @xmath153 ) in an energy - independent manner . using eq . ( [ qdefn ] ) , one can check that @xmath154 is independent of energy - independent phase changes ; thus @xmath150 . next , we turn to the condition given in eq . ( [ bigcondition ] ) , which requires a certain proportionality between @xmath155 and @xmath156 in order to reach the quantum limit . given the state @xmath78 which describes the detector ( eq . ( [ statedefn ] ) ) , the only matrix elements of @xmath14 and @xmath10 which contribute to the zero frequency noise correlators ( c.f . eqs ( [ corrs ] ) ) involve energy - conserving transitions where a scattering state incident from the left reservoir is destroyed while a scattering state incident from the right reservoir is created . since these transitions require an occupied initial state and an unoccupied final state , they can only occur in the energy interval @xmath157 . we are thus interested in the coefficients of the operators @xmath158 appearing in the expansion of @xmath14 and @xmath10 in this energy interval . the proportionality requirement of eq . ( [ bigcondition ] ) thus results in a necessary condition on @xmath159 : @xmath160 , \hspace{0.2 cm } \left[s_{lr } \right]^{*}(e ) s_{ll}^{\phantom { * } } ( e ) = i \cc \nn_{rl}(e ) \label{scondition}\end{aligned}\ ] ] where @xmath77 is a real , energy - independent constant . using eq . ( [ sdecomp ] ) , the imaginary and real parts of the above condition become : @xmath160 , \hspace{0.2 cm } \frac{d}{de } \left(\beta(e ) - \phi(e ) \right ) & = & 0 \label{phasecondition } \\ \frac { \frac{dt}{de}(e ) } { t(e ) ( 1-t(e ) ) } & = & -\frac{4 \pi}{\cc } \label{tcondition}\end{aligned}\ ] ] similar conditions for reaching the quantum limit for this version of the scattering detector were first developed in ref . by directly calculating @xmath0 and @xmath3 ( note there is a sign error in eq . ( 7 ) of ref . which must be corrected to obtain our eq . ( [ phasecondition ] ) ) . @xcite the fulfilling of these conditions does not correspond to symmetries usually considered in mesoscopic systems ; for example , as we will show , the presence of time - reversal symmetry is not a necessary requirement . instead , the conditions of eqs . ( [ phasecondition ] ) and ( [ tcondition ] ) correspond directly to the requirement that there be no missing information in the detector , information which is not revealed in a measurement of @xmath61 . we demonstrate this explicitly in what follows . the first condition ( eq . ( [ phasecondition ] ) ) for reaching the quantum limit requires that the difference between transmission and reflection phases in the scattering matrix be constant in the energy interval defined by the voltage . if it holds , changing the state of the qubit will not modulate this phase difference . ( [ phasecondition ] ) thus constrains information it ensures that the detector does not extract additional information about the qubit which resides in the relative phase between transmission and reflection . such information is clearly not revealed in a measurement of @xmath61 , and would necessarily lead to additional dephasing over and above the measurement rate . in principle , this additional information could be extracted by performing an interference experiment . to be more specific , note that the cross - correlator @xmath161 ( c.f . [ siq ] ) is given by : @xmath162 by definition , the imaginary part of this correlator determines the linear response coefficient @xmath16 ( c.f . ( [ lambdadefn ] ) ) associated with measuring @xmath91 . in contrast , the _ real _ part of this correlator may be interpreted as the linear response coefficient associated with a measurement where one interferes reflected and transmitted electrons ; the factor of @xmath163 corresponds to the fact that the magnitude of this signal will be proportional to the amplitude of both the reflected and transmitted beams . more explicitly , consider the hermitian operator @xmath164 defined by : @xmath165 \label{imoddefn}\ ] ] if one were to now measure @xmath164 , the corresponding linear response coefficient @xmath166 is precisely the real part of @xmath161 ( this can be seen by comparing eqs . ( [ imoddefn ] ) and ( [ idefn ] ) ) . the fact that additional information on the state of the qubit is available in the expectation @xmath167 implies that the qubit is entangling with the detector faster than the measurement rate associated with @xmath61 . this remains true even if one does not explicitly extract this information , as was demonstrated recently in the experiment of sprinzak et . @xcite stepping back , we see that the general condition @xmath168 ( i.e. the required factor of @xmath169 on the rhs of eq . [ bigcondition ] ) needed to reach the quantum limit directly corresponds to the requirement of no missing " information discussed in the previous subsection . in general , a non - vanishing @xmath170 implies that additional information about the qubit s state could be obtained by simultaneously measuring another quantity in addition to @xmath14 ( e.g. , in our case , the quantity @xmath164 ) . note that in the scattering detector , the symmetry required to ensure that eq . ( [ phasecondition ] ) holds ( i.e. that the phases @xmath171 and @xmath152 coincide ) is not one that is usually considered in mesoscopic systems . in particular , _ the presence of time - reversal symmetry is not _ necessary _ to fulfilling the condition of eq . ( [ phasecondition ] ) _ ; time - reversal symmetry only implies that @xmath172 , and specifies nothing on the relation between @xmath173 and @xmath171 . however , as pointed out in ref . , a _ sufficient _ condition for achieving eq . ( [ phasecondition ] ) is that one has parity symmetry , that is _ both _ time - reversal symmetry and left - right inversion symmetry ( the latter condition implies that the two reflection phases in @xmath118 are identical).@xcite note that this is not a necessary condition . we see that the required symmetry here is best understood as being related to information . we now turn to the second condition ( eq . ( [ tcondition ] ) ) needed to have the scattering detector reach the quantum limit , a condition which constrains the energy dependence of the transmission probability @xmath174 . this condition arises from the requirement that the proportionality between @xmath14 and @xmath10 needed for the quantum limit must hold over the entire energy interval defined by the voltage . in general , energy averaging causes a departure from the quantum limit over sufficiently large intervals , the operators @xmath14 and @xmath10 look less and less like one another . like eq . ( [ phasecondition ] ) , eq . ( [ tcondition ] ) can also be interpreted as a requirement of no missing " information . here , the requirement is that energy averaging does not result in the loss of information about the qubit which is encoded in the energy dependence of @xmath174 . while such information is not obtained in a measurement of @xmath61 ( which involves energy averaging , c.f . ( [ sclambda ] ) ) , it could be obtained if one measured the entire function @xmath175 for @xmath176 . as discussed , the presence of any missing information necessarily implies a departure from the quantum limit . interestingly enough , eq . ( [ tcondition ] ) may be understood completely classically , even though it formally results from requiring the proportionality of two quantum operators . to do so , we calculate the classical information capacity @xmath100 ( c.f . ( [ reqn ] ) ) corresponding to two different possible measurements . first , imagine we measure the integrated current @xmath177 , and assume the probability distributions @xmath178 and @xmath179 are gaussian . for weak coupling , one finds for the capacity : @xmath180 in the last line , we have discretized the energy integrals i.e. partitioned the interval @xmath181 $ ] into equal segments of length @xmath182 . if we now imagine we could measure each @xmath183 , where @xmath184 is the contribution to the current from the @xmath185th energy interval , a similar calculation reveals : @xmath186 one can easily check that @xmath187 ; this corresponds to the additional information that is generally available in the energy dependence of @xmath174 . a necessary and sufficient condition for ensuring @xmath188 is precisely the condition of eq . ( [ tcondition ] ) . on a purely classical level , this condition ensures that no information is lost when one averages over energy . how can the problems generally posed by energy averaging be avoided ? one possible solution would be to use voltages small enough that the scattering matrix @xmath118 can be approximated as being linear in energy , that is @xmath189 ( this is the approach of ref . ) . however , as the linear response coefficient @xmath16 is given by the energy derivative of the transmission ( c.f . ( [ sclambda ] ) ) , such a small voltage would imply both a small signal and essentially no gain . the change in current induced by the qubit , @xmath190 , would be much smaller than the current associated with the coupling voltage @xmath11 : @xmath191 even though this smallness of @xmath16 does not theoretically affect the approach to the quantum limit , it does severely limit the detector s practical value for very slow measurement rates , environmental effects on the qubit will become dominant over backaction effects . if we now consider finite voltages and fully energy - dependent scattering , eq . ( [ tcondition ] ) tells us the condition under which energy averaging the transmission does not impede reaching the quantum limit . the solution to eq . ( [ tcondition ] ) has the form : @xmath192 this form for @xmath193 implies that there is no extra information in the energy dependence of @xmath174 which is lost upon energy averaging . amusingly , eq . ( [ tcondition ] ) corresponds _ exactly _ to the energy - dependent transmission of one channel of an adiabatic quantum point contact @xcite . the constant @xmath194 represents the threshold energy of the channel ( i.e. the transverse mode ) , and the constant @xmath77 is given by : @xmath195 where @xmath196 is the transverse width of the constriction at its center , and @xmath100 is the radius of curvature of the transverse confining potential at the constriction center . we now consider the situation where there are @xmath119 channels in each of the two contacts leading to the reservoirs . it is useful to write @xmath118 in terms of its @xmath119 transmission eigenvalues @xmath197 using the standard polar decomposition:@xcite @xmath198 here , @xmath199 are @xmath200 energy - dependent unitary matrices , and @xmath201 and @xmath202 are diagonal matrices having entries @xmath203 and @xmath204 , respectively . in the multichannel case , the backwards gain @xmath54 again vanishes irrespective of the details of @xmath118 as a result of the analytic properties of @xmath118 . the relevant question then to ask is what conditions must be satisfied by @xmath205 so that the proportionality between @xmath14 and @xmath10 required to reach the quantum limit ( i.e. eq . ( [ bigcondition ] ) ) is achieved . as in the single - channel case , the relevant matrix elements of @xmath14 and @xmath10 involve destroying a scattering state incident from the left and creating an equal - energy state describing an incident wave from the right ; the additional complication now is that these transitions could result in a change of transverse mode . one thus needs to examine the coefficients of the operator products @xmath206 appearing in the expansion of @xmath14 and @xmath10 , in the energy interval @xmath207 $ ] . the proportionality condition of eq . ( [ bigcondition ] ) again yields the requirement that eq . ( [ scondition ] ) hold for all energies in this interval ; now , however , both the right and left - hand side of this equation are @xmath200 matrices : @xmath160 , \hspace{0.2 cm } \left[s_{lr}(e ) \right]^{\dagger } s_{ll}^{\pd } ( e ) = i \cc \nn_{rl}(e ) \label{mcscondition}\end{aligned}\ ] ] here , @xmath77 is again an energy - independent real number . using the polar decomposition , one can derive from eq . ( [ mcscondition ] ) two necessary matrix conditions which must hold for all energies in the interval defined by the voltage : @xmath208 @xmath209 these conditions are the multi - channel analogs of eqs . ( [ phasecondition ] ) and ( [ tcondition ] ) . @xmath210 denotes the @xmath200 unit matrix , and we have introduced the generalized phase - derivative " hermitian matrices @xmath211 and @xmath212 : @xmath213 \\ \phi_v(\ve ) & = & -i v^{\dag}(\ve ) \left [ \frac{d}{de } v(\ve ) \right]\end{aligned}\ ] ] these matrices play the role of the energy - derivatives of the phases @xmath171 and @xmath152 in the single channel case . note the evident asymmetry in eq . ( [ mcphasecondition ] ) : the polar decomposition matrices @xmath214 and @xmath215 enter , but the matrices @xmath216 and @xmath217 do not . we comment on this in what follows . the first requirement ( eq . ( [ mcphasecondition ] ) ) places a stringent requirement on the scattering matrix @xmath118 . like the corresponding requirement for the single - channel system , it ensures that there is no additional information on the state of the qubit available in measurable changes of scattering phases . again , _ time - reversal symmetry is not necessary _ to have this condition hold , as time - reversal symmetry only ensures @xmath218 and @xmath219 . however , unlike the single - channel case , even the presence of parity symmetry ( i.e. the combination of both time - reversal symmetry and left - right inversion symmetry ) is _ not sufficient _ to guarantee that eq . ( [ mcphasecondition ] ) is satisfied . the presence of parity symmetry would indeed ensure @xmath220 , but as in general @xmath221 , \left[\sqrt{r},\phi_u \right ] \neq 0 $ ] , this is not enough . in addition to having @xmath222 , one also generally needs either that @xmath211 is diagonal , meaning that the mode index ( i.e. transverse momentum ) is conserved during scattering , or that all the transmission eigenvalues @xmath137 are identical . we thus see that if the transmissions fluctuate , mode - mixing ( e.g. the non - conservation of transverse energy ) also prevents one from reaching the quantum limit of detection . this can be understood from the point of view of information . if the @xmath223 matrices are not purely diagonal , information about the qubit could be gained by looking at changes in how electrons incident in a given mode are partitioned into outgoing modes . such changes would not be detectable if all channels had the same transmission . note that the matrices @xmath216 and @xmath217 appearing in the polar decomposition of @xmath118 ( eq . ( [ polar ] ) ) are irrelevant to reaching the quantum limit . as each transverse mode is equally populated with incoming waves in the state @xmath27 , there is no information associated with the preferred mode structure for incoming waves ( i.e. the eigenvectors of @xmath216 and @xmath217 ) . consider now the condition imposed by eq . ( [ mctcondition ] ) , which constrains the form of the transmissions @xmath224 of the detector . similar to the corresponding condition for the single - channel system , this requirement ensures that there is no additional information available in either the energy _ or _ channel structure of the @xmath225 which is lost upon averaging . one obtains a necessary form for the transmissions , similar to what was found in ref . : @xmath226 note that different modes differ from one another only by their threshold energy @xmath227 ; the constant @xmath77 is the same for each mode . again , this form for the transmissions \{@xmath224 } corresponds exactly to those expected for a multi - channel adiabatic point contact.@xcite the assumption of adiabaticity implies that transverse energy is conserved . thus , if parity symmetry also holds , we reach the surprising conclusion that _ a multi - channel adiabatic point contact remains a quantum limited detector even if the voltage is large enough that several modes contribute to transport_. previous studies have established that point contact detectors reach the quantum limit in the limit of small voltages , where the energy - dependence of scattering can be neglected @xcite . we have shown here that in the adiabatic case , the quantum limit continues to hold even at voltages large enough that the energy dependence of scattering is important . this is significant from a practical standpoint requiring small voltages limits the magnitude of the output current and thus the overall scale of the measurement rate , making the detector more susceptible to environmental effects . for completeness , we give explicit expressions for the noise correlators . writing them in terms of energy dependent @xmath228 matrix kernels ( i.e. @xmath229 $ ] ) we obtain : [ noises ] @xmath230 \left [ \sqrt{t r},\phi_v \right ] \nonumber \\ & & + \left [ \phi_u , t \right ] \left [ t , \phi_u \right ] + \left [ \phi_v , t \right ] \left [ t , \phi_v \right ] \bigg ) \label{mcsqq}\\ \hat{\lambda}(\ve ) & = & -\frac{e^2}{h } \left ( \partial_\ve t \right)\\ \hat{s}_{iq}(\ve ) & = & i \hbar \hat{\lambda}(\ve ) + \frac{e^2}{\pi } \left [ \sqrt{t r } \left ( \phi_u - \phi_v \right ) \right ] \label{multimode}\end{aligned}\ ] ] a similar expression for the charge noise @xmath231 of a mesoscopic conductor was first derived by bttiker @xcite . unlike the expression for the current noise @xmath232 , which can easily be understood in terms of partition noise , it would seem at first that there is no simple , heuristic way to interpret the expression for @xmath231 . however , if we invoke ideas of information , each term in eq . ( [ mcsqq ] ) acquires a simple meaning . the first term represents information associated with the energy dependence of the transmissions ; the second , information associated with the energy dependence of phase differences ; and the last three terms , information associated with the partitioning of electrons into different modes . in general , using eqs . ( [ rates ] ) and ( [ weakia ] ) , we may _ define _ the charge noise in terms of the accessible information @xmath233 in the coupled conductor plus qubit system : @xmath234 while this last expression may seem purely tautological , it is clear that the various contributions to eq . ( [ mcsqq ] ) for the charge noise are best understood in terms of information . note that the accessible information @xmath233 could be obtained directly in the present system by calculating the overlap between the detector states corresponding to the two qubit states . such a calculation would take the form of an orthogonality catastrophe calculation , similar to that presented in ref . . in the remaining part of this paper , we consider a more general version of the mesoscopic scattering detector , showing that the main results of the previous section continue to hold we relax the condition that the state of the qubit modulates a _ uniform _ potential in the scattering region , thus allowing for a wider class of input operators @xmath10 than that given in eq . ( [ qdefn ] ) . in general , we may write : @xmath235 \label{gqdefn}\end{aligned}\ ] ] where @xmath236 is a @xmath237 hermitian matrix having dimensions of inverse energy . the situation considered in the last section corresponds to choosing @xmath238 to be @xmath239 ( eq . ( [ nndefn ] ) ) , which at @xmath240 is just the wigner - smith delay time matrix . by comparing against the current operator @xmath14 ( c.f . ( [ idefn ] ) ) , it is clear that the proportionality condition of eq . ( [ bigcondition ] ) necessary for the quantum limit constrains the diagonal in energy , off - diagonal in lead index part of the potential matrix @xmath238 : @xmath160 , \hspace{0.2 cm } \left[w(e , e)\right]_{r l } = i \frac{1}{\cc } \left[s_{lr } \right]^{\dagger}(e ) s_{ll}^{\pd } ( e ) \label{gscondition}\end{aligned}\ ] ] where @xmath77 is a real constant . we thus see that the required proportionality between @xmath14 and @xmath10 needed to reach the quantum limit at zero temperature leaves a large part of the potential matrix @xmath238 undetermined ( i.e. terms diagonal in the lead index and/or off - diagonal in energy ) . we now show that by considering a form for @xmath238 which is drastically different from @xmath241 , one can make it easier to reach the quantum limit and have a reasonable gain . in particular , one can work at small voltages without necessarily having a vanishing gain . we specialize the discussion to a case which in many ways is the opposite of having global potential coupling . we take the scattering matrix @xmath118 to be energy - independent over the energy interval defined by the voltage , and take @xmath238 to correspond to a local potential : @xmath242 over the energies of interest . in this case , the scattering matrix @xmath118 will have one of two different energy - independent values depending on the state of the qubit : @xmath243 where @xmath244 is the scattering matrix at zero coupling ( @xmath245 ) . the matrix @xmath238 may be directly related to the change in the scattering matrix , @xmath246 ( see appendix b for a derivation ) : @xmath247 note the similarity to the form of @xmath238 in the global - potential coupling case ( where @xmath248 ) ; now , the energy derivative @xmath249 has been replaced by the finite difference @xmath250 . turning to the conditions needed for the quantum limit , we find again that the causality properties of the scattering matrices @xmath251 ensure @xmath150 always . the remaining proportionality requirement of eq . ( [ bigcondition ] ) places constraints on @xmath251 . these have an analogous form to eqs . ( [ mctcondition ] ) and ( [ mcphasecondition ] ) , but now the energy derivative @xmath252 is replaced by the finite difference @xmath253 ( i.e. @xmath254 - x[s_-])/(2 e a)$ ] ) : @xmath255 @xmath256 where @xmath257 , @xmath258 . importantly , _ the above conditions do not involve any energy averaging _ , as we have taken @xmath118 and @xmath238 to be energy independent . nonetheless , there still is a non - vanishing gain @xmath16 determined by both the voltage and the @xmath259 : @xmath260 thus , using a local coupling between the qubit and the scattering detector makes it easier to reach the quantum limit and have a sizeable gain one can use voltages small enough that energy - averaging is not a problem , while still having the qubit modulate the transmissions . note that in the single channel case , all that is needed for the quantum limit is that the state of the qubit not change the difference between reflected and transmitted phases : @xmath261 . also note the various noise correlators are given by eqs . ( [ noises ] ) , with the substitution @xmath262 . we have developed a general set of conditions which are needed for a detector in the linear - response regime to reach the quantum limit of detection . one needs both a restricted proportionality between the input and output operators of the detector ( c.f . eq . ( [ bigcondition ] ) ) , and a causal relation between the output and input ( c.f . ( [ gaincondition ] ) ) . applying the concept of accessible information to the detector , one sees that deviations from the quantum limit imply the existence of missing " information residing in the detector , information which is not being utilized . the general conditions of eqs . ( [ bigcondition ] ) and ( [ gaincondition ] ) ensure the non - existence of such information . applying these concepts to the mesoscopic scattering detector , we find that these general conditions place restrictions on the form of the detector s scattering matrix . these restrictions do not involve symmetry properties usually considered in mesoscopic systems , but are rather best understood as following from the requirement of having no missing information . in the mesoscopic scattering detector , missing information may reside in the relative phase between transmission and reflection , in the energy or mode structure of the transmission probabilities , or in the partitioning of scattered electrons between different modes . surprisingly , we find that an adiabatic point contact conforms to all the conditions needed for the quantum limit , even when the voltage is large enough that many modes are involved in transport , and the energy dependence of scattering is important . we thank m. devoret for useful conversations . this work was partially supported by arda through the army research office , grant number daad19 - 02 - 1 - 0045 , by the nsf under grants no . dmr-0084501 and dmr-0196503 , and by the w. m. keck foundation . in this appendix , we provide a simple proof of eq . ( [ accessible ] ) for the accessible information @xmath111 . given the two states @xmath263 and @xmath264 , the goal is to maximize the classical mutual information @xmath100 ( defined in eq . ( [ reqn ] ) ) over all possible choices of measurements . a given choice of measurement @xmath92 corresponds to a choice of basis ; the probability distributions @xmath265 and @xmath266 are determined by the elements of the corresponding states in this basis . treating the @xmath267 as independent variables restricted to the interval @xmath268 $ ] , and using lagrange multipliers , we minimize @xmath100 subject to the following constraints : @xmath269 the second condition in principle need only be an inequality , with the left - hand side being greater than or equal to the right - hand side ; however , it can be verified that the maximum value of @xmath100 occurs when it is enforced as an equality . also note that without loss of generality , we can choose the inner product appearing in eq . ( [ ipcondition ] ) to be real and positive , as @xmath100 is independent of the relative phase between the states @xmath270 . finally , we have assumed to start that these states have at most @xmath119 non - zero components in the chosen basis . variation with respect to @xmath265 yields the condition : @xmath271 with a similar equation emerging from variation with respect to @xmath272 . @xmath16 , @xmath273 and @xmath274 , are lagrange multipliers ; @xmath275 / 2 $ ] is the averaged distribution . subtracting the @xmath276 and @xmath277 equations yields : @xmath278 where we have defined @xmath279 via @xmath280 @xmath279 may be thought of as the amount of information gained in a measurement _ given _ that the outcome of the measurement is @xmath281 . now , eq . ( [ betaeqn ] ) must hold for each @xmath279 ( @xmath282 ) ; moreover , the function on the right - hand side is symmetric in @xmath279 and monotone decreasing for @xmath283 . it thus follows that for each @xmath169 , @xmath284 the last equality follows from substitution into eq . ( [ ipcondition ] ) . further substitution into eq . ( [ reqn ] ) for @xmath100 yields the expression in eq . ( [ accessible ] ) ; note that the averaged distribution @xmath285 and the relevant number of basis elements @xmath119 do not appear in this expression . one can explicitly check that choosing any of the @xmath267 to be @xmath286 or @xmath287 results in a lower value of @xmath100 ; thus , eq . ( [ accessible ] ) does indeed correspond to the maximum value of @xmath100 and thus , by definition , to the accessible information @xmath111 . the condition eq . ( [ betaeqn2 ] ) required to optimize @xmath100 implies that the amount of information gained via measurement is the same for each of the measurement outcomes @xmath281 . equivalently , each basis element in an optimal basis has the same information content associated with it . this is similar to requirements obtained to have the mesoscopic scattering detector reach the quantum limit ; in that case , each channel and each energy were required to have the same information content ( c.f . ( [ mctcondition ] ) ) . note also that _ there are several distinct choices of bases ( i.e. measurement schemes ) which optimize @xmath100 _ ; this point was not made in ref . . a particularly simple optimal basis can be constructed for @xmath288 . in this basis , the non - zero components of the states @xmath289 are given by : @xmath290 where @xmath291 . by definition , the state @xmath292 leads to the measurement outcome @xmath293 with perfect certainty , while the state @xmath294 leads to the measurement outcome @xmath295 with perfect certainty . in geometric terms , the optimal basis given here is one in which the angle between the two states @xmath296 is bisected by the vector @xmath297 . more generally , consider the form of an optimal basis where @xmath298 ( i.e. there are @xmath299 possible outcomes when a measurement is made on the state @xmath300 or @xmath301 ) . taking @xmath299 to be even for simplicity , and letting @xmath302 denote the basis states , a possible optimal basis is one in which : @xmath303 the fact that there are many possible outcomes of a measurement does not degrade from the optimality of mutual information @xmath100 , as the information associated with each measurement outcome is the same . in this appendix , we provide a brief derivation of eq . ( [ deltaseqn ] ) which relates the coupling potential matrix @xmath238 ( c.f . ( [ gqdefn ] ) ) to the associated change in the scattering matrix , @xmath246 . the latter quantity determines the noise correlators and gain of the local - potential coupling version of the mesoscopic scattering detector . our approach is similar to that used in ref . to relate the scattering matrix of a quantum dot to its hamiltonian . in what follows , we assume ( as in sec . ii b ) that the potential matrix @xmath238 and the zero - coupling scattering matrix @xmath118 are independent of energy on the scales of interest . we start by writing the system hamiltonian in terms of the scattering states of problem at zero coupling , assuming the qubit is frozen in the @xmath276 state : @xmath304\end{aligned}\ ] ] we have assumed a linear dispersion near the fermi energy , with @xmath305 and @xmath306 representing the deviation of the momentum from the fermi momentum . we have also neglected the fact that the effective fermi velocity is channel dependent ( @xmath307 drops out of all final expressions ) . the operator @xmath308 creates a scattering state incident in the lead and transverse mode indexed by @xmath309 . for definiteness , we take our leads ( both left and right ) to be defined only on the half - line @xmath310 , and to be confined in the @xmath98 and @xmath311 directions . further , we assume that the scattering region is situated on @xmath312 . we may write the full electron field operator in terms of the @xmath313 operators , using the zero - coupling scattering matrix @xmath118 . writing @xmath314 , we have : @xmath315 \\ & = & \frac{1}{\sqrt{2}}\sum_m \psi_m(-x ) e^{i k_f x } \phi_m(y , z ) + \nonumber \\ & & \frac{1}{\sqrt{2 } } \sum_{m , n } \psi_m(x ) e^{-i k_f x } \phi_n(y , z ) s_{n m } \label{auxfieldeqn}\end{aligned}\ ] ] in the last line , we have introduced the operators @xmath316 , which are the fourier transforms of the scattering state operators @xmath313 . note again that this expression is only valid for @xmath317 , as the leads are only defined on @xmath317 . we thus see that for @xmath317 , @xmath316 describes an _ outgoing _ ( i.e. left - moving ) wave , while @xmath318 describes an _ incoming _ ( i.e. right - moving ) wave . next , we may express the system hamiltonian in terms of the @xmath316 operators . this in turn leads to an equivalent single - particle schrdinger equation : @xmath319 \label{schrodinger}\end{aligned}\ ] ] here , @xmath320 is a wavefunction which arises when the field operator @xmath316 is expressed in terms of operators corresponding to the eigenmodes of the full hamiltonian @xmath321 . given the relation of @xmath316 to incoming and outgoing waves ( c.f . ( [ auxfieldeqn ] ) ) , we choose the following form for @xmath322 : @xmath323 where @xmath324 . substituting this form into eq . ( [ auxfieldeqn ] ) , we see that the coefficients @xmath325 and @xmath326 do indeed correspond ( respectively ) to the amplitudes of incoming and outgoing waves . integrating eq . ( [ schrodinger ] ) from @xmath327 to @xmath328 , interpreting @xmath329 as @xmath330/2 $ ] , and then using eq . ( [ ansatz ] ) , we find the following relation between the amplitude of incoming and outgoing waves : @xmath331_{n n ' } a_{\textrm{in},n ' } \\ & \equiv & \sum_{n ' } \left[s + a e \delta s \right]_{m n ' } a_{\textrm{in},n'}\end{aligned}\ ] ] in the last line , we indicate that this relation defines the new scattering matrix @xmath332 which includes effects of the additional potential @xmath238 . expanding to lowest order in the dimensionless potential @xmath333 , we find eq . ( [ deltaseqn ] ) as advertised . y. makhlin _ et al . _ , lett . * 85 * , 4578 ( 2000 ) ; _ ibid . phys . * 73 * , 357 ( 2001 ) . m. h. devoret and r. j. schoelkopf , nature ( london ) * 406 * , 1039 ( 2000 ) . d. v. averin , quant - phys/0008114 ; cond - mat/0004364 . d. v. averin and a. n. korotkov , phys . b * 64 * , 165310 l. i. glazman , g. b. lesovik , d. e. khmelnitskii and r. i. shekhter , jetp lett . * 48 * , 238 ( 1988 ) . s. a. gurvitz , phys . b * 56 * , 15215 ( 1997 ) . l. stodolsky , phys . b * 60 * , 5737 ( 1999 ) . a. n. korotkov and d. v. averin , cond - mat/0002203 . l. b. levitin in _ quantum communication and measurement _ ( v. p. belavkin , o. hirota , and r. l. hudson , eds . ) , ( new york : plenum ) , 439 , 1995 . c. a. fuchs , quant - ph/9601020 ( 1996 ) ; c. a. fuchs and a. peres , phys . a * 53 * , 2038 ( 1996 ) ; c. a. fuchs and k. jacobs , phys . rev . a 63 , 062305 ( 2001 ) . for simplicity , we do not consider self - consistency effects which can become important when the current is no longer linear with voltage , see e.g. t. christen and m. bttiker , europhys . lett . * 35 * , 523 ( 1996 ) . note that our eq . ( [ tcondition ] ) has a constant on the right - hand side , while the corresponding equation in ref . ( eq . ( 12 ) ) has an arbitrary energy - dependent function on the right - hand side . this difference results from the fact that ref . considers voltages so small that the effects of energy averaging can be neglected . note that by left - right symmetry , we mean that the _ scalar _ potential @xmath336 in the scattering region is a symmetric function of @xmath337 , where @xmath337 is the transport direction . one can still have an absence of time - reversal symmetry if , e.g. , there is a vector potential in the scattering region .

in this paper we discuss the results of a programme of spectral synthesis modelling of a sample of starburst radio galaxies in the context of scenarios for the triggering of the activity and the evolution of the host galaxies . new optical spectra are also presented for a subset of the objects discussed . the starburst radio galaxies comprising 25% of all powerful extragalactic radio sources frequently show disturbed morphologies at optical wavelengths , and unusual radio structures , although their stellar masses are typical of radio galaxies as a class . in terms of the characteristic ages of their young stellar populations ( ysp ) , the objects can be divided into two groups : those with ysp ages gyr , in which the radio source has been triggered quasi - simultaneously with the main starburst episode , and those with older ysp in which the radio source has been triggered or re - triggered a significant period after the starburst episode . most of former group are associated with large mid- to far - ir ( mfir ) continuum and [ oiii ] emission line luminosities ( l , w ) , while most of the latter have lower luminosities . combining the information on the ysp with that on the optical morphologies of the host galaxies , we deduce that the majority of the starburst radio galaxies have been triggered in galaxy mergers in which at least one of the galaxies is gas rich . however , the triggering ( or re - triggering ) of the radio jets can occur immediately before , around , or a significant period after the final coalescence of the merging nuclei , reflecting the complex gas infall histories of the merger events . although% of starburst radio galaxies are sufficiently bright at mfir wavelengths to be classified as ultra - luminous infrared galaxies ( ulirgs ) , we show that only the most massive ulirgs are capable of evolving into radio galaxies . finally , for a small subset of starburst radio galaxies in rich clusters of galaxies , cooling flows associated with the hot x - ray haloes offer a viable alternative to mergers as a trigger for the radio jet activity . overall , our results provide further evidence that powerful radio jet activity can be triggered via a variety of mechanisms , including different evolutionary stages of major galaxy mergers ; clearly radio - loud agn activity is not solely associated with a particular stage of a unique type of gas accretion event . [ firstpage ] galaxies : active galaxies : jets galaxies : starburst galaxies : interactions

1011.2096

active galactic nuclei ( agn ) are increasingly recognised as a key element in the overall galaxy evolution process ( e.g. * ? ? ? * ) , yet the nature of the link between agn and galaxy evolution is not understood in depth . the link can take two main forms : the triggering of agn as gas is driven into the nuclear regions of galaxies as they evolve ; and the direct impact of the agn - driven outflows on the ism in the galaxy bulges , haloes and any surrounding groups and clusters . the latter agn feedback effect has been invoked to explain the correlations between black hole mass and host galaxy properties ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , as well as the high - end shape of the galaxy luminosity function @xcite . on the other hand , the triggering of agn activity as a result of evolution - driven gas accretion can provide an explanation for the similarity between the redshift evolution of agn activity and that of the global star formation rate of the field galaxy population @xcite . clearly , if we are to incorporate the agn feedback effect accurately into galaxy evolution models , it is important to understand how and when agn are triggered as their host galaxies evolve . deep ground - based imaging observations of powerful radio galaxies show morphological evidence ( tidal tails , fans , bridges and shells etc . ) that a large fraction are triggered in galaxy interactions and mergers @xcite . a merger origin is also supported by the kinematics of the extended emission line regions in some sources @xcite . however , the idea that the activity is always triggered in major galaxy mergers and interactions has been challenged by high resolution hubble space telescope ( hst ) images which provide evidence that the host galaxies of powerful agn of both radio - loud and radio - quiet types are relatively quiescent giant elliptical galaxies @xcite . at least part of this apparent discrepancy between the hst and ground - based results is likely to be a consequence of the fact that shallow hst observations are sensitive to high surface brightness structures close to the nuclei , while the lower resolution ground - based studies are more sensitive to lower surface brightness structures on larger scales ( see discussion in * ? ? ? * ; * ? ? ? while some radio galaxies show clear evidence for triggering in galaxy interactions and mergers , a significant subset have optical morphologies and emission line kinematics that do not support such a triggering mechanism . many such cases are central cluster galaxies surrounded by massive haloes of hot gas @xcite . therefore the infall of warm / cool gas condensing from the hot x - ray haloes in cooling flows has been been suggested as an alternative trigger for the agn activity ( e.g. * ? ? ? in addition , based both on detailed x - ray studies of individual objects and optical studies of large samples of low luminosity radio sources , the direct ( bondi ) accretion of the _ hot _ ism associated with the x - ray haloes has been proposed to explain the fuelling of weak line radio galaxies ( wlrg ) in the local universe @xcite . although the morphologies and emission line kinematics provide important information about the triggering events , this information is generally qualitative . for example , it is difficult to use such information to quantify when agn are triggered in major galaxy mergers relative to the time of coalescence of the merging nuclei . fortunately , another facet of the gas accretion events that trigger agn activity is that they are also likely to be associated with star formation in the host galaxies ; therefore , by studying the young stellar populations in the predominantly early - type host galaxies of powerful radio sources , we can obtain key information about the nature of the triggering events . substantial recent progress has been made in quantifying the degree of recent star formation activity in radio galaxies using both optical / uv observations of direct starlight @xcite and far - ir observations of dust - reprocessed starlight @xcite . concentrating on optical / uv spectroscopic studies of complete samples that take full account of contamination of the continuum by agn - related components ( e.g. scattered or direct agn light , nebular continuum ) , as well as intrinsic reddening of the starlight , young stellar populations ( ysp ) are detected spectroscopically in 15 30% of powerful 2jy and 3cr radio galaxies at low and intermediate redshifts @xcite ; the major uncertainties in these numbers are due to objects in which the strength of the direct or scattered agn component makes detection of the ysp difficult . despite the possibility that a substantial proportion of ysp emission may be absorbed by dust at optical / uv wavelengths , analysis of the mid- to far - ir ( mfir ) continuum properties of powerful radio galaxies derived from deep spitzer observations yields a similar proportion of powerful radio galaxies with energetically significant recent star formation activity ( @xmath015 28% : tadhunter et al . 2007 , dicken et al . , 2008 , 2009)10 less at uv wavelengths than optical wavelengths , such observations are sensitive to relatively low levels of recent star formation activity that may be insignificant in terms of their contribution to the masses and energetics of the host galaxies . ] . the identification of ysp in a significant fraction of nearby radio galaxies has opened the possibility of using spectral synthesis modelling of the stellar populations to investigate the agn triggering events and the evolution of the host galaxies . initial results provided a mixed picture . on the one hand , tadhunter et al . ( 2005 ) and @xcite found evidence for intermediate ysp ages in their small sample of four nearby radio galaxies , pointing to a significant delay between the starbursts and the agn activity , similar to those reported for samples of star forming galaxies and radio - quiet agn in the local universe ( e.g. * ? ? ? * ; * ? ? ? on the other hand , a detailed study by @xcite of two intermediate redshift radio galaxies demonstrated the presence of much younger stellar populations in one object , as well as intermediate age stellar populations in both objects . all the spectral synthesis modelling studies of radio galaxies have stressed the following common features . * * masses . * the masses associated with the ysp detected spectroscopically in radio galaxies are significant ( @xmath5 m@xmath3 ) , amounting to typically 1 40% of the total stellar masses of the host galaxies ; if the starbursts and agn are triggered in mergers , then the systems with the more massive ysp must be associated with relatively major mergers , in which at least one of the merging galaxies is gas - rich . * * spatial distribution . * when substantial ysp are detected in the nuclear regions of the radio galaxies , in general they are also detected across the full extents of the galaxies over which it is possible to make spectroscopic observations with the high s / n required for spectral synthesis modellling ( e.g. * ? ? ? * ) . in some cases , the ysp are detected on scales of 10s of kpc . again this is consistent with simulations of gas rich mergers that predict galaxy wide star formation activity ( e.g. * ? ? * ; * ? ? ? * ) , although it does not preclude other triggering mechanisms . * * reddening . * reddening of the optical / uv light of the ysp by dust is important , especially in the nuclear regions of the galaxies ( typically @xmath6 ) ; failure to take such reddening into account leads to inaccurate ages for the ysp and , in particular , to the over - estimation of the ages of the ysp . in this paper we draw together and summarise the results of all the recent spectral synthesis studies of radio galaxies that are based on high quality optical spectra , including the recent investigation of 12 low and intermediate redshift radio galaxies by @xcite . new spectral synthesis results are also presented for a subset of the sample . the properties of the ysp are then discussed in the context of the morphologies of the host galaxies , in order to investigate the nature , timescales , and order of events of the triggering of the activity of an important subset of the population of galaxies with powerful agn . throughout this paper we assume a cosmology with @xmath7 km s@xmath8 , @xmath9 , and @xmath10 . optical and mfir observations provide evidence for recent star formation activity in up to 30% of all radio galaxies . however , in this paper we concentrate on the @xmath015 25% of the radio galaxy population with clear , unambiguous evidence for young stellar populations at optical wavelengths , which have been modelled using spectral synthesis techniques that take full account of the reddening and potential agn contamination . these latter criteria are important because of the known contamination by scattered or direct agn light and/or nebular continuum from the nlr in some objects , the strong evidence that the ysp in the nuclear regions are significantly reddened in many cases , and potential degeneracy issues when more than one stellar population is present ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for a full discussion of these issues ) . our sample comprises all 18 of the objects fulfilling these selection criteria from our own detailed studies of nearby 2jy , 3cr and b2 radio galaxies @xcite , along with 3c48 , which has been modelled using a similar approach . in addition , we include a further two nearby objects from the literature with good spectroscopic observations of individual star clusters ( centaurus a and fornax a ) ; such clusters do no suffer from the degeneracy issues associated with multiple stellar populations and agn contamination . for convienience we label all the objects in our sample collectively as ` starburst radio galaxies ' , with the caveat that in some cases the star formation histories may have been more complex than a single starburst . although not in any sense complete , our sample includes all 7 ( 15% ) of the 2jy sample of 46 southern radio galaxies with intermediate redshifts ( @xmath11 ) described in dicken et al . ( 2008 ) , and all 5 ( 25% ) of the low redshift ( @xmath12 ) sample of 19 3crr frii radio galaxies described in dicken et al . ( 2010 ) , that show strong evidence for recent star formation activity at optical wavelengths . therefore it is representative of the most extreme star forming radio galaxies in the local universe ; in these objects the ysp typically contribute @xmath1320% of the total light at @xmath04800 . we emphasise that lower levels of star formation activity may be present in a larger fraction of the local radio galaxy population , but not clearly detected at optical wavelengths because of dilution by the light of the old stellar populations of the host galaxies and/or agn - related continuum components . the general properties of the starburst radio galaxies are presented in table 1 , while the properties of their ysp , as derived from spectral synthesis modelling , are detailed in table 2 . note that , with the exception of the individual star clusters in cen a and fornax a , the results presented in table 2 refer to the simplest models that provide an adequate fit to the overall optical / near - uv seds and detailed absorption features ( see * ? ? ? * ; * ? ? ? * for a detailed description of the modelling approach ) . these comprise a reddened ysp combined with an unreddened old stellar population ( osp : @xmath14 gyr ) . because of the potential degeneracy issues , related to the relatively large contribution from osp and reddening of the ysp , more detailed fits are not , in general , warranted . the exception is the case of 3c459 discussed by @xcite in which the dominance of the young stellar component relative to the osp allows two ysp components of different age ( young : @xmath15 gyr ; and intermediate : @xmath16 gyr ) to be fitted to the data , although the results in table 2 for that object refer to the single reddeneded ysp plus osp fit . overall , the results presented in table 2 represent the luminosity - weighted ages of the ysp present in each system . llrrrll + object & redshift & p@xmath17 & l@xmath18}$ ] & l@xmath19 & optical morphology & radio morphology + & & ( w hz@xmath8 ) & ( w ) & ( @xmath20 ) & + + pks0023 - 26 & 0.322 & @xmath21 & @xmath22 & @xmath23 & e , a , ce & css + ngc612 & 0.030 & @xmath24 & @xmath25 & @xmath26 & s0 , d , s & fri / frii + 3c48 & 0.367 & @xmath27 & @xmath28 & @xmath29 & q , 2n , a , tt & css + fornax a & 0.006 & @xmath30 & @xmath31 & @xmath32 & e / s0 , d , s & frii + pks0409 - 75 & 0.693 & @xmath33 & @xmath34 & @xmath35 & e , 2n & frii + b2 0648 + 27 & 0.041 & @xmath36 & @xmath37 & @xmath38 & e , r , de , tt & css + pks0620 - 52 & 0.0511 & @xmath39 & @xmath40 & @xmath41 & e & fri + 3c213.1 & 0.194 & @xmath42 & @xmath43 & @xmath44 & e , de & fri / frii , dd + 3c218 & 0.055 & @xmath45 & @xmath46 & @xmath47 & e , d & fri , hsbi+do + 3c236 & 0.101 & @xmath48 & @xmath49 & @xmath50 & e , s , d , tt , de & frii , dd + 3c285 & 0.079 & @xmath51 & @xmath52 & @xmath26 & e , tt , f , b & frii + centaurus a & 3.4mpc & @xmath53 & @xmath54 & @xmath55 & e / s0 , d , s , r & fri , hsbi+do + pks1345 + 12 & 0.122 & @xmath56 & @xmath57 & @xmath58 & e , 2n , d , tt & gps , hsbi+do + 3c293 & 0.045 & @xmath59 & @xmath60 & @xmath61 & e / s0 , tt , d , b & fri / frii , hsbi+do + 3c305 & 0.041 & @xmath62 & @xmath52 & @xmath63 & e / s0 , tt , d & css + 3c321 & 0.096 & @xmath64 & @xmath34 & @xmath65 & e , 2n , tt , s , f & frii , ssc + pks1549 - 79 & 0.152 & @xmath66 & @xmath67 & @xmath68 & e , tt , d & css / cfs + pks1932 - 46 & 0.231 & @xmath69 & @xmath70 & @xmath71 & e , s , a & frii + 3c433 & 0.106 & @xmath45 & @xmath72 & @xmath73 & e , a , ce , d & fri / frii , ssc + pks2135 - 209 & 0.636 & @xmath74 & @xmath75 & @xmath76 & q , f & css + 3c459 & 0.220 & @xmath77 & @xmath78 & @xmath79 & e , f , de & frii , ssc + + llllll + object & region & ysp age & ysp reddening & comment & references + & & ( @xmath80 ) & e(b - v ) & & + + pks0023 - 26 & n & 0.03 0.05 & 0.9 0.7 & & 1 + ngc612 & n & 0.05 0.1 & 1.4 1.2 & & 1 + & e : a f & 0.04 0.2 & 1.0 0.0 & extended disk & 1 + 3c48 & e : 3c48 a & 0.14 & 0 & secondary nucleus & 2 + & e : a , b , c , g & 0.005 0.114 & & several regions , 4 slits & 3 + fornax a & e & 2.5 3.5 & & luminous star clusters & 4 + & e & 2.0 & & diffuse light , line index work , luminosity weighted & 5 + & n , e & 0 2 & 0 1.5 & diffuse light , sed modelling & 6 + pks0409 - 75 & n & 0.01 0.04 & 0.9 0.5 & & 1 + b2 0648 + 27 & n & 0.2 0.4 & 0.2 0.4 & & 7 + & e : ne & 0.3 0.6 & 0.0 0.1 & & 7 + & e : sw & 0.3 0.6 & 0.0 0.1 & & 7 + pks0620 - 52 & n & 0.001 0.9 & 0.0 1.5 & & 6 + 3c213.1 & n & 0.4 0.8 & 0.1 0.0 & & 10 + 3c218 & n & 0.05 & 0.4 & & 1 + 3c236 & n & 0.04 1.0 & 1.6 0.6 & & 1 + & e & 0.01 , 0.1 1 & & extended star formation complexes , photometric & 8 + 3c285 & n & 0.1 0.5 & 0.2 0.0 & & 1 + centaurus a & e & 0.35 & & blue tidal stream & 12 + & e & 2 8 & & globular cluster system & 13 + pks1345 + 12 & n & @xmath810.06 & 0.8 1.3 & & 15 + & e , e1 & 0.5 1.5 & & & 14 + 3c293 & n & 1.0 2.5 & 0.5 0.2 & & 14 + & e , e1 & 0.1 0.5 & & & 14 + & e , e2 & 1.0 2.5 & & & 14 + 3c305 & n & 0.4 1.0 & 0.8 0.4 & & 14 + & e1 & 0.5 1.5 & & & 14 + & e2 & 0.5 1.0 & & & 14 + 3c321 & n , se & 0.1 1.0 & 0.3 0.1 & & 1 + & n , nw & 0.1 1.4 & 0.3 0.1 & radio source host & 1 + & e & 0.4 1.0 & 0.1 0.0 & extended regions between nuclei and to se & 1 + pks1549 - 79 & n & 0.04 0.1 & 0.75 0.0 & & 9 + & e & 0.04 0.08 & 0.4 0.0 & extended envelope to s & 6 + pks1932 - 46 & e & @xmath820.01 & & extended star forming halo detected in emission lines & 11 + 3c433 & n , sw & 0.03 0.1 & 0.7 0.4 & radio source host & 1 + & n , ne & 0.03 1.4 & 0.9 0.3 & companion galaxy & 1 + & e , mid & 0.05 1.0 & 0.7 0.0 & mid way between 3c433 & companion & 1 + pks2135 - 209 & n & 0.04 0.6 & 0.5 0.2 & quasar nucleus modelled with power - law & 1 + pks2314 + 03 & n & 0.04 0.07 & 0.4 0.0 & & 10 + & e , n & 0.5 0.7 & 0.3 0.0 & fit includes a power - law & 10 + & e , e & 0.5 0.6 & 0.2 0.0 & fit includes a power - law & 10 + + in addition to spectroscopic observations already reported in the literature and described in the appendix , new spectra for some of the starburst radio galaxies were obtained in various runs on the eso very large telescope ( vlt ) , 3.6 m and new technology telescope ( ntt ) telescopes . details of these new observations are given in table 3 . the reduction of these data followed the standard steps of bias subtraction , flat fielding , wavelength calibration , atmospheric extinction correction , flux calibration , and registration of blue and red spectral images . based on measurements of night sky lines , the uncertainty in the wavelength calibration is estimated to be smaller than 0.5 for all datasets ; the relative flux calibration uncertainty , based on the comparison of observations of several spectroscopic standard stars , is better than @xmath835% . prior to the analysis of the data , the spectra were corrected for galactic extinction using values of e(b - v ) derived from @xcite along with the @xcite extinction law . note that , in the cases of the vlt / fors1 observations of pks0023 - 26 and pks1549 - 79 , the data were taken in spectropolarimetric mode with the light passing through a half - wave plate as well as an analysing prism . in the latter two cases , the ` o ' and ` e ' ray images of the long - slit spectra were spatially registered and co - added , and the flux calibration was performed using spectropolarimetric standard stars observed through the same instrument configuration as the target galaxies . for some of the objects discussed in this paper ( fornax a , pks0652 - 20 , pks1549 - 79 ) we have carried out new spectral synthesis modelling in an attempt to improve our knowledge of the properties of the young stellar populations ( ysp ) . in these cases we followed the procedures outlined in @xcite , @xcite , and @xcite : modelling the continuum spectrum using a combination of a reddened ysp ( @xmath84 gyr , @xmath85 ) and an unreddened osp ( @xmath86 gyr ) , taking full account of agn continuum components ( where appropriate ) . in all cases we used instantaneous burst , solar metallicity spectral synthesis templates from @xcite . lllllll + object & date & tel / inst & grism & exposure & slit & seeing + & & & & time ( s ) & width ( arcsec ) & fwhm ( arcsec ) + + pks0023 - 26 & 24/07/03 & vlt / fors1 & gris-600r & 5580 & 1.3 & 0.70 + fornax a & 7/12/02 & ntt / emmi & gr5 & 900 & 1.0 & 1.1 + & & & gr6 & 900 & & + pks0620 - 52 & 16/02/07 & eso3.6m / efosc & gr7 & 3600 & 1.2 & 1.7 + & & & gr4 & 1800 & & + pks1549 - 79 & 24/07/03 & vlt / fors1 & gris-600b & 5880 & 1.3 & 0.65 + + detailed discussions of the individual starburst radio galaxies , including the new spectroscopic results , are presented in the appendix . in this section we consider the general results and patterns obtained for the population of starburst radio galaxies as a whole . it is striking that a large proportion of starburst radio galaxies show unusual radio morphologies that place them outside the regular fri / frii morphological classification for extended radio sources : 7 ( 33% ) are compact steep spectrum ( css ) or gigahertz peaked ( gps ) sources whose radio structures are dominated by structures with a diameter @xmath87kpc ( pks0023 - 26 , 3c48 , b2 0648 + 27 , pks1345 + 12 , 3c305 , pks1549 - 79 , pks2135 - 209 ) ; 3 ( 14% ) show unusually prominent compact steep spectrum core components on a scale @xmath88kpc , even if their radio emission is dominated by radio lobes and hotspots on a larger scale ( 3c321 , 3c433 , 3c459 ) ; 6 ( 29% ) show inner high surface brightness steep spectrum structures along with lower surface brightness outer haloes or double structures ( 3c213.1 , 3c218 , 3c236 , cen a , pks1345 + 12 , 3c293 ) ; and 15 ( 71% ) show one or more of these peculiarities . for comparison , the rate of detection of such features in the southern 2jy sample of radio galaxies with redshifts in the range @xmath11 ( see dicken et al . 2008 for sample definition ) is only 28% . although our small sample size makes it difficult to find clear trends in the detailed radio morphologies with the properties of the stellar populations , it is notable that all but one ( 3c305 ) of the 7 css / gps sources in our sample have nuclear spectra that are consistent with relatively young ages for their ysp ( @xmath15 gyr ) . at optical wavelengths most of the objects in our sample of starburst radio galaxies show morphological peculiarities compared with quiescent elliptical galaxies : 17 ( 80% ) show tidal tails , fans , or highly asymmetric / clumpy outer envelopes at relatively high surface brightness levels ; 14 ( 67% ) show dust features ; 5 ( 23% ) have double nuclei or close companions within 15 kpc ; and 20 ( 95% ) show one or more of these optical peculiarities . this rate of incidence is much higher than in the general population of massive elliptical galaxies observed with similar surface brightness sensitivity ( e.g. * ? ? ? however , at this relatively crude level of morphological classification , a similar rate of morphological disturbance has recently been found in the general population of powerful 2jy radio galaxies at intermediate redshifts ( including non - starburst objects : ramos almeida et al . 2010 ) . the ages of the ysp detected in radio galaxies can provide key information about the order - of - events and the triggering of the agn / jet activity . therefore it is interesting to examine the distribution of luminosity - weighted ysp ages determined from the two component fits to the optical spectra of the full sample of starburst radio galaxies described in section 2 . the top panel of figure 1 shows the distribution of luminosity - weighted ages for the nuclear apertures of the starburst radio galaxies . for comparison we also show the ysp age distribution for a complete sample of ulirgs with redshifts @xmath89 representing the extreme starburst population in the local universe which have been modelled using identical techniques by @xcite . note that , in some apertures of some starburst radio galaxies it proved possible to model the optical spectra with ysp covering a wide range of ages . in such cases we have used the mean age over the range of models that provided good fits . since the upper limiting ages in the latter group were often large ( @xmath131gyr ) , this will tend to skew the age distribution to larger ages . several features are apparent from table 2 and figure 1 . * * nuclear apertures . * considering the nuclear apertures ( figure 1 top ) , 9 ( 43% ) of sources have relatively young ysp ages : @xmath15 gyr . this group is interesting because the typical timescale of the starburst in a major galaxy merger is @xmath00.1 gyr ( e.g. * ? ? ? * ) , and powerful extragalactic radio sources are generally considered to have maximum lifetimes in the range 0.01 0.1 gyr . for such objects the results are consistent with the starburst and agn / jet activity occurring quasi - simultaneously , with little evidence for a major time lag between the starburst and the agn / jet activity . on the other hand , there is also a group of 8 objects ( 36% of sample ) with well - determined older ysp ages ( @xmath90 gyr ) , in which the agn / jet activity appears to have been triggered or re - triggered a significant period _ after _ the major star formation episode in the nuclear regions ( see also the discussion in section 5.1 ) . the latter group includes 3c305 and 3c293 , which were discussed in detail in @xcite . * * extended vs. nuclear apertures . * considering only the apertures with well - determined ysp ages , there is little evidence for a gradient in the ysp ages from the nuclear to the extended regions ; in most cases the nuclear and extended apertures have luminosity - weighted ysp ages that are consistent given the ( sometimes large ) uncertainties . the exception is 3c459 for which the intermediate - aged ysp component becomes more significant off - nucleus @xcite , in common with the trend found for other ulirgs in the local universe @xcite . * * comparison with ulirg ysp ages . * the overwhelming majority of nuclear apertures in the ulirgs modelled by @xcite yield relatively young luminosity - weighted ysp ages : 88% have ysp ages @xmath15 ( figure 1 bottom ) . this is consistent with the group of 9 radio galaxies indentified above with younger ysp ages . indeed , four of the radio galaxies with younger ysp ages 3c48 , pks1345 + 12 , pks1549 - 79 , 3c459 qualify as ulirgs based on their far - infrared luminosities . moreover , our modelling is consistent with young ysp ages for the one other radio galaxy in our sample with a ulirg - like far - ir luminosity pks2135 - 20 even if the ysp age in that case can not be determined with any accuracy , due to the presence of strong nuclear agn emission . * * links with level of agn activity . * it is notable that _ all _ of the objects that have both highly powerful radio emission with @xmath91 w hz@xmath8 and luminous [ oiii ] line emission with @xmath92 } \ge 10^{35}$ ] w luminosity corresponds to luminous quasar - like nuclear activity ( e.g. * ? ? ? ] have spectra consistent with young ysp ages ( @xmath15 gyr ) . all 7 of these objects ( 33% of the sample ) with powerful jets and luminous agn are also luminous at far - ir wavelengths ( @xmath93 w ) . on the other hand , all but one of the objects ( ngc612 is the exception ) that have low radio powers with @xmath94 w hz@xmath8 have spectra consistent with older , intermediate - age ysp . thus , there appears to be a link between the level of agn / jet activity and the age of the ysp , although the relationship is not perfect . objects that do not fit in with the general trend include 3c321 , which has highly luminous [ oiii ] and far - ir emission but intermediate age ysp , and 3c218 which has weak emission lines but much younger ysp . interestingly both of the latter objects fall off the correlation between radio power and emission line luminosity : 3c321 is unusually emission line luminous for its radio power , whereas 3c218 has an unusually low emission line luminosity for its relatively high radio power , perhaps as a consequence of the effect of its rich cluster environment boosting its radio luminosity . we further note that a substantial subset of radio galaxies with quasar - like levels of nuclear activity show no evidence for sgnificant recent star formation activity ; these objects are discusssed further in section 5.4 . llllllll + object & ysp model & mag & ref & @xmath95 & @xmath96 & @xmath97 & ysp mass contr . + & ( age , e(b - v ) ) & type & & ( @xmath98 ) & ( @xmath98 ) & ( @xmath98 ) & + + pks0023 - 26 & ( 0.03,0.9 ) & @xmath99 & 1 & @xmath100 & @xmath101 & @xmath102 & 17% + ngc612 & ( 0.05,1.2 ) & v@xmath103 & 2 & @xmath104 & @xmath105 & @xmath106 & 18% + 3c48 & ( 0.1,0)@xmath107 & f555w & 3 & @xmath108 & @xmath109 & @xmath110 & 2% + fornax a & & r@xmath111 & 4 & & @xmath112 & @xmath112 & + pks0409 - 75 & ( 0.02,0.7 ) & @xmath113 & 1 & @xmath114 & @xmath115 & @xmath105 & 5% + pks0620 - 52 & ( 0.7,0.0 ) & r@xmath116 & 7 & @xmath117 & @xmath118 & @xmath118 & 1% + b20648 + 27 & ( 0.3,0.3 ) & m@xmath119 & 5 & @xmath26 & @xmath120 & @xmath121 & 28% + 3c213.1 & ( 0.6,0.0 ) & @xmath99 & 6 & @xmath47 & @xmath122 & @xmath123 & 11% + 3c218 & ( 0.05,0.4 ) & r@xmath116 & 7 & @xmath124 & @xmath125 & @xmath126 & 4% + 3c236 & ( 0.05,1.4 ) & @xmath99 & 6 & @xmath38 & @xmath109 & @xmath127 & 28% + 3c285 & ( 0.2,0.2 ) & @xmath99 & 6 & @xmath128 & @xmath102 & @xmath129 & 17% + centaurus a & & r@xmath103 & 8 & & @xmath130 & @xmath130 & + pks1345 + 12 & ( 0.05,0.8 ) & r & 11 & @xmath131 & @xmath112 & @xmath132 & 8% + 3c293 & ( 2.0,0.4 ) & @xmath99 & 6 & @xmath104 & @xmath133 & @xmath120 & 57% + 3c305 & ( 0.4,0.6 ) & @xmath99 & 6 & @xmath26 & @xmath134 & @xmath135 & 19% + 3c321 & ( 0.6,0.2 ) & @xmath99 & 6 & @xmath124 & @xmath136 & @xmath137 & 6% + pks1549 - 79 & ( 0.05,0.4 ) & r & 9 & @xmath138 & @xmath104 & @xmath104 & 3% + pks1932 - 46 & & @xmath99 & 1 & & @xmath73 & @xmath73 & + 3c433 & ( 0.05,0.7 ) & v@xmath139 & 10 & @xmath63 & @xmath134 & @xmath140 & 7% + pks2135 - 209 & ( 0.2,0.2 ) & @xmath113 & 1 & @xmath141 & @xmath142 & @xmath143 & 15% + 3c459 & ( 0.05,0.2 ) & @xmath99 & 1 & @xmath144 & @xmath104 & @xmath122 & 8% + + in order to compare the starburst radio galaxies with other types of star forming galaxies in the local universe it is important to consider their stellar masses . we have estimated the total stellar masses by using the monochromatic continuum fluxes derived from the best available total photometric magnitudes to scale the stellar masses derived from our spectral synthesis modelling of the spectra . in each case we used the best - fitting ysp+osp model to derive the stellar masses for the nuclear spectroscopic aperture , then assumed that the same model was appropriate for the entire stellar halo of the galaxy sampled by the photometric aperture . this assumption is justified on the basis that our spectra show little evidence for major gradients in the stellar populations across the halos of most of the galaxies in our sample ( e.g. * ? ? ? * ) . although the major source of uncertainty in this procedure is likely to be related to the choice of a single stellar model for each galaxy , experiments involving the adoption of extreme models ( e.g. assuming that the halos are dominated by osps ) suggest that the derived total stellar masses are accurate to within a factor of two . the results for the starburst radio galaxies are presented in table 4 and figure 2 . they demonstrate that most starburst radio galaxies are giant elliptical galaxies in terms of their total stellar masses . considering the galaxy mass function derived by @xcite m@xmath3 , adapted to our cosmology ] , the majority of starburst radio galaxies are super - m@xmath145 : the mean and median stellar masses are @xmath146 m@xmath3 ( 4@xmath147m@xmath145 ) and @xmath140 m@xmath3 ( 3.3@xmath147m@xmath145 ) . these results are consistent with those derived for the general population of powerful radio galaxies at all redshifts ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? for example , based on the near - ir photometry measurements presented in @xcite , the general population of nearby ( @xmath148 ) 2jy radio galaxies has stellar masses in the range @xmath149 m@xmath3 . however , as highlighted in @xcite , it is dangerous to generalise , and a minority of radio galaxies have stellar masses that are significantly lower ( @xmath150 ) . figure 2 also shows a comparison between the total stellar masses of starburst radio galaxies ( top panel ) and nearby ulirgs from the @xmath89 complete sample of @xcite ( bottom panel ) , derived using identical techniques . in making this comparison it is important to recognise that the estimated total stellar masses of the ulirgs have been derived using osp+ysp models ( modelling combination i in * ? ? ? * ) , which tend to maximise the osp combination ( and hence the total stellar masses ) , whereas in fact the nuclear spectra of many of ulirgs can be adequately modelled by combinations of intermediate age and young ysp , without any contribution from an osp ( combination iii in * ? ? ? * ) , similar to the case of 3c459 amongst the starburst radio galaxies @xcite . despite the fact that the masses of ulirgs in figure 12 may be over - estimated , most radio galaxies have total stellar masses that are greater than those of most ulirgs , although there is a significant overlap between the two distributions . using a k - s two sample test we can reject the null hypothesis that the two samples have the same mass distributions at the @xmath1399.5% level of significance . overall , this comparison provides a clear demonstration of the fact that , although some radio galaxies _ are _ ulirgs and others may have _ evolved from _ ulirgs ( see * ? ? ? * ) , not all ulirgs can evolve into radio galaxies and vica versa ; only the most massive 50% of ulirg systems are capable of harbouring powerful radio jets . in this context it is notable that , out of the complete sample of 26 @xmath89 ulirgs investigated by @xcite , only one is a radio galaxy , and that object pks1345 + 12 ( also in our sample of starburst radio galaxies ) has the largest total stellar mass of all the ulirgs . it is clear that many of the optical morphological features of starburst radio galaxies are consistent with the idea that both the agn / jet and starburst activity have been triggered as the result of the gas infall associated with galaxy mergers . hydrodynamical simulations of mergers reveal that they are complex events , and it can take a gigayear or more for the merger remnant to relax and achieve the appearance of a normal galaxy . the main gas infall and star formation are predicted to occur in two major peaks : one at or just after the first peri - centre passage of the merging nuclei ; the other as the two nuclei finally coalesce ( e.g. * ? ? ? the relative intensity and separation in time of these two peaks depends on several factors such as the morphologies , mass ratios and gas contents of the precursor galaxies , the merger geometries , and the importance of feedback processes associated with the agn and starbursts @xcite . given that most starburst radio galaxies have massive old stellar populations in addition to their ysp , it is likely their precursors had signifcant stellar bulges . moreover , the relatively large masses and proportional mass contributions of the ysps ( see table 5 ) suggests that at least one of the precursors must have been gas - rich . in such circumstances the models predict that the second star formation peak associated with the coalescence of the two nuclei is likely to be the most intense @xcite . if the agn feedback effects are important , and the agn is triggered concurrently with the starburst , then the star formation will be halted shortly after ( @xmath820.1 gyr ) the final coalescence of the nuclei by powerful agn - induced winds . the time lag between the two star formation peaks is of order 0.3 1.5 gyr , while the major phase of activity associated with the second ( more intense ) peak is expected to last of order 0.1 gyr . however , star formation and agn activity may not only be associated with the two main peaks of activity , but may also occur albeit at reduced intensity at other stages , for example , as the tidal debris rain back down on the merger remnant following the coalescence of the two nuclei . many of the most recent simulations demonstrate that the fuel supply to the central supermasive black holes can be highly irregular over the course of a merger @xcite . it is important to emphasise that , as well as the delivery of gas into the _ general vicinity _ of the supermassive back hole ( within @xmath0100 pc the typical resolution of the hydrodynamical sigmulations ) , there may be other timescales involved in the triggering of the activity . these include the time required for the supermassive black holes from the precursor galaxies to merge , and the time required for the accreted gas to lose sufficient angular momentum to move close enough to the remnant supermassive black hole to fuel the activity ( see discussion in * ? ? ? it has also been suggested that the powerful winds associated with the circum - nuclear star formation activity may disrupt the flow of gas into the nuclear regions and delay the onset of agn activity @xcite . unfortunately the current generation of hydrodynamical simulations do not have sufficient resolution to include these processes in any detail . therefore in what follows we will assume that the black hole accretion and star formation rates are as ( crudely ) predicted by the simulations , with the caveat that the reality is likely be more complex . we now consider whether , based on their ysp properties and optical morphologies , the objects discussed in this paper are consistent with the merger scenario for the triggering of the activity . most of the starburst radio galaxies can be fitted into one of the following three categories , each corresponding to a particular phase of a gas - rich merger . this merger sequence is also illustrated in figure 3 . * * pre - coalescence . * these double nucleus systems are observed after the first peri - centre passage , but immediately before ( within @xmath00.1 gyr ) the coalescence of the two nuclei , as they move together . all of these objects have large agn / jet and far - ir luminosities , reflecting the level of agn and starburst activity associated with the gas infall expected at this stage of a merger . the ysp in these systems represent a combination of the younger ysp formed as a consequence of the fresh gas infall associated with coalescence , and intermediate age stellar populations formed in the first peak of star formation , around the time of the first peri - centre passage . the relative importance of the latter two stellar populations will depend on the details of the encounter , and the extent to which the most recent episode of star formation is obscured by dust . alternatively , any intermediate age stellar populations could be associated with the captured disks of the precursor galaxies . examples : pks0409 - 75 , pks1345 + 12 , 3c321 and 3c433 . * * coalescence . * these single nucleus systems are observed in the 0.1 gyr period immediately following the coalescence of the two nuclei . for major mergers involving gas - rich galaxies with significant bulges , the models predict that this should represent the most active phase of starburst and agn activity . due to the fact that the most recent phase of star formation activity has been particularly intense , and is likely to dominate ( in flux ) over any previous generations of stars , the luminosity - weighted ages in the nuclear regions of these objects are relatively young ( @xmath820.1 gyr ) . the far - ir , [ oiii ] and radio luminosities are large in this phase , reflecting the high level of agn and starburst activity associated with the large concentrations of gas and dust in the nuclear regions ; in some cases ( depending on the details of the merger ) the source may attain a ulirg - like luminosity . examples : 3c48 , pks1549 - 79 , pks2135 - 209 and 3c459 . * * post - coalescence . * these single nucleus systems still bear the morphological hallmarks of a merger in the recent past ( tidal tails , dust lanes , distorted outer isophotes etc . ) , but they have intermediate age ysp ( @xmath151 gyr ) , consistent with them being observed a significant period after the main merger - induced starburst ( i.e. the ysp represent post - starburst stellar populations ) . this is the scenario outlined in @xcite and @xcite . unless the agn / jet activity lasts considerably longer than the 0.1 gyr maximum lifetime that is typically estimated for powerful extragalactic radio sources ( e.g. * ? ? ? * ) , there must have been a time lag between the main merger - induced starburst and the triggering of the current phase of activity . examples : b20648 + 27 , 3c213.1 , 3c285 , 3c293 , 3c305 , centaurus a , fornax a. just as it appears likely that radio - loud agn activity can be triggered at different stages during a merger , it is also possible that there is more than one episode of agn / jet activity in the course of a merger . for example , an individual system may undergo a particularly powerful phase of agn / jet activity as the nuclei coalesce , close to the peak of starburst activity ( the coalescence phase above ) , but the agn / jet activity may also be triggered ( or re - triggered ) earlier or later in the merger sequence , depending on the details of the radial gas flows in the merger . indeed , as noted above , six of the starburst radio galaxies in our sample show evidence for re - triggered radio source activity in the form of high surface brightness inner radio structures and more diffuse and extended outer radio structures ( 3c213.1 , 3c218 , cen a , 3c236 , 3c293 , pks1345 + 12 ) . moreover , in the particular case of 3c236 , @xcite have argued for multiple phases of jet activity , based on both the double - double morphology of its radio source , and the evidence for two major epochs of star formation in its host galaxy . we emphasise that not all of the starburst radio galaxies can be readily accommodated within the merger scheme outlined above ; there are some prominent misfits . most notably , the central cluster galaxies 3c218 and pks0620 - 52 do not show clear morphological signs of major mergers , and 3c218 is also unusual in having relatively young ysp ( @xmath1520.05 gyr ) but low emission line and far - ir luminosities ( the age of the ysp in pks0620 - 52 is not well - determined ) . pks0023 - 26 may also represent an ambiguous case in the sense that it lies at the heart of a rich cluster of galaxies , and has a peculiar amorphous outer envelope that is difficult to classify in terms of the merger sequence ( although a merger can not be entirely ruled out ) . in the case of the one system in our sample that shows evidence for large - scale , and relatively settled , gaseous disk ngc612 it is not clear whether the agn / jet activity has been triggered by interactions with massive galaxies on a large scale in the wider galaxy group ( as may be evidenced by the hi observations ) , or by recent mergers / interactions with closer companion galaxies ( see * ? ? ? the latter seems more likely given that ngc612 shows an optical shell structure , along with ysp that are much younger ( @xmath15 gyr ) than the estimated time since its closest approach to the the massive companion galaxy ngc619 ( @xmath153 gyr ) . clearly , the possibility of multiple interactions and mergers in galaxy groups can complicate the interpretation of the galaxy morphologies and ysp ages in terms of a simple merger sequence . finally we note that , while the properties of the most luminous starburst radio galaxies in particular , the ulirg - like coalescence systems are consistent with triggering in major ( similar mass ) gas - rich mergers , it is difficult to rule out minor mergers as the trigger for some of the other systems . indeed , it has been argued that the large - scale features of centaurus a are consistent with a minor ( 1:10 ) merger between the host radio galaxy and a smaller disk galaxy @xcite . as noted in the previous section , some of the starburst radio galaxies that are situated close to the centres of rich clusters of galaxies are not readily accommodated in the merger sequence . for such objects cooling flows k ) x - ray emitting gas of the cluster to cooler phases of the ism in the central cluster galaxy , where it can form stars and/or be accreted by the central black holes . the direct accretion of the hot gas itself by the central black hole will be discussed in the following section . ] associated with the hot x - ray emitting gas may provide an alternative triggering mechanism for both the visible star formation and the agn / jet activity . this mechanism is supported by the irregular morphologies and kinematics of the emission line gas in central cluster galaxies hosting radio sources @xcite . moreover , while the high velocity dispersions of the galaxies in the centres of massive galaxy clusters can hamper gas accretion via major , gas - rich mergers , there is no such problem for the accretion of warm / hot gas via cooling flows , since the cooling gas will naturally fall towards the centre of the the cluster potential well . although star formation has been discussed as a potential sink for the cooling gas , until recently the apparently large differences between the estimated hot gas cooling rates and the star formation rates in the central cluster galaxies suggested that only a small fraction of the the cooling gas ends up forming stars ( e.g. * ? ? ? however , spectroscopic observations with the new generation of x - ray satellites have led to a major downward revision in estimates of the hot gas cooling rates , so that they are now much closer to the star formation rates . therefore , it is plausible that a significant fraction of the cooling gas does in fact form stars ( see * ? ? ? * ) . in the cases of the three cooling flow candidates in our sample , based on their infrared luminosities and the relation of @xcite , the star formation rates are 5 , 4 and 25 m@xmath3 yr@xmath8 for pks0620 - 52 , 3c218(hydra a ) and pks0023 - 36 respectively . of these three , only 3c218 has published high quality x - ray observations , and its hot gas cooling rate of @xmath155 m@xmath3 yr@xmath8 @xcite proves to be within a factor of 4 of the star formation rate . given the agreement between its star formation and hot gas cooling rates , 3c218 is one of the best candidates for an object in which the activity has been triggered by the warm / cool gas condensing out of a cooling flow . however , even in the case of 3c218 we can not entirely rule out the idea that the activity has been triggered in a galaxy merger or interaction , since it is possible for central cluster galaxies to undergo mergers that could , potentially , form star forming gaseous disks similar to that associated with the dust lane in the central regions of 3c218 @xcite . although such disks ( and the merging galaxies that produced them ) may represent only a small fraction of the total masses of the cd galaxies , the associated gas infall rates may be sufficient to fuel the agn / jet activity in the nuclei on the requisite timescales . note that the large - scale morphological signatures of galaxy mergers ( e.g. tidal tails , fans , shells ) are more difficult to detect against the light of the massive stellar haloes of the central cluster galaxies than they are in lower mass elliptical galaxies ; the tidal features are also likely to be erased on a relatively short timescale by ongoing tidal interactions between all the galaxies in the dense central regions of the galaxy clusters . as a class , low - power fri radio galaxies are almost invariably classified as weak line radio galaxies ( wlrg ) at optical wavelengths . however , a significant proportion of frii galaxies also have wlrg spectra , with [ oiii]@xmath1565007 and mid - ir luminosities that are an order of magnitude lower than narrow- and broad - line radio galaxies ( nlrg / blrg ) of comparable radio power @xcite . on the basis of their nuclear x - ray and emission line properties , and the way that their emission line luminosities and absolute magnitudes correlate with radio power , it has been suggested that the nuclear accretion rate or accretion mode in wlrg is distinct from that in their strong - lined radio galaxy counterparts ( e.g. * ? ? ? * ; * ? ? ? in particular , it has been argued that wlrg represent objects in which the agn / jet activity is fuelled via bondi accretion of the hot x - ray - emitting ism , while strong - lined radio galaxies represent objects in which the activity is fuelled by the accretion cooler phases of the ism . for low power radio galaxies the bondi accretion hypothesis is supported by correlations between galaxy / black hole mass and the fraction of galaxies that are radio - loud , the relatively large proportion of galaxies that are radio - loud at large galaxy / black hole mass ( suggesting a high duty cycle ) , and the fact that the energy input of radio jets fuelled by bondi accretion is sufficient to balance radiative losses of the hot x - ray emitting haloes of the host giant elliptical galaxies @xcite . significant star formation is not expected in the case of triggering via bondi accretion of the hot ism . therefore it is interesting that a significant subset of the starburst radio galaxies are classed wlrg ( ngc612 , pks0620 - 52 , 3c213.1 , 3c218 , 3c236 , 3c293 , fornax a , centaurus a ) . although some such objects ( pks0620 - 52 , 3c218 ) are candidates for fuelling via cooling flows ( see section 5.2 above ) , more than 50% of the wlrg in our sample of starburst radio galaxies can be classified as post - coalescence or late post - coalescence systems , with evidence for a rich ism in the circum - nuclear regions in the form of circum - nuclear dust lanes , as well as other signs that they have been involved in mergers in the recent past . this suggests a possible alternative to fuelling via bondi accretion of the hot ism : accretion of the cool ism at a slow rate from the debris disks of galaxy mergers . as the disks settle it is expected that the shocks associated with the gaseous dissipation process will lead to a net infall of gas , and also produce a liner - type emission line spectrum , similar to those observed in wlrg @xcite . in the later post - merger stages the debris disks may also act as large reservoirs of _ potential _ fuel for the agn / jet activity , but external stimuli , such as interactions with neighbouring galaxies or the accretion of satellite galaxies , may be required to perturb the disks and induce radial gas infalls that are sufficient to trigger the agn . this might help to explain late post - coalescence systems such as the iconic fri radio galaxies centaurus a and fornax a , which appear to have been triggered a substantial period after ( @xmath02 3 gyr after in the case of fornax a ) the mergers associated with the formation of the dust lanes . in such cases , the original merger may not have triggered the current phase of nuclear activity directly , but rather delivered a reservoir of cool gas that was then perturbed by more minor galaxy interactions to produce the observed level of activity . in this context it is notable that centaurus a shows evidence for an ongoing satellite galaxy accretion event in the form of a blue tidal stream @xcite . overall , the assumption that _ all _ wlrg are fuelled by bondi accretion of the hot ism is likely to represent an over - simplification ; for a significant fraction including some fri radio galaxies slow accretion of cooler gas from the debris disks of galaxy mergers represents a viable alternative mechanism . this is consistent with the x - ray and mid - ir evidence for compact , cool dust / gas structures in the nuclear regions of a minority of wlrg , including centaurus a @xcite . the discussion above has concentrated on starburst radio galaxies that show strong evidence for recent star formation activity at optical wavelengths . however , such objects comprise only a small proportion ( @xmath015 25% ) of the full population of powerful radio - loud agn in the local universe . clearly it is important to consider how the agn / jet activity is triggered in the 75 85% majority of powerful radio galaxies that do not show strong signs of optical star formation activity . note that many of the latter belong to the class of strong - lined frii radio sources in which the agn / jet activity is generally considered to be triggered by accretion of the cooler phases of the ism @xcite . one possibility is that the non - starburst radio galaxies are objects in which the agn / jet activity has been triggered or re - triggered in the late post - coalescence phase , @xmath131 gyr after the starburst associated with the coalescence of the nuclei in the merger . as argued in @xcite , after @xmath01 gyr it can be difficult to detect a merger - induced starburst at optical wavelengths against the light of the giant elliptical galaxy host , especially if the ysp comprises a relatively small fraction of the total stellar mass and/or suffers a moderate amount of extinction . in the late post - coalescence phase , the morphological signs of mergers also become more difficult to detect , especially for radio galaxies at intermediate / high redshifts . the main problem with this explanation is that , for strong - lined , non - starburst frii radio galaxies , the moderate rates of gas infall associated with the debris disks may not be sufficient to fuel the prodigious nuclear activity . indeed , many of the starburst radio galaxies that we identified above as post - coalescence systems are wlrg . alternatively , it is possible that the activity in the non - starburst radio galaxies has been triggered by close encounters with gas - rich companion galaxies , or following the first peri - center passages of the merging nuclei in galaxy mergers . in the latter case , hydrodynamical simulations show that the tidal forces following the first encounter induce significant radial infalls of gas into the nuclei for an extended period around ( ( @xmath00.5 1.0 gyr : * ? ? ? * ; * ? ? ? * ) ) the time of closest approach . such infalls may capable of fuelling the prodigious agn / jet activity in strong - lined radio galaxies . however , in cases in which the merging nuclei have significant bulges ( highly likely in the case of radio galaxies ) , only a relatively low level of star formation is expected at this stage . this mechanism may be supported by the finding of a relatively high incidence of tidal bridge features and tidally distorted companion galaxies in deep imaging observations of non - starburst radio galaxies in the 2jy sample @xcite ; it is further supported by detailed studies of individual radio galaxies in interacting groups @xcite . a final possibility is that non - starburst radio galaxies are triggered in relatively minor mergers ( 1:3 or less ) , or in major mergers that are relatively gas poor ( `` dry '' ) . in such cases relatively low levels of star formation are expected , which are likely to be difficult to detect against the light of the old stellar populations in the host galaxies . however , without further theoretical work , it is not clear whether minor or gas - poor mergers would be capable of delivering sufficient gas to the nuclear regions to fuel the quasar - like levels of nuclear activity detected in some non - starburst radio galaxies . an interesting aspect of starburst radio galaxies is that they show a high incidence of unusual radio structures : our sample includes several compact css / gps sources , as well as sources with relatively bright steep spectrum core structures , diffuse outer haloes and double - double structures . what ( if any ) is the relationship between these unusual radio structures and the presence of young stellar populations in the host galaxies ? as already discussed above , double - double sources , and sources with compact high surface brightness inner structures combined with diffuse outer haloes , may represent cases in which the radio jet activity has been re - triggered ( but see morganti et al . 1999 and wise et al . 2007 for counter - arguments in the cases of centaurus a and hydra a ) . given the complexity of the gas infall histories of major gas rich mergers , it is certainly plausible that each merging system undergoes more than one phase of agn / jet activity , thus explaining the presence of such sources in our sample . considering the compact ( css / gps ) radio sources , there is direct observational evidence from measurements of hotspot advance speeds that css / gps radio sources are relatively youthful ( @xmath157 @xmath158 yr ) . however , since the css / gps radio sources are generally estimated to be much younger than the ysps detected in their host galaxies ( @xmath159 @xmath160 yr ) , the youth of the compact sources does not necessarily help to explain their relatively high rate of occurrence in our sample of starburst radio galaxies . alternatively , the high incidence of css / gps sources in the starburst radio galaxies sample may be the consequence of an observational selection effect as follows . we expect a relatively rich and dense ism to be present in the nuclear regions of merging systems , especially around the time of nuclear coalescence . a radio source triggered in a merger will interact particularly strongly with this rich ism in the early stages of the radio source history , as the jets expand through the central regions of the host galaxies ; direct evidence for strong jet - cloud interactions in young radio sources is provided by their extreme emission line kinematics @xcite . the strong interactions between the jets and the rich ism associated with the mergers will not only result in extreme emission line kinematics , but may also affect the conversion of jet power into radio luminosity , boosting the radio luminosities of sources . there is already evidence that interaction with the relatively dense , hot x - ray haloes associated with clusters of galaxies boosts the radio luminosities of jets for a given jet power @xcite , and it is plausible that there will be a similar boosting effect when the jets strongly interact with the ( cooler ) ism in the central regions of merger remnants . indeed , strong enhancements in the radio emission are observed at the sites of interactions between radio jets and warm emission line clouds in the haloes of radio galaxies in the local universe ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? for a given intrinsic jet power , this flux boosting will lead to the compact radio sources that are triggered in young , star forming merger remnants being preferentially selected in flux - limited radio surveys . this in turn could explain the relatively high rate of occurrence of compact radio sources amongst the starburst radio galaxies , as well as the bias of the compact sources towards relatively young ysp ages ( @xmath15 gyr ; see section 4.1 ) . the interaction of the jets with the richer gaseous environments present in merger remnants could also help to explain the relatively high incidence of extended radio sources with compact steep spectrum cores in our sample , since such interactions have the potential to boost the radio emission from the jets , even if the radio lobes are well ouside the central regions of the galaxies . in this paper we have discussed the properties of the young stellar populations ( ysp ) in the @xmath015 25% of powerful radio galaxies that show strong evidence for recent star formation activity at optical wavelengths . combined with information about the morphologies of the host galaxies , the ysp properties of most of these starburst radio galaxies are consistent with the triggering of both starburst and agn / jet activity in galaxy mergers in which at least one of the merging galaxies is gas - rich . however , the triggering of the agn / jet activity is not confined to a single evolutionary phase of galaxy mergers . while in a significant subset of objects the agn / jet activity has been triggered within 0.1 gyr of the coalescence of merging nuclei , close to the expected peaks of the merger - induced starbursts , many objects are observed in the post - coalescence phase , @xmath130.2 gyr after the starburst peaks . in the former group the triggering of the activity can be readily identified with the major infalls of gas that are predicted to occur around the time of coalescence . on the other hand , in the latter group the nature of the link between the triggering of the agn / jet and the original starburst - inducing mergers is less clear : late - time infall of merger debris , settling of the debris disks to an equilibrium configuration , and perturbation of the debris disks by minor mergers and encounters , are all possibilities for the triggering of the activity in these , generally lower luminosity , objects . our results clearly demonstrate that luminous radio - loud agn _ can _ be triggered close to the peaks of major gas - rich mergers . however , the relatively low incidence of radio - loud agn in nearby ulirgs suggests that , even under similar conditions of gas infall , it is not inevitable that radio - loud agn activity _ will _ be triggered in such mergers . indeed our comparison of the stellar masses of ulirgs and starburst radio galaxies suggests that the triggering of powerful radio - loud agn activity is most likely to occur in the most massive merger remnants , consistent with other studies of agn host galaxies that provide evidence for a link between host galaxy mass ( and hence black hole mass ) and the incidence of the radio - loud agn population ( e.g. * ? ? ? overall , if we also take into consideration the evidence that some radio - loud agn are triggered by the accretion of gas from the hot x - ray haloes , either via cooling flows or the direct ( bondi ) accretion of hot gas , it appears unlikely that radio - loudness is solely a consequence of a particular mechanism for the delivering of the fuel supply to the nuclear regions . rather , the ability to produce powerful relativistic jets is more likely to be related to the intrinsic properties of energy generating regions ( e.g. the spin and/or mass of the supermassive black hole ) . finally , we emphasise that a considerable fraction of powerful radio galaxies ( @xmath075 85% ) do not show evidence for energetically significant starburst activity at either optical or mfir wavelengths . therefore , in order to complete our understanding of triggering of the activity in radio - loud agn , in the future it will be important to compare the host galaxy properties and environments of the non - starburst majority of radio galaxies , with those of the starburst minority discussed in this paper . cra , jh and kji acknowledge support from the uk science and technology facilities council ( stfc ) , de nederlandse organisatie voor wetenschappelijk onderzoek ( nwo ) , and the emmy noether programme of the german science foundation ( dfg ) respectively . we thank the anonymous referee for useful comments that have helped to improve the manuscript . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory . based on observations made with eso telescopes at the la silla and paranal observatories under programmes 70.b-0663(a ) , 71.b-0320(a ) , 078b-0660(a ) . davies , r.i . , snchez , f. , mueller , f. , genzel , r. , tacconi , l.j . , hicks , e.k.s . , friedrich , s. , sternberg , a. , 2007 , apj , 671 , 1388 de koff , s. , baum , s.a . , sparks , w.b . , biretta , j. , golombek , d. , macchetto , f. , mccarthy , p. , miley , g.k . , 1996 , apjs , 107 , 621

we present a series of quantum states that are characterized by dark solitons of the nonlinear schrdinger equation ( i.e. the gross - pitaevskii equation ) for the one - dimensional ( 1d ) bose gas interacting through the repulsive delta - function potentials . the classical solutions satisfy the periodic boundary conditions and we simply call them classical dark solitons . through exact solutions we show corresponding aspects between the states and the solitons in the weak coupling case : the quantum and classical density profiles completely overlap with each other not only at an initial time but also at later times over a long period of time , and they move together with the same speed in time ; the matrix element of the bosonic field operator between the quantum states has exactly the same profiles of the square amplitude and the phase as the classical complex scalar field of a classical dark soliton not only at the initial time but also at later times , and the corresponding profiles move together for a long period of time . we suggest that the corresponding properties hold rigorously in the weak coupling limit . furthermore , we argue that the lifetime of the dark soliton - like density profile in the quantum state becomes infinitely long as the coupling constant approaches zero , by comparing it with the quantum speed limit time . thus , we call the quantum states _ quantum dark soliton states_.

1602.08329

the experimental technique of trapped one - dimensional atomic gases @xcite has now become a fundamental tool for revealing nontrivial effects in quantum many - body systems @xcite . for the interacting bose gas in one dimension ( 1d ) , the first set of exact results goes back to the pioneering work of girardeau @xcite on the impenetrable bose gas where the strong interacting limit is considered . the 1d bose gas interacting with the delta - function potentials , i.e. the lieb - liniger ( ll ) model , gives a solvable model for interacting bosons in 1d @xcite , where it is integrable even when the interaction parameter is generic . for the impenetrable bose gas which corresponds to the tonks - girardeau ( tg ) limit , i.e. the strong interacting limit of the ll model , the one - body reduced density matrix is derived and successfully expressed in terms of the determinant of a fredholm operator @xcite . the exact result is followed by several important developments in mathematical physics @xcite . furthermore , dynamical correlation functions of the ll model are now systematically derived @xcite . soliton - like localized excitations in a periodic 1d bose gas have attracted much interest recently and have been studied theoretically @xcite . here we remark that dark solitons are created experimentally in cold atoms @xcite , for instance , by the phase - imprinting method @xcite ( see also @xcite ) . localized quantum states are important for investigating dynamical responses of interacting quantum systems . quantum dark solitons in confining potentials are studied by semiclassical quantization @xcite , and those in thermal equilibrium of a quasi-1d bose gas by generating classical field ensembles @xcite . however , it is not clear even at zero temperature how we can construct quantum states associated with dark solitons in the many - body system of the ll model . let us consider the gross - pitaevskii ( gp ) equation , which describes bose - einstein condensation ( bec ) in the mean - field approximation @xcite . we also call it the nonlinear schrdinger equation . the gp equation has dark soliton solutions for the repulsive interactions , while it has bright soliton solutions for the attractive interactions @xcite . it was conjectured that dark solitons are closely related to lieb s type - ii excitations , i.e. one - hole excitations , by carefully studying the dispersion relations @xcite . the dispersion relations of the ll model are briefly compared with those of the classical nonlinear schrdinger equation in the weak coupling limit @xcite . however , it has not been shown how one can construct such a quantum state that leads to a dark soliton in the classical limit or what kind of physical quantity can show a property of a dark soliton for some quantum state . here we remark that each of the type - ii eigenstates has a flat density profile since the bethe ansatz eigenstates are translationally invariant . moreover , we consider dark solitons under the periodic boundary conditions , which are expressed in terms of elliptic functions @xcite . in this paper we demonstrate that a quantum state derived from the bethe ansatz eigenvectors of the ll model by taking the fourier transform of the type - ii excitations over a branch @xcite has many aspects closely related to classical dark solitons of the gp equation under the periodic boundary conditions . we call the state the _ quantum dark soliton state _ and a classical dark soliton under the periodic boundary conditions simply a _ classical dark soliton_. through the bethe ansatz we show several corresponding aspects in the weak coupling regime . firstly , the density profile of the quantum dark soliton state at an initial time is consistent with that of a classical dark soliton . here we call the graph of the expectation value of the local density operator for a given state versus the position coordinate the density profile of the state , and for a quantum dark soliton state we simply call it the density profile of quantum dark soliton ; we call the graphs of the square amplitude and phase in the complex scalar field of a classical dark soliton versus the position coordinate the density profile and phase profile of classical dark soliton , respectively . secondly , in time evolution the density profile of quantum dark soliton coincides with that of the corresponding classical dark soliton over the whole graph and they move together with the same velocity for a long period of time . thirdly , for the matrix element of the bosonic field operator between two quantum dark soliton states where one state has @xmath0 particles and another @xmath1 particles , the profiles of the square amplitude and phase at the initial time graphically agree with those of classical dark soliton , respectively . in time evolution the profiles of square amplitude and phase overlap with those of classical dark soliton , respectively , over the whole region and the corresponding profiles move together in time for a long period of time . here we remark that a classical dark soliton parametrized by elliptic functions becomes a standard dark soliton with open boundaries by simultaneously sending the system size to infinity and the elliptic modulus to unity . furthermore , in order to illustrate the method for constructing quantum dark solitons , in the 1d free fermions we show from the anti - commutation relations that a notch appears in the density profile of some superposition of one - hole excitations . interestingly , the density profile of the fermionic state coincides with that of quantum dark soliton for the 1d bose gas in the large coupling case , i.e. near the tg limit , not only at an initial time but also during the quantum dynamics for some period of time . the time evolution of the expectation value of the local density operator in the 1d bose gas should be important also from the renewed interest on the fundamentals of quantum statistical mechanics @xcite . the density profile of a quantum dark soliton state has initially a localized notch but collapses slowly to a flat profile in time evolution @xcite . the relaxation behavior is consistent with the viewpoints of equilibration of an isolated quantum system @xcite and thermalization due to the typicality of quantum states @xcite . we now argue that the density profile of quantum dark soliton has a finite lifetime but it is much longer than the quantum speed limit time of the quantum state in the weak coupling case . here we remark that the lifetime of a generic state is given by the quantum speed limit time for it @xcite . by observing the exact time evolution of the density profile of quantum dark soliton we estimate its lifetime . we shall show in section 5 that the observed life time is inversely proportional to the coupling constant . we thus suggest that the localized density profile of a quantum dark soliton is much more stable than the density profile of a generic quantum state in the weak coupling case . here we remark that girardeau and wright discussed a permanent quantum soliton for impenetrable bosons in 1d @xcite , which corresponds to the infinite coupling case of the ll model . we also argue that the behavior of the wavefunctions is nontrivial in the weak coupling limit for the ll model . the wavefunctions do not simply become close to such wavefunctions of the 1d free bosons that could be consistent with the mean - field picture , when the coupling constant approaches zero but it takes a nonzero value . the exact bethe ansatz wavefunctions consist of a large number of terms such as @xmath2 . furthermore , we can show that many - body correlations increase if we increase the particle number @xmath1 while we keep the coupling constant small but fixed and the particle density constant . for instance , the zero mode fraction in the ground state of the 1d bose gas , which we regard as the condensate fraction of bec , becomes small and decreases to zero if the number of particles increases while the coupling constant and the particle density are fixed @xcite . we suggest that it is also the case for quantum dark solitons i.e. , that some properties of quantum dark solitons corresponding to classical dark solitons do not necessarily hold if we increase the particle number @xmath1 while we keep the coupling constant small but fixed and the density of particles fixed . in fact , it is not _ a priori _ clear how valid the mean - field approximation is for quantum dark solitons even in the weak coupling case . here we remark that the breakdown of mean - field theory is addressed for impenetrable bosons @xcite , and a long - wavelength theory beyond the mean - field approximation is discussed for low - dimensional bose liquids @xcite . let us consider the hamiltonian of the ll model for @xmath1 bosons with coupling constant @xmath3 @xmath4 here we impose periodic boundary conditions with length @xmath5 . we employ a system of units with @xmath6 , where @xmath7 is the particle mass . we introduce the canonical bose field operator @xmath8 with the commutation relations @xcite @xmath9=\delta(x - x ' ) , \\ & [ \hat { \psi}(x , t ) , \hat { \psi}(x',t)]=[\hat { \psi}^{\dagger}(x , t ) , \hat { \psi}^{\dagger}(x',t)]=0 . \end{aligned}\ ] ] the second - quantized hamiltonian of the ll model is written in terms of the field operator @xmath8 as @xmath10 , \label{h2}\end{aligned}\ ] ] where @xmath11 is the chemical potential . the heisenberg equation of motion of this system has the form @xmath12 in the classical limit where the quantum field operator @xmath8 is replaced by a complex @xmath3-number field @xmath13 , equation ( [ eq : qnse ] ) becomes the following partial differential equation @xmath14 we call it the nonlinear schrdinger equation . the equation ( [ eq : nse ] ) can be solved by the inverse scattering method and has soliton solutions @xcite ( see also @xcite ) . we recall that it has dark soliton solutions for the repulsive interactions @xmath15 , while it has bright soliton solutions for the attractive interactions @xmath16 . for the attractive case , bright solitons are analytically derived from the quantum wave packets constructed through the fourier transform of the bound states @xcite . the construction was extended for optical fibers @xcite . time evolution of quantum states of bright solitons was investigated theoretically @xcite and experimentally @xcite . the effect of quantum noises on the quantum states of bright solitons was also studied @xcite . the contents of the paper consists of the following . in section 2 we briefly introduce the notation for the bethe ansatz , characterize the type ii excitation branch , and then construct quantum dark soliton states . we argue that the structure of quantum dark soliton states is analogous to the dirichlet kernel , and show for free fermions that the density profile of a superposition of one - hole excited states has a notch . it illustrates the construction of quantum dark soliton states . in section 3 we derive one - soliton solutions of the nonlinear schrdinger equation ( [ eq : nse ] ) under the periodic boundary conditions . we express the classical dark solitons in terms of elliptic functions similarly as in ref . @xcite . by sending the elliptic modulus @xmath17 to 1 , the classical dark solitons under the periodic boundary conditions become the standard classical dark solitons of the nonlinear schrdinger equation with open boundaries . furthermore , we calculate the logarithmic corrections to them , in particular , to the chemical potential and the soliton velocity , by which we can confirm the numerical estimates of the soliton parameters shown in section 4 at least approximately . in section 4 we show several corresponding aspects between the quantum dark soliton states and the corresponding classical dark solitons , which hold in the weak coupling case and should be rigorously valid when the coupling constant approaches zero with the density and the particle number @xmath1 fixed . firstly , we show that the density profile of quantum dark soliton at an initial time is consistent with that of classical dark soliton . secondly , we show that in exact time evolution the density profile of quantum dark soliton as a whole coincides with that of classical dark soliton for a long period of time . thirdly , we show that both the square amplitude and phase profiles for the matrix element of the quantum field operator @xmath18 in the ll model between the quantum dark soliton states with @xmath0 and @xmath1 particles overlap with those of classical dark soliton , respectively , all over the profiles . the square amplitude and phase profiles move in time evolution with exactly the same speed as those of classical dark soliton , respectively . furthermore , we remark that the density profile of a sum of one - hole excitations in the 1d free fermions overlap with that of quantum dark soliton in the 1d bose gas with the large coupling constant , i.e. in a regime close to the tg limit . for the matrix element of the fermionic field operator between two such fermionic states the square amplitude and phase profiles have several features in common with those of quantum dark soliton in the 1d bose gas for the large coupling case , respectively , although they do not overlap each other . we thus obtain useful tools to describe approximately the profiles of quantum dark soliton in the large coupling case . however , it is still quite nontrivial that the profiles of quantum dark soliton in the weak coupling case are consistent with those of classical dark soliton . in section 5 we argue that the lifetime of a notch in the density profile of quantum dark soliton is much longer than that of a generic state when the coupling constant @xmath3 is very small . we compare the lifetime of a notch in the density profile of quantum dark soliton with the quantum speed limit time , and show that the ratio of the former to the latter increases as the coupling constant decreases . it becomes infinitely large as the coupling constant approaches zero . thus , although the localized density profile collapses in time evolution , we call the quantum states quantum dark soliton states . in the framework of the bethe ansatz method , the bethe ansatz wavefunction which is expressed in terms of quasi - momenta @xmath19 @xmath20 , \nn\db & a_\sg=(-1)^\sg \prod_{j>\ell}^n\[k_{\sg_j}-k_{\sg_\ell}-\i c \ , \sign(x_j - x_\ell)\ ] \end{aligned}\ ] ] gives an eigenstate of the ll model , if the quasi - momenta satisfy the bethe ansatz equations @xmath21 in the logarithmic form , we have @xmath22 here @xmath23 s are integers for odd @xmath1 and half - odd integers for even @xmath1 . we call them the bethe quantum numbers . the total momentum @xmath24 and energy eigenvalue @xmath25 are written in terms of the quasi - momenta as @xmath26 in the case of repulsive interaction @xmath15 , the bethe equations have a unique real solution @xmath27 if we specify a set of bethe quantum numbers @xmath28 @xcite . the configuration of the bethe quantum numbers for the ground state is given by the regular array around the center : @xmath29 it is depicted in the top panel of figure [ ij1h1p ] ( a ) in the case of @xmath30 . we can construct low - lying excited states systematically by putting holes and placing particles in the regular array of the bethe quantum numbers for the case of the ground state . the arrays of the bethe quantum numbers for one - hole excitations are shown in the second panel of figure [ ij1h1p ] ( a ) . in each sequence there exists an unoccupied number in the middle of the sequence of the five red circles , which we call the one - hole . we remark that the one - hole excitations are also called the type ii excitations @xcite . similarly as in the second panel , the arrays of the bethe quantum numbers for one - particle excitations are shown in the third panel of figure [ ij1h1p ] ( a ) . the right most red particle corresponds to the one - particle in each array . the one - particle excitations are also called the type i excitations @xcite . the dispersion relation of the one - particle excitations and that of the one - hole excitations are plotted with filled circles and filled squares , respectively , in figure [ ij1h1p ] ( b ) . the former branch corresponds to the type i excitations and the latter branch the type ii excitations . . ( b ) dispersion relations of the 1-hole excitations and 1-particle excitations in the case of @xmath30 . a triangle represents the ground state , big circles represent 1-particle excitations , and squares represent 1-hole excitations . small dots are the 1-particle 1-hole excitations . ] we now explain how we construct a series of quantum states @xcite through the type ii excitations . furthermore , we call them quantum dark soliton states . in the type ii branch , for each integer @xmath31 in the set @xmath32 , we consider momentum @xmath33 . we denote the normalized bethe eigenstate of @xmath1 particles with total momentum @xmath24 by @xmath34 , which we call a one - hole excited state . the bethe quantum numbers @xmath23 s of the one - hole excitation @xmath35 are given by @xmath36 for any value of @xmath37 satisfying @xmath38 we define the coordinate state @xmath39 by the discrete fourier transformation : @xmath40 shows a density notch as will be shown in section 4 . the formal structure of quantum dark soliton states is analogous to a truncated series of the delta function . for periodic functions with period @xmath5 the delta function is expressed in the form of an infnite series @xmath41 we now truncate it by keeping only such terms with integers @xmath42 satisfying @xmath43 @xmath44 it is called the dirichlet kernel in the fourier series analysis . it has the peak around at @xmath45 with peak height @xmath46 and oscillating behavior in the region for @xmath47 with period @xmath48 . here we assume that @xmath49 is independent of @xmath5 . for @xmath50 , the oscillations in the graph of the dirichlet kernel become of very short wavelength , and we may call them ripples . the integral of the squared amplitude of the dirichlet kernel over the period @xmath5 is given by @xmath51 we therefore normalize the function @xmath52 by dividing it by the square root @xmath53 @xmath54 in terms of the normalized momentum eigenfunctions @xmath55 where @xmath33 the normalized function @xmath56 is expressed as @xmath57 it is analogous to the structure of the states of quantum dark solitons , where the number @xmath58 in ( [ eq : xp ] ) corresponds to the number of states @xmath1 in , and the single - particle momentum eigenstates @xmath59 to the one - hole excitations @xmath60 consisting of @xmath1 particles . by considering the phase factors @xmath61 in the graphs are shifted by @xmath37 in the @xmath62-direction . we have @xmath63 for the truncated series of the delta function modified with the phase factors . through the anti - commutation relation of the field operators we now show that a notch appears in the density profile of a linear combination of one - hole excited states in the 1d free fermions . under the periodic conditions of length @xmath5 the momenta are given by @xmath64 with some integers @xmath65 in the 1d free fermions . we denote by @xmath66 and @xmath67 for some integers @xmath68 the creation and annihilation operators of the one - dimensional free fermions with momenta @xmath69 , respectively . we assume that an infinite series of integers @xmath65 for @xmath70 cover all integers . the field operator and its conjugate are given by @xmath71 let us take a set of arbitrary integers @xmath72 and consider the state of @xmath73 fermions with momenta @xmath74 s @xmath75 for simplicity , we may assume that @xmath76 is the ground state . by applying the field operator @xmath77 to it , we have @xmath78 here we have assumed the ordering of fermionic operators as @xmath79 . we regard states @xmath80 as ` one - hole excitations ' of @xmath81 . we define the state @xmath82 by the sum over all one - hole excitations derived from @xmath83 @xmath84 let us define the local density @xmath85 for the state @xmath86 of the 1d free fermions by @xmath87 we now show that the local density vanishes at the origin : @xmath88 for the state . we first recall that the square of the field operator at the origin vanishes , @xmath89 , due to the anti - commutation relations of the field operators . thus , we have @xmath90 we can show that the integral of the local density over the entire interval @xmath5 is given by a large positive value such as @xmath91 @xmath92 it follows that the local density can not be always equal to zero , although it vanishes at the origin . therefore , for the state @xmath82 which is given by the sum over all the one - hole excited states of @xmath83 , the local density @xmath93 has a notch at least at the origin . in section 4 we shall construct the free - fermionic analog of quantum dark soliton states for the 1d free fermions and show that the density profile coincides with those of the ll model if the coupling constant is large enough such as @xmath94 . let us assume the traveling wave solution with velocity @xmath95 : @xmath96 to the nonlinear schrdinger equation . hereafter in section 3 we denote @xmath97 simply by @xmath98 . we denote the time derivative by @xmath99 and the second spatial derivative by @xmath100 , where @xmath101 denotes @xmath102 with @xmath103 . we therefore have @xmath104 we denote the solution of eq . ( [ eq2 ] ) by @xmath105 as a function of @xmath62 , for simplicity . we express the complex scalar field @xmath105 in terms of the amplitude @xmath106 and the phase @xmath107 , both of which are real , as @xmath108 by substituting ( [ eq : ap ] ) into equation we derive a pair of coupled equations from the real and imaginary parts , respectively . @xmath109 it follows from equation that we have @xmath110 . by integrating it once with respect to @xmath62 , we have @xmath111 here @xmath112 denotes a constant of integration . substituting ( [ vpd ] ) into and multiplying it by @xmath113 , we have @xmath114 integrating this equation with respect to @xmath62 , we have @xmath115 where @xmath116 is another constant of integration . thus , we have @xmath117 where the potential function @xmath118 is given by @xmath119 let us assume that the cubic equation : @xmath120 has three distinct real roots , @xmath121 , @xmath122 , and @xmath123 . we put them in increasing order : @xmath124 . the coefficients of the cubic equation are expressed in terms of the roots as follows . @xmath125 we now derive a solution @xmath126 to equation ( [ eq : u ] ) such that it satisfies @xmath127 in the interval of @xmath62 with @xmath128 . we set the initial condition : @xmath129 , for simplicity . it thus follows from equation ( [ eq : u ] ) that we have @xmath130 let us define the modulus of jacobi s elliptic functions , @xmath17 , by @xmath131 it is clear that @xmath132 . through the transformation of variables from @xmath133 to @xmath134 @xmath135 we express the integral ( [ eq : integral - x ] ) in terms of the elliptic integral of the first kind : @xmath136 we therefore have the solution to eq . ( [ eq : u ] ) in terms of jacobi s elliptic function @xmath137 @xmath138 we now integrate equation . here we remark that the constant @xmath112 is given by @xmath139 . in terms of the elliptic integral of the third kind we have @xmath140 where @xmath141 corresponds to the constant of integration . we now express @xmath121 , @xmath122 and @xmath123 in terms of the complete elliptic integrals . here we consider the periodic boundary conditions ( pbc ) for the square amplitude @xmath126 and the phase @xmath142 of the classical complex scalar field @xmath143 , and the normalization condition of the square amplitude @xmath126 . then , the elliptic modulus @xmath17 is related to the depth of the classical dark soliton , i.e. the minimum value @xmath121 of the square amplitude @xmath126 . we express the soliton velocity @xmath95 in terms of the elliptic integral of the second kind , and determine the chemical potential @xmath11 . suppose that the square amplitude @xmath126 of a classical dark soliton increases from @xmath121 to @xmath122 and then returns back to @xmath121with period @xmath5 . we therefore assign the conditions at the initial and the middle points as @xmath144 it follows from that in terms of the complete elliptic integral of the first kind , @xmath145 , we have @xmath146 we have @xmath147 . here and hereafter we often denote @xmath145 by @xmath148 , suppressing the modulus @xmath17 . we assume the periodic boundary condition for the phase : @xmath149 . suppose that the soliton velocity is positive : @xmath150 . it follows that we have @xmath151 and the velocity @xmath95 is expressed in terms of the complete elliptic integral of the third kind as @xmath152 we impose the normalization condition to the square amplitude @xmath126 @xmath153 where @xmath154 is the density of the number of particles @xmath1 , i.e. we have @xmath155 . by the formula we transform equation ( [ eq : rho ] ) in terms of jacobi s @xmath156 function into the following . @xmath157 thus , by the formula in terms of the complete elliptic integral of the second kind @xmath158 we have @xmath159 by solving equations , and with respect to the roots @xmath121 , @xmath122 and @xmath123 , we express them in terms of the complete elliptic integrals as follows : @xmath160 since the minimum value of the amplitude must be non - negative , i.e. @xmath161 , we assign the following condition @xmath162 thus , if the coupling constant @xmath3 , the system size @xmath5 , the density @xmath42 and the elliptic modulus @xmath17 satisfy condition ( [ eq : nonnegative ] ) , the classical dark soliton exists as a solution of the nonlinear schrdinger equation ( i.e. the gp equation ) . the soliton velocity @xmath95 and the chemical potential @xmath11 are given by @xmath163 let us explain how we evaluate the parameters of a dark soliton such as @xmath42 and @xmath17 in order to compare the profiles of the dark soliton with those of a given quantum dark soliton . here we remark that when a quantum state is given , parameters @xmath3 , @xmath5 and the particle number @xmath1 have been specified . we determine the elliptic modulus @xmath17 from the depth of the dark soliton , i.e. @xmath121 : we solve equation as an equation for modulus @xmath17 . we assign the value @xmath164 on the parameter of density @xmath42 for the density profile of a quantum state . however , we determine the density @xmath42 numerically for the squared amplitude profile of the matrix element of the bosonic field operator between different quantum states , since it may take a value smaller than @xmath164 . for the density profile of a quantum state @xmath165 , the integral of the expectation value @xmath166 of the local density operator over the whole space is equal to @xmath164 . however , for the matrix element @xmath167 of the quantum field operator @xmath168 between two states @xmath165 and @xmath169 , the integral of the squared amplitude @xmath170 over the whole space may be smaller than the value of @xmath164 . in this case we determine the parameter @xmath42 numerically by taking the integral of the squared amplitude @xmath171 over the whole space . here we remark that the symbols will be defined in section 4.1.1 . in summary , after we fix the value of density @xmath42 , we solve equation as a function of modulus @xmath17 and estimate its numerical value very precisely . in section 4 we shall make the profiles of classical dark soliton for given density profiles of quantum dark soliton by the method explained in the above . we define the function @xmath172 by @xmath173 $ ] . it appears in the third term of the expression for the chemical potential @xmath11 . we can show that it is a monotonically decreasing function of modulus @xmath17 , which yields the inequality @xmath174 . this gives an upper bound for the absolute value of soliton velocity @xmath95 : @xmath175 , where @xmath176 here we remark that the critical velocity is defined for the system of a finite size . we shall denote by @xmath177 the critical velocity for an infinite system . we now express a classical dark soliton explicitly in terms of the complete elliptic integrals . the square amplitude is given by @xmath178 we shall show that the second term in the right - hand side of equation leads to the logarithmic correction associated with the conservation of the number of particles . let us introduce a parameter @xmath179 by @xmath180 we denote the argument @xmath181 in by @xmath182 . then , the square amplitude is expressed as follows : @xmath183 we now show that the phase @xmath142 is expressed in terms of jacobi s theta function . let us define a pure imaginary number @xmath184 with @xmath185 by @xmath186 by formula ( [ eq : ei3 ] ) we evaluate the elliptic integral of the third kind with jacobi s zeta and theta functions @xcite , and we express the phase @xmath142 as the logarithm of a ratio of theta functions @xmath187 let us consider the limit of sending the system size @xmath5 to infinity ( @xmath188 ) and the modulus @xmath17 to 1 ( @xmath189 ) simultaneously so that the ratio @xmath190 is kept constant . we denote the ratio by @xmath191 : @xmath192 . here we remark that we keep the density @xmath154 constant . we define @xmath193 by the simultaneous limit of sending @xmath5 to infinity and @xmath17 to 1 with @xmath194 . @xmath195 let us introduce the complementary modulus @xmath196 by @xmath197 . we define @xmath198 and @xmath199 by @xmath200 and @xmath201 , respectively . we remark that @xmath148 and @xmath202 give quarter periods of jacobi s elliptic functions . we introduce a parameter @xmath31 by @xmath203 , which is small when modulus @xmath17 is close to 1 . making use of formulae , , and legendre s relation we expand @xmath148 s and @xmath25 s with respect to @xmath31 as follows : @xmath204 we have @xmath205 , since we have fixed the ratio @xmath194 . we recall that the soliton velocity @xmath95 is expressed in terms of the complete elliptic integral of the third kind . through we express it in terms of jacobi s zeta function as follows : @xmath206 here we recall that parameter @xmath68 has been defined by . let us now evaluate @xmath207 . we first express the left - hand side of in terms of @xmath179 as @xmath208 when the modulus @xmath17 is close to 1 , by making use of jacobi s imaginary transformation of @xmath209 function : @xmath210 and then by expressing it in terms of jacobi s eta and theta functions , we show @xmath211 through we express the value of zeta function @xmath212 in terms of @xmath184 as follows . @xmath213 applying the expansions and to , we evaluate the velocity @xmath95 upto the order of logarithmic terms @xmath214 and through we evaluate the chemical potential up to the order of logarithmic terms @xmath215 it follows that chemical potential @xmath11 approaches @xmath216 when we send system size @xmath5 to infinity . we denote the limiting value by @xmath217 . here we remark that the logarithmic corrections correspond to the finite size corrections since we have @xmath218 . moreover , the chemical potential @xmath11 is larger than @xmath219 when the system size is finite since we have @xmath220 for @xmath221 . we can confirm such behavior by checking the numerical estimates for the parameters of classical dark solitons listed in section 4 . through jacobi s imaginary transformation for @xmath156 function we have @xmath222 we therefore have @xmath223 it is easy to show that the asymptotic expansion of the square amplitude up to the order of @xmath224 , i.e. the order of @xmath225 , satisfies the normalization condition . at @xmath226 , the square amplitude @xmath126 is expressed in terms of @xmath227 by @xmath228 thus , the critical velocity @xmath229 with the infinite system size ( @xmath230 ) is given by @xmath231 . here we recall @xmath232 . it follows that we have @xmath233 \right\ } . \end{aligned}\ ] ] here @xmath193 is expressed as @xmath234 . we have thus derived the square amplitude of one - soliton solution due to tsuzuki @xcite . applying to we derive the following expansion in terms of @xmath235 for any system size @xmath5 : @xmath236 making use of we derive the expansion of @xmath142 up to the order of the inverse of @xmath237 as follows : @xmath238 therefore , in the limit of sending @xmath17 to 1 , we obtain the phase of the dark soliton solution by tsuzuki @xcite : @xmath239 thus , we have shown that the classical dark soliton under the pbcs approaches the dark soliton solution by tsuzuki @xcite through the simultaneous limit of sending the system size to infinity and the modulus @xmath17 to 1 . we define the local density operator @xmath240 by @xmath241 , in the second - quantized system of the 1d bose gas interacting through the delta - function potentials . here we denote by @xmath242 the field operator at the initial time @xmath243 : @xmath244 . we now consider the graph of the expectation value of the local density operator for a quantum state @xmath245 ( i.e. @xmath246 ) versus the position coordinate @xmath62 . here we recall that we call it the density profile of the state @xmath245 . we evaluate the expectation value of the local density operator for the state @xmath247 by the form factor expansion . we express the expectation value as the sum over the form factors between the bethe eigenstates @xmath248 \bra p ' , n|\hat{\rho}(0)|p , n\ket \ , . \label{eq : dp - formfactor}\end{aligned}\ ] ] here @xmath34 and @xmath249 are the normalized bethe eigenstates of @xmath1 particles in the type ii branch ( i.e. they are one - hole excitations ) and have total momentum @xmath33 and @xmath250 , respectively . here the form factors @xmath251 are effectively calculated @xcite by the determinant formula for the norms of bethe eigenstates @xcite and that of the form factors of the density operator @xcite @xmath252 where the quasimomenta @xmath253 and @xmath254 lead to the eigenstates @xmath255 and @xmath256 , respectively . we use the abbreviations @xmath257 and @xmath258 . the kernel @xmath259 is defined by @xmath260 . the matrix @xmath261 is called the gaudin matrix , whose @xmath262 th element is given by @xmath263-{\hat k}(k_{j,\ell } ) \quad \mbox{for } \quad j , \ell=1 , 2 , \cdots , n . \label{eq : gaudin}\ ] ] the matrix elements of the @xmath264 by @xmath264 matrix @xmath265 are given by @xmath266 + \frac{\prod^n_{a=1}(k'_a - k_j)}{\prod^n_{a\neq j}(k_a - k_j ) } \({\hat k}(k_{j,\ell})-{\hat k}(k_{n,\ell})\ ) . \label{eq : matrixu_density}\end{aligned}\ ] ] , i.e. @xmath267 versus @xmath62 , is shown by a red solid line to each value of @xmath3 with @xmath268 and 100 . the density profile of classical dark soliton @xmath269 with parameters shown in table [ density_table20-t1 ] is plotted with a blue broken line to each value of @xmath3 with @xmath268 and 100 . ] , i.e. @xmath270 versus @xmath62 , is shown by a red solid line to each value of @xmath3 with @xmath268 and 100 . the density profile of classical dark soliton @xmath271 with parameters shown in table [ density_table500-t2 ] is plotted with a blue broken line to each of @xmath3 with @xmath268 and 100 . ] .parameters for the density profiles of the classical dark solitons with @xmath272 plotted by blue broken lines in fig . [ density20 ] and in the first and the second columns of fig . [ density_dynamics ] . [ cols="^,^,^,^,^",options="header " , ] the square amplitude profile @xmath273 of quantum dark soliton is consistent with the density profile of classical dark soliton in the weak coupling case of @xmath274 , as shown in the upper left panel of fig . [ field_amp20 ] . moreover , the phase profiles @xmath275 $ ] of quantum dark soliton perfectly agree with those of classical dark soliton for @xmath274 and @xmath276 , as graphs . here , no deviation is found between the quantum and classical profiles . we thus suggest that both the profiles of square amplitude @xmath273 and phase @xmath275 $ ] for the matrix element @xmath277 between the quantum dark soliton states should be consistent with those of classical dark soliton , respectively , in the weak coupling limit : @xmath278 . for @xmath279 the profiles of square amplitude @xmath273 and phase @xmath275/\pi$ ] are plotted in figs [ field_amp500 ] and [ field_phase500 ] , respectively . the density profile @xmath271 and phase profile @xmath280/\pi = \varphi(x - l/2)/\pi$ ] of classical dark soliton are shown by broken blue lines . the numerical estimates of the soliton parameters are given in table [ density_table500 ] . they are obtained by the numerical method of section 3.2.4 . the phase profiles @xmath275/\pi$ ] of the matrix element @xmath277 perfectly agree with those of the corresponding classical dark solitons for the cases of @xmath281 and 1 , respectively . on the other hand , the square amplitude profile @xmath273 of the matrix element @xmath277 is not necessarily in very good agreement with the density profile of classical dark soliton even for the case of @xmath274 . we suggest that it is due to the effect of many - body correlations in the 1d bose gas such as observed in the study of the bec fraction @xcite as mentioned in introduction . we also suggest that if the coupling constant @xmath3 becomes much smaller , the square amplitude profiles @xmath273 should be in much better agreement with the density profiles of classical dark soliton @xmath282 . we now show the time evolution of the square amplitude profile and the phase profile of the matrix element of the bosonic quantum field operator with respect to the quantum dark soliton state with @xmath1 particles and that of @xmath0 particles . we denote by @xmath283 the matrix element of the bosonic quantum field operator @xmath284 between the quantum dark soliton states @xmath39 and @xmath285 as follows . @xmath286 here we recall that the two quantum states have different particle numbers but have the same system size @xmath5 . the snapshots at times @xmath287 for the profiles of the square amplitude and the phase of the matrix element @xmath283 of the bosonic quantum field operator @xmath284 between the quantum dark soliton states @xmath39 and @xmath285 are plotted by red solid lines for the cases of @xmath288 and @xmath289 in figs . [ field_n20_dynamics ] and [ field_n500_dynamics ] , respectively . here , we consider two values of the coupling constant such as @xmath274 and @xmath290 to each of the particle number @xmath1 in the cases of @xmath291 and 500 . for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] of quantum dark soliton with @xmath288 for @xmath274 are shown by red solid lines in the first and second columns from the left , respectively . the snapshots at times @xmath287 of the density profile @xmath294 and phase profile @xmath295/\pi$ ] of classical dark soliton are plotted by blue broken lines in the first and second columns from the left , respectively . here , the soliton parameters are given in table [ density_table20 ] . ( ii ) snapshots at times @xmath287 for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] of quantum dark soliton with @xmath288 for @xmath290 are shown by red solid lines in the third and fourth columns , respectively . the snapshots of the density profile @xmath294 and phase profile @xmath295/\pi$ ] of classical dark soliton are shown by blue broken lines in the third and fourth columns from the left , respectively . here , the soliton parameters are shown in table [ density_table20 ] . ] the quantum and classical profiles are in good agreement in many aspects of the time evolution as shown in figs . [ field_n20_dynamics ] and [ field_n500_dynamics ] . we now describe three aspects as follows . firstly , in the weak coupling case of @xmath274 , both the time evolution of the square amplitude profile @xmath296 and that of the phase profile @xmath293/\pi$ ] are consistent with those of the corresponding classical dark soliton , @xmath294 and @xmath295/\pi$ ] , respectively . the consistency of the quantum profiles of the square amplitude and the phase with the classical profiles of the corresponding classical dark soliton holds over a long period of time such as at least up to @xmath297 as shown in the first and second columns of panels in fig . [ field_n20_dynamics ] from the left . secondly , the quantum profiles move with the same speed as the corresponding classical profiles , as shown in all the eight columns of panels in figs . [ field_n20_dynamics ] and [ field_n500_dynamics ] . in the first and third columns of figs . [ field_n20_dynamics ] the positions of the notches in the square amplitude profiles @xmath292 are in very good agreement with those of the corresponding classical dark solitons . in the second and fourth columns of fig . [ field_n20_dynamics ] , the step positions in the phase profiles @xmath293/\pi$ ] are consistent with those of the phase profiles @xmath295/\pi$ ] of the corresponding classical dark solitons . thirdly , as far as the phase profiles are concerned , the quantum and classical profiles coincide with each other very well for almost all cases as shown in figs . [ field_n20_dynamics ] and [ field_n500_dynamics ] . in fact , not only the step position in the profile of @xmath293 $ ] moves together with that of the phase @xmath295 $ ] of the corresponding classical dark soliton , but also the whole profile itself coincides with the corresponding classical one . \(i ) case of @xmath291 for @xmath274 with @xmath274 and @xmath288 , i.e. the weak coupling case , for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] of the matrix element @xmath283 , the snapshots at times @xmath287 are plotted by red solid lines in the first and second columns of panels of fig . [ field_n20_dynamics ] from the left , respectively . here , five snapshots are shown in each column of panels for @xmath298 and 40 . the snapshots at times @xmath287 for the density profile @xmath299 and phase profile @xmath295/\pi = \varphi(x - vt - l/2)/\pi$ ] of classical dark soliton are shown by blue broken lines in the first and second columns , respectively . the numerical estimates of the soliton parameters are given in table [ density_table20 ] . \(ii ) case of @xmath291 for @xmath290 with @xmath290 and @xmath288 , for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] of the matrix element @xmath283 , the snapshots at times @xmath287 are plotted by red solid lines in the third and fourth columns of panels of fig . [ field_n20_dynamics ] from the left , respectively . here , five snapshots are shown in each column of panels for @xmath300 and 8 . the density profile @xmath299 and phase profile @xmath295/\pi = \varphi(x - vt - l/2)/\pi$ ] of classical dark soliton are shown by blue broken lines in the third and fourth columns , respectively . the numerical estimates of the soliton parameters are given in table [ density_table20 ] . we observe that by comparing the snapshots of @xmath274 with those of @xmath290 , the density notch in the square amplitude profile @xmath292 at @xmath243 decays in a shorter period of time as the coupling constant becomes larger , as shown in the first and third columns of fig . [ field_n20_dynamics ] for @xmath274 and @xmath290 , respectively . however , the position of the density notch moves together with that of classical dark soliton , although the notch itself becomes wider rather rapidly in time . in the phase profiles , the abrut step becomes softer and milder gradually in time , but the change in the phase profile is much slower than in the square amplitude profile . \(iii ) case of @xmath279 for @xmath274 with @xmath274 and @xmath289 , for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] of the matrix element @xmath283 , the snapshots at times @xmath287 are plotted by red solid lines in the first and second columns of panels of fig . [ field_n500_dynamics ] from the left , respectively . here , five snapshots are taken with much longer intervals of time than in the case of @xmath288 in fig . [ field_n20_dynamics ] such as for @xmath301 and 4000 in the case of the square amplitude @xmath292 and for @xmath302 and 16000 in the case of the phase @xmath293 $ ] . the profiles of square amplitude @xmath303 and phase @xmath295/\pi = \varphi(x - vt - l/2)/\pi$ ] are shown by blue broken lines in the first and second columns of fig . [ field_n500_dynamics ] from the left , respectively . the numerical estimates of the soliton parameters are given in table [ density_table500 ] . we observe that the square amplitude @xmath292 for the matrix element of the quantum field operator @xmath284 approaches zero in time evolution . the decaying behavior is clear for @xmath290 as shown in the third column from the left of figs . [ field_n20_dynamics ] and [ field_n500_dynamics ] . in the first column of fig . [ field_n500_dynamics ] from the left , we also observe it for @xmath274 and @xmath289 over a very long period of time up to @xmath304 . however , for the density profiles of quantum dark soliton , the density approaches a nonzero constant value in time evolution , as shown in fig . [ density_dynamics ] . we suggest that the quantum dark soliton states with different particle numbers such as @xmath1 and @xmath0 do not have the same set of energy eigenvalues in common , so that the off - diagonal matrix element @xmath305 vanishes in time , finally . \(iv ) case of @xmath279 for @xmath290 with @xmath290 and @xmath289 , for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] in the matrix element @xmath283 , the snapshots at times @xmath287 are plotted by red solid lines in the third and fourth columns of panels of fig . [ field_n500_dynamics ] from the left , respectively . here , five snapshots are taken for almost the same intervals of time as in the case of fig . [ field_n20_dynamics ] such as for @xmath306 and 12 in both cases of the profiles of square amplitude @xmath292 and phase @xmath293 $ ] . the density profile @xmath299 and phase profile @xmath295/\pi = \varphi(x - vt - l/2)/\pi$ ] of classical dark soliton are shown by blue broken lines in the third and the fourth columns of fig . [ field_n500_dynamics ] from the left , respectively . the numerical estimates of the soliton parameters are given in table [ density_table500 ] . here we remark that when we evaluate the matrix element @xmath283 straightforwardly by making use of the form factor formula , the phase of the matrix element @xmath283 increases with a constant velocity in time , and the phase profile moves downwards with a constant velocity in the animation . we suggest that it is related to the difference between the zero point energy of the eigenstates with @xmath0 particles and that of @xmath1 particles . in figs . [ field_n20_dynamics ] and [ field_n500_dynamics ] we plotted the phase profiles which are subtracted by the constant shift in time . for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] of quantum dark soliton with @xmath289 for @xmath274 are shown by red solid lines in the first and second columns from the left , respectively . the snapshots at times @xmath287 for the density profile @xmath294 and phase profile @xmath295/\pi$ ] of classical dark soliton are plotted by bule broken lines in the first and second columns from the left , respectively . here , the soliton parameters are given in table [ density_table500 ] . ( ii ) snapshots at times @xmath287 for the profiles of square amplitude @xmath292 and phase @xmath293/\pi$ ] of quantum dark soliton with @xmath289 for @xmath290 are shown by red solid lines in the third and fourth columns , respectively . the snapshots at times @xmath287 for the density profile @xmath294 and phase profile @xmath295/\pi$ ] of classical dark soliton are shown in the third and fourth columns from the left , respectively . here , the soliton parameters are shown in table [ density_table500 ] . ] as the quantum dark soliton states for the 1d free fermions with @xmath1 and @xmath0 particles we consider the following @xmath307 we calculate the matrix element of the field operator between the two states as @xmath308 here , the exponential sum is given by the dirichlet kernel . for instance , when we take @xmath309 and momenta @xmath310 for @xmath311 , -1 , and @xmath312 where @xmath313 we have @xmath314 the square - amplitude and phase profiles for the matrix element of the fermionic field operator have several features in common with those of the 1d bose gas in the large coupling case shown in figs . [ field_amp20 ] , [ field_phase20 ] , [ field_amp500 ] , and [ field_phase500 ] , although they do not overlap with those of the 1d bose gas even in the large coupling case . for instance , the square - amplitude profile vanishes at the origin , and it has a density notch around at the origin and small ripples at the shoulders of the notch ; the phase profiles change signs at the origin , etc .. for a given quantum state we denote the expectation value of the hamiltonian measured from the ground state energy by @xmath315 , and the square root of the variance of the expectation value of the hamiltonian by @xmath316 . we define the quantum speed limit time @xmath317 @xcite by @xmath318 we remark that the quantum speed limit time gives a typical lifetime of a generic quantum state . let us evaluate the quantum speed limit time @xmath317 for the quantum dark soliton state @xmath319 in the weak coupling limit where we send the coupling constant @xmath3 to zero . we solve the bethe ansatz equations ( [ bae ] ) in the weak coupling limit . here we recall that the bethe quantum numbers for one - hole excitations are defined by eq . ( [ eq : ij - one - hole ] ) . let us consider an expansion of function @xmath320 which is valid if the absolute value of @xmath62 is larger than 1 . @xmath321 here @xmath322 for @xmath323 while @xmath324 for @xmath325 . applying this expansion we show in the weak coupling case ( @xmath326 ) @xmath327 we have the solution of one - hole excited state @xmath328 approximately as follows . @xmath329 therefore , the one - hole excitation @xmath328 has the energy eigenvalue @xmath330 . let us denote by @xmath331 the expectation value of the energy eigenvalues for the quantum dark soliton state @xmath319 . we have @xmath332 and hence @xmath333 since the ground - state energy is given by 0 in the weak coupling limit , we denote it also by @xmath315 . we calculate the variance of the energy eigenvalues @xmath319 as @xmath334 . we thus have the square root of the variance , @xmath316 , as @xmath335 it follows that the quantum speed limit time for the quantum dark soliton state in the weak coupling limit is given by that for the inverse of the square root of variance @xmath336 @xmath337 here we remark that the quantum speed limit time @xmath317 is constant in the weak coupling limit where only the coupling constant @xmath3 approaches zero while the density @xmath42 and the particle number @xmath1 are kept constant . the estimates of the relaxation time ( or collapse time ) for the density notch in the density profile of quantum dark soliton for the case of @xmath338 are plotted against coupling constant @xmath3 in the double logarithmic scale in fig . [ rt_n20 ] . by observing the time evolution of the density profile of quantum dark soliton for a given value of the coupling constant @xmath3 we estimated the relaxation time by the time when the smallest value of the density notch in the density profile reaches the value of @xmath339 . here we recall that at the initial time @xmath243 the density profile of quantum dark soliton is consistent with that of classical dark soliton , and then the bottom level of the density profile of quantum dark soliton increases gradually in time evolution . in fig . [ rt_n20 ] we observe that the collapse time @xmath340 increases as the coupling constant @xmath3 approaches zero . it is almost inversely proportional to the coupling constant @xmath3 in the weak coupling regime : @xmath341 hence we suggest that it becomes infinitely large in the weak coupling limit . the quantum speed limit time @xmath317 for quantum dark soliton states approaches a non - zero constant value when we send the coupling constant @xmath3 to zero , if we keep the system size @xmath5 and the number of particles @xmath1 fixed . in fact , in the weak coupling limit , all the solutions of the bethe ansatz equations are given by some integral multiples of @xmath342 . hence , the quantum speed limit time of any quantum state is given by some finite value . on the other hand , the collapse time of the density notch in the density profiles of quantum dark soliton is approximately given by the inverse of the coupling constant , @xmath343 , in the very weak coupling case . therefore , the collapse time of the density notch in the density profile of quantum dark soliton is much longer than the quantum speed limit time when the coupling constant @xmath3 is very small . for the quantum dark soliton state with @xmath1 particles in the system size @xmath5 we have approximately @xmath344 in the case of @xmath288 we have a rough estimate : @xmath345 . here we recall that we have set @xmath346 . in fig . [ rt_n20 ] the collapse time is given by approximately 10.0 for @xmath276 . thus , for @xmath347 the collapse time of the density profile in the quantum dark soliton state is much longer than the quantum speed limit time . we conclude that the density notch in the density profile of quantum dark soliton , i.e. that of state @xmath319 , has a much longer lifetime than a generic quantum state when the coupling constant @xmath3 is very small but nonzero in the 1d bose gas interacting through the repulsive delta - function potentials . the authors would like to thank dr . a. del campo for useful comments , in particular , on the quantum speed limit . the present research is partially supported by jsps kakenhi ( grants - in - aid for scientific research ) grant no . 15k05204 . we define the elliptic integrals of the first , second and third kind , denoted by @xmath348 , @xmath349 and @xmath350 , respectively , as follows . @xmath351 the complete elliptic integrals of the first , second and third kind , denoted by @xmath145 , @xmath158 and @xmath352 , respectively , are derived by substituting @xmath353 in ( [ ei1 ] ) , ( [ ei2 ] ) and ( [ ei3 ] ) , i.e. @xmath354 , @xmath355 and @xmath356 , respectively . we define the hypergeometric function @xmath357 in terms of the gamma function @xmath358 by @xmath359 the complete elliptic integrals of the first and second kind , @xmath145 and @xmath158 , are expressed as @xcite @xmath360 jacobi s @xmath209 function is defined by the inverse of the elliptic integral of the first kind @xmath361 we therefore have @xmath362 . jacobi s @xmath363 and @xmath156 functions are related to @xmath364 as @xmath365 hereafter we shall often suppress the modulus @xmath17 . jacobi s amplitude function @xmath366 is defined by the inverse of @xmath367 as @xmath368 we express @xmath369 and @xmath370 as @xmath371 and @xmath372 . making the transformation of variables @xmath373 in the integral expression of @xmath158 such as we have @xmath374 due to the periodicity and the parity invariance of @xmath156 function , we have @xmath375 we introduce the complementary modulus @xmath196 by @xmath376 . we define @xmath198 and @xmath199 by @xmath377 and @xmath378 , respectively . here we remark that jacobi s elliptic functions have quarter periods @xmath148 and @xmath379 . we can show legendre s relation @xcite @xmath380 let us define @xmath381 and @xmath382 by @xmath383 and @xmath384 , respectively @xcite . we have @xmath385 we can show @xmath386 by making transformation of variable @xmath387 in eq . , we have @xmath388 we define jacobi s zeta function @xmath389 by @xmath390 we introduce jacobi s theta function by @xmath391 in terms of @xmath392 we have @xmath393 by taking the logarithmic derivative of jacobi s theta function we have @xmath394 let us take a complex number @xmath395 satisfying the following relation : @xmath396 we can express the elliptic integral of the third kind in terms of jacobi s zeta function as follows . @xmath397 we express the complete elliptic integral of the third kind as follows . @xmath398 we define jacobi s eta function by @xmath399 in terms of eta and theta functions jacobi s @xmath209 , @xmath363 and @xmath156 functions are given by @xmath400 jacobi s imaginary transformation for @xmath209 , @xmath363 and @xmath156 functions is given by @xmath401 jacobi s imaginary transformation of the elliptic integral of the second kind is given by @xmath402 we can derive jacobi s imaginary transformation of zeta function from @xmath403 it follows that we have @xmath404 therefore , in terms of @xmath405 , we have @xmath406 by taking the logarithmic derivative of we expand jacobi s zeta function through as follows . in order to reduce the short - wavelength oscillations or ripples such as shown in the graph of the dirichlet kernel @xmath52 we may modify the truncated sum with some weights . let us denote by @xmath17 the wavenumbers @xmath408 for integers @xmath42 . we take an integer @xmath409 and set @xmath410 as a reference wavenumber . we define @xmath411 by @xmath412 and denote the difference of wavenumbers by @xmath413 . with the gaussian weights @xmath414 we modify the truncated sum as @xmath415 in terms of @xmath411 we express it as @xmath416 we can show that the sum over @xmath411 in does not become very large in the region of @xmath62 satisfying @xmath417 . if @xmath418 , each term in the sum has its absolute value larger than 1 , and hence the sum can be much larger than @xmath58 which is the number of the terms in the sum . here we remark that the bound @xmath419 is given by @xmath420 if we set @xmath421 . we may therefore assume that the sum over @xmath411 in is approximately constant in some region near the origin such as @xmath422 . evaluating the sum by an integral we approximate the gaussian weighted sum @xmath423 as @xmath424 here the error function @xmath425 is given by @xmath426 typically , by putting @xmath427 , i.e. half the period of the short - distance oscillations or ripples in the graph of the function @xmath52 , the ripples disappear , and the gaussian weighted sum @xmath428 approximately becomes a smooth gaussian wave packet with the standard deviation @xmath429 . d. c. wadkin - snaith and d. m. gangardt , phys . * 108 * , 085301 ( 2012 ) . t. karpiuk , p. deuar , p. bienias , e. witkowska , k. pawlowski , m. gajda , k. rzazewski and m. brewczyk , phys . * 109 * , 205302 ( 2012 ) . a. j. leggett , rev . * 73 * ( 2001 ) 307356 . m. rigol , v. dunjko and m. olshanii , nature * 452 * , 854 ( 2008 ) . s. goldstein , t. hara and h. tasaki , new j. phys . * 17 * , 045002 ( 2015 ) l. mandelstam and i. g. tamm , j. phys . ( ussr ) * 9 * , 249 ( 1945 ) . n. margolus and l. b. levitin , physica d * 120 * , 188 ( 1998 ) . v. giovannetti , s. lloyd , and l. maccone , phys . a * 67 * , 052109 ( 2003 ) . m. d. girardeau and e. m. wright , phys . . lett . * 84 * , 5691 ( 2000 ) . j. sato , e. kaminishi , and t. deguchi , arxiv:1303.2775 [ cond-mat.quantum-gas ] . girardeau and e.m . wright , phys . 84 * , 5239 ( 2000 ) . e. b. kolomeisky , t. j. newman , j. p. straley , and x. qi , phys . lett . * 85 * , 1146 ( 2000 ) . t. kawata and h. inoue , j. phys . 44 * , 1722 ( 1978 ) . c. r. nohl , ann . phys . * 96 * , 234 ( 1976 ) ; m. wadati , m. sakagami , j. phys . jpn . * 53 * , 1933 ( 1984 ) ; m. wadati , a. kuniba , t. konishi , j. phys . . jpn . * 54 * , 1710 ( 1985 ) ; m. wadati , a. kuniba , j. phys . soc . jpn . * 55 * , 76 ( 1986 ) . y. lai et al . , phys . a * 40 * , 854 ( 1989 ) . s. j. carter et al . , phys . 58 * , 1841 ( 1987 ) . m. rosenbluh and r.m . shelby , phys . lett . * 66 * , 153 ( 1991 ) ; p. d. drummond et al . , nature * 365 * , 307 ( 1993 ) . k " artner et al . , phys . a * 53 * , 454 ( 1996 ) . j. f. corney et al . , j. opt . b * 18 * , 153 ( 2001 ) . a. d. martin and j. ruostekoski , phys . lett . * 104*,194102 ( 2010 ) . new j. phys . * 12 * , 055018 ( 2010 ) . e. t. copson , _ an introduction to the theory of functions of a complex variable _ ( oxford university press , oxford , 1978 ) . m. gaudin , `` _ la fonction donde de bethe _ '' , masson ( paris ) ( 1983 ) ; v. e. korepin , commun . math . phys . * 86 * , 391 ( 1982 ) . n. a. slavnov , teor . * 79 * , 232 ( 1989 ) ; * 82 * , 389 ( 1990 ) . p. calabrese and j .- s . caux , j. stat . ( 2007 ) p08032 . t. kojima , v.e . korepin , n.a . slavnov , commun . * 188 * , 657 ( 1997 ) j .- s . caux , p. calabrese and n. a. slavnov , j. stat . mech . p01008 ( 2007 ) .

we present submillimeter polarimetry at 850 toward the filamentary star - forming region associated with the reflection nebulosity ngc 2068 in orion b. these data were obtained using the james clerk maxwell telescope s scuba polarimeter . the polarization pattern observed is not consistent with that expected for a field geometry defined by a single mean field direction . there are three distinct distributions of polarization angle , which could represent regions of differing inclination and/or field geometry within the filamentary gas . in general , the polarization pattern does not correlate with the underlying total dust emission . the presence of varying inclinations against the plane of the sky is consistent with the comparison of the 850 continuum emission to the optical emission from the palomar optical sky survey , which shows that the western dust emission lies in the foreground of the optical nebula while the eastern dust emission originates in the background . percentage polarizations are high , particularly toward the north - east region of the cloud . the mean polarization percentage in the region is 5.0% with a standard deviation of 3.1% . depolarization toward high intensities is identified in all parts of the filament .

astro-ph0201348

the study of polarized emission from molecular clouds is of great interest since , at long wavelengths ( @xmath0 ) , it effectively traces the orientation on magnetic fields local to star - forming regions @xcite . magnetic fields have been shown to be energetically comparable to gravity and kinetic motions within molecular clouds @xcite and are theorized to provide vital support to clouds , preventing global collapse ( * ? ? ? * ; * ? ? ? * and references therein ) . such support is necessary to explain the low star - forming efficiencies observed in molecular clouds , including orion b , where the efficiency is @xmath1% @xcite . the orion b cloud , at a distance of @xmath2 pc @xcite , is the closest giant molecular cloud . within it , star formation is concentrated into five distinct regions : ngc 2071 , ngc 2068 , lbs 23 ( hh 24 ) , ngc 2024 and ngc 2023 , as determined from unbiased surveys for young stellar objects @xcite and dense cs gas @xcite . large scale 850 dust emission from these regions has been mapped by @xcite and @xcite . maps of the polarized emission from ngc 2071 , lbs 23 , and ngc 2024 have already appeared in the literature ( see ( * ? ? ? * hereafter paper iii ) ) . in order to compare field geometries across orion b , we have now mapped the polarized emission from ngc 2068 . the submillimeter emission from ngc 2068 lies south of the reflection nebula ( see figure [ p4:pass+scuba ] ) which is seen optically and contains an infrared cluster @xcite . the overall structure of the dust emission is that of a `` clumpy filament '' in which 18 distinct compact sources are identified . most of the submillimeter sources fall outside the boundary of the cluster identified in cs by @xcite . star formation is ongoing in this region , as evidenced by detections of bipolar outflows by @xcite around oribn 51 , from which evidence of outflow previously existed @xcite , and also around oribn 35 and oribn 36 peaks ( see their fig . 1c or figure [ p4:map ] below ) , although the sources of these outflows are ambiguous due to the close proximity of oribn 33 and oribn 37 to the positions of high velocity gas . oribn 39 has a 2mass infrared source associated with it , and thus should be a source of outflow . evidence for redshifted gas near oribn 47 also suggests some outflow from this source . there are no prior observations of polarized emission from the ngc 2068 dust emitting region . however , @xcite measured linear polarization from scattering against the reflection nebula north of the molecular condensations . based on the centrosymmetric pattern observed , @xcite ruled out the presence of aligned grains within the reflection nebula . they infer the presence of a foreground assembly of grains illuminated from behind solely by the star hd 38563n . the position of this star is @xmath3 north - east of the dust emission on which we report in this work , concident with the near - ir cluster observed by @xcite . polarization data probe only two directions of the magnetic field geometry those on the plane of the sky . additionally , they provide no information about the strength of the magnetic field . where field geometries are simple and the direction of the magnetic field does not vary through the cloud depth , the polarized emission detected is perpendicular to the mean magnetic field and the latter can be inferred simply by rotating the polarization vectors by @xmath4 . if the field has a more complex , non - uniform geometry , then interpretation becomes more difficult . in such cases , it is best to compare directly the polarization maps with polarization patterns predicted from a physical model of a magnetized cloud . it is important to recognize that the polarizations measured are vector sums along a particular line of sight through the cloud observed , weighted by column density . @xcite present a model for a filamentary cloud in which a helical magnetic field threads the filament and plays an important role in determining the radial density structure . this model predicts an @xmath5 density profile , which has been observed in several clouds , including the integral - shaped filament @xcite and several clouds in cygnus @xcite . the helical field geometry also predicts depolarization toward the axis of a filament due to cancellation effects on either side of the axis . @xcite present predicted polarization patterns for cases in which the field is either poloidally or toroidally dominated as well as for various filament inclinations . qualitative extensions to these models have been shown to reproduce observed polarization patterns in the filamentary clouds omc-3 ( * ? ? ? * hereafter paper ii ) and ngc 2024 ( see paper iii ) . the filamentary structure of ngc 2068 is therefore of particular interest as a further test of axially symmetric magnetic field geometries . this paper is the fourth in a series which seeks to compare the polarization patterns ( and inferred magnetic field geometries ) in different star - forming regions . the observations and data reduction techniques are described in @xmath6[p4:obs ] . the polarization data are presented and analyzed in @xmath6[p4:poldata ] . the implications of these data for the local magnetic field geometry and that of orion b as a whole are discussed in @xmath6 [ p4:disc ] , and @xmath6 [ p4:summ ] summarizes our results . using the uk / japan polarimeter with the scuba ( submillimeter common user bolometer array ) detector at the james clerk maxwell telescope ( jcmt ) , we have mapped polarized thermal emission from dust at 850 toward a filamentary dust cloud associated with the star - forming region ngc 2068 . the observations were taken from 11 to 16 october 1999 , and additional data were added on 18 february 2000 . the polarizer and general reduction techniques are described in @xcite and @xcite . more information on data reduction and systematic errors can be found in paper ii . six different scuba fields were required to map the entire ngc 2068 filament . initially , four fields were used , but then the number of fields was increased and the positions shifted ; hence 10 independent field centers were used . the field centers and other observing parameters are summarized in table [ p4:observing_parameters ] . the data were obtained using a 16 point jiggle - map mode , in which the telescope is `` jiggled '' in order to completely sample the scuba field . chopping to a reference position was done to remove sky effects . the maximum chop throw possible is 180 . we typically used 150 to ensure we were not chopping onto polarized emission . the extinction of the atmosphere , @xmath7 ( 225 ghz ) , ranged from 0.03 to 0.09 over the observations , but @xmath8% were taken when tau(225 ghz ) was 0.06 or 0.07 . we have reduced the data using the starlink software package polpack , designed to include polarization data obtained with bolometric arrays . the data have been corrected for an error in the scuba computer s clock which placed incorrect lst times in the data headers from july 1999 to may 2000 . this error did not affect the telescope s acquisition or tracking but affects data reduction since the elevation and sky rotation are calculated from the lst times in the data files . the evolution of the magnitude of this error over time has been tracked and can therefore be corrected retroactively by adjusting the times in the headers ( see the jcmt website for details ) . the error in timing after this adjustment is @xmath9s . after extinction correction , noisy bolometers were identified and removed from the data sets . the data were sky subtracted using bolometers with mean values close to zero , but not those which were significantly negative . ( the area mapped by each bolometer in a single jiggle map can be identified with reasonably accuracy , since little sky rotation occurs during the 16-point jiggle . hence , by examining the flux levels of pixels within a bolometer , those in which a majority of adjacent pixels are negative were considered significantly negative and not used . those for which negative and positive values seem equally distributed can be considered to contain zero flux and be used for sky subtraction . ) at 850 , the sky is highly variable on timescales of seconds . this variability must be measured and removed from the data . chopping removes the effects of slow sky variability ; however , fast variations remain in the data , which require sky subtraction using array bolometers devoid of significant flux . we used between one and three bolometers to determine the sky variability , using the existing scan maps of total intensity at 850 @xcite to help select empty bolometers . the methods of sky subtraction are discussed in detail in @xcite . as part of our analyses of other regions , we determined that sky subtraction can be done effectively with a single bolometer ( see paper ii ) and that the polpack reduction process is extremely robust with respect to the selection of sky bolometers ( see paper iii ) . we have not added the mean flux removed by sky subtraction back into the ngc 2068 maps , since the flux in the sky positions was sufficiently close to zero ( as determined from the large scale intensity map of @xcite ) . once the instrumental polarizations ( ips ) were removed from each bolometer , all the data sets were combined to produce a final map in three stokes parameters , @xmath10 , @xmath11 and @xmath12 , where @xmath10 is the total , unpolarized intensity and @xmath11 and @xmath12 are two orthogonal components of linearly polarized light . these three stokes parameters were then combined to yield the polarization percentage , @xmath13 , and polarization position angle , @xmath14 , in the map according to the relations : @xmath15 the uncertainties in each of these quantities are given by : @xmath16 } ; \ \ \ d\theta = 28.6^\circ / \sigma_p \label{p4:uncertainties}\ ] ] where @xmath17 is the signal - to - noise in @xmath13 , or @xmath18 . a bias exists which tends to increase the @xmath13 value , even when @xmath11 and @xmath12 are consistent with @xmath19 , because @xmath13 is forced to be positive . the polarization percentages were debiased according to the expression : @xmath20 future references to polarization percentage , or @xmath13 , refer to the debiased value . in order to improve the @xmath17 , and hence @xmath21 , the data were binned to 15 sampling . this rebinning improves @xmath17 by a factor of five over the unbinned 3 sampled data . good polarization vectors were selected to have @xmath22 ( which implies @xmath23 ) , @xmath24% , @xmath25% and be coincident with positions where the total unpolarized 850 flux exceeds 20% of the faintest of the six compact peaks , oribn 35 . the filtering of polarizations less than 1% accounts for the uncertainties in the ip values , as well as for any contamination due to sidelobe polarization , as discussed in @xcite . sidelobe contamination refers to a polarized signal measured within the main beam due to a source within a sidelobe . if there is enough polarized flux ( @xmath26 ) from a source in a sidelobe position , a signal can be produced in the main beam despite the fact that the power in the jcmt sidelobes at 850 is typically @xmath27% that of the main beam . @xcite derive the minimum believable polarization percentage , @xmath28 , based on : the ratio of power in the sidelobe to the main beam , @xmath29 ; the ratio of unpolarized , total flux of the source in the sidelobe to that in the main beam , @xmath30 ; and the ip estimate at the sidelobe , @xmath31 : @xmath32 we can estimate the extreme value of @xmath28 for the entire ngc 2068 region by calculating the worst case scenario . this occurs for the field centered at @xmath33312 and @xmath340000255 , in which the brightest source in our map , oribn 51 , was approximately 54from the center of the array . the beam responses and ip values are obtained from maps of unpolarized planets . using polarization maps of saturn from 14 october 1999 , 16 october 1999 and 18 february 2000 , the ratios of power in the sidelobe versus main beam are deduced to be 0.009 , 0.01 , and 0.004 respectively . the corresponding mean ip values at this sidelobe position are 4.5% , 4.7% , and 4.3% . the ratio of the flux of ori bn 51 to the center of the field is @xmath35 . hence , using equation ( [ p4:pcrit ] ) , the @xmath28 values are determined to be 0.57% , 0.66% , and 0.24% . thus , by selecting only values of @xmath36% , the polarizations can not be attributable to sidelobe contamination . figure [ p4:map ] shows the polarization data toward the ngc 2068 filamentary cloud . these data are displayed over a portion of the @xcite map of orion b north . the polarization data are binned to 15 , slightly greater than the 14 beamwidth of the jcmt at 850 . thus , each vector can be treated as an independent measurement of the local polarization at 850 . the data are presented in tabular form in appendix [ appendixa ] . by eye , one can readily discern that three different populations of position angles exist within figure [ p4:map ] . separation of the data into three regions based on boundaries in ra can generally separate these populations . figure [ p4:pahist ] shows the distributions in position angle over the whole map , and for each of the three subsections . region 1 encompasses the eastern part of the map , including all vectors east of the j2000 position of @xmath37398 . region 2 is west of region 1 , extending to ra @xmath38338 ( j2000 ) . region 3 covers the western section of the map , including all vectors west of @xmath39338 ( j2000 ) . ngc 2068 contains six compact dust condensations @xcite , as well as a dozen more amorphous , fainter condensations . in the omc-3 filament of orion a , the polarization pattern was no different in the presence of embedded cores than elsewhere along the filament ( paper ii ) . however , in figure [ p4:map ] , polarization vectors near cores are smaller and more variable in direction , particularly near the cores oribn 35 , 47 and especially 51 . in the latter case , no polarization is detected toward the core . toward each of these cores , there is evidence of outflow ; they are not pre - protostellar in nature . we note however that across the outflow source oribn 39 and the compact core oribn 38 , the polarization vectors are consistent with the patterns surrounding the cores . the effect of the cores on the polarization pattern indicates that the filament is not dominating the polarized emission as was the case in omc-3 in orion a ( see @xcite and paper ii ) . in region 1 , the mean polarization position angle is @xmath40 east of north , which is equivalent to @xmath41 in linear polarization ( since vectors offset by @xmath42 are indistinguishable ) . the amorphous cores near the north - east have the highest polarizations , and the vector orientations there are roughly perpendicular to the projected filamentary axis , aligned in a roughly north - east to south - west orientation . interestingly , the vectors along the three cores oribn 34 , 35 and 39 align well with those to the north - east , but now appear to lie parallel to the string of three cores . in region 2 , which includes the cores oribn 37 , 38 and part of 36 , the vectors also follow the string of cores . finally , in region 3 , which includes all cores west of oribn 42 , the polarization vectors are distributed about a mean of @xmath43 , although the projected filament direction varies considerably . the vectors thus tend to follow the filament near oribn 42 to 48 , and then lie perpendicular to the filamentary dust emission along the whole eastern edge . overall , there is no strong correlation between the filament orientation and the vector position angles . table [ p4:poltable ] summarizes the statistics of the three data subsets plus the whole data set for the ngc 2068 region , as well as the results of single gaussian fits to each of the position angle distributions shown in figure [ p4:pahist ] . the mean position angle is denoted @xmath44 , while the width of the distribution is @xmath45 . the reduced chi - squared value is also listed and in each case is @xmath46 . the fit to region 3 gives a mean of @xmath47 with a width of @xmath48 . region 2 gives a mean of @xmath49 with a width of @xmath50 , and region 1 yields a mean of 123@xmath51 and width of @xmath52 . the results of the fits are consistent with the statistical mean and standard deviation in the position angles . the results of a basic statistical analysis of the polarization percentage distributions in the ngc 2068 cloud and its subset of three regions are reported in table [ p4:poltable ] . the mean polarization percentage in the whole filamentary cloud is 5.0% with a standard deviation of 3.1% . similar values are found for each separate region within ngc 2068 . no systematic variation in polarization percentage is obvious from examination of table [ p4:poltable ] . interestingly , region 2 has the highest median value of @xmath13 , and it is the only region where the median exceeds the mean , which indicates that in this region , polarization is high . the mean is not biased by only a few vectors of particularly high polarization percentage . one of the most interesting properties of dust emission polarimetry has been the measurement of declining polarization percentage with increasing unpolarized intensity . figure [ p4:depol ] shows this same trend in the ngc 2068 region and its subregions that has been observed in other regions . over the entire data set , the variation of @xmath13 with total intensity follows a power law of the form : @xmath53 , where @xmath54 and @xmath55 . no systematic variation in polarization percentage between regions is evident from examination of figure [ p4:depol ] . we do note , however , that the effect is strongest in region 3 . in region 1 , the vectors surrounding the cores oribn 31 and 32 are shown as filled triangles , while the rest of the region s vectors are plotted as open triangles . it is clear that the highest polarizations are observed at the extreme north - east of the ngc 2068 region . as discussed in paper ii , a depolarization effect can be produced systematically by chopping the telescope during observing . however , based on that analysis , the steep profiles shown in figure [ p4:depol ] could not be explained by this effect unless there were _ significant _ polarized flux in the chop position . for example , if the reference , or chop , position had 25% of the peak flux in the source field and was polarized to twice the degree of the source field , then a slope of @xmath56 could be produced on a log @xmath13 versus log @xmath10 plot _ in regions of low total intensity_. even under this extreme scenario , in regions of high flux , we would not expect such a sharp decline in polarization percentage with intensity . we thus conclude that the depolarization effect in this region is not produced by systematic effects of chopping . figure [ p4:pcompare ] shows the distributions of polarization percentage in the regions of orion b observed with the scuba polarimeter . these distributions have been normalized to the total number of vectors measured in each region , and the counts expressed as a fraction of the total . the error bars represent the @xmath57 statistical uncertainty in the number of counts , also normalized to the total in each map . the distribution for ngc 2068 is comparable to those measured in the other star - forming regions of orion b north ngc 2071 and lbs23n and with measurements toward the omc-3 filament in orion a ( paper ii ) . the ngc 2024 region is dominated by low percentage polarizations , with the majority between 1 - 2% . this result could imply ngc 2024 has weaker magnetic fields , poorer grain rotation or alignment , a different grain composition , or some combination of these factors . the fact that ngc 2071 , ngc 2068 and lbs 23n show distributions which are reasonably flat from 1 - 6% could indicate similar grain properties , field strengths , and degrees of grain aligment , but this can not be proven with these data . the densities toward the ngc 2024 cores have been estimated to be abnormally high ( @xmath58 @xmath59 @xcite ) , but @xcite suggest the values are more typical of cores ( @xmath60 @xmath59 ) . polarization data at other wavelengths ( such as 350 ) could help constrain the dust properties according to models ( see @xcite and references therein ) , and observations of dense molecular tracers like oh for zeeman splitting could provide more detailed information about the field geometries in these four regions . toward all four regions of orion b observed in polarized emission , the depolarization effect is detected ( see fig . 5 of paper iii and figure [ p4:depol ] ) . paper iii shows that ngc 2024 exhibits a similar depolarization signature to the northern core ngc 2071 . however , lbs 23n has a significantly steeper slope of @xmath61 , which is more consistent with that of region 3 in our data set . this trend has been observed in many other regions as well , including massive cores such as omc-1 @xcite and protostellar and starless cores @xcite . the filamentary dust emission of ngc 2068 likely arises from dust at different depths in the orion b cloud . the comparison between the palomar observatory sky survey optical data and the dust emission has been made by @xcite . we produce a similar image in figure [ p4:pass+scuba ] , which shows that while the western dust emission lines up well with the optically dark dust lane , the eastern dust emission has no corresponding dark lane . ( north of the region lies another dust condensation not yet observed with the polarimeter . ) hence , it is likely that the dust emission may lie on the outer edge of the reflection nebula with the western material in the foreground and the eastern material behind . thus , it is clear that this filament does not lie in the plane of the sky , and that the inclination on the sky is likely variable . an obvious question is whether or not all the dust emission arises from material which is spatially related . it is noteworthy that the core oribn 34 does not appear to line up well with the filamentary material of figure [ p4:map ] . in fact , our comparison of the optical and dust emission shows that the location of oribn 34 is optically dark . this could be a coincidence , or oribn 34 could be closer than the dust emission appearing next to it on the scuba image . oribn 39 also appears dark and could be in the foreground . @xcite present @xmath62co @xmath63 toward ngc 2068 and a partial map of the north - eastern region in c@xmath64o @xmath63 ( see their figs . 6 and 7 ) . the @xmath62co contours are closely correlated to the 850 dust emission . the same is true for the c@xmath64o emission where data exist . these maps are integrated over velocity ranges from 5 km s@xmath65 @xmath66 15 km s@xmath65 and 7 km s@xmath65 @xmath66 13 km s@xmath65 , respectively . thus , the emission is confined to the orion b cloud . the fact that oribn 34 and 39 show similar polarization position angles as the eastern area but may be in the foreground would argue against these regions being completely spatially distinct ( unless the cores are not contributing as much polarized emission as the background diffuse material . if the material is spatially separated , then the similar orientations of the polarization position angles could suggest that the field , if not defined by a single mean field direction , could be organized on spatial scales at least as large as the separation between them . the 2.4 emission from the ngc 2068 reflection nebula spans approximately 6 in spatial extent @xcite . this is consistent with the physical size of the nebula in optical emission as shown in figure [ p4:pass+scuba ] , and at a distance of 415 pc corresponds to 0.7 pc . if the nebula is as deep as it is wide , then the source of dust emission at the north - east could be more than a parsec displaced spatially from the cores at the west ( taking 7 as an estimate of their separation in projection ) . the vectors of ngc 2068 do not support a single field orientation in ngc 2068 . if one assumes that the magnetic field geometry throughout a cloud s depth is reasonably well defined by a mean field direction , then the direction of the field can be obtained by rotating emission polarization data by @xmath4 . doing this in the ngc 2068 polarization map will clearly not produce aligned vectors , any more than the polarization data themselves are aligned . therefore , we can rule out a uniform , unidirectional field across all of ngc 2068 . this was also the case in omc-3 in orion a ( see paper ii ) ; however , in that region , the filamentary axis was easy to define , and a strong correlation between the axis and polarization position angle was observed along 75% of the filament . the vectors of ngc 2068 do not show an obvious alignment of polarization position angle with the filament orientation . we can compare the polarization position angles across those regions observed thus far in orion b. in the three regions of orion b north , no evidence for similar polarization orientations exists . toward ngc 2071 , the polarization vectors align generally with the prominent outflow from the irs 1 source at a position angle of @xmath67 east of north . within the lbs 23n string of cores to the south of ngc 2068 , the vectors are generally aligned north - south ( position angle @xmath68 ) although the scatter in this faint region is considerable . neither of these orientations is dominant in ngc 2068 . thus , there is no support for a mean field direction in ngc 2068 or across the three star - forming regions of orion b north . the one region observed in orion b south is ngc 2024 , and the polarization pattern from that region has been modeled as arising either from a helical field geometry or from the expansion of the ionization front due to the associated region in paper iii . the latter geometry is favored since it is most compatible with the total physics and geometry of ngc 2024 . it is clear that the polarization patterns across these star - forming regions can be strongly correlated with the dust and gas structures of a particular region , and that each region must be modeled separately . as discussed above , there is evidence that the dust arises from a connected gas structure . however , the spatial separation between the filament edges could be @xmath69 pc . the regions of different position angles could indicate regions of different field geometry or inclinations . we propose two model geometries which could potentially explain the variable position angles observed . since the cores are mainly aligned along a filamentary structure , a helical field geometry is appealing . it can explain the confinement of gas in the filament , the fragmentation to cores and the elongation of the asymmetric concentrations along the axis of the filament , which is certainly the case for oribn 31 , 32 , 33 , 41 and 49 . @xcite show possible polarization patterns for different helical field conditions ( i.e. poloidally - dominated versus toroidally - dominated ) . these models are developed for straight filaments , and @xcite find that for such filaments , the polarization vectors align either parallel or perpendicular to the projected filament axis regardless of the inclination of the filament . however , paper ii presents a model for a helical field threading a bent filament . in this case , the vectors may adopt any orientation relative to the filament due to the asymmetries in the filament . the relative alignment depends on the filament s inclination and rotation on the plane of the sky . thus , as inclination and angle in the plane of the sky vary , a helically - threaded filament should produce different polarization position angles relative to the projected filament orientation . for instance , if the polarization positions are aligned with the filament , a toroidally - dominated helical field could wrap the filament locally . conversely , where the vectors appear perpendicular to the filament , a helical field would be poloidally - dominated , or the filament and field could be significantly bent . at the north - western edge near the cores oribn 33 and 41 , the vectors are neither parallel nor perpendicular to the filament . this could also indicate that the filament is bent along its length in this area . modeling such a complex filamentary structure will be difficult , since more than one filament / field geometry can produce similar polarization patterns . one additional test for the presence of a helical field geometry is the radial profile of the total , unpolarized emission . using the 850 map of @xcite , a radial profile can be built up by taking several slices across the filament . we have attempted to confine the cuts to the regions between cores along ngc 2068 . figure [ p4:slice ] shows the profiles through several such slices , taken between the oribn 44 and 48 , between 32 and 39 and between 35 and 36 . for the second cut , both sides of the profile were used ; for the other two , one side of the slice was discarded due to the presence of extended emission from the filament near oribn 47 ( for the cut between oribn 44 and 48 ) and extended emission from core oribn 34 ( for the cut between oribn35 and 36 ) . the fluxes have been normalized to the maximum through the cuts ( respectively 0.07 jy beam@xmath65 , 0.32 jy beam@xmath65 , and 0.28 jy beam@xmath65 ) ; this position is assumed to be the axis of the filament . the ngc 2068 filament is so narrow and faint , it is difficult to interpret the profiles . to guide the eye , we have drawn on figure [ p4:slice ] lines corresponding to slopes of -0.5 and -1 . near the axis , the flux falls off with a distribution consistent with a power law index of @xmath70 . toward lower fluxes , the index could be closer to @xmath71 ; however , at these low flux levels , interpretation of the index becomes difficult . the flux profile of @xmath72 corresponds to a density profile of @xmath73 for an isothermal filament . the helical field geometry predicts a density profile with index @xmath74 for an isothermal equation of state . for other equations of state ( e.g. the logotrope , @xcite ) , a shallower range of density profiles can be generated . unmagnetized , isothermal filaments predict much steeper profiles , with indices of @xmath75 , which are clearly not consistent even with these poor profiles . a second possible field geometry is suggested by the position of the molecular filament so near the periphery of the reflection nebula of ngc 2068 . paper iii presents a model of the polarization data toward the ngc 2024 dense ridge of cores in which the field , compressed by the expansion of the ngc 2024 region , is then moulded around the dense ridge as the ionization front approaches the cores . this picture accounts for both the polarization pattern at 850 and the measurements of the line - of - sight field direction and strength measured by @xcite . the expansion of the reflection nebula could produce a similar effect in ngc 2068 , although the pattern produced is much more complex . in ngc 2024 , the ridge is entirely located on the far - side of the region , whereas in ngc 2068 the comparison of optical and 850 dust emission in figure [ p4:pass+scuba ] indicates that the dust emission arises from both the near and far - side of the nebula . a recent publication on the formation of quiescent cores through turbulent flows @xcite provides a third possible interpretation of the polarization pattern . this work discusses the potential formation of cores due to the presence of super - sonic turbulent flows within molecular clouds . in this model , cores form by accretion along filamentary structures , with the brightest cores forming at the loci of intersecting shocks . the core oribn 47 lies at the intersection of three filamentary segments . as predicted by @xcite , this core exhibits depolarization , although at the distance of orion b , it is difficult to achieve a good sampling of polarization vectors across this core . thus , we do not see the large changes in position angle predicted near cores ( see their fig . 3 ) and routinely observed in bok globules and starless cores in closer regions of star formation ( e.g. see @xcite ) . however , the polarization pattern along the larger scale filamentary structure is well sampled in our map . in the turbulent flows model , the polarization vectors along the filamentary structures are seen to align well with the filaments axes . in ngc 2068 , only the filament segment east of oribn 48 exhibits this behavior . the segments to the north - west and south - west show vectors oriented roughly perpendicular to the filamentary structure , which does not agree well with the turbulent flows picture . a simulation of turbulent flows along filaments of higher density may agree better with our observations in ngc 2068 . if a threshold in extinction exists beyond which grains are not effectively aligned @xcite , then denser turbulent flows could exhibit less correlation between the filamentary axis and the inferred polarization . we have detected polarized emission from aligned dust grains at 850 with the scuba polarimeter . these data reveal strong degrees of polarization ( up to 17% ) toward the ngc 2068 region , with higher polarization percentages detected toward regions of fainter total intensity . this depolarization effect has been observed in most regions of polarized emission . significant depolarization has been detected toward several compact cores along the filament . within the cores of lbs 23n and some cores along the ngc 2024 dense ridge , significant depolarization was also measured ( see paper iii ) we note that this was not the case in the omc-3 filament of orion a , where the polarization patterns behaved consistently in the presence or absence of embedded cores ( see paper i and paper ii ) . this was attributed to the fact that filamentary dust dominated the polarization pattern in orion a , but that this is not the case in orion b. the polarization pattern is inconsistent with that expected for a uniform field geometry , for which vectors should be aligned with one another across the entire region . the polarization position angles are roughly organized into three distinct distributions which do not align with the orientations of the filamentary total intensity emission . this is also different than the pattern observed in the omc-3 part of the integral - shaped filament in paper ii where the vectors aligned with the filament orientation along 75% of the filament s length . taken in comparison with the polarization data for three other star - forming regions in orion b ( paper iii ) , there is also no evidence for a single field orientation across the orion b cloud as a whole . comparison of optical and 850 dust emission establishes that the dust filament is not lying in the plane of the sky , but must be twisted such that its western edge is in the foreground of the optical reflection nebula , while the eastern edge lies behind . these changes in inclination can help explain the polarization pattern observed . paper ii shows that a bent filament threaded by a helical field geometry can produce polarization vectors offset from the filament orientations by varying degrees . material lying on the edge of an expanding reflection nebula could also exhibit different magnetic field orientations at different locations depending on how effectively the field has been compressed by the expansion . the complexity of this filamentary structure will make modeling difficult , since there may be many degenerate choices of filament / field configurations which can produce the observed polarization pattern . these are the first observations of polarized emission toward this filament . the strength of the polarization percentages detected suggests that polarized emission should be detectable at other wavelengths . for instance , observations at 350 with the hertz polarimeter would provide some information on the polarization spectrum for this region . this could lead to constraints on the grain population along the filament . additionally , zeeman data on @xmath76 field would provide significant constraint to the three - dimensional field structure . the authors thank the members of the canadian consortium on star formation for the 850 scan map of the orion b north cloud . thanks to j. greaves , t. jenness , and g. moriarty - schieven at the jcmt for assistance during and after observing . the research of bcm and cdw has been supported through grants from the natural sciences and engineering research council of canada . bcm acknowledges funding from ontario graduate scholarships during the period of this research . alves , j. , lada , c.j . , & lada , e.a . 1999 , , 515 , 265 anthony - twarog , b. j. 1982 , , 87 , 1213 basu , s. 2000 , , 540 , 103 carpenter , j. m. 2000 , , 120 , 3139 crutcher , r. m. , 1999 , , 520 , 706 crutcher , r.m . , roberts , d.a . , troland , t.h . , & goss , w.m . 1999 , , 515 , 275 edwards , s. & snell , r.l . 1984 , , 281 , 237 fiege , j.d . , & pudritz , r.e . 2000a , , 311 , 85 fiege , j.d . , & pudritz , r.e . 2000b , , 544 , 830 gibb , a.g . , & little , l.t . 2000 , , 313 , 663 girart , j.m . , crutcher , r.m . , & rao , r. 1999 , , 525 , 109 greaves , j.s . , holland , w.s . , jenness , t. , moriarty - schieven , g. , chrysostomou , a. , berry , d.s . , murray , a.g . , nartallo , r. , ade , p.a.r . , gannaway , f. , haynes , c.v . , tamura , m. , momose , m. , & morino , j 2001 , , submitted greaves , j.s . , jenness , t. , chrysostomou , a.c . , holland , w.s . , & berry , d.s . 2000 , in imaging at radio through submillimeter wavelengths , eds . mangum & s.j.e . radford , asp - cs 217 , 18 hildebrand , r.h . 1988 , , 29 , 327 hildebrand , r. h. , davidson , j. a. , dotson , j. l. , novak , g. , & vaillancourt , j. e. 2000 , , 112 , 1215 jenness , r. , lightfoot , j.f . , & holland , w.s . 1998 , proc . spie , 3357 , 548 johnstone , d. , & bally , j. 1999 , , 510 , l49 lada , c.j . , alves , j. , & lada , e.a . 1999 , , 512 , 250 lada , e. a. , bally , j. , & stark , a. a. 1991a , , 368 , 432 lada , e. a. , depoy , d. l. , evans , n. j. , & gatley , i. 1991b , , 371 , 171 mannion , m.d . , & scarrott , s.m . 1984 , , 208 , 905 matthews , b. c. , & wilson , c. d. 2000 , , 531 , 868 ( paper i ) matthews , b.c . , fiege , j.d . , & moriarty - schieven , g. 2001a , submitted to ( paper iii ) matthews , b.c . , wilson , c.d . , & fiege , j.d . 2001b , , 562 , 400 ( paper ii ) mckee , c. f. , zweibel , e. g. , & goodman , a. a. 1993 , in _ protostars and planets iii _ , 327 mclaughlin , d. e. , & pudritz , r. e. 1996 , , 469 , 194 mezger , p. g. , chini , r. , kreysa , e. , wink , j. w. , & salter , c. j. 1988 , , 191 , 44 mitchell , g.f . , johnstone , d. , moriarty - schieven , g. , fich , m. , & tothill , n.f.h . 2001 , , 556 , 215 motte , f. , andr , p. , ward - thompson , d. , & bontemps , s. 2001 , , 372 , 41 mouschovias , t. c. 1987 , in _ physical processes in interstellar clouds _ , 453 myers , p.c . , & goodman , a.a . 1988 , , 326 , l27 padoan , p. , goodman , a.a . , draine , b. , juvela , m. , nordlund , . , & rgnvaldsson , . 2001 , , 559 , 1005 schleuning , d.a . , dowell , c.d . , hildebrand , r.h . , platt , s.r . , & novak , g. 1997 , , 109 , 307 schulz , a. , gsten , r. , zylka , r. , & serabyn , e. 1991 , , 246 , 570 sellgren , k. 1984 , , 277 , 623 ward - thompson , d. , kirk , j.m . , crutcher , r.m . , greaves , j.s . , holland , w.s . , & andr , p. 2000 , , 537 , 135 table [ p4:allthedata ] contains the polarization data plotted in figure [ p4:map ] . positional offsets are given from the j2000 coordinates @xmath77338 and @xmath7801430 ( @xmath79000 and @xmath8000000 in b1950 ) . vectors are binned to 15 sampling and all have @xmath22 , @xmath25% , and an absolute uncertainty in polarization percentage , @xmath24% . the total intensity at the vector position exceeds 20% of the faintest compact peak , oribn 35 . this minimizes the chances of systematic effects from chopping to a reference position , as discussed in appendix a of paper ii . cccc @xmath38265 & @xmath810000219 & 40 & 3 + @xmath38305 & @xmath820001484 & 90 & 3 + @xmath38376 & @xmath810000331 & 155 & 3 + @xmath38479 & @xmath810001003 & 155 & 3 + @xmath38257 & @xmath810000349 & 40 & 17 + @xmath38287 & @xmath820001553 & 90 & 18 + @xmath38312 & @xmath820000255 & 165 & 18 + @xmath38392 & @xmath810000489 & 155 & 13 + @xmath38447 & @xmath810000335 & 155 & 12 + @xmath38503 & @xmath810001231 & 155 & 11 + [ p4:observing_parameters ] cccccccccc 1 & 68 & 4.9 & 3.6 & 3.6 & 109.2 & 41 & 123 & 24 & 0.7 + 2 & 26 & 4.7 & 5.2 & 2.2 & 93.4 & 16 & 93 & 16 & 0.4 + 3 & 69 & 5.2 & 4.8 & 2.9 & 74.3 & 35 & 66 & 18 & 0.7 + all & 163 & 5.0 & 4.3 & 3.1 & 91.4 & 40 & 95 & 41 & 0.2 + [ p4:poltable ] rrrrrrr @xmath83 & @xmath84 & 11.22 & 0.96 & 11.7 & @xmath85 & 2.5 + @xmath83 & @xmath86 & 1.65 & 0.28 & 5.8 & @xmath87 & 4.9 + @xmath88 & @xmath86 & 2.39 & 0.47 & 5.1 & @xmath89 & 5.6 + @xmath90 & @xmath91 & 8.62 & 0.85 & 10.1 & @xmath92 & 2.8 + @xmath88 & @xmath91 & 1.24 & 0.20 & 6.2 & @xmath93 & 4.6 + @xmath94 & @xmath91 & 5.17 & 0.59 & 8.7 & @xmath95 & 3.3 + @xmath96 & @xmath97 & 2.01 & 0.57 & 3.5 & @xmath98 & 8.1 + @xmath88 & @xmath97 & 1.96 & 0.31 & 6.2 & @xmath99 & 4.6 + @xmath94 & @xmath97 & 3.40 & 0.44 & 7.8 & @xmath100 & 3.7 + @xmath101 & @xmath97 & 6.16 & 0.65 & 9.5 & @xmath102 & 3.0 + @xmath94 & @xmath103 & 6.91 & 0.41 & 16.9 & @xmath104 & 1.7 + @xmath101 & @xmath103 & 4.46 & 0.45 & 10.0 & @xmath105 & 2.9 + @xmath106 & @xmath103 & 7.21 & 0.62 & 11.6 & @xmath107 & 2.5 + @xmath94 & @xmath108 & 4.11 & 0.43 & 9.5 & @xmath109 & 3.0 + @xmath101 & @xmath108 & 4.75 & 0.30 & 15.7 & @xmath110 & 1.8 + @xmath106 & @xmath108 & 2.38 & 0.39 & 6.1 & @xmath111 & 4.7 + @xmath94 & @xmath112 & 6.16 & 0.46 & 13.5 & @xmath113 & 2.1 + @xmath101 & @xmath112 & 2.65 & 0.27 & 9.9 & @xmath114 & 2.9 + @xmath106 & @xmath112 & 5.59 & 0.37 & 15.2 & @xmath115 & 1.9 + @xmath88 & @xmath116 & 4.26 & 0.52 & 8.2 & @xmath117 & 3.5 + @xmath94 & @xmath116 & 3.10 & 0.29 & 10.5 & @xmath118 & 2.7 + @xmath101 & @xmath116 & 3.86 & 0.25 & 15.2 & @xmath119 & 1.9 + @xmath106 & @xmath116 & 2.28 & 0.33 & 7.0 & @xmath120 & 4.1 + @xmath83 & @xmath121 & 6.03 & 0.43 & 14.0 & @xmath122 & 2.0 + @xmath88 & @xmath121 & 5.70 & 0.32 & 17.6 & @xmath123 & 1.6 + @xmath94 & @xmath121 & 3.22 & 0.23 & 14.1 & @xmath124 & 2.0 + @xmath101 & @xmath121 & 2.86 & 0.15 & 19.6 & @xmath125 & 1.5 + @xmath106 & @xmath121 & 2.06 & 0.19 & 10.9 & @xmath126 & 2.6 + @xmath96 & @xmath127 & 8.99 & 0.46 & 19.5 & @xmath128 & 1.5 + @xmath83 & @xmath127 & 6.78 & 0.59 & 11.4 & @xmath129 & 2.5 + @xmath94 & @xmath127 & 5.12 & 0.42 & 12.2 & @xmath130 & 2.4 + @xmath101 & @xmath127 & 1.12 & 0.15 & 7.7 & @xmath131 & 3.7 + @xmath106 & @xmath127 & 3.22 & 0.22 & 14.5 & @xmath132 & 2.0 + @xmath133 & @xmath127 & 4.21 & 0.62 & 6.8 & @xmath134 & 4.2 + @xmath135 & @xmath127 & 7.42 & 0.80 & 9.3 & @xmath136 & 3.1 + @xmath90 & @xmath137 & 6.00 & 0.63 & 9.5 & @xmath138 & 3.0 + @xmath96 & @xmath137 & 4.32 & 0.52 & 8.3 & @xmath139 & 3.5 + @xmath101 & @xmath137 & 12.43 & 0.63 & 19.7 & @xmath140 & 1.5 + @xmath106 & @xmath137 & 14.80 & 0.56 & 26.5 & @xmath141 & 1.1 + @xmath142 & @xmath137 & 11.26 & 0.74 & 15.3 & @xmath143 & 1.9 + @xmath133 & @xmath137 & 2.80 & 0.39 & 7.2 & @xmath144 & 4.0 + @xmath135 & @xmath137 & 4.12 & 0.34 & 12.3 & @xmath145 & 2.3 + @xmath146 & @xmath137 & 5.16 & 0.76 & 6.8 & @xmath147 & 4.2 + @xmath148 & @xmath149 & 3.66 & 0.69 & 5.3 & @xmath150 & 5.4 + @xmath90 & @xmath149 & 1.78 & 0.45 & 3.9 & @xmath151 & 7.3 + @xmath133 & @xmath149 & 4.33 & 0.49 & 8.9 & @xmath152 & 3.2 + @xmath135 & @xmath149 & 1.18 & 0.35 & 3.4 & @xmath153 & 8.4 + @xmath146 & @xmath149 & 4.88 & 0.59 & 8.3 & @xmath154 & 3.5 + @xmath155 & @xmath156 & 5.22 & 0.67 & 7.8 & @xmath157 & 3.7 + @xmath148 & @xmath156 & 3.30 & 0.45 & 7.4 & @xmath158 & 3.9 + @xmath142 & @xmath156 & 6.96 & 0.71 & 9.8 & @xmath159 & 2.9 + @xmath133 & @xmath156 & 4.77 & 0.33 & 14.3 & @xmath160 & 2.0 + @xmath135 & @xmath156 & 5.11 & 0.28 & 18.1 & @xmath99 & 1.6 + @xmath146 & @xmath156 & 7.09 & 0.48 & 14.9 & @xmath161 & 1.9 + @xmath162 & @xmath156 & 7.00 & 0.91 & 7.7 & @xmath163 & 3.7 + @xmath164 & @xmath165 & 7.74 & 0.82 & 9.4 & @xmath166 & 3.0 + @xmath167 & @xmath165 & 2.41 & 0.61 & 4.0 & @xmath168 & 7.2 + @xmath169 & @xmath165 & 1.42 & 0.30 & 4.8 & @xmath170 & 6.0 + @xmath171 & @xmath165 & 1.72 & 0.38 & 4.6 & @xmath172 & 6.3 + @xmath173 & @xmath165 & 1.39 & 0.46 & 3.0 & @xmath174 & 9.5 + @xmath175 & @xmath165 & 1.47 & 0.39 & 3.8 & @xmath176 & 7.6 + @xmath177 & @xmath165 & 2.88 & 0.48 & 6.0 & @xmath95 & 4.8 + @xmath178 & @xmath165 & 2.97 & 0.46 & 6.4 & @xmath179 & 4.5 + @xmath180 & @xmath165 & 4.06 & 0.46 & 8.9 & @xmath181 & 3.2 + @xmath123 & @xmath165 & 4.06 & 0.56 & 7.3 & @xmath182 & 3.9 + @xmath183 & @xmath165 & 4.47 & 0.67 & 6.6 & @xmath184 & 4.3 + @xmath185 & @xmath165 & 7.57 & 0.74 & 10.2 & @xmath186 & 2.8 + @xmath187 & @xmath165 & 7.18 & 0.62 & 11.5 & @xmath188 & 2.5 + @xmath155 & @xmath165 & 5.39 & 0.46 & 11.7 & @xmath189 & 2.4 + @xmath148 & @xmath165 & 5.09 & 0.45 & 11.3 & @xmath190 & 2.5 + @xmath90 & @xmath165 & 3.51 & 0.71 & 4.9 & @xmath191 & 5.8 + @xmath142 & @xmath165 & 6.93 & 0.62 & 11.2 & @xmath192 & 2.6 + @xmath135 & @xmath165 & 3.84 & 0.42 & 9.1 & @xmath193 & 3.1 + @xmath146 & @xmath165 & 10.14 & 0.70 & 14.4 & @xmath194 & 2.0 + @xmath164 & @xmath195 & 3.63 & 0.49 & 7.4 & @xmath196 & 3.9 + @xmath167 & @xmath195 & 2.48 & 0.29 & 8.6 & @xmath197 & 3.3 + @xmath169 & @xmath195 & 1.39 & 0.15 & 9.0 & @xmath198 & 3.2 + @xmath171 & @xmath195 & 2.18 & 0.23 & 9.5 & @xmath199 & 3.0 + @xmath173 & @xmath195 & 1.80 & 0.41 & 4.4 & @xmath200 & 6.5 + @xmath175 & @xmath195 & 1.44 & 0.31 & 4.6 & @xmath201 & 6.2 + @xmath202 & @xmath195 & 2.54 & 0.26 & 9.8 & @xmath154 & 2.9 + @xmath177 & @xmath195 & 1.59 & 0.27 & 5.9 & @xmath203 & 4.9 + @xmath178 & @xmath195 & 2.14 & 0.29 & 7.4 & @xmath204 & 3.9 + @xmath180 & @xmath195 & 5.08 & 0.26 & 19.2 & @xmath160 & 1.5 + @xmath123 & @xmath195 & 1.43 & 0.31 & 4.7 & @xmath205 & 6.2 + @xmath183 & @xmath195 & 3.34 & 0.27 & 12.3 & @xmath204 & 2.3 + @xmath185 & @xmath195 & 4.38 & 0.27 & 16.3 & @xmath206 & 1.8 + @xmath187 & @xmath195 & 5.37 & 0.32 & 16.9 & @xmath207 & 1.7 + @xmath155 & @xmath195 & 5.23 & 0.33 & 15.7 & @xmath208 & 1.8 + @xmath148 & @xmath195 & 6.04 & 0.42 & 14.3 & @xmath209 & 2.0 + @xmath90 & @xmath195 & 4.19 & 0.84 & 5.0 & @xmath210 & 5.7 + @xmath133 & @xmath195 & 7.35 & 0.65 & 11.3 & @xmath211 & 2.5 + @xmath135 & @xmath195 & 12.61 & 0.70 & 18.0 & @xmath212 & 1.6 + @xmath213 & @xmath214 & 4.22 & 0.86 & 4.9 & @xmath83 & 5.9 + @xmath164 & @xmath214 & 5.38 & 0.57 & 9.4 & @xmath215 & 3.0 + @xmath167 & @xmath214 & 2.83 & 0.42 & 6.7 & @xmath216 & 4.3 + @xmath169 & @xmath214 & 3.91 & 0.34 & 11.6 & @xmath217 & 2.5 + @xmath171 & @xmath214 & 4.35 & 0.41 & 10.5 & @xmath218 & 2.7 + @xmath173 & @xmath214 & 4.84 & 0.53 & 9.2 & @xmath219 & 3.1 + @xmath175 & @xmath214 & 2.47 & 0.33 & 7.4 & @xmath220 & 3.9 + @xmath202 & @xmath214 & 1.55 & 0.19 & 8.1 & @xmath221 & 3.5 + @xmath177 & @xmath214 & 1.10 & 0.19 & 5.7 & @xmath222 & 5.0 + @xmath180 & @xmath214 & 3.64 & 0.24 & 15.1 & @xmath223 & 1.9 + @xmath123 & @xmath214 & 2.84 & 0.31 & 9.2 & @xmath224 & 3.1 + @xmath183 & @xmath214 & 1.70 & 0.33 & 5.2 & @xmath225 & 5.6 + @xmath185 & @xmath214 & 1.94 & 0.34 & 5.8 & @xmath226 & 5.0 + @xmath187 & @xmath214 & 9.12 & 0.39 & 23.5 & @xmath227 & 1.2 + @xmath155 & @xmath214 & 5.51 & 0.48 & 11.5 & @xmath228 & 2.5 + @xmath148 & @xmath214 & 9.28 & 0.96 & 9.7 & @xmath229 & 3.0 + @xmath106 & @xmath214 & 5.40 & 0.52 & 10.4 & @xmath230 & 2.7 + @xmath142 & @xmath214 & 6.52 & 0.64 & 10.3 & @xmath231 & 2.8 + @xmath232 & @xmath233 & 5.51 & 0.95 & 5.8 & @xmath234 & 5.0 + @xmath213 & @xmath233 & 4.84 & 0.54 & 8.9 & @xmath235 & 3.2 + @xmath164 & @xmath233 & 4.28 & 0.53 & 8.0 & @xmath236 & 3.6 + @xmath167 & @xmath233 & 2.82 & 0.43 & 6.6 & @xmath237 & 4.3 + @xmath169 & @xmath233 & 2.89 & 0.37 & 7.7 & @xmath238 & 3.7 + @xmath171 & @xmath233 & 3.55 & 0.53 & 6.7 & @xmath239 & 4.3 + @xmath173 & @xmath233 & 3.93 & 0.67 & 5.9 & @xmath240 & 4.9 + @xmath202 & @xmath233 & 1.53 & 0.29 & 5.2 & @xmath241 & 5.5 + @xmath178 & @xmath233 & 1.75 & 0.35 & 5.0 & @xmath242 & 5.7 + @xmath180 & @xmath233 & 3.05 & 0.25 & 12.0 & @xmath239 & 2.4 + @xmath123 & @xmath233 & 1.66 & 0.28 & 6.0 & @xmath243 & 4.7 + @xmath185 & @xmath233 & 5.60 & 0.82 & 6.8 & @xmath244 & 4.2 + @xmath187 & @xmath233 & 7.16 & 0.88 & 8.1 & @xmath245 & 3.5 + @xmath106 & @xmath233 & 2.51 & 0.35 & 7.2 & @xmath89 & 4.0 + @xmath142 & @xmath233 & 4.12 & 0.40 & 10.3 & @xmath246 & 2.8 + @xmath213 & @xmath247 & 2.21 & 0.62 & 3.6 & @xmath248 & 8.0 + @xmath164 & @xmath247 & 2.74 & 0.37 & 7.4 & @xmath221 & 3.9 + @xmath167 & @xmath247 & 3.06 & 0.26 & 11.5 & @xmath249 & 2.5 + @xmath169 & @xmath247 & 5.40 & 0.36 & 15.1 & @xmath250 & 1.9 + @xmath171 & @xmath247 & 10.42 & 0.72 & 14.4 & @xmath251 & 2.0 + @xmath173 & @xmath247 & 7.38 & 0.80 & 9.2 & @xmath252 & 3.1 + @xmath175 & @xmath247 & 5.88 & 0.73 & 8.1 & @xmath253 & 3.5 + @xmath202 & @xmath247 & 5.63 & 0.64 & 8.8 & @xmath254 & 3.3 + @xmath177 & @xmath247 & 3.20 & 0.45 & 7.1 & @xmath255 & 4.1 + @xmath178 & @xmath247 & 3.38 & 0.40 & 8.4 & @xmath256 & 3.4 + @xmath180 & @xmath247 & 1.90 & 0.29 & 6.5 & @xmath257 & 4.4 + @xmath123 & @xmath247 & 1.95 & 0.26 & 7.4 & @xmath258 & 3.9 + @xmath101 & @xmath247 & 5.90 & 0.66 & 9.0 & @xmath259 & 3.2 + @xmath106 & @xmath247 & 3.76 & 0.36 & 10.5 & @xmath260 & 2.7 + @xmath142 & @xmath247 & 3.60 & 0.53 & 6.8 & @xmath261 & 4.2 + @xmath213 & @xmath262 & 13.06 & 0.96 & 13.6 & @xmath263 & 2.1 + @xmath164 & @xmath262 & 6.99 & 0.53 & 13.2 & @xmath264 & 2.2 + @xmath167 & @xmath262 & 6.16 & 0.31 & 20.1 & @xmath265 & 1.4 + @xmath169 & @xmath262 & 8.99 & 0.34 & 26.1 & @xmath266 & 1.1 + @xmath171 & @xmath262 & 17.07 & 0.70 & 24.5 & @xmath267 & 1.2 + @xmath178 & @xmath262 & 8.63 & 0.86 & 10.1 & @xmath268 & 2.8 + @xmath180 & @xmath262 & 2.83 & 0.56 & 5.0 & @xmath96 & 5.7 + @xmath183 & @xmath262 & 5.62 & 0.92 & 6.1 & @xmath269 & 4.7 + @xmath101 & @xmath262 & 6.53 & 0.75 & 8.7 & @xmath270 & 3.3 + @xmath106 & @xmath262 & 2.93 & 0.52 & 5.6 & @xmath271 & 5.1 + @xmath142 & @xmath262 & 8.23 & 0.90 & 9.1 & @xmath272 & 3.1 + @xmath164 & @xmath273 & 8.20 & 0.77 & 10.6 & @xmath274 & 2.7 + @xmath167 & @xmath273 & 8.08 & 0.43 & 18.6 & @xmath156 & 1.5 + @xmath169 & @xmath273 & 15.63 & 0.51 & 30.5 & @xmath151 & 0.9 + @xmath232 & @xmath275 & 3.27 & 0.83 & 3.9 & @xmath276 & 7.3 + @xmath213 & @xmath275 & 6.25 & 0.58 & 10.8 & @xmath277 & 2.7 + @xmath164 & @xmath275 & 6.74 & 0.47 & 14.3 & @xmath165 & 2.0 + @xmath167 & @xmath275 & 6.48 & 0.54 & 12.1 & @xmath278 & 2.4 + @xmath169 & @xmath275 & 7.45 & 0.95 & 7.8 & @xmath279 & 3.7 + @xmath213 & @xmath280 & 6.72 & 0.73 & 9.2 & @xmath281 & 3.1 + @xmath164 & @xmath280 & 12.65 & 0.59 & 21.6 & @xmath282 & 1.3 + @xmath167 & @xmath280 & 15.56 & 0.93 & 16.7 & @xmath283 & 1.7 + [ p4:allthedata ]

we theoretically study counterflow instability and turbulence in a spin-1 spinor bose einstein condensate by the gross pitaevskii equation and the bogoliubov de gennes equation . our study considers ( i ) the dynamics induced by the counterflow of two components with different magnetic quantum numbers , which leads to turbulence with spin degrees of freedom , and ( ii ) the properties of the turbulence . for ( i ) , the behavior of the condensate induced by the counterflow strongly depends on whether the spin - dependent interaction is ferromagnetic or antiferromagnetic , leading to different behaviors for the dispersion relation and the spin density vector , . for ( ii ) , we numerically calculate the spectrum of the spin - dependent interaction energy , which also depends on the spin - dependent interaction . the spectrum of the spin - dependent interaction energy in the ferromagnetic case exhibits a power law , whereas that in the antiferromagnetic case does not . the power law can be explained by scaling analysis .

1112.2443

turbulence is one of the most important topics in modern physics . turbulence in classical fluids has been studied for a long time @xcite , but turbulence is also observed in diverse fields such as low - temperature physics , plasma physics , and astronomy , @xmath0 . kolmogorov studied turbulence in classical fluids in 1941 @xcite and significantly advanced the study of turbulence . he found that the kinetic energy spectrum obeyed a @xmath2 power law , which became known as the kolmogorov @xmath2 power law . several methods have been used for generating turbulence in classical fluids . reynolds performed a famous experiment in 1883 in which he created turbulence by using flow through a circular pipe . another method uses hydrodynamic instabilities , which have been investigated for a long time . there are many kinds of instabilities such as kelvin helmholtz ( kh ) and rayleigh taylor ( rt ) instabilities @xcite which can lead to turbulence . hydrodynamic instability has recently been studied in atomic bose einstein condensates ( becs ) . atomic becs exhibit different dynamics from classical fluids because they are quantum fluids . quantized vortices occur in becs ; these vortices are nucleated through the hydrodynamic instability . kh and rt instabilities have been studied in two - component atomic becs . for kh instability , quantized vortices nucleate at the boundary layer between the two components @xcite , whereas rt instability generates the shape of mushroom in the condensate winded by the quantized vortex ring @xcite . moreover , counterflow instability in two - component atomic becs results in vortices that nucleate by the collapse of solitons @xcite . thus , hydrodynamic instability has been extensively investigated in two - component atomic becs . unlike two - component atomic becs , spinor becs have spin degrees of freedom and exhibit phenomena characteristic of spin . collisions of spin-1 spinor becs have been investigated numerically and show different results from one- and two - component becs @xcite . the hydrodynamic equation for spinor becs has recently been studied @xcite . however , hydrodynamic instability in spinor becs has been investigated less than in two - component becs . we expect that hydrodynamic instability in spinor becs will exhibit unique behavior due to their spin degrees of freedom and form turbulent states in which the spin density vector has different directions . quantum turbulence has been investigated for a long time in superfluid @xmath3 and @xmath4 @xcite , being currently studied in atomic becs too @xcite . numerical studies predict that the kolmogorov @xmath2 power law , which was first observed in classical fluids , also holds in atomic becs @xcite . turbulence in two - component becs has been investigated @xcite . we expect that the turbulent state in a spin-1 spinor bec will exhibit properties characteristic of the spin degrees of freedom , which one- and two - component becs do not show . this is one of the major themes of this paper . in this paper , we focus on counterflow instability in spin-1 spinor becs in a homogeneous two - dimensional system . there are three reasons for studying the instability in this system . the first reason is that the dynamics of spin-1 spinor becs induced by counterflow exhibit characteristic behaviors . dynamics peculiar to counterflow have been observed in two - component becs @xcite . however , there are distinct differences between two - component becs and spinor becs . a spinor bec has spin degrees of freedom . the number of particles of each component is conserved in a two - component bec , whereas it is not conserved in a spinor bec because of the spin - dependent interaction . therefore , we expect that a spin-1 spinor bec will exhibit dynamics characteristic of not only the counterflow but also the spin degrees of freedom . the second reason for studying counterflow instability in spin-1 spinor becs is that counterflow instability can generate turbulence in spin-1 spinor becs . as stated above , one aim of this study is to investigate the behavior of turbulence in spin-1 spinor becs . the third reason is that counterflow in two - components becs has been experimentally investigated @xcite . we expect that it may be possible to experimentally study counterflow of spin-1 spinor becs . hence , we study the counterflow instability of spin-1 spinor becs using the gross pitaevskii ( gp ) equation and the bogoliubov de gennes ( bdg ) equation . we consider counterflow between the @xmath5 and @xmath6 components , where @xmath7 is the magnetic quantum number . our main purposes are to investigate phenomena characteristic of the spin degrees of freedom induced by the counterflow instability and the properties of a turbulent state in spin-1 spinor becs . for the instability , we investigate the pattern of the particle number density and the time dependence of the magnitude of the spin density vector , @xmath0 . for the turbulence of a spin-1 spinor bec , we focus on statistical quantities such as the probability density function ( pdf ) of the magnitude of the spin density vector and the spectrum of the spin - dependent interaction energy . the results obtained reveal that the behaviors of the instability and the turbulent state induced by the counterflow depend greatly on the sign of the spin - dependent interaction . we calculate the dispersion relation obtained from the bdg equation , the pdf of the magnitude of the spin density vector and the spectrum of the spin - dependent interaction energy ; these quantities exhibit different behaviors depending on whether the spin - dependent interaction is ferromagnetic or antiferromagnetic . as for the spectrum of the spin - dependent interaction energy , the @xmath1 power law is clearly found in the ferromagnetic case , but not in the antiferromagnetic case . the @xmath1 power law can be understood by the scaling analysis of the time development equation of the spin density vector . this paper is organized as follows . section ii describes the formulation . in sec . iii , we analytically calculate the bdg equation . section iv presents the numerical results related to the dynamical instability induced by counterflow . the turbulent state in a spin-1 spinor bec is treated in sec . v. in sec . vi , we discuss some problems of our study . finally , we summarize the findings in sec . we consider a spin-1 spinor bec in a homogeneous two - dimensional system at zero temperature because this system is easy to study theoretically and is well described by macroscopic wave functions @xmath8 ( @xmath9 ) . for simplicity , we do not treat a magnetic field or the dipole dipole interaction . the wave functions @xmath8 then obey the gp equation @xcite : @xmath10 where @xmath11 is the mass of a particle . the total density @xmath12 and the spin density vector @xmath13 are respectively given by @xmath14 @xmath15 where @xmath16 are the spin-1 matrices : @xmath17 @xmath18 @xmath19 the parameters @xmath20 and @xmath21 are the coefficients of the spin - independent and dependent interactions for two - dimensional system . the total energy @xmath22 is given by @xmath23 d \bm{r } \\ & + & \frac{c_{0}}{2 } \int n^{2 } d \bm{r } + \frac{c_{1}}{2 } \int \bm{s } ^{2 } d \bm{r}.\end{aligned}\ ] ] the spin - dependent interaction energy is the last term with the coefficient @xmath21 on the right - hand side of eq . ( 7 ) . the ground state in a homogeneous system without a magnetic field is ferromagnetic for @xmath24 and polar for @xmath25 . the total particle number and the spin in the @xmath26 direction are conserved in the gp model without a magnetic field and the dipole - dipole interaction , namely @xmath27 @xmath28 with @xmath29 . equations ( 8) and ( 9 ) show that @xmath30 has the same time evolution as @xmath31 and that the change of @xmath32 is related to that of @xmath30 and @xmath31 ; this is important for understanding the instability induced by the counterflow ( see sec . iv ) . we consider the counterflow between the @xmath5 and @xmath6 components with a relative velocity @xmath33 in a homogeneous two - dimensional system , where @xmath34 is a unit vector along the x direction . the initial state is expressed by @xmath35 \\ 0 \\ { \rm{exp}}[-i(\frac{m}{2 \hbar}\bm{v}_{r}\cdot \bm{r } + \frac{\mu _ { -1}}{\hbar}t ) ] \end{pmatrix } , \ ] ] where @xmath36 is the total density and @xmath37 and @xmath38 are the chemical potentials and are equal to @xmath39 . we use this initial state to investigate the counterflow instability and the turbulent state in a spin-1 spinor bec . we use the crank nicholson method to numerically calculate the gp equation starting from the initial state of eq . the coordinate is normalized by the coherence length @xmath40 and the box size is @xmath41 . space in the @xmath42 and @xmath43 directions is discretized into @xmath44 bins . the time is normalized by @xmath45 . we add some small white noise to the initial state of eq . ( 10 ) ; without this noise , the instability can not be generated . in our calculations , the noise forms the particles of the @xmath46 component , which accounts for @xmath47 of the total particle number . this is consistent with experimental results @xcite . in this section , we consider a small deviation @xmath48 from the initial state of eq . ( 10 ) , whose dispersion relation can be obtained by linear analysis . and @xmath49 . the upper and lower graphs are respectively the imaginary parts of @xmath50 and @xmath51 . the vertical and horizontal axes are respectively the relative velocity and wave number in the @xmath42 direction , which is normalized by the sound velocity @xmath52 and the coherence length @xmath40 . the imaginary part of @xmath51 has finite values for any finite relative velocity @xmath53 , while that of @xmath50 has finite values only for a relative velocity larger than the critical value @xmath54 . , title="fig:",width=264 ] and @xmath49 . the upper and lower graphs are respectively the imaginary parts of @xmath50 and @xmath51 . the vertical and horizontal axes are respectively the relative velocity and wave number in the @xmath42 direction , which is normalized by the sound velocity @xmath52 and the coherence length @xmath40 . the imaginary part of @xmath51 has finite values for any finite relative velocity @xmath53 , while that of @xmath50 has finite values only for a relative velocity larger than the critical value @xmath54 . , title="fig:",width=264 ] and @xmath49 . the upper and lower graphs are respectively the imaginary parts of @xmath50 and @xmath51 . the vertical and horizontal axes are respectively the relative velocity and wave number of the @xmath42 direction , which is normalized by the sound velocity @xmath52 and the healing length @xmath40 . the imaginary parts of @xmath50 and @xmath51 have finite values for any relative velocity @xmath53 . therefore , the initial state is unstable even for @xmath55 . this is in contrast to fig . 1 , which shows the antiferromagnetic case.,title="fig:",width=264 ] and @xmath49 . the upper and lower graphs are respectively the imaginary parts of @xmath50 and @xmath51 . the vertical and horizontal axes are respectively the relative velocity and wave number of the @xmath42 direction , which is normalized by the sound velocity @xmath52 and the healing length @xmath40 . the imaginary parts of @xmath50 and @xmath51 have finite values for any relative velocity @xmath53 . therefore , the initial state is unstable even for @xmath55 . this is in contrast to fig . 1 , which shows the antiferromagnetic case.,title="fig:",width=264 ] we can write the wave functions as @xmath56 our system is homogeneous so that we can express the small deviation @xmath48 by plane waves as @xmath57 @xmath58 where @xmath59 is @xmath60 . substituting eqs . ( 11)(13 ) into eq . ( 1 ) and neglecting the quadratic terms of the small deviation , we obtain the following equations : @xmath61 @xmath62 where @xmath63 @xmath64 with @xmath65 and @xmath66 . it follows that the small deviations of the @xmath67 components couple with each other , but that the small deviation of the @xmath46 component develops independently of the small deviation of the @xmath67 components . this is because the initial state of eq . ( 10 ) does not have any @xmath46 component . since eqs . ( 14 ) and ( 15 ) are eigenvalue problems , we can obtain the dispersion relations : @xmath68 @xmath69 where @xmath51 and @xmath50 are the eigenfrequencies of eq . ( 14 ) and ( 15 ) , respectively . the dispersion relations of eqs . ( 18 ) and ( 19 ) have dynamically unstable regions where the imaginary parts of @xmath50 and @xmath51 become finite . figures 1 and 2 show the imaginary parts of the dispersion relations for the antiferromagnetic ( @xmath70 ) and ferromagnetic cases ( @xmath71 ) , respectively . here , the wave number and the velocity are normalized by the coherence length @xmath40 and the sound velocity @xmath52 . experiments typically use @xmath72na and @xmath73rb atoms . the interaction parameters for @xmath72na atoms satisfy @xmath74 and @xmath75 , and those for @xmath73rb atoms satisfy @xmath76 and @xmath75 . if we had used @xmath76 and @xmath75 , it would take much longer for the instability to occur . in this study , to extract the dynamics characteristic of the ferromagnetic interaction , we use @xmath77 and @xmath75 . in the following , we explain the character of the dispersion relations and show the kinds of dynamics that they are expected to give , which can be confirmed by numerical calculations based on the gp equation ( see sec . iv ) . in the antiferromagnetic case , the imaginary part of @xmath51 has finite values for all relative velocities @xmath53 except @xmath55 , whereas the imaginary part of @xmath50 has finite values only for relative velocities larger than some critical value @xmath54 . the critical velocity @xmath78 of @xmath50 in fig . 1 ( i.e. , the lowest relative velocity for which the imaginary part of @xmath50 is finite ) is @xmath79 . for @xmath80 , only @xmath51 has an imaginary part . this means that the density is modulated only for the @xmath46 component . however , we expect that the density will be modulated for the @xmath67 components too . the instability occurs for the @xmath67 components even though the imaginary part of @xmath50 in fig . 1 vanishes because eqs . ( 8) and ( 9 ) show that increasing the particle number of the @xmath46 component reduces the particle number of the @xmath81 components . consequently , the instability of the @xmath81 components can occur . when the relative velocity @xmath53 exceeds @xmath54 , the amplitude of the imaginary part of @xmath50 is larger than that of @xmath51 . thus , the instability of the @xmath81 components occurs faster than that of the @xmath46 component . in the ferromagnetic case , the dispersion relations for @xmath50 and @xmath51 have finite imaginary parts at any arbitrary relative velocity @xmath53 . this is in contrast to the antiferromagnetic case and it implies that the initial state of eq . ( 10 ) is unstable even without a counterflow . in the case @xmath82 , the instability of all components occurs nearly at the same time because the amplitude of the imaginary parts of @xmath50 is almost same as that of @xmath51 . on the other hand , in the case @xmath83 , the instability of the @xmath81 components grows faster than that of the @xmath46 component , which reflects the amplitude of the imaginary part of @xmath50 and @xmath51 . finally , we discuss the isotropy and anisotropy of the dispersion relation about the direction of the relative velocity . equation ( 18 ) shows that @xmath51 depends not on the direction of the relative velocity but on its amplitude . on the other hand , @xmath50 depends on the direction of the relative velocity as well as its amplitude . thus , the dynamics obtained by the gp equation exhibits different behaviors depending on which eigenfrequencies give larger imaginary parts . if @xmath51 ( @xmath50 ) is dominant , the early dynamics will be isotropic ( anisotropic ) . these results obtained from the dispersion relation are consistent with numerical calculations based on the gp equation ( see sec . components for the antiferromagnetic case at @xmath84 ( a ) @xmath85 , ( b ) @xmath86 , and ( c ) @xmath87 . ( d ) and ( e ) profiles of the spin density vector corresponding to ( b ) and ( c ) , respectively . the field of view of each image is @xmath88 . the shading of the arrows in ( d ) and ( e ) denote the magnitude of the spin density vector . these results are for numerical calculations with @xmath89 , @xmath75 , and @xmath90 . , we investigate the dynamics of a spin-1 spinor bec induced by counterflow by performing numerical calculations based on the gp equation . the dynamics is mainly classified according to whether the spin - dependent interaction is antiferromagnetic or ferromagnetic . in this section , we present the detailed dynamics for both cases . the dynamics for the antiferromagnetic interaction induced by counterflow instability are shown . the dynamics strongly depends on whether @xmath80 or @xmath91 . as shown in sec . iii , in the former case , the instabilities for all three components are expected to grow simultaneously , whereas the instability for the @xmath67 components is expected to grow faster than that for the @xmath46 component in the latter case . ( @xmath92 ) and the particle number @xmath93 ( @xmath92 ) for the antiferromagnetic case . @xmath94 ( @xmath92 ) is defined by eqs.(20 ) and ( 21 ) . these results are for numerical calculations with @xmath89 , @xmath75 , and @xmath90.,width=340 ] first , we present the dynamics for the case @xmath80 . figure 3 shows the density profiles of all components obtained by numerical calculations with @xmath90 . in this case , the critical velocity @xmath78 is about @xmath95 . figure 3(a ) shows the initial state in which the @xmath67 components have a velocity @xmath96 relative to the counterflow . figure 3(b ) shows the density profiles at @xmath97 , where instabilities occur in all components . the instability in fig . 3(b ) appears as an isotropic density modulation independent of the direction of the relative velocity @xmath96 : the observed density modulation is circular . these results can be understood by considering the imaginary part of the dispersion relation of eqs . ( 18 ) and ( 19 ) . for @xmath80 , the instability is induced by the imaginary part of the dispersion relation of the @xmath46 component , which is independent of the direction of the relative velocity @xmath53 . in addition , as pointed out in sec . iii , the instability of the @xmath46 component causes the instability of the @xmath81 components through eq . thus , the instability of the @xmath67 components is caused by isotropic density modulation of the @xmath46 component and thus it also exhibits isotropic modulation , as shown in fig . hence , all components exhibit circular density modulation . after a certain time , the density modulation of all components becomes very complicated , as shown in fig . after fig . 3(b ) , the circular modulation expands in every component . this results in a narrow path in the low density region . thus , in the transition from fig . 3(b ) to ( c ) , density modulation with various wave numbers grows with increasing time . as a result , the circular density modulation in fig . 3(b ) disappears . components for the antiferromagnetic case at ( a ) @xmath84 ( a ) @xmath98 and ( b ) @xmath99 . ( c ) and ( d ) profiles of the spin density vector corresponding to ( a ) and ( b ) , respectively . the field of view of each image is @xmath88 . the shading of the arrows in ( c ) and ( d ) denotes the magnitude of the spin density vector . these results for numerical calculations with @xmath89 , @xmath75 , and @xmath100.,width=340 ] ( @xmath92 ) and particle number @xmath93 ( @xmath92 ) for the antiferromagnetic case . @xmath94 ( @xmath92 ) is defined by eqs . ( 20 ) and ( 21 ) . these results for numerical calculations with @xmath89 , @xmath75 , and @xmath100.,width=340 ] the behavior of the spin density vector is shown in figs . 3(d ) and ( e ) , which correspond to the density profile in figs . 3(b ) and ( c ) , respectively . the spin density vector in fig . 3(d ) almost lies in the @xmath101 plane . in the early stages of the instability at @xmath97 , the density modulation of the @xmath102 component almost overlaps that of the @xmath103 component . thus , the @xmath26 component of the spin density vector is very small . as time progresses , the modulation of high wave numbers increases , and the @xmath67 components become less overlapped . therefore , the @xmath26 component grows as shown in fig . 3(e ) . to understand the time dependence of the magnitude of the spin density vector , we numerically calculate the following quantities : @xmath104 @xmath105 where @xmath106 is the area of the system . their time dependence is shown in fig . the result of @xmath107 means that the instability of all components occurs at @xmath108 because growth of @xmath109 and @xmath110 requires the instability of @xmath46 component and that of @xmath111 requires the instability of the @xmath67 components . figure 4(b ) shows that the particle number of the @xmath46 component increases rapidly at @xmath108 , which corresponds to the occurrence of density modulation in fig . hence the instability induced by the counterflow starts to exchange the particle number among the three components and causes the spin density vector to increase . next , we present the dynamics for @xmath91 . figure 5 shows the density profile for @xmath100 . figures 5(a ) shows that the density modulation of the @xmath81 components is much greater than that of the @xmath46 component because @xmath112 $ ] is larger than @xmath113 $ ] in fig . 1 . this density profile differs from that for @xmath80 because the density modulation in fig . 5(a ) is anisotropic . this is understood by the dispersion relation for @xmath50 in eq . ( 19 ) which depends on the direction of the relative velocity @xmath96 . the low density region of the @xmath81 components in fig . 5(a ) is nucleated due to the growth of the density stripe perpendicular to the relative velocity @xmath96 . the interval of the stripe is consistent with the most unstable wave number obtained for @xmath112 $ ] in fig . 1 . through this stripe in the low density region in fig . 5(a ) , the phase of each wave function rapidly changes by about @xmath114 , whose structure is similar to solitons in one - component becs . this soliton - like structure soon collapses and the density modulation of the @xmath81 components become complicated , as shown in fig . 5(b ) . through this collapse , the density modulation of the @xmath115 components becomes isotropic . similar dynamics to that shown in figs . 5(a ) and ( b ) has been reported in two - component becs @xcite . even after a long time , the density modulation of the @xmath46 component does not increase . however , this result depends on the initial noise ( see sec . iv c ) . the spin density vector behaves very differently from the case when @xmath80 , as shown in figs . 5(c ) and ( d ) . in fig . 5(c ) , density modulation occurs only in the @xmath81 components ( fig . 5 ( a ) ) , so that the spin density vector is oriented in only @xmath116 directions . in addition , the vector in fig . 5(c ) exhibits the stripe structure as that in the density profiles in fig . 5(a ) . as time passes , the structure in fig . 5(c ) collapses because the density profiles lose the stripe structure as shown in fig . then , as the density modulation of the @xmath46 component does not increase , the spin density vector can not lean in the @xmath101 plane ( fig . therefore , domains in which the spin density vector points in the @xmath117 or @xmath118 direction are formed through the instability and they move around . the behaviors of the magnitude of the spin density vector and the particle number of each component for @xmath91 differ greatly from those for @xmath119 too . figure 6 shows the time dependence of @xmath120 @xmath121 and the particle number of each component . @xmath122 grows rapidly at @xmath123 , but the particle number of each component remains almost the same as that of the initial state . this means that the instability of the @xmath81 components occurs , but that of the @xmath46 does not , which is consistent with the density profiles in figs . 5(a ) and ( b ) . this dynamics is almost the same as the behavior of the two - component bec @xcite . components in the ferromagnetic case at @xmath124 ( a ) @xmath125 and ( b ) @xmath99 . ( c ) and ( d ) profiles of the spin density vector corresponding to ( a ) and ( b ) , respectively . the field of view of each image is @xmath88 . the shading of the arrows in ( c ) and ( d ) indicates the magnitude of the spin density vector . these results are for numerical calculation with @xmath126 , @xmath75 , and @xmath127,width=340 ] . ( @xmath92 ) and the particle number @xmath93 ( @xmath92 ) for the ferromagnetic case . @xmath120 ( @xmath92 ) is defined by eqs . ( 20 ) and ( 21 ) . these results are for numerical calculation with @xmath126 , @xmath75 , and @xmath127.,width=340 ] this subsection presents the dynamics with the ferromagnetic interaction induced by the counterflow instability . in this case , the imaginary parts of @xmath50 and @xmath51 in fig . 2 do not exhibit a critical velocity , unlike the case for the antiferromagnetic interaction . hence , the dynamics greatly depends on whether the relative velocity is @xmath128 or @xmath129 . as shown in sec . iii , in the former case , the instability of all components grows , whereas in the latter case the instability of the @xmath81 components grows faster than that of the @xmath46 component . we present the dynamics for the case @xmath130 . in this case , there is no counterflow , but instability occurs because the initial state is unstable even for @xmath130 . figure 7 shows the density profiles of each component obtained by numerical calculations with @xmath130 . as expected , the density profiles do not exhibit anisotropy ( fig . 7(a ) ) since the initial state with @xmath130 is isotropic . isotropic density modulation occurs ; it corresponds to the most unstable wave number in fig . 2 . the instability grows considerably , as shown in fig . 7(b ) . the spin density vector corresponding to the density profiles in figs . 7(a ) and ( b ) are shown in figs . 7(c ) and ( d ) , respectively . when the instability occurs at @xmath131 , the density modulation of the @xmath102 component overlaps with that of the @xmath103 component . thus , the spin density vector almost lies in the @xmath101 plane , as shown in fig . 7(c ) . with increasing time , the overlap decreases , so that the vector points in various directions , as shown in fig . this behavior of the spin density vector is similar to that for the antiferromagnetic case with @xmath132 ( see figs . 3(d ) and ( e ) ) . note that the magnitude of the vector in the ferromagnetic case is larger than that in the antiferromagnetic case ( indicated by the shading of the arrows in the vector plots ) ; this is discussed later . figure 8 shows the time dependences of @xmath120 @xmath121 and the particle number . these results show that the instabilities for all components occur almost simultaneously . the density profile for each component in fig . 7 is consistent with these results . we present the dynamics for the case @xmath129 . figures 9 and 10 show numerical results for @xmath133 . figure 9(a ) shows the instability of the @xmath67 components ; density modulation of the @xmath46 component does not occur because @xmath112 > { \rm{im}}[\omega _ { 0}]$ ] . the low density regions of the @xmath81 components in fig . 9(a ) are the soliton - like structure , which collapse ( fig . as time progresses , the instability of the @xmath46 component develops , as shown in fig . 9(c ) . the spin density vector for this case is shown in figs . 9(d ) and ( e ) , which correspond to figs . 9(a ) and ( c ) , respectively . in the early stages of the instability , density modulation occurs only in the @xmath81 components , so that the spin density vector points in the @xmath116 directions , which is similar to fig . as time increases , the @xmath46 component grows . thus , the @xmath42 and @xmath43 components of the spin density vector become large so that the spin density vector points in various directions , as shown in fig . these behaviors of the vector are consistent with the time dependences of the @xmath134 and the particle number of each component shown in fig these results reveal that there are obvious differences in the behaviors of the magnitude of the spin density vector for the antiferromagnetic and ferromagnetic cases . figures 4 , 6 , 8 , and 10 show that in the antiferromagnetic case @xmath135 tends to decrease , whereas it tends to increase in the ferromagnetic case . this gives rise to the different behaviors of the pdf of the magnitude of the spin density vector , which is very important for the turbulence of a spin-1 spinor bec . the details are described in sec . v. components for the ferromagnetic case at @xmath84 ( a ) @xmath136 , ( b ) @xmath137 , and ( c ) @xmath138 . ( d ) and ( e ) profiles of the spin density vector corresponding to ( a ) and ( c ) , respectively . the field of view of each image is @xmath88 . the shaping of the arrows in ( d ) and ( e ) indicates the magnitude of the spin density vector . these results are for numerical calculations with @xmath126 , @xmath75 , and @xmath139.,width=340 ] ( @xmath92 ) and the particle number @xmath93 ( @xmath92 ) for the ferromagnetic case . @xmath94 ( @xmath92 ) is defined by eqs . ( 20 ) and ( 21 ) . these results are for numerical calculations with @xmath126 , @xmath75 , and @xmath139.,width=340 ] we discuss the dependence of the dynamics on the initial noise . in our numerical calculations , small white noise is added to the initial state . different samples of the white noise are available , although the magnitude is fixed . our numerical results can be classified into four categories : ( i ) antiferromagnetic case with a small relative velocity ( fig . 3 ) ; ( ii ) antiferromagnetic case with a large relative velocity ( fig . 5 ) ; ( iii ) ferromagnetic case with a small relative velocity ( fig . 7 ) ; ( iv ) ferromagnetic case with a large relative velocity ( fig . 9 ) . we performed some numerical calculations and we empirically observed a strong dependence on the noise sample only for ( ii ) . this strong dependence is related to the growth of the @xmath46 component . in this case , the instability of the @xmath81 components occurs first ( figs . 5 and 6 ) , which is independent of the noise . however , numerical calculations reveal that as time increases , the @xmath46 component may grow depending on the noise sample used . we can not control whether growth occurs or not . in this paper , we consider the results that do not depend on the initial noise ; thus , figs . 5 and 6 show only the results obtained before the @xmath46 component grows . for cases ( i ) , ( iii ) , and ( iv ) , the dynamics is qualitatively unchanged even when the initial noise sample is varied . we find that , in the ferromagnetic case , the spectrum of the spin - dependent interaction energy in the turbulent state obeys the @xmath1 power law , whereas that in the antiferromagnetic case does not . this result was obtained by numerical calculations and it can be understood in terms of scaling analysis . this section mainly considers the spectrum of the spin - dependent interaction energy and its time dependence . the methodology usually used for analyzing classical turbulence is applied to a spin-1 spinor becs in this study . here , we briefly review turbulence in classical fluids @xcite . many studies focus on statistical quantities and laws because they reflect properties characteristic of the complicated motion in turbulence . one quantity often investigated is the kinetic energy spectrum . kolmogorov proposed that the kinetic energy spectrum of incompressible fluid obeys a @xmath2 power law in fully developed homogeneous isotropic turbulence ; this can be demonstrated by making several assumptions . one assumption is that the kinetic energy flux is independent of the wave number . this means that the kinetic energy is transported to a high wave number with a constant energy flux in the wave number region that obeys the @xmath2 power law . this result has been confirmed by many numerical calculations and experiments . the inertial term in the navier stokes equation was found to play a dominant role in energy transfer . these results show the self - similarity of the velocity field in wave number space . on the other hand , in real space , a richardson cascade is believed to occur in which large vortices become smaller through reconnections of vortices . however , this has not been confirmed yet . we focus on the spectrum of the spin - dependent interaction energy in the turbulence of a spin-1 spinor bec . in analogy with classical turbulence , the flux of the spin - dependent interaction energy is expected to be independent of the wave number , which gives rise to the spectrum characteristics of this system . we derive an expression for the spectrum of the spin - dependent interaction energy . the spin - dependent interaction energy @xmath140 per unit area is given by @xmath141 we expand the spin density vector @xmath142 with plane waves : @xmath143 the spin - dependent interaction energy @xmath140 is represented by @xmath144 as @xmath145 therefore , the energy spectrum of spin - dependent interaction energy is given by @xmath146 where @xmath147 is @xmath148 for a system size @xmath149 . our numerical results reveal that the spectrum of the spin - dependent interaction energy in the ferromagnetic case obeys the power law shown in fig . this spectrum is numerically calculated at @xmath150 , which is a sufficiently long time after the occurrence of the instability . this is found by the time dependence of @xmath120 ( @xmath92 ) shown in fig . the spectrum in fig . 11 ( a ) has two regions that are separated by the wave number @xmath151 , which corresponds to the spin coherence length @xmath152 , ( i.e. , the characteristic scale of spin structures such as domain walls and polar core vortex @xcite ) . in the low wave number region ( @xmath153 ) , the spectrum obeys the power law , whereas it does not in the high wave number region ( @xmath154 ) . the origin of this power law in the low wave number region is discussed below . ( upper ) and time dependences of the quantities @xmath120 ( @xmath92 ) ( lower ) for the ferromagnetic case at @xmath150 . in the graph of the spectrum , the red squares and blue solid and black dashed lines show the numerical results for the spectrum , a line proportional to @xmath155 , and the boundary of the wave number corresponding to the spin coherent length , respectively . this spectrum is obtained by numerical calculations with @xmath77 , @xmath75 , @xmath156 . , title="fig:",width=340 ] ( upper ) and time dependences of the quantities @xmath120 ( @xmath92 ) ( lower ) for the ferromagnetic case at @xmath150 . in the graph of the spectrum , the red squares and blue solid and black dashed lines show the numerical results for the spectrum , a line proportional to @xmath155 , and the boundary of the wave number corresponding to the spin coherent length , respectively . this spectrum is obtained by numerical calculations with @xmath77 , @xmath75 , @xmath156 . , title="fig:",width=340 ] . this was obtained by performing numerical calculations with the following parameters : @xmath77 , @xmath75 , @xmath156.,width=340 ] we will obtain the power exponent of the spectrum of the spin - dependent interaction energy in the low wave number region ( @xmath153 ) using the gp equation . the calculation used to obtain fig . 11 ( a ) was for the ferromagnetic case , where the magnitude of the spin density vector is expected to have some large value in their pdf . to confirm this , we calculated the pdf of the magnitude of the spin density vector . the pdf in fig . 12 has a sharp peak , indicating that the magnitude of the spin density vector tends to be @xmath36 in the turbulent state . this observation allows us to write the macroscopic wave function in the turbulent state with the ferromagnetic interaction as @xmath157 where @xmath158 , @xmath159 , and @xmath160 are the euler angles in spin space and @xmath161 is the total density of the initial state . this wave function leads to @xmath162 in addition , the total density is assumed to be time - independent because @xmath163 . in this condition , the spin - independent interaction energy is larger than the kinetic and spin - dependent interaction enegies , so that the total density @xmath12 has a weak time dependence . we can obtain the time evolution equation of @xmath164 from eqs . ( 1 ) and ( 26 ) : @xmath165,\ ] ] @xmath166 with @xmath167 @xcite . in our case , the total density is approximately @xmath36 in the initial state of eq . ( 10 ) , so that @xmath168 vanishes . therefore , the following equation can be used to calculate the power exponent of the spectrum : @xmath169 where space and time are respectively normalized by @xmath170 and @xmath171 ( @xmath172 , @xmath173 ) . in the early stages of the instability , the wave function has vortices and a soliton - like structure . the velocity @xmath174 can be large in their neighborhoods . this is similar to the vicinity of vortices in one - component becs , where the velocity is larger than the sound velocity in the vortex core region whose size is equal to the coherence length . however , the number of such defects tends to decrease with time . therefore , the velocity @xmath174 is lower than the sound velocity @xmath175 almost everywhere after a sufficiently long time after the instability occurs . we confirm that the pdf of the magnitude of the velocity has a peak at about 1/10th of the sound velocity . we thus expect that the term on the right - hand side of eq . ( 30 ) is important for transporting the spin - dependent interaction energy to a higher wave number . this is different from classical turbulence for which the inertial term in the navier stokes equation is dominant for energy transfer . for the ferromagnetic case at @xmath84 ( a ) @xmath137 , ( b ) @xmath176 , ( c ) @xmath177 , and ( d ) @xmath178 . in the spectrum , the red squares , blue solids , and black dashed lines indicate the numerical results for the spectrum , the line proportional to @xmath155 , and the boundary of the wave number corresponding to the spin coherent length , respectively . this was obtained by performing numerical calculations with @xmath77 , @xmath75 , @xmath156.,width=340 ] to understand the behavior of the spectrum in fig . 11 , we consider spin density vector in wave number space . we can express eq . ( 30 ) by neglecting the second term on the left - hand side by using @xmath179 : @xmath180 in the following , we apply kolmogorov - type dimensional analysis to eq . ( 31 ) @xcite under two assumptions : equation ( 31 ) is invariant under the scale transformation and the flux of the spin - dependent interaction energy is independent of the wave number . we perform the scale transformation @xmath181 , @xmath182 in eq . ( 31 ) . then , if @xmath183 is transformed to @xmath184 , eq . ( 31 ) will be invariant . we can then write the dependence @xmath183 on @xmath185 and @xmath186 as @xmath187 we assume that the flux of the spin - dependent interaction energy is independent of the wave number , which is equivalent to assuming the existence of a region in which the energy is constantly transported . this result can be expressed by @xmath188 where @xmath189 and @xmath190 are respectively the energy flux and the characteristic time . using eqs . ( 25 ) , ( 32 ) , and ( 33 ) , we obtain the @xmath1 power exponent : @xmath191 this result agrees with the numerical result in fig . 11 , where the blue line denotes @xmath155 . the same assumption has been used for classical turbulence ; the wave number region that obeys the @xmath2 power law is known as the inertial range since it originates from the inertial term in the navier stokes equation @xcite . however , in our system , the inertial term is not important for energy transfer , so that the region obeying the power law in fig . 11 can not be termed the inertial range . the power law in the spectrum of the spin - dependent interaction energy originates from the fact that the energy flux induced by the first term of the right - hand side of eq . ( 28 ) is independent of the wave number . this nonlinear term contains the second derivative and it differs from the first - derivative inertial term in the navier stokes equation . the different nonlinear terms in the two equations are responsible for the different exponents of the energy spectrum . therefore , the @xmath1 power law of the spectrum is peculiar to turbulence of a spin-1 spinor bec . figure 13 shows how the @xmath1 power spectrum develops . in the early stages of the instability , the spectrum has a peak corresponding to the most unstable wave number of the dispersion relation given by eqs . ( 18 ) and ( 19 ) , as shown in fig . this is confirmed by fig . 2 , from which it follows that the most unstable wave numbers @xmath192 is approximately equal to @xmath193 . actually , the density modulation at @xmath194 has a stripe structure that resembles those in figs . 5(a ) and 9(a ) , and the unstable wave number corresponds to the wave number of the stripe . after the instability occurs , the spectrum changes to that in fig . the density then no longer sustains the stripe structure and excites modulation of various wave numbers . as time increases , the spectrum starts to obey the @xmath1 power law , as shown in fig . after that , the spectrum continues to obey this power law . this is confirmed by fig . 13(d ) showing the spectrum at @xmath150 . we calculate the time dependence of the power exponent by the least - squares method @xcite . the deviation @xmath195 from the straight line obtained by the method is also calculated . the results with @xmath90 , @xmath196 , and @xmath197 are shown in fig . 14 , which exhibits that the power exponent is approximately @xmath1 over a long time . note that since our system is not stationary , the power exponent asymptotically approaches @xmath1 in fig . 14 . also , fig . 14(a ) shows how the time development of the power exponent depends on the relative velocity @xmath53 . when the relative velocity is small , it takes long time for the value of @xmath12 to approach @xmath1 . this is confirmed by comparing the cases @xmath90 and @xmath196 . on the other hand , the spectrum behaves differently when @xmath139 . in this case , the value of @xmath12 approaches @xmath1 from above ( fig . this can be understood from the imaginary part of the dispersion relation in fig . 2 , which shows that the unstable region of @xmath50 is broad for a large relative velocity . this means that , unlike the case for a small velocity , the instability contains various wave numbers . thus , in this case , the spectrum in the low wave number region @xmath198 becomes flat in the early stages of the instability , leading to the time dependence of @xmath12 . in summary , we observed a @xmath1 power law in spin-1 spinor becs with the ferromagnetic interaction . this law is independent of the relative velocity @xmath53 in our numerical calculations based on the gp equation for @xmath199 . however , if the relative velocity is much greater than the sound speed @xmath175 , this law will not hold because the total particle density @xmath12 may be inhomogeneous . in ( a ) is calculated by the least - squares method . the deviation @xmath195 in ( b ) is defined as @xmath200 , where @xmath201 , @xmath202 , and @xmath203 are the line obtained by the least - squares method , the number , and the data set . in this calculation , the wave number region ( @xmath204 ) is limited whit @xmath205 . this was obtained by performing numerical calculations with @xmath77 , @xmath75 , @xmath206.,width=340 ] unlike the ferromagnetic case shown in fig . 11 , the spectrum of the spin - dependent interaction energy in the antiferromagnetic case does not obey the @xmath1 power law . this is confirmed by fig . 15 which shows the spectrum of the spin - dependent interaction energy for the antiferromagnetic case at @xmath207 when the spin density vector is greatly disturbed . however , this spectrum may show a power law in the narrow range @xmath208 . as shown in the following , we can not estimate the power exponent by the simple scaling analysis applied to the ferromagnetic case . the @xmath1 power law for the ferromagnetic case was obtained by applying a scaling argument under certain assumptions , which are not applicable for the antiferromagnetic case . this can be confirmed by checking the pdf of the magnitude of the spin density vector ( fig . there are two distinct differences between figs . 12 and 16 : the peak width and the magnitude of the spin density vector corresponding to the peak both differ . the peak width in fig . 16 for the antiferromagnetic case is larger than that in fig . 12 for the ferromagnetic case . the magnitude of the spin density vector corresponding to the peak for the ferromagnetic case is approximately @xmath36 , whereas that for the antiferromagnetic case is smaller than @xmath36 . these results imply that eq . ( 26 ) is not valid for the antiferromagnetic case . thus , eq . ( 34 ) , which is based on eq . ( 26 ) , can not be applied to the antiferromagnetic case . therefore , the spectrum for the antiferromagnetic case can not be analyzed by the simple scaling analysis applied to the ferromagnetic case . for the antiferromagnetic case at @xmath207 . in the spectrum , the red squares and blue solid and black dashed lines indicate the numerical results for the spectrum , the line proportional to @xmath155 , and the boundary of the wave number corresponding to the spin coherent length , respectively . this was obtained by performing numerical calculations with the following parameters : @xmath89 , @xmath75 , @xmath156.,width=340 ] . this was obtained by performing numerical calculations with the following parameters : @xmath89 , @xmath75 , @xmath156.,width=340 ] in this section , we discuss the following three topics : ( i ) comparison between turbulence in spin-1 spinor becs and other kinds of turbulence ; ( ii ) possibility of observing the @xmath1 power law , and ( iii ) how to generate counterflow in spin-1 spinor becs . we compare the turbulence in spin-1 spinor becs with that in one - component becs and classical incompressible fluids . through this comparison , we discuss some characteristic properties of turbulence in spin-1 spinor becs . there are some differences between turbulence in one - component becs and that in spin-1 spinor becs . in the former system , the kolmogorov @xmath2 power law spectrum is obtained only if an external force is applied . this force injects energy on large scales and generates quantized vortices . if the external force is weak and the number of vortices is not enough , the incompressible kinetic energy decreases and the kolmogorov spectrum tends to disappear @xcite . thus , the existence of the quantized vortices is considered to be important for sustaining the kolmogorov @xmath2 power law in the turbulence of one - component becs . on the other hand , the @xmath1 power law in spin-1 spinor becs is sustained without the application of an external force , as confirmed by fig . 14 . in the ferromagnetic case ( see figs . 8 ( a ) and 10 ( a ) ) , the absolute value of the spin - dependent interaction energy of eq . ( 22 ) tends to increase without the application of an external force due to the ferromagnetic interaction . this is in contrast with the reduction in the incompressible kinetic energy in the turbulence of one - component becs when no external force is applied . we conjecture that this increase allows the system to obey the @xmath1 power law even when no external force is applied . the self - similarity in wave number space can be strongly related to the structures in real space , which is very important for understanding the behavior of turbulence in real space . in the turbulence of the one - component bec , the kolmogorov @xmath2 power law is believed to be related to the richardson cascade of quantized vortices , where larger vortices become smaller through reconnections of vortices . this cascade may correspond to the energy cascade in wave number space , where the incompressible kinetic energy is transported to a high wave number . we expect that there are some spin structures characteristic of turbulence with the @xmath1 power law in a spin-1 spinor bec . like the richardson cascade of quantized vortices , some larger spin structures may become smaller to transport the energy to a high wave number . however , we currently do not know what kinds of spin structures are essential for the @xmath1 power law . we compare turbulence in spin-1 spinor becs with that in classical fluids . in classical turbulence where the reynolds number becomes infinite , the inertial term of the navier stokes equation becomes dominant in the inertial range , which causes the system to obey the @xmath2 power law . we do not know any quantities that correspond to the reynolds number in spin-1 spinor becs . however , as our system becomes turbulent , the term on the right - hand side of eq . ( 30 ) becomes dominant . hence , the spin - dependent interaction energy in the turbulence is transported to a high wave number mainly by this term ; this differs from the turbulence mechanism in incompressible classical fluids . this is very significant for the @xmath1 power law , as pointed out in sec . v. from the above discussion , it follows that the turbulence in spin-1 spinor becs has some properties that other kinds of turbulence do not have . there are several unknowns associated with this turbulence such as the kinetic energy spectrum , the spin structure in real space , the interaction between the velocity and the spin field . these will be studied in the future . the possibility of observing the @xmath1 power law is discussed . we consider that the @xmath1 power law may be experimentally observed if the following three conditions are satisfied . the spectrum of trapped systems with the ferromagnetic interaction is expected to manifest the @xmath1 power law less clearly than that the spectrum of homogeneous systems . this is because we neglect the term containing @xmath209 in eq . ( 28 ) in our derivation of the @xmath1 power law . variation of the total particle density @xmath12 should affect the spectrum in trapped systems . however , for large trapped systems , this is expected to have a small effect on the spectrum . therefore , large trapped systems are preferable for observing the @xmath1 power law . this is the first condition . all components of the spin density vector have been experimentally observed by a phase contrast imaging method @xcite . the expressions for the spectrum of the spin - dependent interaction energy show that it is possible to obtain a spectrum if the spin density vector is observed everywhere . this is because eq . ( 25 ) contains only the fourier component of the vector @xmath144 . thus , the second condition is the observation of the spin density vector . the experimental resolution in wave number space is important for observing the @xmath1 power law . our numerical calculations reveals that the spectrum of the spin - dependent interaction energy obeys the @xmath1 power law in the low wave number region ( @xmath210 ) , as shown in fig . 11 . in trapped systems , we expect that the @xmath1 power law holds in the region @xmath211 , where @xmath212 is the wave number corresponding to the system size . typically , the system size is approximately equal to the thomas fermi radius . thus , experiments for observing the @xmath1 power law need to have a resolution up to the spin coherence length @xmath213 in real space . since sadler @xmath214 @xmath215 . @xcite observed a polar core vortex , the resolution of their experiments seems to be sufficient for observing the @xmath1 power law . the third condition is the observation with a resolution up to the spin coherence length @xmath213 . note that the larger a system size is , the broader the region that obeys the @xmath1 power law will be , because @xmath212 is inversely proportional to the system size . thus , larger systems are more suitable for observing the @xmath1 power law ( this is also consistent with the first condition ) . in summary , we conclude that the @xmath1 power law may be experimentally observed if the above three conditions are satisfied . we discuss the two methods for generating counterflow of spin-1 spinor becs in trapped systems . the first method is to use a double well potential . a numerical study has already investigated this @xcite . in this case , the @xmath81 components are separately trapped in each well . the central barrier between two wells is then removed , which generates counterflow in spin-1 spinor becs . the second method is to utilize a magnetic field gradient . this method was used to generate counterflow of two - component becs @xcite . in the initial state , the @xmath81 components are trapped in a harmonic trap . as a magnetic field gradient is applied , one component moves in the the field gradient direction , while the other component moves in the opposite direction . the magnetic field gradient is switched off when the two components are sufficiently separated , which causes the @xmath67 components to move toward the center of the trap , generating a counterflow . in future , we may use these methods to study counterflow in spin-1 spinor becs in a trapped system . this paper addressed two main topics : the dynamics induced by counterflow of the spin-1 spinor becs in a homogeneous two - dimensional system and the turbulence generated by the counterflow . these themes are investigated using the gp and bdg equations . the results reveal that the properties of the dynamics and turbulence in this system are strongly dependent on whether the spin - dependent interaction is ferromagnetic or antiferromagnetic . we summarize the results below . the dynamics induced by the counterflow in the spin-1 spinor bec was investigated by performing analytical calculations using the bdg equation and numerical calculations using the gp equation . we obtain the dispersion relations of eqs . ( 18 ) and ( 19 ) from the bdg equation ; these relations show the dynamical instability . the dispersion relations depend on the spin - dependent interaction , so that @xmath216 $ ] and @xmath217 $ ] for the antiferromagnetic case differ from those for the ferromagnetic case ( figs . 1 and 2 ) . the numerical calculations reveal that , in the early stages of the instability , the dynamics can be understood in terms of the dispersion relations . the stripe width of the density modulation in fig . 5(a ) and 9(a ) is approximately equal to the most unstable wavelength obtained by the dispersion relations . in addition , the isotropy and anisotropy of the density modulation in figs . 3(b ) and 5(a ) can be explained in terms of the dispersion relations . the distinct difference between the ferromagnetic and antiferromagnetic cases appears in the magnitude of the spin density vector , as shown in figs . 3 @xmath218 10 . this is very important for the spectrum of the spin - dependent interaction energy in the turbulence , as pointed out in sec . v. these results reveal dynamics peculiar to the spin degrees of freedom . we studied the turbulence generated by the counterflow in spin-1 spinor becs by performing numerical calculations using the gp equation and scaling analysis . in the ferromagnetic case , the spectrum of the spin - dependent interaction energy obeys the @xmath1 power law ( fig . the power exponent @xmath1 is obtained by the scaling analysis of eq . ( 31 ) when we make the following three assumptions : the wave functions @xmath219 are approximately expressed by eq . ( 26 ) , eq . ( 31 ) are invariant under the scale transformation , and the flux of the spin - dependent interaction energy is independent of the wave number . on the other hand , for the antiferromagnetic case , the spectrum does not exhibit the @xmath1 power law . this is probably attributable to the breakdown of the validity of eq . ( 26 ) . in this case , the pdf of the magnitude of the spin density vector ( fig . 16 ) differs from that for the ferromagnetic case ( fig . 12 ) , so that such scaling analysis used for the ferromagnetic case is not applicable . however , in the narrow range of the wave number , the spectrum in fig . 15 may show a power law , which can not be estimated by the scaling analysis . these results are characteristic of the spin degrees of freedom and are thus not exhibited by turbulence in one - component becs and classical fluids . there are some unresolved problems associated with turbulence in spin-1 spinor becs has ( see sec . vi ) ; we intend to study these in the near future . m. t. acknowledges the support of a grant - in - aid for science research from jsps ( grant no . 21340104 ) . u. frisch , _ turbulence _ ( cambridge university press , cambridge , 1995 ) . a. n. kolmogorov , dokl . nauk sssr * 30 * , 9 ( 1941 ) ; * 32 * , 19 ( 1941 ) ; proc . a * 434 * , 9 ( 1991 ) ; * 434 * , 15 ( 1991 ) . s. chandrasekhar , _ hydrodynamics and hydromagnetic stability _ ( clarendon press , oxford , 1961 ) . h. takeuchi , n. suzuki , k. kasamatsu , h. saito , and m. tsubota , phys . b * 81 * , 094517 ( 2010 ) . k. sasaki , n. suzuki , d. akamatsu , and h. saito , phys . a * 80 * , 063611 ( 2009 ) . h. takeuchi , s. ishino , and m. tsubota , phys . lett . * 105 * , 205301 ( 2010 ) . s. ishino , m. tsubota , and h. takeuchi , phys . rev . a , * 83 * , 063602 ( 2011 ) . m. guilleumas , b. julia - diaz , j. mur - petit , and a. polls , europhys . lett . * 84 * , 60005 ( 2008 ) . a. lamacraft , phys . a * 77 * , 063622 ( 2008 ) . k. kudo and y. kawaguchi , phys . a * 82 * , 053614 ( 2010 ) . k. kudo and y. kawaguchi , phys . rev . a * 84 * , 043607 ( 2011 ) . w. p. halperin and m. tsubota , eds . , _ progress in low temperature physics _ ( elsevier , amsterdam , 2009 ) , vol . 16 . e. a. l. henn , j. a. seman , g. roati , k. m. f. magalhaes , and v. s. bagnato , phys . lett . * 103 * , 045301 ( 2009 ) . j. a. seman @xmath214 @xmath215 . , laser phys . * 8 * , 691 ( 2011 ) . m. kobayashi and m. tsubota , phys . a * 76 * , 045603 ( 2007 ) . horng , c .- h . hsueh , s .- w . su , y .- m . kao , and s .- c . gou , phys . a * 80 * , 023618 ( 2009 ) . c. hamner , j. j. chang , p. engels , and m. a. hoefer , phys . lett . * 106 * , 065302 ( 2011 ) . m. a. hoefer , j. j. chang , c. hamner , and p. engels , phys . a * 84 * , 041605 ( 2011 ) . t. ohmi and k. machida , j. phys . . jpn . * 67 * , 1822 ( 1998 ) . t. -l . ho , phys . rev . lett . * 81 * , 742 ( 1998 ) . l. e. sadler , j. m. higbie , s. r. leslie , m. vengalattore , and d. m. stamper - kurn , nature * 443 * , 312 ( 2006 ) . m. ottaviani and j. a. krommes , phys . lett . * 69 * , 2923 ( 1992 ) . t. watanabe , h. fujisaka , and t. iwayama , phys . e , * 55 * , 5575 ( 1997 ) . the value of @xmath12 in fig . 14 was calculated by the least - squares method in the wave number region @xmath220 , where @xmath221 is @xmath222 for a system size @xmath223 . we employ this method irrespective of whether the spectrum exhibits a power law or not . thus , the value of @xmath12 in the early stages of the instability does not necessarily reflect the actual power exponent because the deviation @xmath195 is large . the values of @xmath12 are meaningful as a power exponent when the spectrum exhibits a power law ( e.g. , when @xmath224 due to the small deviation @xmath195 ) . let us mention the finite size effect which can affect the power exponent . we observe that , in the least square method , the deviation of @xmath12 from @xmath1 becomes large when @xmath221 is smaller than @xmath222 . this may be caused by the finite size effect . thus , in fig . 14 , we use @xmath225 to reduce this effect . m. kobayashi and m. tsubota , phys . lett . * 94 * , 065302 ( 2005 ) . m. kobayashi and m. tsubota , j. phys . . jpn . * 74 * , 3248 ( 2005 ) . h. saito , y. kawaguchi , and m. ueda , phys . a * 75 * , 013621 ( 2007 ) .

a new computational technique based on the symbolic description utilizing kneading invariants is proposed and verified for explorations of dynamical and parametric chaos in a few exemplary systems with the lorenz attractor . the technique allows for uncovering the stunning complexity and universality of bi - parametric structures and detect their organizing centers codimension - two t - points and separating saddles in the kneading - based scans of the iconic lorenz equation from hydrodynamics , a normal model from mathematics , and a laser model from nonlinear optics .

1204.3278

a great deal of analytical and experimental studies , including modeling simulations , have been focused on the identification of key signatures to serve as structural invariants that would allow dynamically alike nonlinear systems with chaotic dynamics from diverse origins to be united into a single class . among these key structures are various homoclinic and heteroclinic bifurcations of low codimensions that lie at the heart of the understanding of complex behaviors because of their roles of organizing centers of dynamics in parameterized dynamical systems . dynamical systems theory has aimed to create purely abstract approaches that are further proceeded by creation and the development of applicable tools designed for the search and identification of such basic invariants for simple , morse - smale , systems and ones with complex chaotic dynamics . one such a [ computationally justified ] approach for studying complex dynamics capitalizes on the concept of sensitivity of deterministic chaos . sensitivity of chaotic trajectories can be quantified in terms of the divergence rate evaluated through the largest lyapunov characteristic exponent . the approach has been proven to work exceptionally well for various systems with chaotic and simple dynamics . in several low - order dissipative systems , like the rssler model , the computational technique based on the largest lyapunov characteristic exponent reveals that they possess common , easily recognizable patterns involving spiral structures in bi - parametric planes @xcite . such patterns turn out to be ubiquitously alike in both time - discrete @xcite and time - continuous systems @xcite , and they are easily located when the spiral patterns have regular and chaotic spiral `` arms '' in the systems with the shilnikov saddle - focus @xcite . application of the lyapunov exponents technique fails , in general , to reveal fine structures embedded in the bi - parametric scans of lorenz - like systems . as such it can not deliver the desired insights into intrinsic bifurcations because regions of chaotic dynamics appear to be uniformly this basically means that the instability of the lorenz attractors does not vary noticeably as control parameters of the system are varied . this holds true too when one attempts to find the presence of characteristic spiral structures that are known to exist theoretically in the lorenz - like systems @xcite and therefore could only be identified using accurate bifurcation continuation approaches @xcite . such spirals in a bi - parametric parameter plane of the system in question are organized around the so - called t[erminal]-points corresponding to codimension - two or -higher heteroclinic connections between two or more saddle equilibria . for @xmath0-symmetric systems with the lorenz attractor , the degeneracy is reduced to a cod-2 bifurcation of a closed heteroclinic connection involving two saddle - foci and a saddle at the origin . t - points have been located in various models of diverse origins including electronic oscillators @xcite and nonlinear optics @xcite , etc . ) , the number of turns of the one - dimensional separatrix of the saddle , @xmath1 , around the saddle - focus increases by one in the homoclinic loop and becomes infinite at the t - point . ] figure [ figtpoint ] sketches an idea of the structure of the bifurcation unfolding near an ordinary t - point in the parameter plane in a symmetric system @xcite ; various t - points configurations for other heteroclinic connections were examined in detail in @xcite . here , the heteroclinic connection is formed by a pair of symmetric saddle - foci and a saddle whose one - dimensional stable ( incoming ) and unstable ( outgoing ) , respectively , separatrices merge only at the codimension - two t - point in the bi - parametric space of the lorenz model . it follows from the theoretical analysis that the unfolding of this bifurcation includes three main curves ending at the t - point : a spiraling bifurcation curve , ( 1-hom @xmath1 ) corresponding to two simultaneous ( due to the symmetry ) homoclinic loops of the saddle , a curve , ( 1-hom @xmath2 ) , corresponding to two simultaneous homoclinic loops of the saddle - foci , and another codimension - one bifurcation curve , ( 1-het @xmath2 ) , corresponding to a heteroclinic correction between both saddle - foci . in addition , the unfolding includes infinitely many subsidiary t - points in between 1-hom @xmath2 and 1-het @xmath2 . despite a rather overwhelming number of studies reporting the occurrence of various spiral structures , there is yet unproportionately little known about construction details and generality of underlying bifurcation scenarios giving rise to such patterns . in this paper we introduce a novel computational toolkit capitalizing on the idea of the symbolic representation for the dynamics of lorenz - like systems that employs the kneading invariants @xcite . we will then show how the toolkit can be used for detecting various structures in bi - parametric scans of such systems . it is our intention to enhance further the technique thus allowing for systematic studies of the stunning complexity and universality of spiral structures in models of diverse dynamics and origins . the paper is organized as follows : in section [ sec:2 ] we review the homoclinic bifurcation theory and discuss the routes to chaos and the formation stages of the strange attractor in the lorenz model ; in section [ sec:3 ] we introduce basics of kneading theory ; in section [ sec:4 ] we demonstrate of the computational technique using kneading invariants to reveal hidden structures of biparametric chaos in the iconic lorenz model ; in section [ sec:5 ] we apply the technique to uncover colorfulness of the bifurcation diagrams of the shimizu - morioka model . finally , in section [ sec:6 ] the proposed technique will be tested on a 6d model of the optically pumped , far infrared red three - level laser @xcite to confirm the universality of the patters produced by the deterministic chaos in the lorenz like systems . the strange chaotic attractor in the lorenz equation from hydrodynamics has become a de - facto proof of deterministic chaos . the butterfly - shaped image of the iconic lorenz attractor , shown in fig . [ fig1l ] , has become the trademark of chaos theory and dynamical systems . this theory elaborates on complex trajectory behaviors in nonlinear systems from mathematics , physics , life sciences , finance , etc . universality of the methods along with bifurcation tools has made them spread wide and deep across all other disciplines of the modern science . the lorenz equation @xcite is a system of three differential equations : @xmath3 with three positive bifurcation parameters : @xmath4 being the prandtl number quantifying the viscosity of the fluid , @xmath5 being a positive constant of magnitude order 1 which originates from the nonlinearity of the boussinesq equation , and @xmath6 being a reynolds number that characterizes the fluid dynamics . notice that eqs . ( [ lorenz ] ) are @xmath0-symmetric , i.e. @xmath7 @xcite . and @xmath8 : plotted are the coordinates , @xmath9 , of the limit trajectories ( equilibria , periodic and homoclinic orbits ) against the bifurcation parameter @xmath6 . ] an exploration of primary bifurcations in the lorenz equation begins with a single - parameter examination of the dynamics , where @xmath6 serves as the bifurcation parameter increasing from laminar to weekly turbulent magnitudes around 25 , while the two other parameters are set to the original saltzman values : @xmath10 ( for the water , and @xmath11 for the air ) and @xmath8 . this would give a uni - parametric cut through the lorenz equation that was originally explored in these independent studies @xcite ( see fig . [ figure1 ] ) : * for @xmath12 the only equilibrium state @xmath13 at the origin is a global attractor in the 3d phase space of the lorenz equation . * this equilibrium state undergoes a pitch - fork bifurcation at @xmath14 , and for @xmath15 becomes a saddle so that the stability is transferred to two symmetric stable foci . * at @xmath16 the unstable separatrices of the saddle return to the origin thus forming a homoclinic butterfly . this causes a `` homoclinic explosion '' in the phase space of the model that becomes filled in at once with countably many saddle periodic orbits that would further compose the skeleton of the lorenz attractor . * for @xmath17 the model exhibits transient chaos with two ultimate attractors : stable foci @xmath2 . such transient chaos is associated with a pre - turbulence regime . * the value @xmath18 corresponds to the onset of the lorenz attractor coexisting with stable foci @xmath19 and @xmath20 . the attraction basins of @xmath2 are bounded by the 2d cylindrically - shaped stable manifolds of two saddle periodic orbits that have earlier bifurcated from the homoclinic loops of the saddle at @xmath21 . * as @xmath6 increases to @xmath22 the saddle periodic orbits shrink the attraction basins and collapse onto the stable foci @xmath2 through a subcritical andronov - hopf bifurcation . * for @xmath23 the lorenz equation possesses a genuinely strange chaotic attractor , known as the lorenz attractor , containing no stable orbits . the pilot study of the dynamics of the lorenz equation needs to be further enhanced by the bi - parametric examination of the model , including at large parameter values @xcite . we will start off with the canonical bifurcation diagram , shown in the left panel of fig . [ fig2l ] ( courtesy of @xcite ) , of the lorenz equation that depicts the basic bifurcation curves in the @xmath24-parameter plane with fixed @xmath8 . the right panel of fig . [ fig2l ] sketches en route fragments of the formation of the lorenz attractor on the pathway , @xmath10 @xcite through the bifurcation curves . for @xmath12 , ( [ lorenz ] ) has a single stable equilibrium state at the origin . this equilibrium state undergoes a pitch - fork bifurcation at @xmath25 , so that for @xmath15 the origin becomes a saddle , @xmath1 , of the topological type ( 2,1 ) due to the characteristic exponents @xmath26 . the saddle has a 1d unstable manifold , @xmath27 is made of @xmath1 itself and a pair of 1d unstable separatrices , @xmath28 and @xmath29 ( due to @xmath30 ) entering the saddle as @xmath31 , and a 2d stable manifold , @xmath32 , containing the leading ( due to @xmath33 ) invariant @xmath34-axis ; the eigenvector due to @xmath35 determines the non - leading or strongly stable direction , @xmath36 in @xmath32 . after the pitch - fork bifurcation , the separatrices of the saddle tend to two symmetric attractors equilibrium states , @xmath37 ( fig . [ fig2l](a ) ) that become the global attractor for all trajectories in the phase space of the lorenz equation other than in @xmath32 . a homoclinic butterfly bifurcation occurs in the lorenz equation when both separatrices , @xmath28 and @xmath29 , of the saddle come back to the origin along the @xmath34-axis ( fig . [ fig2l](b ) ) . in virtue of the symmetry of the lorenz equation , such homoclinic loops are always formed in pairs , and therefore constitute bifurcations of codimension - one , in general . the bifurcation of the homoclinic butterfly takes place on the curve @xmath38 in the @xmath24-parameter plane . bifurcations of the separatrices of the saddle at the origin are crucial for the lorenz attractor . the very first homoclinic butterfly made of two separatrices looping a single round about the equilibrium states @xmath2 , causes the homoclinic explosion in the phase space of the lorenz equation . this bifurcation gives rise to a onset of countably many saddle periodic orbits that forming an unstable chaotic ( saddle ) set , which is not an attractor yet . for this explosion to happen , the so - called _ saddle value _ @xmath39 , which is the sum of the leading characteristic exponents of the saddle , must be positive ; alternatively , the saddle index @xmath40 . otherwise , if @xmath41 ( @xmath42 ) the homoclinic butterfly produces a symmetric figure-8 periodic orbit in the aftermath of the gluing bifurcation through which two stable periodic orbits merge after flowing into the separatrix loops . l. shilnikov @xcite pointed out two more conditions , in addition to the primary one ( 1 ) : @xmath43 , or @xmath44 , needed for the separatrix loop of the saddle in @xmath45 and higher dimensions to produce a single saddle periodic orbit ; they are : ( 2 ) @xmath46 , i.e. the separatrix comes back to the saddle along the leading direction , ( 3 ) the so - called separatrix value ( @xmath47 in the mapping eq . ( [ map ] ) below ) does not vanish , its sign determines whether the separatrix loop is oriented or twisted , and hence the stable manifolds of the saddle periodic orbit are homeomorphic to a cylinder or a mbius band . otherwise , he predicted @xcite that the lorenz attractor could be born right near such codimension-2 bifurcations , termed resonant saddle , orbit- and inclination - flip , correspondingly @xcite , as it occurs in the shimizu - morioka and similar models @xcite below . out of many saddle periodic orbits , which exploded the phase space of the lorenz equation , two are the special , @xmath48 , ones as they demarcate the thresholds of the interior " of the chaotic unstable set ( fig . [ fig2l](c ) ) . after the homoclinic butterfly bifurcation , in the region between the bifurcation curves @xmath38 and @xmath49 in the ( @xmath50)-parameter plane , the separatrices @xmath28 and @xmath29 of the saddle switch the targets : now the right / left separatrix tends the left / right stable focus @xmath51 . in order for the unstable chaotic set to become the lorenz attractor , it must become invariant , i.e. a closed set containing all @xmath52-limit orbits , and hence no loose of trajectories escaping to stable foci @xmath2 this occurs on the bifurcation curve , @xmath49 , in the parameter space ( fig . [ fig2l](d ) ) . to the right of the curve , the basin of the lorenz attractor is shielded away from those of the stable equilibrium states by the 2d cylinder - shaped stable manifolds of the two threshold " saddle orbits , @xmath48 that have simultaneously emerged from both separatrix loops , @xmath53 at the homoclinic explosion on the curve @xmath38 in the parameter plane . as one moves further to the right in the parameter plane , the saddle orbits , @xmath48 , keep narrowing the attraction basins of the foci @xmath2 , and on the bifurcation curve @xmath54 they collapse into the stable equilibria . the equilibrium states become saddle - foci of the ( 1,2)-type through a subcritical andronov - hopf bifurcation @xcite . the topological ( 1,2)-type means that each saddle - focus has 2d unstable and 1d stable manifolds ; the latter is formed by two incoming separatrices . some local properties of the saddle - foci can be revealed without evaluating their characteristic exponents explicitly . let @xmath55 stand for the real stable exponent of @xmath2 , and @xmath56 stand for a complex conjugate pair such that @xmath57 . observe that the divergence of the vector field defined by eqs . ( [ lorenz ] ) is given by @xmath58 $ ] , which equals @xmath59 . this implies @xmath60 , i.e. , the complex conjugate pair is closer to the imaginary axis than the real negative exponent , and hence the saddle - foci meet the shilnikov condition @xcite . therefore , as soon as the saddle - focus possesses a homoclinic loop , such a bifurcation causes the abundance of periodic orbits nearby . those periodic orbits constantly undergo saddle - node and period doubling bifurcations as the parameters are varied . moreover , since the divergence of the vector field of the lorenz equation is always negative , saddle - node bifurcations give rise to stable periodic orbits near the homoclinic saddle - focus bifurcation . under fulfillment of some global conditions , a single shilnikov saddle - focus bifurcation can lead to the formation of a spiral or screw - like attractor . however , a strange attractor due to the shilnikov saddle - focus in a 3d system with a negative divergence is no genuinely chaotic set in the sense that it contains stable periodic orbits within . because of that , such a chaotic attractor is called quasi - attractor thus referring to that because besides stable periodic orbits with weak basins it may have structurally unstable or non - transverse homoclinic orbits @xcite . note that systems in higher dimensions can possess genuinely strange attractors with the shilnikov loop without stable periodic orbits , the so - called wild chaotic attractors @xcite . . orange spheres on the butterfly wings indicating the turning points around the right and left saddle - foci define the kneading sequence entries , @xmath61 , respectively . ( b ) a typical time evolution of the @xmath62-coordinate of the right separatrix of the saddle . ] the lorenz attractor is non - hyperbolic because it includes the singularity at the origin the saddle equilibrium state with an 1d unstable manifold while all other saddle periodic orbits on the attractor have 2d stable and unstable manifolds . moreover , the manifolds of those orbits in the lorenz attractor ( self ) cross transversally in the 3d phase space thereby producing only structurally stable ( transverse ) homoclinic and heteroclinic trajectories . this condition imperative for the lorenz attractor will not always hold for larger parameter values and lead to homoclinic tangencies of the manifolds which are followed by saddle - node bifurcations in the newhouse regions @xcite of the parameter plane of the system . thus , in order for the lorenz attractor to be strange and chaotic with no stable orbits , it must not include the [ homoclinic ] saddle - foci , @xmath2 , as well as contain only structurally stable homoclinic orbits due to transverse intersections of the manifolds of saddle periodic orbits . chaos can be quantified by several means . one customary way is through the evaluation of the topological entropy . the greater the value of topological entropy , the more developed and unpredictable the chaotic dynamics become . another practical approach for measuring chaos in simulations capitalizes on evaluations of the largest ( positive ) lyapunov exponent of a long yet finite - time transient on the chaotic attractor . after the stable foci have lost the stability through the subcritical andronov - hopf bifurcation , the lorenz equation exhibits the strange attractor of the iconic butterfly shape . the wings of the butterfly are marked with two symmetric eyes containing the saddle - foci isolated from the trajectories of the lorenz attractor . this attractor is structurally unstable @xcite as it undergoes bifurcations constantly as the parameters of the lorenz equation are varied . the primary cause of structural and dynamical instability of chaos in the lorenz equation and similar models is the singularity at the origin the saddle with two one - dimensional outgoing separatrices . both separatrices fill in densely two spatially symmetric , [ @xmath7 ] , wings of the lorenz attractor in the 3d phase space ( see fig . [ fig1l ] ) . the lorenz attractor undergoes a homoclinic bifurcation when the separatrices of the saddle change the flip - flop pattern of switching between the butterfly wings centered around the saddle - foci . at such a change , the separatrices comes back to the saddle thereby causing additional homoclinic explosions in phase space @xcite . the time progression of the right " ( or symmetrical left " ) separatrix of the origin can be described geometrically and categorized in terms of the number of flip - flops around the equilibrium states @xmath64 and @xmath65 in the 3d phase space of the lorenz equation ( fig . [ fig1l ] ) . or , alternatively , can be reduced to the time - evolution of the @xmath62-coordinate of the separatrix , as shown in panel b of fig . [ fig1l ] . the sign - alternation of the @xmath62-coordinate suggests the introduction of a @xmath61-based alphabet to be employed for the symbolic description of the separatrix . namely , whenever the right separatrix turns around @xmath19 or @xmath20 , we write down @xmath66 or @xmath67 , respectively . for example , the time series shown in panel b generates the following kneading sequence starting with @xmath68 etc . in what follows we will introduce and demonstrate a new computational toolkit for the analysis of chaos in the lorenz - like models . the toolkit is inspired by the idea of kneading invariants introduced in @xcite . a kneading invariant is a quantity that is intended to uniquely describe the complex dynamics of the system that admit a symbolic description using two symbols , here @xmath66 and @xmath67 . the kneading invariant is supposed to depend monotonically on the governing parameter so that any two systems can be compared and differentiated , or equivalently , ordered in terms of @xmath69 and @xmath70 , by the kneadings . two systems with the lorenz attractors are topologically conjugate when they share the same kneading invariant , see @xcite and the references therein . moreover , for the lorenz - like systems , the topological entropy can be evaluated though the kneading invariants by reducing consideration to piecewise monotone mappings of a closed interval @xcite . such mappings are closely related to the geometric models of the lorenz attractor which are 1d and 2d poincar return mappings defined on some cross - section transverse to trajectories of the lorenz attractor . the basic idea behind either geometric model capitalizes on extended properties of the local poincar mapping near a homoclinic butterfly of the saddle @xcite . the mapping is assumed to meet a few other conditions related to the global properties of the flow far from the homoclinic butterfly . such a 1d constrained mapping shown in fig . [ 1dmapping](a ) can be written as : @xmath71 here @xmath72 is the saddle index , @xmath73 controls the distance between the returning separatrices , @xmath74 , and the saddle , and @xmath47 is a scalar whose sign determines whether the homoclinic loops at @xmath75 are oriented if @xmath76 , or twisted when @xmath77 , see @xcite for more details . loosely speaking , this geometric model ( [ map ] ) should have no critical point on both branches , and moreover possess a property of strong stretching with a rate of expansion more than @xmath78 @xcite . this would guarantee that the lorenz attractor will be densely filled in by the forward iterates of the separatrices , @xmath79 with no holes lacunas containing isolated periodic orbits inside , stable or not . strong stretching would also imply a monotone dependence of the kneading invariants on a governing parameter . in a symmetric system with the lorenz attractor , the kneading invariant is assigned to quantify the symbolic description of either separatrix ; in the asymmetric case one should consider two kneading invariants for both separatrices of the saddle . thus , in respect it reflects quantitatively a qualitative change in the separatrix behavior , such as flip - flopping patterns , as the parameter of the system is changed . by construction , kneading invariants are proposed to serve as moduli of the topological equivalence that are employed to compare or contrast between any two lorenz attractors or , equivalently , any two lorenz - like systems . due to the symmetry of the lorenz mapping @xmath80 from eq . ( [ map ] ) , forward iterates of the right separatrix , @xmath81 , of the discontinuity point ( resp . the saddle ) are detected to generate a _ kneading sequence _ @xmath82 defined by the following rule : @xmath83 here @xmath84 is the @xmath85-th iterate of the right separatrix @xmath81 of the origin . the condition @xmath86 is interpreted as a homoclinic loop , i.e. the separatrix returns to the origin after @xmath85 steps . the kneading invariant for the separatrix is defined in the form of a formal power series : @xmath87 setting @xmath88 make the series ( [ fs ] ) convergent . the smallest zero , @xmath89 , if any , of the graph of ( [ fs ] ) in the interval @xmath88 yields the topological entropy , @xmath90 , of the 1d mapping ( [ map ] ) . let us next draw a parallelism between the geometric model and the lorenz equation : the kneading sequence @xmath91 comprised of only @xmath66s of the mapping ( [ map ] ) corresponds to the right separatrix converging to the stable equilibrium state , @xmath19 ( or possibly , a periodic orbit with @xmath92 ) . the corresponding kneading invariant is maximized at @xmath93 . when the right separatrix converges to an @xmath94-limit set with @xmath95 , like the left stable focus , @xmath65 then the kneading invariant is given by @xmath96 because the first entry @xmath66 in the kneading sequence is followed by infinite @xmath67s . thus , @xmath97 $ ] yield the range of the kneading invariant values ; for instance @xmath98 $ ] . two samples of the separatrix pathways shown in fig . [ figtwo ] generating the following kneading invariants @xmath99 illustrate the parallelism . . [ fig3l ] to conclude this section we point out that there is another approach for constructing 1d return mappings through the evolution of the @xmath34-variable of the separatrix of the lorenz equation . the mapping generated by the turning points where @xmath100 on the attractor has a distinct cusp - shaped form depicted in fig . [ 1dmapping](b ) . the point corresponding to the cusp is used for the initiation of the kneading sequence . due to its unimodal graph with two , increasing and decreasing , segments the corresponding formal power series is then defined as follows : @xmath101 i.e. @xmath102 . in this section we carry the concept of the kneading invariants over to numerical studies of fine structures of chaos in the lorenz equation . for sake of simplicity in this pilot phase , we employ a rough idea for defining the kneading sequences of @xmath66s and @xmath67s that relies on whether the right separatrix of the saddle makes a revolution around the right equilibrium state , @xmath19 , or the left one , @xmath20 , respectively , in the ( @xmath103)-projection of the 3d phase space . one can utilize even a simpler approach where the sign of the current kneading entry in the sequence is determined by the sign of the @xmath62-coordinate of the separatrix at the off - lying turning points , @xmath104 ( on the butterfly wings ) , given by @xmath105 , see the trace in fig . [ fig1l](b ) . there are other alternative ways for defining kneading entries , involving cross - sections , @xmath106 etc , that are not free of certain limitations either . in the future we plan to enhance the current kneading algorithms by utilizing the incoming 1d separatrices of the saddle - foci and finding the winding numbers around them instead . in this computational study of the lorenz equation and two other models below , we consider partial kneading power series truncated after the first 50 entries : @xmath107 . the choice of the number of entries is not motivated by numerical precision , but by simplicity , as well as by a resolution of the bitmap mappings for the bi - parametric scans of the models . one has also to figure the proper value of @xmath108 : setting it too small makes the convergence fast so that the tail of the series has a little significance and hence does not differentiate the fine dynamics of the lorenz equation on longer time scales . at the first stage of the routine , we perform a bi - parametric scan of the lorenz equation within a specific range in the @xmath109-plane . the resolution of scans is set by using mesh grids of @xmath110 $ ] equally - distanced points . next by accurately integrating the separatrix using taylor series software tides @xcite we identify and record the sequences @xmath111 for each point of the grid in the parameter plane . then we define the bi - parametric mapping : @xmath112 with some appropriately chosen @xmath108 , the value that determines the depth of the scan . the mapping is then colorized in matlab by using various built - in non - linear spectra ranging between to @xmath113 and @xmath114 , respectively . in the mapping , a particular color in the spectrum is associated with a persistent value of the kneading invariant on a level curve . such level curves densely foliate the bi - parametric scans . -bifurcation diagram of the lorenz equation depicting the existence region ( shaded ) of the lorenz attractor . the second bifurcation curve , @xmath115 , passing through the primary t - point , @xmath116 , crosses the first boundary , @xmath117 ( @xmath49 in fig . [ fig2l ] ) , at a point , labeled by @xmath118 , thus closing the existence region . crossing the branch @xmath117 rightward above the point @xmath118 does lead to the emergence of the lorenz attractor : after a long chaotic transient , the separatrix of the saddle will be attracted to either stable foci , @xmath19 or @xmath20 . courtesy of @xcite . , scaledwidth=40.0% ] figure [ fig3l ] represents the kneading - based color scan of the dynamics of the lorenz equation mapped onto a fragment of the ( @xmath119)-parameter plane . in the scan , a window of a solid color corresponds to a constant kneading invariant . in such windows the dynamics of the lorenz equation are dominated by simple attractors such as stable equilibria or stable periodic orbits to which the separatrix converge . a quick examination of the kneading definition ( [ fs ] ) reveals that the kneading invariant does not vary at a supercritical andronov - hopf bifurcation and a pitch - fork bifurcation describing continuous transitioning between stable symmetric and asymmetric periodic orbits . this holds true for a period - doubling bifurcation too , because the kneading sequence , say @xmath120 , inherits the same block repeated twice @xmath121 in the code for the periodic orbit of a double period and so forth . while the kneading technique does not detect such safe bifurcation boundaries @xcite having crossed through which the phase point does not run far from an old attractor to a new one , though it detects dangerous boundaries well , including a generic saddle - node bifurcation , homoclinic bifurcations , and some other . a borderline between two solid - color regions corresponds to a bifurcation through a dangerous boundary which is associated with a jump between the values of the kneading invariant . for example , the borderline in figure [ fig3l ] between the brown region with the kneading sequence @xmath122 and the blue region , with the kneading sequence @xmath123 , corresponds to the primary homoclinic butterfly of the saddle . the second borderline of the blue region corresponds to the onset of the lorenz attractor existing on the right from it . one can see that above @xmath124 this border is adjoined by windows of solid colors thus indicating that the separatrices start converging to stable equilibria after chaotic transients , long or short . indeed , crossing the curve , @xmath49 ( or @xmath117 in fig . [ fig2l ] ) , above @xmath125 does not imply that the lorenz equation will have a strange attractor , but a chaotic saddle set @xcite . what the proposed kneading technique does extraordinary well , compared to the bi - parametric screening based on the finite - time lyapunov exponent approach ( see below ) , and what we developed it for , is the detection of bifurcations within the lorenz strange attractor . the corresponding yellow - reddish regions in the parameter plane in fig . [ fig3l ] clearly demonstrate the evidence of the parametric chaos that , like in turbulence , enriched by vortices of multiple t - points . panel b of this figure depicts the kneading mapping near the left - bottom corner of the bifurcation diagram in panel a. in it , the black ( blue in panel a ) region corresponds to the chaotic saddle dynamics transitioning into the lorenz attractor after the mapping color shifts to the yellow - reddish spectrum . blue curves in panel b are the bifurcation curves of some homoclinic orbits with short admissible kneadings . one can see from this panel that the mapping spectrum is clearly foliated by the kneading invariant level curves of the colors gradually progressing from red to yellow . this indicates that the new born lorenz attractor , while being structurally unstable and sensitive to parameter variations , persists initially the pseudo - hyperbolic property because the foliation remains uniform , and transverse to the classical pathway @xmath10 ( fig . [ fig2l ] ) . the homogeneous foliation starts breaking around a saddle point after which one singled bifurcation curve spirals onto the primary t - point . far from this point , the curve corresponds to the formation of the homoclinic loop with the kneading @xmath126 ; i.e. the right separatrix makes one excursion around the saddle - focus @xmath19 , followed by three revolutions around the saddle - focus @xmath20 , and then returns to the saddle at the origin . while moving along the spiraling curve toward the t - points , the separatrix makes progressively more turns around @xmath20 , or more precisely around the 1d incoming separatrix of this saddle - focus . with each incremental turn around @xmath20 , the separatrix comes closer to @xmath20 while the bifurcation curve becomes one scroll closer to the t - point simultaneously . due to this feature the t - point @xmath127 is called a terminal point . the t - point corresponds to the following symbolic sequence @xmath128 . in virtue of the symmetry of the lorenz equation , the t - point actually corresponds to a closed heteroclinic connection involving all three saddle - equilibria , see figs . [ figtpoint ] and [ fig1l ] . the merger of the right ( left ) separatrix of the saddle @xmath1 with the incoming separatrix of the saddle - focus @xmath129 , increases the codimension ( degeneracy ) of this heteroclinic bifurcation to two ; note that intersections of the 2d unstable manifolds of the saddle - foci , with the 2d stable manifold of the saddle at the origin are transverse in the 3d phase space in general . breaking the 1d heteroclinic connection gives rise to a primary homoclinic orbit to the saddle - focus , as well as to a heteroclinic connection between both saddle - foci ( see the sketch of the bifurcation unfolding of the t - point in fig . [ figtpoint ] ) . the corresponding bifurcation curves of these homoclinic and heteroclinic bifurcations originating rightward from the t - point bound a sector containing subsidiary t - points @xcite . each new t - point produces other self - similar structures scaled like fractals . panel c shows two such identified t - points : primary @xmath130 and secondary @xmath131 . the primary codimension - two t - point in the lorenz equation was originally discovered by yudovich @xcite . finite - time largest - lyapunov exponent , @xmath132 , scan of the lorenz equation showing no sign of spiral structures in the @xmath24-parameter plane . the dark region corresponds to trivial attractors , where @xmath133 , while the red color indicates @xmath134 in chaotic regions . the red dot points out the location of the primary t - point . ] as soon as the saddle - foci and their bifurcations become involved in the dynamics of the lorenz equation near the primary t - point , the lorenz attractor loses the purity of the genuine chaotic attractor that used to have neither stable periodic orbits nor non - transverse homoclinic trajectories . it transforms into a quasi - chaotic attractor with weakly stable orbits of small basins and nontransverse homoclinic orbits . the idea of non - transversality or tangency was employed in @xcite to numerically identify the second boundary , @xmath135 , in addition to the first one , @xmath117 ( @xmath49 in fig . [ fig2l ] ) , that bounds the existence region of the lorenz attractor in the parameter plane , see fig . [ fig4l ] . note that @xmath135 crosses the initial boundary , @xmath117 . this means that above the intersection point , crossing @xmath117 ( @xmath49 in fig . [ fig2l ] ) rightwards does not guarantee that the basin of the lorenz attractor will necessarily be isolated from the basins of the stable foci , @xmath2 by the stable manifolds of the saddle periodic orbits , @xmath48 ( fig . [ fig2ln ] ) . this implies that the separatrices of the saddle will demonstrate chaotic transient behavior , long or short , prior to them converging to @xmath2 , see @xcite for full details . the other feature of the boundary , @xmath135 , is that it passes through the primary t - point , thereby separating the existence region of the lorenz attractor from bifurcations of the saddle - foci , and consequently from all subsidiary t - points existing on the right from it in the parameter plane . here the chaotic dynamics of the lorenz equation become even wilder and less predictable @xcite . the indication to that is panel a of fig . [ fig3l ] that reveals , through the kneading scan , a parametric turbulence in the @xmath24-parameter plane with fractal explosions in the forms of multiple spiral structures tornado eyes " centered around t - points . note that basins of spiraling t - points are separated by corresponding saddles . one can spot self - similar smaller - scale spiral structures within large - scale ones and so forth . the richness of such fractal structures in the parameter plane results from the synergy of the lorenz - like dynamics amplified by chaos induced by the shilnikov saddle - foci . to conclude this section we contrast the scans of the lorenz equation obtained using the proposed kneading technique with the sweeps based on the evaluation of the largest lyapunov exponent , @xmath136 , for the separatrices of the saddle evaluated over a finite time interval @xcite . figure [ fig4la ] shows a fragment of the typical bi - parametric sweep of the lorenz equation : the dark region at small values of the reynolds number , @xmath6 , is where @xmath136 is negative on the stable foci . the sweep yields a clear borderline between the regions of the simple and chaotic attractors . the chaotic region ( yellow - reddish ) is characterized by a small positive lyapunov exponent , variations of which are not significant enough to reveal fine structures , as spiraling t - points . the method can detect well stability islands in the biparametric diagram which correspond to emergent stable periodic orbits . two panels in fig . [ fig10 ] represent the expanded scans of the of lorenz equation : panel ( a ) is the kneading invariant mapping on the grid of @xmath110 $ ] points and panel ( b ) shows @xmath136-based sweeping of the @xmath24-parameter plane . while panel a reveals a chain of large - scale spiral hubs around t - points , panel b reveals none but stability windows ( dark ) . we would like to point out that the stability windows can also be detected in the kneading scan in panel a showing the border of the solid color ( dark yellow ) island stretched horizontally at small values of the parameter @xmath4 . next we perform the kneading - based bi - parametric scanning of another classical three - dimensional system called the shimizu - morioka model @xcite : @xmath137 here , @xmath52 and @xmath138 are positive bifurcation parameters . like the lorenz equation , this @xmath0-symmetric model has 3 equilibrium states : a saddle of the ( 2,1)-topological type at the origin , and two symmetric stable - foci or saddle - foci of the ( 1,2)-topological type . this model was originally introduced to examine a pitch - fork bifurcation of the stable figure-8 periodic orbit that gives rise to multiple cascades of period doubling bifurcations in the lorenz equation at large values of the reynolds number @xmath6 . it was proved in @xcite that the eqs . ( [ sm ] ) are a universal normal form for several codimension - three bifurcations of equilibria and periodic orbits on @xmath0-central manifolds . the model turned out to be very rich dynamically : it exhibits various interesting global bifurcations @xcite including t - points for heteroclinic connections . while the model inherits all basic properties of the lorenz equation , in addition , and of special interest , are two homoclinic bifurcations of codimension - two : resonant saddle with the zero saddle value or the saddle index @xmath139 , and the orbit - flip bifurcation where the separatrix value @xmath47 in eq . ( [ map ] ) vanishes . recall that the sign of the separatrix value determines whether the homoclinic loop , here double - pulsed , is oriented or flipped . these codimension - two points globally organize the structure of the compact ( @xmath140)-parameter region of the shimizu - morioka model , including structural transformations of the lorenz attractor in the model , including the emergence of stability islands lacunae inside the attractor . detailed @xmath141-parameter plane of the shimizu - morioka model obtained by the parameter continuation method ( courtesy of @xcite ) and by the bi - parametric scan based on the kneading invariants . the scan revealing multiple t - points and saddles that globally organize complex chaotic dynamics of the model . solid - color regions associated with constant values of the kneading invariant correspond to simple dynamics dominated by stable equilibria ( brown ) or stable periodic orbits ( blue ) . the border between the brown and blue regions corresponds to the bifurcation curve of the homoclinic butterfly . the codimension two point , @xmath142 , gives rise to loci of bifurcation curves including @xmath143 below which the lorenz attractor exists . bifurcation loci of the other codimension two point , @xmath144 ( orange zone ) giving rise to subsidiary orbit - flip bifurcations on turns of spirals around t - points , are separated by saddles ( two large scale ones ) in the parameter plane . , scaledwidth=95.0% ] figure [ fig5l ] represents a partial ( @xmath140)-diagram of the shimizu - morioka model and depicts primary bifurcation curves and the corresponding phase portraits of the separatrix behaviors . among those are the andronov - hopf bifurcation curve , @xmath145 , above which the equilibrium states , @xmath2 are stable , and saddle - foci below . this bifurcation is primarily supercritical , but becomes subcritical at smaller values of the parameter @xmath52 . the bifurcation curve , @xmath146 , corresponds to the formation of the homoclinic butterfly , or figure-8 at larger values of @xmath52 . the transition between the branches of the curve is no bifurcation like the inclination - switch because the saddle value , @xmath4 is negative here . the pathway at @xmath147 demonstrates the evolution of the simple morse - smale dynamics of the model from the stable equilibrium states to a stable symmetric periodic orbit . this orbit emerges through a gluing bifurcation after that two stable periodic orbits existing below the supercritical andronov - hopf curve @xmath145 form the homoclinic butterfly with @xmath148 . the saddle value becomes positive to left from the codimension - two point , labeled by @xmath142 on the bifurcation curve @xmath146 . the left segment of the curve @xmath146 is similar to the bifurcation curve , @xmath38 , of the lorenz equation ( fig . [ fig2l ] ) ; namely , the homoclinic butterfly with @xmath149 causes the homoclinic explosion in the phase space of the shimizu - morioka model . the curve @xmath150 ( @xmath143 in fig . [ fig5l ] ) being an analog of the curve @xmath49 in the bifurcation diagram in fig . [ fig2l ] is the upper boundary of the existence region of the lorenz attractor is the parameter plane of this model . below @xmath150 the separatrices of the saddle no longer converge to stable periodic orbits but fill in the strange attractor , like in fig . [ fig2ln ] . the bifurcation unfolding of the homoclinic resonant saddle includes various bifurcation curves : among them in the figure we depict , in addition to @xmath150 , the curve , @xmath151 , corresponding to saddle - node of merging stable ( through the supercritical andronov - hopf bifurcation ) and saddle ( through the homoclinic bifurcation on @xmath146 ) periodic orbits . the curve labeled by @xmath152 corresponds to a pair of the double - pulsed homoclinic loops . continued further away from the codimension - two point , @xmath142 , the curve @xmath152 frames the chaotic region in the parameter plane . the point , @xmath144 , on this curve is a codimension - two orbit - flip homoclinic bifurcation : to the left of it , the loops become flipped like the median of a mbius band . the dashed line , @xmath153 originating from the point @xmath144 is alike the curve @xmath154 in fig . [ fig4l ] for the lorenz equation . this curve is the second boundary of the existence region ( above ) of the lorenz attractor in the shimizu - morioka model , that separates wild chaos with homoclinic tangencies ( below ) . this like passes through the primary t - point in the @xmath155-parameter plane . figure [ fig5l ] depicts a few more bifurcation curves originating from the point , @xmath144 : two four - pulses homoclinic curve terminating at subsidiary t - points and the curve labeled by `` @xmath67 '' corresponding to a period doubling bifurcation . an intersection of the dashed line with a homoclinic bifurcation curve ( see fig . [ fig5l ] ) corresponds to another orbit - flip bifurcation and so forth . zoom of the @xmath141-parametric mapping in fig . [ fig5l](b ) near the primary t - point revealing self - similar structures embedding smaller - scale spirals around secondary t - points in the shimizu - morioka model . ] indeed the skeleton of the bifurcation set of the shimizu - morioka is more complex . the detailed bifurcation diagram is shown in the top panel of fig . [ fig5l ] . it reveals several t - points , the pitch - fork bifurcation curve , @xmath156 , among other bifurcation curves for various homoclinic and heteroclinic connections . the detailed description of the bifurcation structure of the shimizu - morioka model is out of scope of this paper . the curious reader can find a wealth of information on bifurcations of the lorenz attractor in the original papers @xcite . we point out that those bifurcation curves were continued in the @xmath141-parameter plane of the model using a. shilnikov s home - made software also based on the symbolic kneading toolbox . the bottom panel of fig . [ fig5l ] is a _ de - facto _ proof of the new kneading invariant mapping technique . the panel represents the bi - parametric scan in color of the dynamics of the shimizu - morioka model that is based on the evaluation of the first 50 kneadings of the separatrix of the saddle on the grid of @xmath157 points in the @xmath141-parameter region . getting the mapping took about one hour on a high - end workstation without any parallelization efforts . the color scan reveals a plethora of large - scale t - points , as well as nearby smaller ones ( fig . [ fig6l ] ) invisible in the given parameter range , as well as the saddles separating spiral structures . the solid - color zones in the mapping correspond to simple morse - smale dynamics in the model . these trivial dynamics are due to either the separatrix converging to the stable focus @xmath19 ( @xmath20 ) and emergent periodic orbit with the same kneading invariant ( brown region ) , or to the symmetric and asymmetric stable figure-8 periodic orbits ( dark blue region ) . the borderlines between the simple and complex dynamics in the shimizu - morioka model are clearly demarcated . on the top it is the curve , @xmath143 , ( see the top panel of fig . [ fig5l ] ) . the transition from the stable 8-shaped periodic orbits to the lorenz attractor ( through the boundary , @xmath158 ) is similar though more complicated as it involves a pitch - fork bifurcation and bifurcations of double - pulsed homoclinics , see @xcite for details . one can clearly see the evident resemblance between both diagrams found using the bifurcationaly exact numerical methods and by scanning the dynamics of the model using the proposed kneading invariant technique . the latter reveals a richer structure providing finer details . the structure can be enhanced further by examining longer tails of the kneading sequences . this allows for the detection of smaller - scale spiral structures within scrolls of the primary t - vortices , as predicted by the theory @xcite . figure [ zoom ] shows a magnification of the scan of the shimizu - morioka model near the primary t - point that depicts several other small - scale t - points . finally , we compare the new kneading scanning apparatus with the customary bi - parametric sweeping ( shown in fig . [ simo ] ) of the shimizu - morioka model that is based on the evaluation of the lyapunov exponent spectrum computed over a finite time interval @xcite . likewise the case of the lorenz model , the sweeping based on the lyapunov exponents shows no sign of spiral or saddle structures inside the region of deterministic chaos . the regions of the solid colors are associated with the sign of the largest lyapunov exponent , @xmath132 : negative @xmath132 values correspond to steady state attractors in the green region ; @xmath159 corresponds to periodic attractors in the blue region ; and @xmath134 is associated with chaotic dynamics in the model in the region . note blue islands in the red - colored region that correspond to stability windows in chaos - land . in such windows the lorenz attractor has an emergent lacuna containing , initially , a single symmetric saddle periodic orbit . the orbit then undergoes a pitch - fork bifurcation that makes it stable . the basin of the stable orbit , which is first bounded by the 2d stable manifold of two asymmetric saddle periodic orbits , increases so that the stable orbit starts to dominate over chaotic dynamics in the corresponding stability window . these bifurcations underlie the transition from simple dynamics ( blue region ) due to the symmetric stable periodic orbit to chaos through the curve , @xmath152 , as the parameter @xmath52 in decreased . the coexistence of multiple t - points and accompanying fractal structures in the parameter plane is a signature for systems with the lorenz attractor . a question though remains whether the new computational technique will work for systems of dimensions higher than three . in fact , to apply the technique to a generic lorenz - like system , only wave forms of a symmetric variable progressing in time , that consistently starts from the same initial condition near the saddle is required . next is an example from nonlinear optics a 6d model of the optically pumped , far infrared red three - level molecular laser @xcite given by @xmath160 here , @xmath52 and @xmath5 are the rabi flopping quantities representing the electric field amplitudes at pump and emission frequencies . the parameter @xmath52 is a natural bifurcation parameter as it is easily varied experimentally . the second bifurcation parameter , @xmath5 , can be varied to some degree at the laboratory by the addition of a buffer gas . this system presents , like the lorenz equations , a symmetry @xmath161 . the laser model has either a single central equilibrium state , @xmath1 ( with @xmath162 ) , or additionally , through a pitch - fork bifurcation , a pair of symmetric equilibrium states , @xmath2 ( with @xmath163 ) ; the stability of the equilibria depends on the parameter values . ( a ) lorenz attractor with a lacuna in the laser model at @xmath164 , @xmath165 , @xmath166 and @xmath10 . ( b ) @xmath167-bifurcation diagram of the model for @xmath168 and @xmath169 . @xmath170 and hb here denote the pitch - fork and andronov - hopf bifurcations , respectively . @xmath171 and @xmath172 denote the branches of the primary homoclinic ( of the saddle ) and heteroclinic orbits ( of the saddle - foci ) . @xmath173 is the codimension two khorozov - taken point for the equilibrium state with double zero eigenvalues , and @xmath174 is the primary terminal point . the spiraling curve connects the t - point with the homoclinic resonant saddle on @xmath171 , near which separatrix loops are double pulsed ones . courtesy of @xcite . , title="fig : " ] ( a ) lorenz attractor with a lacuna in the laser model at @xmath164 , @xmath165 , @xmath166 and @xmath10 . ( b ) @xmath167-bifurcation diagram of the model for @xmath168 and @xmath169 . @xmath170 and hb here denote the pitch - fork and andronov - hopf bifurcations , respectively . @xmath171 and @xmath172 denote the branches of the primary homoclinic ( of the saddle ) and heteroclinic orbits ( of the saddle - foci ) . @xmath173 is the codimension two khorozov - taken point for the equilibrium state with double zero eigenvalues , and @xmath174 is the primary terminal point . the spiraling curve connects the t - point with the homoclinic resonant saddle on @xmath171 , near which separatrix loops are double pulsed ones . courtesy of @xcite . , title="fig : " ] optically pumped , far infrared lasers are known to demonstrate a variety of nonlinear dynamic behaviors , including lorenz - like chaos @xcite . an acclaimed example of the modeling studies of chaos in nonlinear optics is the two level laser haken model @xcite to which the lorenz equation can be reduced . a validity that three level laser models would inherently persist the lorenz dynamics were widely questioned back then . it was comprehensively demonstrated in @xcite in 1991 that this plausible laser model possesses a plethora of dynamical and structural features of the lorenz - like systems , including the lorenz attractor _ per se _ ( with lacunae as well ) , similar andronov - hopf , @xmath0 pitchfork , various homoclinic and heteroclinic bifurcations including codimension - two ones t - points , see fig . [ laser1 ] . similar structures were also discovered in another nonlinear optics model for a laser with a saturable absorber which can be reduced to the shimizu - miorioka model near a steady state solution with triple zero exponents @xcite ( a ) bi - parametric scan of the laser model featuring the t - points and saddles typical for the lorenz - like systems , mapping the dynamics of the 6d optically pumped far - infrared - red laser model onto the ( electric - field - amplitude , omission - frequency)-diagram at @xmath175 and @xmath176 . solid - color windows and fractal regions correspond to trivial and chaotic dynamics generated by the laser model . ( b ) partial bifurcation diagram though the parameter continuation showing the curves for pitch - fork ( @xmath177 ) and andronov - hopf ( @xmath178 ) bifurcations for the equilibrium state , @xmath1 , and another similar supercritical one for @xmath2 . the homoclinic curve , @xmath179 begins from the codimension - two point , @xmath180 for the khorozov - takens bifurcation and ends up at the resonant saddle point . ( b ) elevating @xmath181 makes the @xmath179 turned by a saddle point in the parameter plane and terminate at the primary t - point . ] the laser model ( [ laseh ] ) is quite rich in bifurcations ; the list also includes a super - critical andronov - hopf bifurcation of the central equilibrium state that gives rise to a stable figure-8 shaped periodic orbit ( in proper projections ) for the parameter values to the left of the bifurcation curve , @xmath178 , in the bifurcation diagrams shown in panels b and c of fig . [ laser2 ] . observe from the diagram that the curve @xmath178 originates from the point labeled , @xmath180 . this point corresponds to a codimension - two khorozov - takens bifurcation of an equilibrium state with two zero lyapunov exponents . the bifurcation is an extension of the bogdadov - takens bifurcation on a symmetric central manifold . the unfolding of this bifurcation includes two more curves : @xmath182 standing for the andronov - hopf bifurcation for the secondary equilibrium states , @xmath2 ; and @xmath179 standing for a homoclinic connection made of two symmetric separatrix loops bi - asymptotic to the saddle , @xmath1 . the continuation of the curve , @xmath179 , away from @xmath180 reveals rather peculiar details that substantially organize the bifurcation diagram of this laser model . near @xmath180 the curve , @xmath179 , corresponds to a homoclinic figure-8 of the saddle with a negative saddle value , on the @xmath0-symmetric 2d central manifold in the 6d phase space of the model . recall that the figure-8 homoclinic connection stands for the case where the 1d unstable separatrices come back to the saddle , @xmath1 from the symmetrically opposite directions along the eigenvector corresponding to the leading stable exponent at the equilibrium state . this bifurcation gluing two stable periodic orbits emerging from @xmath2 through the supercritical andronov - hopf bifurcation gives rise to the stable symmetric figure-8 periodic orbit existing nearby @xmath180 . as the curve , @xmath179 , is continued further away from @xmath180 , the stable leading direction at the saddle , @xmath1 , changes : it becomes the invariant @xmath138-axis ( like the @xmath34-axis in the shimizu - morioka model ) so that the separatrix loops start returning tangent to each other and hence forming the homoclinic butterfly . nevertheless , this is a gluing bifurcation , not a codimension - two bifurcation of the change of the leading direction ( inclination switch ) as the saddle value remains negative on this branch of the curve , @xmath179 . the saddle value vanishes , making the saddle resonant at the codimension - two point , and becomes positive further down on @xmath179 . as the curve is continued , the homoclinic butterfly undergoes another codimension - two orbit - flip bifurcation so that the separatrices loops of the saddle , @xmath1 become non - orientated . as the result , further down the curve , each loop gains an extra turn around the incoming separatrix of the opposite saddle - focus , i.e. becomes a double - pulsed one with the @xmath183 kneading . depending on the parameter cut @xmath4 there are two scenarios for termination of the curve , @xmath179 , in the @xmath167 diagram ( compare the bifurcation diagrams in panels b and c of fig . [ laser2 ] ) : first , when @xmath169 it terminates at the codimension - two t - point corresponding to the heteroclinic connection between all saddle equilibria , @xmath1 and @xmath2 as shown in panel b. the second scenario for is less predictable at @xmath184 : the branch , @xmath179 , of double - pulsed separatrix loops ends up at the codimension - two point of the resonant saddle with the zero saddle - value ( panel c ) . the answer to the question what makes the curve change its destination is a saddle point in the parameter diagram that the kneading scan reveals in fig . [ fig5l ] . by varying the @xmath4-parameter cut in the 3d parameter space , this bifurcation curve is destined by the saddle to finish at either terminal point , see details in @xcite . in the case where it spirals onto the t - point , there is another bifurcation curve corresponding to the same @xmath183 kneading that connects the codimension - two orbit - switch point and the point corresponding to the resonant saddle located on the curve @xmath179 . the panel a in fig . [ laser2 ] represents the kneading scans of the dynamics of the laser model which is mapped onto the @xmath185-parameter plane with @xmath175 and @xmath176 . the scan is done using the same 50 kneading entries . it has the regions of chaotic dynamics clearly demarcated from the solid color windows of persistent kneadings corresponding to trivial attractors such as stable equilibria and periodic orbits . the region of chaos has a vivid fractal spiral structure featuring a chain of t - points . observe also a thin chaotic layer bounded away from the curve @xmath179 by a curve of double - pulsed homoclinics with the kneading @xmath186 connecting the codimension - two points : the resonant saddle and the orbit - flip both on @xmath179 . one feature of these points is the occurrence of the lorenz attractor with one or more lacunae @xcite . such a strange attractor with a single lacuna containing a figure-8 periodic orbit in the phase space of the given laser model is shown in panel a of fig.[laser1 ] . we have demonstrated the new proposed computational toolkit for thorough explorations of chaotic dynamics in three exemplary models with the lorenz attractor . the algorithmically easy though powerful toolkit in based on the scanning technique that maps the dynamics of the system onto the bi - parametric plane . the core of the approach is the evaluation of the kneading invariants for regularly or chaotically varying flip - flop patterns of a single trajectory the separatrix of the saddle singularity in the system . in the theory , the approach allows two systems with the structurally unstable lorenz attractors to be conjugated with the mean of a single number the kneading invariant . by using ready - for - use tools in matlab , we could have the parameter plane of the model in question be foliated by the level curves of distinct colors corresponding to distinct values of the kneading invariants . the kneading scans revel unambiguously the key features in the lorenz - like systems such as a plethora of underlying spiral structures around t - points , separating saddles in intrinsically fractal regions corresponding to complex chaotic dynamics . we point out that no other techniques , including approaches based on the lyapunov exponents , can reveal the discovered parametric chaos with such stunning clarity and beauty . figure [ sm3 ] for the shimizu - morioka model shows that a fine scan based on the finite - time lyapunov exponents is able to indicate some presence of spiral and saddle structures and differentiate between the chaotic dynamics due to the lorenz attractor and those due to additional degrees of instability brought in by saddle - foci . fine scan based on the lyapunov exponents indicating the presence of the saddle ( in the lorenz attractor region shown in cold blue ) and spiral structures ( in the reddish regions with larger lyapunov exponents for wild chaos due to saddle - foci ) in the ( @xmath187)-parameter space of the shimizu - morioka model.,scaledwidth=85.0% ] the kneading based methods should be beneficial for detailed studies of other systems admitting reasonable symbolic descriptions . it bears an educational aspect as well : the kneading based scanning can be used for in - class presentation to reveal the fascinating complexity of low - order models in the cross - disciplinary field of nonlinear dynamics . the bi - parametric mapping technique can be easily adopted for parallel multicore gpu platforms allowing for ultra - fast simulations of models in questions . additional implementation of high - precision computations of long transients shall thoroughly reveal multi - layer complexity of self - similar fractal patterns near t - point vortices . in future research we intend to enhance and refine the toolkits for exploration of other symmetric and asymmetric @xcite systems of differential and difference equations , like 3d poincar mappings @xcite , including 4d models with saddle - foci , that require two and more kneading invariants for the comprehensive symbolic description . this work is supported by the spanish research project mtm2009 - 10767 ( to r.b . ) , and by nsf grant dms-1009591 , rffi grant no . 08 - 01 - 00083 , gsu brain & behaviors program , and mesrf project 14.740.11.0919 ( to a.s ) as well as by the grant 11.g34.31.0039 of the government of the russian federation for state support of scientific research conducted under supervision of leading scientists in russian educational institutions of higher professional education ( to l.p.s ) . we thank dima turaev for stimulating discussions , yuri kuznetsov for sharing the data used in fig . [ fig3l ] . we thank aaron kelley and jeremy wojcik for careful proofreading the manuscript . barrio , r. , blesa , f. , serrano , s. & shilnikov , a. [ 2011b ] `` global organization of spiral structures in bi - parameter space of dissipative systems with shilnikov saddle - foci , '' _ phys . e _ * 84 * , 035201 .

all the models of elementary particles and their interactions derived from string theory involve a compact six dimensional internal space . its volume and shape should be fixed or stabilized , since otherwise massless scalar fields ( moduli ) reflecting their deformations appear in our four dimensional space time , with sizable effects on known particles and fields . we propose a strategy towards stabilizing the compact space without fluxes of three form fields from closed strings . our main motivation and goal is to proceed insofar as possible within conventional string world sheet theory . as we shall see , d - branes with magnetic flux ( `` magnetized d - branes '' ) and the forces between them can be used to this end . we investigate here some necessary ingredients : open string one loop vacuum amplitudes between magnetized d - branes , magnetized d - branes fixed at orbifold singularities , and potential energies among such d - branes in the compact space that result from tree level closed string exchanges . * * 1.2 towards the stabilization of extra dimensions by brane dynamics noriaki kitazawa department of physics , tokyo metropolitan university , + hachioji , tokyo 192 - 0397 , japan + e - mail : [email protected]

1409.8401

superstring theory in a flat ten dimensional space time can provide a framework to describe the elementary particles and their interactions including gravity beyond the standard model . the heterotic string theory , with its elegant inclusion of gauge symmetry , can naturally accommodate models with grand unification of the gauge interactions ( see for example @xcite ) , while the structure of yukawa couplings in the standard model can find a natural setting in type iia / iib superstring theory with d - branes ( see for example @xcite ) . both scenarios , however , rest on a six dimensional compact internal space whose volume and shape should be fixed or stabilized ( moduli stabilization ) . aside from the wide arbitrariness in the choice of vacuum , this remains a difficult problem in string theory . in this paper we focus on the moduli stabilization problem in type iib theory , which has been extensively studied only in the low energy effective field theory , the type iib supergravity , introducing three form fluxes from massless modes of the corresponding closed string ( `` flux compactifications '' ) @xcite , while it is generally not accessible with string methods . therefore , it appears important to explore insofar as possible the actual predictions of string theory while keeping within the reach of conventional world sheet theory . a scenario without three form fluxes was already proposed in type i theory @xcite . it rests on the magnetic flux of the d9-brane gauge field in orientifold models @xcite , and the basic idea of this scenario is simple . once the distribution of magnetic fluxes on d9-branes is fixed by supersymmetry conditions , or minimum energy conditions , volume and shape of the compact space can be fixed as a result of quantization conditions of magnetic fluxes . since magnetic fluxes on d - branes can be analyzed within conventional world sheet theory , this type of scenario grants calculability . although it seems difficult to stabilize all moduli by this simple mechanism , it is therefore worth exploring this idea further . we investigate the volume stabilization of orbifold compact spaces with magnetic fluxes on d - branes in more general situations without supersymmetry . the d - branes are not necessarily space time filling , while supersymmetry may be broken , for instance , by `` brane supersymmetry breaking '' @xcite while retaining , for the bulk , supersymmetric compactifications . in addition to the above idea related to magnetic fluxes , we examine volume stabilization ( khler moduli stabilization ) resulting from the balance of attractive and repulsive forces between d - branes in the compact space . we propose a mechanism that can in principle stabilize the volumes of some orbifold spaces with fixed shapes , and thus lacking complex structure moduli , as for example @xmath0 , @xmath1 , @xmath2 ( see fig.[fig : balance ] for a schematic picture ) . orbifold and the three fixed points ( left ) . balance of the forces between three objects located at three different fixed points ( right ) . as a result , the area of the fundamental region is fixed , or stabilized . , width=529 ] in these contexts , mutual attractive forces could result from the simultaneous presence of d - branes and anti - d - branes , while in principle the non - bps - branes of @xcite and the fractional non - bps states of @xcite could provide additional repulsive contributions . a concrete model whose compact space is stabilized by this scenario , even if incomplete , could allow a related discussion of early cosmology , and thus of cosmic inflation , within string theory . in particular , the fields describing brane displacements away from their balanced locations could play the role of inflatons , while the vacuum energy of balanced configurations could be the origin of dark energy . models of this type also possess the attractive feature of linking spontaneous gauge symmetry breaking in string theory to geometrical d - brane displacements , along the lines of @xcite . one might also conceive of turning the constraint that the compact internal space be stabilized , which is unavoidable in a strict sense , into a constructive principle to build realistic models of elementary particles and their interactions . in this paper we concentrate on a compact six dimensional orbifold of @xmath2 , which possesses three volume moduli ( untwisted khler moduli , corresponding to the areas of the three two tori of @xmath3 ) , 81 blow up moduli ( 3 twisted khler moduli for each of the 27 fixed points ) and no complex structure moduli . we investigate the behavior of d@xmath4- , d@xmath5- and d@xmath6-branes in this compact space , where the index identifies the d5-branes whose world volumes _ include _ @xmath7-th torus , and the d7-branes whose world volume _ do not include _ it . since the numbers of dirichlet neumann directions of the open string stretched between d@xmath8- and d@xmath9-branes are not multiples of 4 , the system breaks supersymmetry and has a tachyonic ground state . however , as we shall see the inclusion of appropriate magnetic fluxes on d@xmath6-branes can make the system supersymmetric ( namely the lowest energy state ) while removing the tachyon instability ( see also the related work of @xcite ) . moreover , this configuration fixes the total volume of the first and second tori in the @xmath2 orbifold , the inclusion of a d3-brane pose an additional constraint on the magnetic flux on the d@xmath6-brane , so that finally the radii of the first and second tori are both fixed . the d3-brane should be located far enough from the d@xmath5-brane , in order to exclude tachyonic ground states for d@xmath4-d@xmath5 open strings . the magnetized d@xmath5-d@xmath6 system can be placed at a @xmath2 orbifold fixed point , while satisfying twisted ramond ramond tadpole cancelation conditions for the consistency at the quantum level . the simplest system of this type with overall ramond ramond tadpole cancelation involves a magnetized d@xmath5-d@xmath6 and its anti system placed at two different orbifold singularities . if these are separated in the third torus , its radius is driven to shrink by mutual attractive forces , so that other objects are needed to stabilize the internal volume . in this paper we propose a non trivial treatment of the tadpole problem , ubiquitous for d - branes in compact spaces , which can lead to this physically reasonable result . the paper is organized as follows . in section [ one - loop ] we provide a brief review of one loop vacuum amplitudes for open strings ending on magnetized d - branes . a number of basic facts that are scattered in many articles are collected for later convenience . in section [ stabilization-1 ] the system of magnetized d@xmath5- and d@xmath6-branes is investigated in detail . the vacuum energy determined by tree level closed string exchange is calculated both in the low energy effective theory and in the string world sheet theory . we shall see in detail that the vacuum energy vanishes for a certain configuration of magnetic fluxes , when the system possesses supersymmetry . the resulting configuration of magnetic fluxes stabilizes the overall volume of the first and second tori , while the inclusion of a d3-brane completes the stabilization of their radii . in section [ singularity ] we put the system of magnetized d@xmath5- and d@xmath6-branes at a singularity of the @xmath2 orbifold . the twisted ramond ramond tadpole cancelation is non trivial due to the magnetic fluxes , and supersymmetry is broken . in section [ stabilization-2 ] we discuss the force , or potential energy , between the d@xmath5-d@xmath6 system and its anti system lying at a different singularity separated in third torus , and we also propose a non trivial treatment on the sum of the open string winding modes in the third torus . in section [ conclusion ] we provide a summary of this work and briefly address some future problems . many techniques in this paper are familiar to string theorists , but we take the freedom to show them in detail for the benefit of others who might develop further these ideas in more realistic settings for the purpose of model building . let us first investigate in detail the one loop vacuum amplitude of open strings between d@xmath5- and d@xmath6-branes without magnetic flux in a @xmath2 orbifold compactification . the extension of the arguments to more general configurations should be straightforward . among the coordinates of ten dimensional space time @xmath10 with @xmath11 those of the non compact four dimensional space time bear labels @xmath12 , while those of the compact directions correspond to @xmath13 . the compact @xmath14 factorizes as @xmath15 , and the three pairs @xmath16 , @xmath17 and @xmath18 correspond to the first , second and third tori with radii @xmath19 , @xmath20 and @xmath21 , respectively . we use the su@xmath22 lattice of fig.[fig : balance ] for all the tori with metric @xmath23 the twist vectors of the two @xmath24 transformations , @xmath25 specify the angles of discrete rotations in each torus . note that the simultaneous action of these two @xmath24 corresponding to the twist vector @xmath26 results in a @xmath0 orbifold that has 9 volume moduli ( untwisted khler moduli ) , while the present @xmath2 orbifold has only 3 volume moduli corresponding to the radii of three tori . on the other hand , there are 81 twisted khler moduli in @xmath2 , while there are only 27 twisted khler moduli in the @xmath0 orbifold . there are _ no _ complex structure moduli in both orbifolds . in this paper we do not consider the stabilization of twisted khler moduli against the blow up of orbifold singularities . the world sheet fields of the open string , @xmath10 and @xmath27 , satisfy neumann neumann boundary condition for @xmath28 , dirichlet neumann boundary condition for @xmath17 , and dirichlet dirichlet boundary condition for @xmath18 . the open string has kaluza klein modes in the first torus , @xmath16 , and winding modes in the third torus , @xmath18 . although the directions corresponding to @xmath17 are compact , dirichlet neumann boundary conditions do not allow kaluza klein or winding modes . the one loop vacuum amplitude of open strings between d@xmath5- and d@xmath6-branes separated by a distance @xmath29 with @xmath30 in the third torus is @xmath31 with @xmath32(0,\tau ) } \over { \eta(\tau ) } } \right ) \times { 1 \over { ( \eta(\tau))^2 } } \left ( { { \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 \end{array } \right](0,\tau ) } \over { \eta(\tau ) } } \right ) \nonumber\\ & \times & \left ( { { \theta \left [ \begin{array}{c } \alpha/2 - 1/2 \\ \beta/2 \end{array } \right](0,\tau ) } \over { \theta \left [ \begin{array}{c } 0 \\ 1/2 \end{array } \right](0,\tau ) } } \right ) \times { 1 \over { ( \eta(\tau))^2 } } \left ( { { \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 \end{array } \right](0,\tau ) } \over { \eta(\tau ) } } \right ) \ , \label{d5-d7-no - flux}\end{aligned}\ ] ] where @xmath33 and @xmath34 . the sign @xmath35 enforces fermi statistics in the open string ramond sector , while the sign @xmath36 determines the ramond ramond charge of the closed string , and reflects a non trivial gliozzi olive parity of some neveu schwarz ( ns ) sectors the open string . the remaining three factors in the first line of eq . ( [ d5-d7-no - flux ] ) result from the integration of continuous momenta in the non compact directions , from the summation of kaluza klein momenta in the @xmath37 directions , and from the summation of open string winding contributions in the @xmath38 directions . the four factors in the second and third lines of eq . ( [ d5-d7-no - flux ] ) represent the contribution of string vibration modes under neumann neumann boundary conditions in non compact directions ( @xmath39 ) ( the contribution from , say @xmath40 , are canceled by ghost contributions ) , neumann neumann boundary conditions in the first torus ( @xmath16 ) , dirichlet neumann boundary condition in the second torus ( @xmath17 ) , and dirichlet dirichlet boundary conditions in the third torus ( @xmath18 ) , respectively . the contribution of vibration modes with neumann neumann boundary condition is the same as with dirichlet dirichlet boundary condition , and it is different from that with dirichlet neumann boundary condition . the first systematic understanding of the one loop open string vacuum amplitudes for each combinations of dirichlet and neumann boundary conditions was given in @xcite . having defined the building blocks of the open string one loop amplitude , it is simple to understand the one loop vacuum amplitudes of the open strings between d@xmath6- and d@xmath6-branes and d@xmath5- and d@xmath5-branes . @xmath41(0,\tau ) } \over { \eta(\tau ) } } \right)^4 \ , \label{d7-d7-no - flux}\end{aligned}\ ] ] @xmath42(0,\tau ) } \over { \eta(\tau ) } } \right)^4 \ . \label{d5-d5-no - flux}\end{aligned}\ ] ] now we introduce the magnetic flux on d@xmath6-brane in @xmath37 directions ( first torus ) , with @xmath43 , and in @xmath44 directions ( second torus ) , with @xmath45 . these magnetic fluxes in compact spaces are quantized as @xmath46 where @xmath47 and @xmath48 are areas of the first and second tori . for the quantization it is convenient to define new world sheet fields using zweibeins so that they form orthonormal bases in the compact space , letting @xmath49 where @xmath50 and concretely @xmath51 with @xmath52 , @xmath53 . the boundary conditions for the @xmath37 directions on the d@xmath6-brane are @xmath54 with derivatives with respect to world sheet coordinates @xmath55 , where @xmath56 and @xmath57 here @xmath58 and @xmath59 are world sheet coordinates , @xmath60 , and for @xmath61 eqs . ( [ bc - with - flux - x ] ) reduce to neumann neumann boundary conditions . the same happens to world sheet fermion fields . @xmath62 with @xmath63 , and @xmath64 for the d@xmath6-d@xmath6 open string the virasoro generator @xmath65 , which determines the spectrum of open string vibrations , is not affected by the magnetic flux , but only the quantization condition of momenta in the first torus changes to @xmath66 because @xmath67 so that @xmath68 this is an interesting result given in @xcite in the case of a constant background @xmath69-field . then we have @xmath70(0,\tau ) } \over { \eta(\tau ) } } \right)^4 \ , \label{d7-d7}\end{aligned}\ ] ] where @xmath71 , and @xmath72(0,\tau ) } \over { \eta(\tau ) } } \right)^4 \ . \label{d5-d5}\end{aligned}\ ] ] for the d@xmath5-d@xmath6 open string vacuum amplitude the quantization of the string vibration modes is modified by magnetic fluxes , because the ends of the open string feel different magnetic fields . the calculation is straightforward and the result is @xmath73(0,\tau ) } \over { \eta(\tau ) } } \right ) \times \left ( { { e^{-i \pi \beta ( { \alpha \over 2 } - \alpha_1 ) } \theta \left [ \begin{array}{c } \alpha/2-\alpha_1 \\ \beta/2 \end{array } \right](0,\tau ) } \over { { e^{-i \pi ( { 1 \over 2 } - \alpha_1 ) } \theta \left [ \begin{array}{c } 1/2-\alpha_1 \\ 1/2 \end{array } \right](0,\tau ) } } } \right ) \nonumber\\ & \times & \left ( { { e^{-i \pi \beta ( { \alpha \over 2 } - \alpha_2 ) } \theta \left [ \begin{array}{c } \alpha/2-\alpha_2 \\ \beta/2 \end{array } \right](0,\tau ) } \over { { e^{-i \pi ( { 1 \over 2 } -\alpha_2 ) } \theta \left [ \begin{array}{c } 1/2-\alpha_2 \\ 1/2 \end{array } \right](0,\tau ) } } } \right ) \times { 1 \over { ( \eta(\tau))^2 } } \left ( { { \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 \end{array } \right](0,\tau ) } \over { \eta(\tau ) } } \right ) \ , \label{d5-d7}\end{aligned}\ ] ] where @xmath74 and @xmath75 with @xmath76 . note that there are no kaluza klein modes in the first torus because of the magnetic flux on the d@xmath6-brane . there are many phase factors in the contributions from the @xmath77 directions because of the shift of zero modes by magnetic fluxes on the d@xmath6-brane . therefore , the limit of no magnetic fluxes , @xmath78 and @xmath79 , does not coincide with the amplitude of no magnetic fluxes . we can now investigate the tree level closed string exchange amplitude between d@xmath5- and d@xmath6-branes , which is the potential energy between these d - branes , or a contribution to the vacuum energy . it is instructive to investigate it first in the low energy effective field theory that only includes the massless states of closed string : dilaton , graviton , b - field and ramond ramond fields . for simplicity , we leave aside the effects of compactification . the couplings of these fields to a d@xmath80-brane are described by the effective action in einstein frame @xmath81 where the integrations are over the d@xmath80-brane world volume , @xmath82 is the d@xmath80-brane tension , @xmath83 and @xmath84 are pull back tensors of space time metric and b field on d@xmath80-brane , @xmath85 is the field strength of the u@xmath86 gauge field on the d@xmath80-brane . the differential forms , @xmath87 , @xmath88 and @xmath89 ( @xmath90 ) , in the second term refer to the gauge field , to the b - field and to the ramond ramond fields , respectively . the linear term in each field describes its tadpole coupling to d@xmath80-brane . we obtain the propagator of each field in ten dimensional space time from the type iib supergravity action in einstein frame . the introduction of the magnetic flux , a constant @xmath85 , changes tadpole couplings . for example , the b - field acquires a tadpole coupling with a magnetic flux . for the d@xmath5-brane there are tadpole couplings of graviton , dilaton and ramond ramond fields @xmath91 . for the magnetized d@xmath6-brane in the previous section there are tadpole couplings of graviton , dilaton , b - field @xmath92 and @xmath93 and ramond ramond fields @xmath94 , @xmath91 and @xmath95 . the amplitude resulting from exchanges of graviton , dilaton and @xmath91 is in momentum space reads @xmath96 \ , \ ] ] where @xmath97 and @xmath98 are magnetic fluxes defined as in the previous section , @xmath99 is the gravitational constant in ten dimensional space time , and @xmath100 is the momentum vector in 8,9 directions of the space . the first term in the square brackets is the contribution of the ns ns sector , while the second term is the contribution of the ramond ramond sector . the former gives an attractive force while the latter gives a repulsive force between d@xmath5- and d@xmath6-branes that can be separated in the 8,9 directions . notice that there are solutions of magnetic fluxes that guarantee the balance of the forces , or a vanishing amplitude : @xmath101 assuming @xmath102 . since the quantity inside square brackets is positive semi definite , the solutions correspond to continuously degenerate vacua . this simple analysis in the low energy effective theory is useful for estimating the forces between d - branes with various constant background b - fields and magnetic fluxes . after obtaining this result in the low energy effective theory , let us return to the amplitude of eq . ( [ d5-d7 ] ) . this amplitude vanishes indeed on account of the identity of eq . ( [ jacobi ] ) , if @xmath103 , or @xmath104 . @xmath105(0,\tau ) \right)^2 \nonumber\\ & & \times e^{-i \pi \beta ( { \alpha \over 2 } - \alpha_1 ) } \theta \left [ \begin{array}{c } \alpha/2-\alpha_1 \\ \beta/2 \end{array } \right](0,\tau ) \ , e^{-i \pi \beta ( { \alpha \over 2 } - \alpha_2 ) } \theta \left [ \begin{array}{c } \alpha/2-\alpha_2 \\ \beta/2 \end{array } \right](0,\tau ) \nonumber\\ & = & q^{\alpha_1 ^ 2/2 } q^{\alpha_2 ^ 2/2 } { 1 \over 2 } \sum_{\alpha=0,1 } \sum_{\beta = 0,1 } ( -1)^\alpha ( -1)^{(1-\alpha)\beta } \left ( \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 \end{array } \right](0,\tau ) \right)^2 \nonumber\\ & & \times \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 \end{array } \right](-\alpha_1 \tau,\tau ) \ , \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 \end{array } \right](-\alpha_2 \tau,\tau ) \nonumber\\ & = & q^{\alpha_1 ^ 2/2 } q^{\alpha_2 ^ 2/2 } \prod_{i=1}^4 \theta \left [ \begin{array}{c } 1/2 \\ 1/2 \end{array } \right](x_i,\tau ) = 0 \ , \end{aligned}\ ] ] with @xmath106 note that @xmath107 and @xmath108 for @xmath103 and @xmath109(0,\tau ) = 0\ .\ ] ] we conclude that a contribution to the vacuum energy by tree level exchange of closed strings @xmath110 has continuous supersymmetric local minima for magnetic fluxes @xmath103 . ( here , the factor @xmath111 is introduced to represent the existence of left and right modes of closed string . ) notice that the quantization conditions of magnetic fluxes , eqs . ( [ quantization_1 ] ) and ( [ quantization_2 ] ) , fix the total volume of the first and second tori , which is proportional to the product @xmath112 . this is the mechanism of volume stabilization by magnetic fluxes that we had anticipated . here , we do not require supersymmetry , but we require that the system should be in a lowest energy state . namely , we require that the contribution to the energy of this subsystem to that of the total system , which is not necessary supersymmetric , should be minimum . let us now introduce a d@xmath4-brane and let us consider the d@xmath4-d@xmath6 open string vacuum amplitude . it is exactly the same of eq . ( [ d5-d7 ] ) , except for a difference in the definition of @xmath113 : @xmath114 , which reflects the change of boundary condition from neumann neumann to dirichlet neumann . since the amplitude vanishes for @xmath103 , we have @xmath115 . note that @xmath116 is a solution in this case , because the combination of d@xmath4- and d@xmath6-branes is originally supersymmetric without magnetic flux . with the solution of eq . ( [ solution_d5d7 ] ) , the radii of the first and second tori are fixed according to @xmath117 the radius of the third torus should be stabilized by some brane dynamics , and we shall discuss some possibilities in the remainder of this paper . before closing this section let us mention the fate of the open string tachyon excitation between d@xmath5- and d@xmath6-branes with magnetic fluxes . without magnetic fluxes the tachyon state is a ramond vacuum state ( level 0 ) , because the vacuum constant in the corresponding virasoro operator @xmath65 is @xmath118 . @xmath119 where @xmath120 is the spin in the 6,7 directions . the fluxes modify the constant in @xmath65 as @xmath121 , and the states with lowest level in the ns sector are @xmath122 and/or @xmath123 . these are massless states with @xmath103 , and therefore there is no tachyon state . we need to fix the magnetized d@xmath5-d@xmath6 system at an orbifold singularity to stabilize the radius of the third torus , following the idea anticipated in section [ introduction ] . the system of d@xmath4-branes at orbifold singularities are extensively discussed in @xcite . the projection operator of the @xmath124 transformation should be inserted in the traces of one loop open string amplitudes . it is @xmath125 where operators @xmath126 and @xmath127 generate first and second @xmath24 transformations , respectively . since repeating twice a @xmath24 operation is equivalent to the opposite of a single operation , the corresponding amplitudes are related by hermitian conjugation . therefore , there are four independent twisted sectors corresponding to @xmath128 we specify each twisted sector by two numbers , ( @xmath129 , @xmath130 ) , which are powers of @xmath126 and @xmath127 resulting in eq . ( [ twisted - sectors ] ) , from left to right , in @xmath131 , @xmath132 , @xmath133 and @xmath134 sectors . to achieve ramond ramond tadpole cancelation in each twisted sector , we need to introduce multiple d@xmath5- and d@xmath6-branes with a non trivial action of @xmath124 on their chan paton indexes . the states of the open string between d@xmath5- and d@xmath6-branes should carry a chan paton matrix @xmath135 with @xmath136 and @xmath137 , where @xmath138 and @xmath139 are numbers of d@xmath5- and d@xmath6-branes on an orbifold singularity , respectively . the simple actions of the first and second @xmath24 on the chan paton matrix are @xmath140 respectively . the @xmath141 twisted sector one loop vacuum amplitudes of open strings between d@xmath5- and d@xmath6-branes are @xmath142(0,\tau ) } \over { \eta(\tau ) } } \right ) \left ( { { e^{-i 2 \pi ( { \alpha \over 2 } - \alpha_1 ) ( { \beta \over 2 } + v_1 ) } \theta \left [ \begin{array}{c } \alpha/2-\alpha_1 \\ \beta/2 + v_1 \end{array } \right](0,\tau ) } \over { { e^{-i 2 \pi ( { 1 \over 2 } - \alpha_1 ) ( { 1 \over 2 } + v_1 ) } \theta \left [ \begin{array}{c } 1/2-\alpha_1 \\ 1/2 + v_1 \end{array } \right](0,\tau ) } } } \right ) \nonumber\\ & \times & \left ( { { e^{-i 2 \pi ( { \alpha \over 2 } - \alpha_2 ) ( { \beta \over 2 } + v_2 ) } \theta \left [ \begin{array}{c } \alpha/2-\alpha_2 \\ \beta/2 + v_2 \end{array } \right](0,\tau ) } \over { { e^{-i 2 \pi ( { 1 \over 2 } - \alpha_2 ) ( { 1 \over 2 } + v_2 ) } \theta \left [ \begin{array}{c } 1/2-\alpha_2 \\ 1/2 + v_2 \end{array } \right](0,\tau ) } } } \right ) \nonumber\\ & \times & ( -2 \sin \pi v_3 ) { { \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 + v_3 \end{array } \right](0,\tau ) } \over { \theta \left [ \begin{array}{c } 1/2 \\ 1/2 + v_3 \end{array } \right](0,\tau ) } } \ , \label{d5-d7-at - singularity}\end{aligned}\ ] ] where @xmath143 with twist vectors defined in eq . ( [ twist - vectors ] ) . here , we set the distance between d@xmath5- and d@xmath6-branes to zero . this amplitude gives eq . ( [ d5-d7 ] ) which is the contribution of the exchange of untwisted closed string ( untwisted sector ) . the argument of the volume stabilization in previous section applies to this sector . the effect of twisted sectors ( the contributions of the exchanges of twisted closed strings ) to the energy for the volume stabilization will be discussed at the end of this section after solving twisted ramond ramond tadpole cancelation conditions . ] the corresponding amplitudes for open strings between d@xmath6- and d@xmath6-branes and d@xmath5- and d@xmath5-branes read @xmath144(0,\tau ) } \over { \eta(\tau ) } } \prod_{i=1}^3 \left [ ( -2 \sin \pi v_i ) { { \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 + v_i \end{array } \right](0,\tau ) } \over { \theta \left [ \begin{array}{c } 1/2 \\ 1/2 + v_i \end{array } \right](0,\tau ) } } \right ] \ , \label{d7-d7-at - singularity}\end{aligned}\ ] ] @xmath145(0,\tau ) } \over { \eta(\tau ) } } \prod_{i=1}^3 \left [ ( -2 \sin \pi v_i ) { { \theta \left [ \begin{array}{c } \alpha/2 \\ \beta/2 + v_i \end{array } \right](0,\tau ) } \over { \theta \left [ \begin{array}{c } 1/2 \\ 1/2 + v_i \end{array } \right](0,\tau ) } } \right ] \ . \label{d5-d5-at - singularity}\end{aligned}\ ] ] although eqs . ( [ d7-d7-at - singularity ] ) and ( [ d5-d5-at - singularity ] ) vanish by the @xmath146-function identity of appendix [ laplace ] with @xmath147 , eq . ( [ d5-d7-at - singularity ] ) does not vanish and supersymmetry is broken . it is well known that when we translate these amplitudes into corresponding tree level closed string exchanges , they acquire an additional factor if there are compact directions with neumann neumann boundary condition . suppose that there are two compact directions with neumann neumann boundary condition parameterized by a complex coordinate @xmath148 . in the absence of an orbifold twist the freely moving open string end returns to the same place after its one loop motion : @xmath149 on the other hand , with an orbifold twist the open string may come back to the point up to the identification of the orbifold transformation generated by @xmath150 with twisted vector @xmath151 : @xmath152 the physical meaning of this factor is to divide by the number of fixed points in the compact directions , and it is necessary to consistently obtain the same twisted ramond ramond tadpole cancelation conditions in non magnetized d@xmath4- and d@xmath9-branes at an orbifold singularity , for example , from d@xmath4-d@xmath4 and d@xmath4-d@xmath9 amplitudes and from d@xmath9-d@xmath9 and d@xmath4-d@xmath9 amplitudes . if @xmath153 , this factor should be replaced by unity , as should be clear from the preceding arguments . for the d@xmath5-d@xmath5 amplitude the two directions of the first torus bear neumann neumann boundary conditions , and the standard factor of @xmath154 is included . on the other hand , for the first torus in the d@xmath5-d@xmath6 amplitude we do not include the factor , because the magnetic flux on the d@xmath6-brane modifies open string boundary conditions . it is straightforward to obtain twisted tadpole cancelation conditions using d@xmath5-d@xmath5 and d@xmath5-d@xmath6 amplitudes . for the @xmath131 twisted sector , since the second torus is untwisted , the effect of the winding modes in the second torus can not be forbidden , and @xmath155 since the phase of the second term can not be produced by powers of @xmath156 , this condition actually requires @xmath157 for the @xmath132 twisted sector , since the first torus is untwisted , the effect of kaluza klein modes in the first torus can not be forbidden , @xmath158 this requires @xmath159 for the same reasons as above . for the @xmath133 twisted sector @xmath160 and for the @xmath134 twisted sector @xmath161 let us discuss two typical solutions of these twisted tadpole cancelation conditions and the corresponding massless spectra . the simplest solution is @xmath162 in the non compact four dimensional world volume of the d@xmath5-brane there are the @xmath163 gauge multiplet of u@xmath164u@xmath165u@xmath166 gauge symmetry and chiral multiplets in the representation @xmath167 , @xmath168 and @xmath169 , which are beautifully represented by a quiver diagram , where @xmath170 and @xmath171 denote fundamental and anti fundamental representations of the u@xmath172 gauge symmetry . the same structure appears for the d@xmath6-brane . no massless state arises from the d@xmath5-d@xmath6 open string . the second solutions is @xmath173 @xmath174 in the non compact four dimensional world volume of the d@xmath5-brane there is the @xmath163 gauge multiplet of u@xmath164u@xmath165u@xmath166 gauge symmetry , and there is no chiral multiplet . the same structure appears for the d@xmath6-brane with u@xmath175u@xmath176u@xmath177 gauge symmetry . the d@xmath5-d@xmath6 open string yields no massless fermion fields , but the following massless scalar fields appear : [ cols="^,^,^,^,^,^ " , ] note that originally the system of d@xmath5- and d@xmath6-branes has no supersymmetry without appropriate magnetic fluxes , and the amplitude of eq . ( [ d5-d7-at - singularity ] ) does not vanish . we can now discuss the possibility of d - branes moving away from the singularity while keeping @xmath124 invariance . it may be possible for triples of d@xmath5-branes and/or d@xmath6-branes with identification by @xmath124 , without violating twisted ramond ramond tadpole cancelations @xcite . in the non compact four dimensional portions of the world volumes of such d - branes there is an @xmath178 gauge multiplet . for the first solution , if the three scalar fields on the @xmath179 d@xmath5-branes have the same vacuum expectation values , the u@xmath164u@xmath165u@xmath166 gauge symmetry is broken to a single u@xmath180 with an adjoint chiral multiplet , which can form an @xmath178 gauge multiplet with a @xmath163 u@xmath180 gauge multiplet . this means that @xmath179 d@xmath5-branes may move away from a singularity in such a way that three @xmath181 d@xmath5-branes are identified by @xmath24 . the same is true for the d@xmath6-branes of the first solution . in the second solution there is no such scalar field that suggests d - brane movement away from the singularity . note that fixing d - branes at a singularity means stabilization of d - brane moduli , since the conditions of twisted tadpole cancelations forbid d - branes to move away from the place of the singularity . the flat directions associated to such motions are not present in the world volume field theory on the d - branes @xcite . this is also related with the stabilization of twisted khler moduli , since the vacuum expectation values of twisted khler moduli fields are related to fayet - iliopoulos terms on the d - branes . once the d - brane moduli are stabilized in the way as above , the emergences of fayet - iliopoulos terms simply increase the energy of the system and the twisted khler moduli should also be stabilized . this problem is also related to the understanding of spontaneous breaking of gauge symmetry losing its rank @xcite . before closing this section we discuss the effects of twisted sectors of eq . ( [ d5-d7-at - singularity ] ) to the stabilization of the radii of the first and second tori . in case of the solution of eq.([gamma - solution ] ) only the @xmath133 twisted sector among four twisted sectors contributes non - trivially . since the corresponding amplitude depends on the magnetic fluxes , it provides additional potential energy for volume stabilization in addition to that from untwisted sector . unfortunately , the amplitude diverges in the integration over the modulus @xmath182 due to the existence of tadpoles in ns - ns sector of twisted closed string . ( there is no tachyon state even though the system is not supersymmetric . ) therefore , without properly applying fischler - susskind mechanism @xcite or tadpole resummations @xcite we can not obtain a definite answer . the is a difficult generic problem in models without supersymmetry , and finding the solution is beyond the scope of this paper . in this paper we simply assume that the contribution does not disturb the volume stabilization . we investigate potential energies between d - branes separated in compact spaces . there is a general problem of tadpole divergence due to the contribution of kaluza klein modes and winding modes @xcite , and this is usually taken to indicate the need for some redefinition of the closed string background . in the following we interpret the problem differently and propose a procedure to obtain potential energies with the appropriate periodicities in the compact spaces . as an example , we consider the system of magnetized d@xmath5- and d@xmath6-branes at a singularity and its anti system at a different singularity of @xmath2 orbifold , and investigate the potential energy between these branes and anti branes . the d - brane configuration is shown in fig.[fig : config ] . - and d@xmath6-branes and its anti - system at @xmath2 orbifold singularities . blobs indicate singularities . d@xmath6-brane is at the first singularity in the third torus and includes nine singularities in its world volume . three d@xmath5-branes are at the first singularity in the third torus and at the first , second and third singularities in the second torus , respectively . each of the d@xmath5-brane includes three singularities in its world volume . at the second singularity in the third torus there is the same system with all anti - d - branes . , width=529 ] the system is globally consistent provided all ramond ramond tadpoles cancel , and we still have some empty singularities that could host , in principle , construct realistic models of elementary particles . the distance between the d - brane system at the first singularity and the anti - d - brane system at the second singularity is described by a vector in the third torus . @xmath183 with @xmath184 and @xmath185 . the potential energy density in four dimensional space time is a function of @xmath186 , or @xmath21 obtained as follows : @xmath187 \ .\ ] ] here , all one loop open string amplitudes are those for the untwisted sector , @xmath188 where the amplitudes on the right hand sides are given in eqs . ( [ d7-d7 ] ) , ( [ d5-d5 ] ) and ( [ d5-d7 ] ) with the introduction of the distance between two d@xmath6-branes and between two d@xmath5-branes . the potential is described as @xmath189 where the @xmath190 include theta functions and the sums of kaluza klein and winding modes in the first and second tori . one can take the @xmath191 limit ( or @xmath192 limit with @xmath193 ) to investigate the contributions of massless closed string modes . in this limit @xmath194 where @xmath195 is a positive constant ( provided @xmath19 and @xmath20 are fixed by magnetic fluxes ) and @xmath196 if we take further the limit using the poisson summation formula , @xmath197 and this is the `` d@xmath198-brane '' tadpole divergence that were encountered in @xcite . this is the general problem to investigate the potential energies or forces between d - branes that are separated in compact spaces . ignoring that 8,9 directions are a compact torus , we have @xmath199 where we use the argument of appendix [ laplace ] , namely , we define the divergent integral subtracting an infinite constant . note that a definition of the divergent integral is necessary to obtain this physically motivated result . therefore , it is reasonable to consider a procedure to define the divergent integral of eq . ( [ potential - massless ] ) . the summation over the open string winding modes in the third torus should translate into the summation of the tree level propagations of closed string winding in the third torus . therefore it is reasonable to consider eq . ( [ potential - massless ] ) as an infinite sum of logarithmic potentials that preserves torus periodicity . following the concrete procedure in appendix [ laplace ] , we obtain the potential energy by massless closed string exchange : @xmath200 ( { { \tilde b } / { 2 \pi r_3}},\tau ) } \over { \eta(\tau ) } } \right\vert^2 - { { 2 \pi \left ( { \rm i m } ( { { \tilde b } / { 2 \pi r_3 } } ) \right)^2 } \over { { \rm i m } \tau } } \right\ } \ , \ ] ] where @xmath201 since @xmath202 is a constant independent of @xmath21 , the potential energy is also a constant , and the exchanges of massless modes of the closed string do not affect the size of the third torus . the effect of the exchanges of massive modes of the closed string can be generally investigated replacing the constant @xmath195 by an appropriate factor depending on @xmath182 as @xmath203 with a positive integer @xmath204 . in the present d - brane system , the coefficient @xmath205 is positive , the integral is not divergent and @xmath206 where @xmath207 is a modified bessel function . since the contributions of larger winding modes are exponentially smaller , we may approximate this result as @xmath208 the factor @xmath209 is the mass , because @xmath210 this potential energy indicates that the third torus is forced to shrink by the attractive force between d - brane and anti d - brane systems , although this force decays exponentially with @xmath211 . for a real stabilization of the third torus , it is inevitable to resort to some other objects capable of producing repulsive forces , as for instance the non bps branes of @xcite and the fractional non bps branes of @xcite . we leave this analysis to future work . we have proposed a strategy that can contribute to volume stabilization of internal compact spaces , via a combination of magnetic fluxes on d - branes and some d - brane dynamics that all lie within the reach of conventional string world sheet theory , and we have explored its features to some extent . the relevant compact spaces are special ones that possess only a small number of khler moduli ( volume moduli ) but no complex structure moduli . however , they include some familiar examples , since for instance the @xmath2 orbifold is of this type . since we have not fully addressed the stabilization of twisted khler moduli , which correspond to the blow up of orbifold singularities , our scenario is not complete and deserves further investigations . nonetheless , we have shown that magnetic fluxes on d@xmath6-branes in the presence of non magnetized d@xmath5- and d@xmath4-branes can stabilize the volume moduli of the first and second tori of the @xmath2 orbifold , as a result of quantization conditions and minimum vacuum energy requirements . an additional result of this paper concerns systems of magnetized d@xmath5-d@xmath6 fractional branes , which are fixed at orbifold singularities of the @xmath2 orbifold , whose d@xmath4 counterparts were discussed in @xcite . we have also computed the potential energy responsible for the force between this d@xmath5-d@xmath6 system and its anti system lying at different orbifold singularities . in this computation we have proposed a physically motivated prescription to deal with the tadpole divergences that generally appear for d - branes in compact spaces , which reproduces the expected tendency of the third torus to shrink due to their mutual attractive forces . it is conceivable that repulsive forces introduced by other objects like the non bps branes of @xcite and the fractional non bps branes of @xcite could stabilize the last volume modulus of @xmath2 orbifold , but we leave a detailed analysis to future work . the problem of non - geometric dilaton stabilization has not been addressed , and we also have to leave this to future work . the author would like to thank augusto sagnotti for helpful discussions , suggestions and a careful reading of the manuscript . the author also would like to thank satoshi iso for encouraging discussions . this work was supported in part by grant - in - aid for scientific research ( # 26400253 ) from mext japan , by infn ( i.s . stefi ) , by the erc grant n. 226455 ( superfields ) and by scuola normale superiore . the author would like to thank scuola normale superiore for the kind hospitality extended to him while this work was in progress . the theta functions used in this paper are defined as @xmath212(z,\tau ) \equiv e^{2 \pi i a ( z + b ) } q^{{1 \over 2}a^2 } \prod_{n=1}^\infty ( 1 - q^n ) \prod_{m=1}^\infty ( 1 + q ^{m + a - { 1 \over 2 } } e^{2 \pi i ( z + b ) } ) ( 1 + q ^{m - a - { 1 \over 2 } } e^{- 2 \pi i ( z + b ) } ) \ , \ ] ] or @xmath212(z,\tau ) \equiv \sum_{n=-\infty}^\infty q^{{1 \over 2 } ( n + a)^2 } e^{2 \pi i ( n + a ) ( z + b ) } \ , \ ] ] where @xmath213 . this definition is the same as in @xcite . for two integers @xmath204 and @xmath214 , @xmath215(z,\tau ) = e^{2 \pi i a n } \theta \left [ \begin{array}{c } a \\ b \end{array } \right](z,\tau ) \ , \ ] ] and for @xmath216 @xmath217(0,\tau ) = \theta \left [ \begin{array}{c } a \\ b \end{array } \right](0,\tau ) \ .\ ] ] it is easy to show that @xmath218(z,\tau ) = \theta \left [ \begin{array}{c } a \\ b + z \end{array } \right](0,\tau ) \ , \ ] ] and @xmath219(z,\tau ) = e^{2 \pi i \omega ( z + b ) } q^{\omega^2/2 } \theta \left [ \begin{array}{c } a \\ b \end{array } \right](z + \omega \tau,\tau ) \ .\ ] ] there famous riemann identities read @xmath220(x_i,\tau ) \nonumber\\ & = & \prod_{i=1}^4 \left [ \begin{array}{c } 0 \\ 0 \end{array } \right](y_i,\tau ) - \prod_{i=1}^4 \left [ \begin{array}{c } 0 \\ 1/2 \end{array } \right](y_i,\tau ) - \prod_{i=1}^4 \left [ \begin{array}{c } 1/2 \\ 0 \end{array } \right](y_i,\tau ) + \prod_{i=1}^4 \left [ \begin{array}{c } 1/2 \\ 1/2 \end{array } \right](y_i,\tau ) \ , \label{jacobi}\end{aligned}\ ] ] where @xmath221 the modular transformation @xmath212(z,\tau ) = ( -i \tau)^{-1/2 } e^{2 \pi i a b - i \pi z^2/\tau } \theta \left [ \begin{array}{c } -b \\ a \end{array } \right](z/\tau,-1/\tau ) \ .\ ] ] obtains via a poisson resummation , and when combined with the dedekind @xmath222 function @xmath223 yields the useful formula @xmath224(0,\tau ) } \over { \eta(\tau ) } } = { { \theta \left [ \begin{array}{c } \beta/2 \\ \alpha/2 \end{array } \right](0,-1/\tau ) } \over { \eta(-1/\tau ) } } \ , \ ] ] where @xmath225 . in the calculation of the open string one loop vacuum energy between parallel d@xmath226 and d@xmath226-branes separated by a distance @xmath227 in flat ten dimensional space time , where @xmath228 is a two dimensional vector perpendicular to the world volume of d@xmath226-brane with components @xmath229 and @xmath230 , one encounters the divergent integral @xmath231 which is a green function of the laplace equation in two dimensional space , so that @xmath232 the integral is defined subtracting an infinite constant using the exponential integral function @xmath233 where @xmath234 . namely , one is resorting to the definition @xmath235 = - { 1 \over { 2\pi } } \ln \left ( { y \over \sqrt{2\pi\alpha ' } } \right ) \ .\ ] ] this is the standard result for the `` potential energy '' in two dimensional space . if one direction perpendicular to d@xmath226-brane world volume ( @xmath229 direction ) is compact and corresponds to a circle of radius @xmath236 , one encounters an infinite summation of logarithmic potentials that results from contributions of open string winding states , @xmath237 where @xmath238 with unit @xmath239 . this possesses formally circle periodicity , @xmath240 , but is divergent one needs to define properly the summation . using the product formula @xmath241 for complex @xmath148 , one can show that @xmath242 and resorting to the analytic continuation of the riemann zeta function @xmath243 finally leads to @xmath244 the first term is the expected green function of the laplace equation on a cylinder , while the second is an infinite constant that is independent of the radius @xmath236 . therefore we define @xmath245 thus recovering the known green function of the laplace equation on a cylinder with the corresponding periodicity @xmath240 . if the two dimensional space perpendicular to the d@xmath226-brane world volume is compactified on a torus with modulus @xmath246 and radius @xmath236 , one is led to a doubly infinite summation of logarithmic potentials , @xmath247 although this formula possesses formally the torus periodicities @xmath240 and @xmath248 , the sums are again divergent and one needs to define them appropriately . first one can apply the definition used in the case of the cylinder to the summation over @xmath204 , @xmath249 the analytic continuation of the hurwitz zeta function , which is commonly used to define the virasoro generator @xmath65 , @xmath250 leads to the formal relation @xmath251 with which @xmath252 can be expressed as a theta function making use of euler s formula for the @xmath253 function in eq . ( [ g_t_intermediate ] ) . the end result is @xmath254(z,\tau ) } \over { \eta(\tau ) } } \right\vert + { { ( { \rm i m } z)^2 } \over { 2 { \rm i m } \tau } } \ , \label{green - torus}\ ] ] where the second term recovers the torus periodicity that was lost in the manipulations of infinite products . this is the well known green function for the laplace equation on a torus , such that @xmath255 where the second term on the right hand side cancels the charge at @xmath256 in the compact torus space , and is the origin of the second term of eq . ( [ green - torus ] ) . m. b. green , j. h. schwarz and e. witten , `` superstring theory , '' 2 vols . , cambridge , uk : cambridge univ . press ( 1987 ) . m. cvetic , g. shiu and a. m. uranga , `` chiral four - dimensional n=1 supersymmetric type iia orientifolds from intersecting d6 branes , '' nucl . b * 615 * ( 2001 ) 3 [ hep - th/0107166 ] . g. aldazabal , l. e. ibanez , f. quevedo and a. m. uranga , `` d - branes at singularities : a bottom up approach to the string embedding of the standard model , '' jhep * 0008 * ( 2000 ) 002 [ hep - th/0005067 ] . s. kachru , r. kallosh , a. d. linde and s. p. trivedi , `` de sitter vacua in string theory , '' phys . d * 68 * ( 2003 ) 046005 [ hep - th/0301240 ] . s. kachru , r. kallosh , a. d. linde , j. m. maldacena , l. p. mcallister and s. p. trivedi , `` towards inflation in string theory , '' jcap * 0310 * ( 2003 ) 013 [ hep - th/0308055 ] . i. antoniadis and t. maillard , `` moduli stabilization from magnetic fluxes in type i string theory , '' nucl . b * 716 * ( 2005 ) 3 [ hep - th/0412008 ] . i. antoniadis , a. kumar and t. maillard , `` magnetic fluxes and moduli stabilization , '' nucl . b * 767 * ( 2007 ) 139 [ hep - th/0610246 ] . a. sagnotti , `` open strings and their symmetry groups , '' in cargese 87 , `` non - perturbative quantum field theory '' , eds . g. mack et al . ( pergamon press , 1988 ) , p.521 [ hep - th/0208020 ] . g. pradisi and a. sagnotti , `` open string orbifolds , '' phys . lett . b * 216 * ( 1989 ) 59 . p. horava , `` strings on world sheet orbifolds , '' nucl . b * 327 * ( 1989 ) 461 . p. horava , `` background duality of open string models , '' phys . b * 231 * ( 1989 ) 251 . m. bianchi and a. sagnotti , `` on the systematics of open string theories , '' phys . b * 247 * ( 1990 ) 517 . m. bianchi and a. sagnotti , `` twist symmetry and open string wilson lines , '' nucl . b * 361 * ( 1991 ) 519 . m. bianchi , g. pradisi and a. sagnotti , `` toroidal compactification and symmetry breaking in open string theories , '' nucl . b * 376 * ( 1992 ) 365 . a. sagnotti , `` a note on the green - schwarz mechanism in open string theories , '' phys . b * 294 * ( 1992 ) 196 [ hep - th/9210127 ] . s. sugimoto , `` anomaly cancellations in type i d-9 - anti - d-9 system and the usp(32 ) string theory , '' prog . * 102 * ( 1999 ) 685 [ hep - th/9905159 ] . i. antoniadis , e. dudas and a. sagnotti , `` brane supersymmetry breaking , '' phys . b * 464 * ( 1999 ) 38 [ hep - th/9908023 ] . c. angelantonj , `` comments on open string orbifolds with a nonvanishing b(ab ) , '' nucl . b * 566 * ( 2000 ) 126 [ hep - th/9908064 ] . g. aldazabal and a. m. uranga , `` tachyon free nonsupersymmetric type iib orientifolds via brane - anti - brane systems , '' jhep * 9910 * ( 1999 ) 024 [ hep - th/9908072 ] . c. angelantonj , i. antoniadis , g. dappollonio , e. dudas and a. sagnotti , `` type i vacua with brane supersymmetry breaking , '' nucl . b * 572 * ( 2000 ) 36 [ hep - th/9911081 ] . a. sen , `` stable nonbps states in string theory , '' jhep * 9806 * ( 1998 ) 007 [ hep - th/9803194 ] . a. sen , `` stable nonbps bound states of bps d - branes , '' jhep * 9808 * ( 1998 ) 010 [ hep - th/9805019 ] . m. r. gaberdiel and a. sen , `` nonsupersymmetric d - brane configurations with bose - fermi degenerate open string spectrum , '' jhep * 9911 * ( 1999 ) 008 [ hep - th/9908060 ] . e. dudas , j. mourad and a. sagnotti , `` charged and uncharged d - branes in various string theories , '' nucl . b * 620 * ( 2002 ) 109 [ hep - th/0107081 ] . n. kitazawa and s. kobayashi , `` spontaneous gauge symmetry breaking in a non - supersymmetric d - brane model , '' phys . b * 720 * ( 2013 ) 373 [ arxiv:1211.1777 [ hep - th ] ] . c. angelantonj , i. antoniadis , e. dudas and a. sagnotti , `` type i strings on magnetized orbifolds and brane transmutation , '' phys . b * 489 * ( 2000 ) 223 [ hep - th/0007090 ] . r. blumenhagen , m. cvetic , f. marchesano and g. shiu , `` chiral d - brane models with frozen open string moduli , '' jhep * 0503 * ( 2005 ) 050 [ hep - th/0502095 ] . w. fischler and l. susskind , phys . b * 171 * ( 1986 ) 383 . w. fischler and l. susskind , phys . b * 173 * ( 1986 ) 262 . e. dudas , g. pradisi , m. nicolosi and a. sagnotti , nucl . b * 708 * ( 2005 ) 3 [ hep - th/0410101 ] . n. kitazawa , phys . b * 660 * ( 2008 ) 415 [ arxiv:0801.1702 [ hep - th ] ] .

in this paper , we consider the numerical approximation of a general second order semilinear stochastic partial differential equation ( spde ) driven by multiplicative and additive noise . our main interest is on such spdes where the nonlinear part is stronger than the linear part also called stochastic reactive dominated transport equations . most numerical techniques , including current stochastic exponential integrators lose their good stability properties on such equations . using finite element for space discretization , we propose a new scheme appropriated on such equations , called stochastic exponential rosenbrock scheme ( sers ) based on local linearization at every time step of the semi - discrete equation obtained after space discretization . we consider noise that is in a trace class and give a strong convergence proof of the new scheme toward the exact solution in the root - mean - square norm . numerical experiments to sustain theoretical results are provided . example.eps gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore

1610.06790

the strong numerical approximation of an it stochastic partial differential equation defined in the bounded domain @xmath1 is analyzed . the domain @xmath2 is assumed to be a convex polygon , or has smooth boundary . boundary conditions on the domain @xmath2 are typically neumann , dirichlet or robin conditions . more precisely , we consider in the abstract setting the following stochastic partial differential equation @xmath3dt+b(x(t))dw(t ) , \quad x(0)=x_0 , \quad t\in[0,t],\end{aligned}\ ] ] on @xmath4 , @xmath5 is a final time , @xmath6 and @xmath7 are nonlinear functions , @xmath8 is the initial data which may be random , @xmath9 is a linear operator , unbounded , not necessarily self adjoint , and the generator of an analytic semigroup @xmath10 the noise @xmath11 is a @xmath12 wiener process defined in a filtered probability space @xmath13 . the filtration is assumed to fulfill the usual conditions ( see ( * ? ? ? * definition 2.1.11 ) ) . we assume that the noise can be represented as @xmath14,\end{aligned}\ ] ] where @xmath15 , @xmath16 , @xmath17 are respectively the eigenvalues and the eigenfunctions of the covariance operator @xmath18 , and @xmath19 are independent and identically distributed standard brownian motion . precise assumptions on @xmath6 , @xmath7 , @xmath8 and @xmath9 will be given in the next section to ensure the existence of the unique mild solution @xmath20 of ( [ model ] ) which has the following representation ( see @xcite ) @xmath21.\end{aligned}\ ] ] in few cases , exact solutions are explicitly available , so numerical techniques are the only tools to provide good approximations in more general cases ( see for examples @xcite ) . approximations are done at two levels , spatial approximation and temporal approximation . for the spatial approximation , the finite difference , the finite element method and the galerkin spectral method are usually used @xcite . as for pdes , standard explicit time stepping methods for spdes are usually unstable for stiff problems and therefore severe time step constraint is needed . to overcome that drawback , standard implicit euler methods are usually used @xcite . although standard implicit euler methods are stable , their implementation requires significantly more computational effort , specially full implicit methods , as newton method is usually used to solve nonlinear algebraic equations at each time step . recently , stochastic exponential integrators @xcite were appeared as non standard explicit methods efficient for spde ( [ model ] ) . all stochastic exponential integrators analyzed in the literature for spdes @xcite are bounded on the nonlinear problem as in ( [ model ] ) where the linear part @xmath9 and the nonlinear function @xmath6 are explicitly known a priori . such approach is justified in situations where the nonlinear function @xmath6 is small . indeed when @xmath6 is small the linear operator @xmath9 drives the spde ( [ model ] ) and the good stability of the exponential integrators and semi - implicit method are ensured . in fact , in more realistic applications the nonlinear function @xmath6 can be stronger . typical examples are semilinear advection diffusion reaction equations with stiff reaction term . in such cases , the spde ( [ model ] ) is driven by the nonlinear operator @xmath6 and both exponential integrators @xcite and semi - implicit euler @xcite will behave as explicit euler - maruyama scheme , therefore their good stability properties are lost . to overcome this issue we propose in this work a new scheme called stochastic exponential rosenbrock scheme ( sers ) . coupled with finite element for space discretization , the new scheme is based on a local linearization of the drift term at each time step in the corresponding semi - discrete problem of ( [ model ] ) . the local linearization therefore weaken the nonlinear part of the drift such that the linearized semi - discrete problem is driven by its linear part , which change at each time step . the standard stochastic exponential scheme @xcite is applied at the end to that linearized semi - discrete problem and the corresponding scheme is our new scheme . the challenge here is to deal with the new discrete semigroup which indeed is a semigroup process , called stochastic perturbed semigroup . the key ideal comes from the deterministic exponential rosenbrock method @xcite . note that similar schemes for stochastic differential equation in finite dimensional have been proposed in @xcite . using some deterministic tools from @xcite , we propose a strong convergence proof of the new schemes where the linear operator @xmath9 is not necessarily self adjoint . note that the orders of convergence are the same with stochastic exponential schemes proposed in@xcite . the deterministic part of this scheme is order @xmath22 in time and has been proven to be efficient and robust in comparison to standard schemes in many applications @xcite where the perturbed semigroup and related matrix functions have been computed using the krylov subspace technique @xcite and fast leja points technique @xcite . for new our stochastic scheme , numerical simulations show the good stability behavior of the new scheme compared with a stochastic exponential scheme proposed in @xcite , where the stochastic perturbed semigroup and related matrix functions are computed using krylov subspace technique . the rest of this paper is organized as follows . section [ wellposed ] is devoted to the mathematical setting , the numerical method and the main result . in section [ convergenceproof ] some preparatory results and the proof of the main result are provided . we end the paper in section [ experiment ] with some numerical experiments to sustain our theoretical results . before we state the well posedness result , let us define keys functional spaces , norms and notations that we will be use in the rest of the paper . let @xmath23 be a separable hilbert space . for all @xmath24 and for a banach space @xmath25 , we denote by @xmath26 the banach space of all equivalence classes of @xmath27 integrable @xmath25-valued random variables . we denote by @xmath28 the space of bounded linear mappings from @xmath25 to @xmath29 endowed with the usual operator norm @xmath30 . by @xmath31 , we denote the space of hilbert schmidt operators from @xmath25 to @xmath29 . for simplicity we use the notations @xmath32 and @xmath33 . we assume that the covariance operator @xmath34 is positive and self - adjoint . throughout this paper @xmath35 is a @xmath18-wiener process . the space of hilbert schmidt operator from @xmath36 to @xmath29 is denoted by @xmath37 with the corresponding norm @xmath38 defined by @xmath39 where @xmath40 is an orthonormal basis of @xmath29 . this definition is independent of the orthonormal basis of @xmath29 . + in order to ensure the existence and the uniqueness of solution of ( [ model ] ) we make the following assumptions . * [ linear operator @xmath9 ] * [ assumption1 ] @xmath42 is a negative generator of an analytic semigroup @xmath43 . * [ initial value @xmath8 ] * [ assumption2 ] we assume that @xmath45 , @xmath46 . as in the current literature on deterministic exponential rosenbrock - type methods @xcite , we make the following assumption on the nonlinear term . * [ nonlinear term @xmath6 ] * [ assumption3 ] we assume that the nonlinear mapping @xmath48 is lipschitz continuous and frchet derivable with its derivative uniformly bounded , i.e. there exists a constant @xmath49 such that @xmath50 as a consequence , there exists a constant @xmath51 such that @xmath52 for all @xmath53 . following ( * ? ? ? * chapter 7 ) or @xcite we make the following assumption on the diffusion term . * [ diffusion term ] * [ assumption4 ] we assume that the operator @xmath54 satisfies the global lipschitz condition , i.e. there exists a positive constant @xmath55 such that @xmath56 as a consequence , there exists a positive constant @xmath51 such that @xmath57 for all @xmath53 . to establish our @xmath0 strong convergence result , we will also need the following further assumption on the diffusion term when @xmath58 , which was also used in @xcite . [ assumption5 ] we assume that there exist two positive constants @xmath59 , and @xmath60 $ ] very small enough such that @xmath61 and @xmath62 for all @xmath63 , where @xmath64 is the parameter defined in assumption [ assumption2 ] . we equip @xmath65 , @xmath66 with the norm @xmath67 , for all @xmath68 . it is well known that @xmath69 is a banach space @xcite . let us recall in the following proposition some semigroup properties of the operator @xmath70 generated by @xmath9 that will be useful in the rest of the paper . * * [ smoothing properties of the semigroup]**@xcite [ prop1 ] let @xmath71 , @xmath72 and @xmath73 , then there exists a constant @xmath49 such that @xmath74 where @xmath75 , and @xmath76 . + if @xmath77 then @xmath78 . * * [ well posedness result ] * * ( * ? ? ? * theorem 7.4 ) + [ theorem1 ] let assumption [ assumption1 ] , assumption [ assumption3 ] and assumption [ assumption4 ] be satisfied . if @xmath8 is a @xmath79- measurable @xmath29 valued random variable , then there exists a unique mild solution @xmath20 of problem ( [ model ] ) taking the form ( [ mild1 ] ) and satisfying the following @xmath80=1,\end{aligned}\ ] ] and for any @xmath24 there exists a constant @xmath81 such that @xmath82}\mathbb{e}\vert x(t)\vert^p\leq c(1+\mathbb{e}\vert x_0\vert^p).\end{aligned}\ ] ] furthermore from ( * ? ? ? * theorem 1 ) or ( * ? ? ? * theorem 2.6 ) it holds that for all @xmath83 , for all @xmath24 there exists a positive constant @xmath55 such that for all @xmath84 $ ] we have @xmath85 in the rest of this paper , to simplify the presentation , we assume that the linear operator @xmath9 a second order . more precisely , we assume that our spde ( [ model ] ) is a second order semilinear parabolic and take the form @xmath86dt\nonumber\\ & + & b(x , x(t , x))dw(t , x ) , \quad x\in\lambda , \quad t\in[0,t],\end{aligned}\ ] ] where the functions @xmath87 and @xmath88 are continuously differentiable with globally bounded derivatives . in the abstract framework ( [ model ] ) , the linear operator @xmath9 takes the form @xmath89 where @xmath90 , @xmath91 . we assume that there is a positive constant @xmath92 such that @xmath93 the functions @xmath94 and @xmath95 are defined by @xmath96 for all @xmath97 , @xmath68 , @xmath98 , with @xmath4 . for an appropriate eigenfunctions @xmath99 such that @xmath100<\infty$ ] , it is well known ( * ? ? ? * section 4 ) that the nemystskii operator @xmath6 related to @xmath101 and the multiplication operator @xmath7 associated to @xmath102 defined in ( [ nemystskii ] ) satisfy assumption [ assumption3 ] and assumption [ assumption4 ] respectively . as in @xcite , we introduce two spaces @xmath103 and @xmath104 , such that @xmath105 ; the two spaces depend on the boundary conditions of @xmath106 and the domain of the operator @xmath9 . for dirichlet ( or first - type ) boundary conditions we take @xmath107 for robin ( third - type ) boundary condition and neumann ( second - type ) boundary condition , which is a special case of robin boundary condition , we take @xmath108 @xmath109 where @xmath110 is the normal derivative of @xmath111 and @xmath112 is the exterior pointing normal @xmath113 to the boundary of @xmath9 given by @xmath114 using the green s formula and the boundary conditions , the corresponding bilinear form associated to @xmath115 is given by @xmath116 for dirichlet and neumann boundary conditions , and @xmath117 for robin boundary conditions . using the garding s inequality ( @xcite ) we obtain @xmath118 by adding and substracting @xmath119 on the right hand side of ( [ model ] ) , we have a new linear operator that we atill call a corresponding to the new bilinear form that we still call @xmath120 such that the following coercivity property holds @xmath121 note that the expression of the nonlinear term @xmath6 has changed as we included the term @xmath122 in a new nonlinear term that we still denote by @xmath6 . the coercivity property ( [ ellip2 ] ) implies that @xmath9 is sectorial on @xmath123 i.e. there exists @xmath124 such that @xmath125 where @xmath126 ( see @xcite ) . then @xmath9 is the infinitesimal generator of a bounded analytic semigroup @xmath127 on @xmath123 such that @xmath128 where @xmath129 denotes a path that surrounds the spectrum of @xmath130 . the coercivity property ( [ ellip2 ] ) also implies that @xmath115 is a positive operator and its fractional powers are well defined for any @xmath131 by @xmath132 where @xmath133 is the gamma function ( see @xcite ) . let s now turn to the discretization of our problem ( [ model ] ) . we start by spltting the domain @xmath106 in finite triangles . let @xmath134 be the triangulation with maximal length @xmath135 satisfying the usual regularity assumptions , and @xmath136 the space of continuous functions that are piecewise linear over the triangulation @xmath134 . we consider the projection @xmath137 from @xmath4 to @xmath138 defined for every @xmath139 by @xmath140 the discrete operator @xmath141 is defined by @xmath142 like @xmath9 , @xmath143 is also a generator of a semigroup @xmath144 . as any semigroup and its generator , @xmath143 and @xmath145 satisfy the smoothing properties of proposition [ theorem1 ] but with a uniform constant @xmath55 , independent of @xmath135 . following @xcite , we characterize the domain of the operator @xmath146 as follow @xmath147 the semi - discrete version of problem ( [ model ] ) consists to find @xmath148 , @xmath149 $ ] such that @xmath150 and @xmath151dt+p_hb(x^h(t))dw(t ) , \quad t\in(0,t].\end{aligned}\ ] ] we note that @xmath143 and @xmath152 satisfy the same assumptions as @xmath9 and @xmath6 respectively . we also note that @xmath153 satisfies assumption [ assumption4 ] . therefore , theorem [ theorem1 ] ensures the existence of the unique mild solution @xmath154 of ( [ semi1 ] ) such that @xmath155.\end{aligned}\ ] ] this mild solution of ( [ semi1 ] ) is given by @xmath156 the following lemma will be useful in our convergence analysis . [ lemma1 ] the following inequality holds @xmath157 from the equivalence of norms ( see ( * ? ? ? * ( 3.12 ) ) ) we have @xmath158 note that @xmath159 where @xmath160 stands for the weak derivative . let @xmath161 be the set of functions @xmath162 with compact support in @xmath106 . let @xmath163 , for all @xmath164 , we have @xmath165 where @xmath166 is a duality pairing between @xmath167 and @xmath161 , and @xmath168 is the derivative of @xmath169 in the classical sense . from ( * remark 2.1 ) we have @xmath170 since @xmath171 is a linear operator . so , the equality ( [ weak2 ] ) yields @xmath172 since ( [ weak3 ] ) holds for all @xmath164 , it follows that @xmath173 in the weak sense . inserting this latter relation in ( [ weak1 ] ) , using the fact that the projection @xmath137 is bounded with respect to the norm @xmath174 and again the equivalence of norm ( * * ( 3.12 ) ) yields @xmath175 we therefore have @xmath176 note that ( [ inter1 ] ) remains true if we replace @xmath177 by @xmath178 . by interpolation theory we have @xmath179 let us recall the following well known lemma . * * [ it isometry ] * * ( * ? ? ? * proposition 2.3.5 ) + for any @xmath180 $ ] and for any @xmath181-valued predictable process @xmath182 , @xmath183 $ ] the following equality holds @xmath184= \mathbb{e}\left[\int_0^t\vert\phi(s ) \vert^2_{l^0_2}ds\right].\end{aligned}\ ] ] the following two lemmas provide space and time regularity results of the mild solution of the semi - discrete problem ( [ semi1 ] ) . these lemmas play an important role in our convergence analysis . more results on the regularity of the mild solution of problem ( [ model ] ) can be found in @xcite . * [ space regularity of the mild solution @xmath154 ] * + [ lemma3 ] let assumption [ assumption1 ] , assumption [ assumption2 ] , assumption [ assumption3 ] and assumption [ assumption4 ] be fulfilled with @xmath186 , then for all @xmath187 $ ] , @xmath188 . moreover , there exists a positive constant @xmath55 independent of @xmath135 such that @xmath189 the proof follows the sames lines as that of ( * ? ? ? * lemma 2.6 ) or ( * ? ? ? * theorem 1 ) or ( * ? ? * theorem 3.1 ) by making use of lemma [ lemma1 ] . * [ time regularity of the mild solution @xmath154 ] * [ lemma4 ] let @xmath190 be the mild solution of ( [ semi1 ] ) . if assumption [ assumption1 ] , assumption [ assumption2 ] , assumption [ assumption3 ] and assumption [ assumption4 ] are fulfilled with the corresponding @xmath191 . for @xmath192 there exists a positive constant @xmath55 independent of @xmath135 such that for @xmath193 $ ] , @xmath194 , we have @xmath195 moreover , if assumption [ assumption5 ] is fulfilled with @xmath196 , then there exists a positive constant @xmath55 such that @xmath197 the proof follows the same lines as that of ( * ? ? ? * lemma 2.7 ) or ( * ? ? ? * theorem 1 ) or ( * ? ? ? * theorem 4.1 ) by making use of lemma [ lemma1 ] . for the time discretization , we consider the one - step method which provides the numerical approximated solution @xmath198 of @xmath199 at discrete time @xmath200 , @xmath201 . the method is based on the continuous linearization of ( [ semi1 ] ) . more precisely we linearize ( [ semi1 ] ) at each time step as @xmath202dt+p_hb(x^h(t))dw(t ) , \end{aligned}\ ] ] for all @xmath203 , where @xmath204 is the frchet derivative of @xmath152 at @xmath198 and @xmath205 is the remainder at @xmath198 . both @xmath206 and @xmath205 are random functions and are defined for all @xmath207 by @xmath208 before build the new numerical scheme , let us recall the following important lemma . [ lemma5 ] for all @xmath209 and all @xmath210 , the random linear operator @xmath211 is the generator of a strongly continuous semigroup @xmath212 , uniformly bounded on @xmath213 $ ] called random ( or stochastic ) perturbed semigroup . furthermore the following estimate holds @xmath214 where @xmath55 is a positive constant independent of @xmath135 , @xmath215 , @xmath216 , @xmath217 and the sample @xmath218 . the proof follows the same lines as ( * ? ? ? * lemma 3.6 ) or ( * ? ? ? * lemma 4 ) . here our jacobian depends on @xmath207 , the constant @xmath55 is independent of the sample @xmath218 since there exists @xmath51 such that @xmath219 according to assumption [ assumption3 ] . given the solution @xmath199 and the numerical solution @xmath198 at @xmath220 , we obtain from ( [ semi ] ) the following mild representation form of @xmath221 @xmath222 we note that ( [ semi2 ] ) is the exact solution of ( [ semi1 ] ) at @xmath223 . to establish our numerical method we use the following approximations @xmath224 therefore the deterministic integral part of ( [ semi2 ] ) can be approximated as follows @xmath225 inserting ( [ constr1 ] ) in ( [ semi2 ] ) and using the approximation @xmath226 give the following approximation @xmath227 of @xmath221 , called stochastic exponential rosenbrock scheme ( sers ) @xmath228 with @xmath229 . the numerical scheme ( [ erem ] ) can be rewritten in the following equivalent form , which is efficient for implementation @xmath230,\end{aligned}\ ] ] where @xmath231 , @xmath232 are independent , standard , normally distributed random variables with mean @xmath178 and variance @xmath233 , and @xmath234 we note that the operator @xmath235 is uniformly bounded , i.e independently of @xmath135 , @xmath215 and @xmath218 ( see e.g ( * ? ? ? * lemma 2.4 ) ) . note that the corresponding standard stochastic exponential scheme presented in @xcite is given by @xmath236.\end{aligned}\ ] ] this scheme will be called setd1 and will be used in our numerical simulations for comparison with sers scheme . if the deterministic part is also approximated as the diffusion part ( [ napp ] ) , we will obtain the following new scheme @xmath237,\end{aligned}\ ] ] our main result is also valid for scheme ( [ schme0 ] ) and the extension of our proof to that scheme is done as in @xcite without any issue . having the numerical method in hand , our goal is to analyze its strong convergence toward the exact solution in the root - mean - square @xmath0 sense . in the following subsection we state our strong convergence result , which is in fact our main result . throughout this paper we take @xmath238 $ ] , where @xmath239 for @xmath240 , @xmath241 , @xmath242 is fixed , @xmath55 is a generic constant that may change from one place to another . the main result of this paper is formulated in the following theorem . [ mainresult1 ] let @xmath243 and @xmath198 be respectively the mild solution ( [ mild1 ] ) and the numerical approximation given by ( [ erem ] ) at @xmath200 . let assumption [ assumption1 ] , assumption [ assumption2 ] , assumption [ assumption3 ] and assumption [ assumption4 ] be fulfilled . for @xmath244 the following error estimate holds @xmath245 moreover , under a strong regularity of the initial data , that is assumption [ assumption2 ] and assumption [ assumption5 ] are fulfilled with @xmath246 , the following error estimate holds @xmath247 as in ( * ? ? ? * remark 2.9 ) , strong assumptions on the nonlinear functions @xmath6 and @xmath7 can allow to achieve a spatial error of order @xmath248 [ remark2 ] for additive noise , smooth noise with further assumptions on the nonlinear term @xmath6 should improve the time accuracy as in @xcite . [ remark3 ] note that the semi - discrete problem ( [ semi1 ] ) can be replaced by the following semi - discrete problem where the noise is truncated @xmath249dt+p_hb(x^h(t))p_hdw(t ) , t\in[0,t].\end{aligned}\ ] ] it was shown in @xcite that in the case of additive noise with smooth covariance operator kernel , this truncation can be done severely without loosing the spatial accuracy of the finite element method . applying our stochastic exponential rosenbrock scheme to ( [ semi3 ] ) yields @xmath250 we note that theorem [ mainresult1 ] also holds for the numerical scheme ( [ serem2 ] ) . parts of @xcite can be used . before prove our main result , some preparatory results are needed . [ lemma6 ] let @xmath251 be defined by ( [ remainder2 ] ) satisfies the global lipschitz condition with a uniform constant , i.e. there exists a positive constant @xmath49 , independent of @xmath135 , @xmath215 and @xmath218 such that @xmath252 using assumption [ assumption3 ] and relations ( [ remainder1])-([remainder2 ] ) , the proof is straightforward . we introduce the riesz representation operator @xmath253 defined by @xmath254 it is well known ( see @xcite ) that @xmath9 and @xmath143 are related by @xmath255 . under the regularity assumptions on the triangulation and in view of the @xmath104-ellipticity ( [ ellip2 ] ) , it is well known ( see @xcite ) that for all @xmath256 the following errors estimates hold @xmath257 let us consider the following deterministic linear problem : find @xmath258 such that @xmath259.\end{aligned}\ ] ] the corresponding semi - discrete problem in space consist to find @xmath260 such that @xmath261.\end{aligned}\ ] ] let s define the following operator @xmath262 so that @xmath263 . the estimate ( [ ritz1 ] ) was used in @xcite to prove the key part of the following lemma . [ lemma6a ] the following estimate holds @xmath264,\quad \alpha\leq r,\,\ , t\in ( 0,t].\end{aligned}\ ] ] the proof of lemma [ lemma6a ] for @xmath265 $ ] can be found in ( * ? ? ? * lemma 3.1 ) . using the stability property of @xmath70 and @xmath145 , and the fact that the projection @xmath137 is bounded , it follows that @xmath266 inequality ( [ addis2 ] ) shows that ( [ addis1 ] ) holds for @xmath267 . interpolating between @xmath267 and @xmath268 completes the proof of lemma [ lemma6a ] . [ lemma7 ] let assumption [ assumption1 ] , assumption [ assumption2 ] , assumption [ assumption3 ] and assumption [ assumption4 ] be fulfilled with @xmath269 . then the mild solutions @xmath270 and @xmath154 given respectively by ( [ mild1 ] ) and ( [ mild2 ] ) satisfy the following error estimate @xmath271.\end{aligned}\ ] ] further , if assumption [ assumption5 ] is fulfilled with @xmath272 . then the following error estimate holds @xmath271.\end{aligned}\ ] ] ( * ? ? ? * theorem 6.1 ) [ lemma8 ] under assumption [ assumption1 ] , for all @xmath210 the stochastic perturbed semigroup @xmath273 satisfies the following stability property * for @xmath274 we have @xmath275 . \\\end{aligned}\ ] ] * for @xmath276 , we have @xmath277,\,\ , \gamma_1\geq 0,\end{aligned}\ ] ] * for @xmath276 , such that @xmath278 and @xmath279 we have @xmath280 . \end{aligned}\ ] ] where @xmath55 is a positive constant independent of @xmath135 , @xmath215 , @xmath217 and the sample @xmath218 . we recall that the perturbed semigroup satisfies the following variation of parameters formula ( see ( * ? ? ? * chapter 3 , corollary 1.7 ) or ( * ? ? ? * section 3.1 , page 77 ) ) @xmath281 for all @xmath68 and all @xmath282 . then it follows that @xmath283 it is obvious that @xmath284 for all @xmath68 . then , replacing @xmath111 in ( [ vp1 ] ) by @xmath285 and pre - multiplying both right hand side of ( [ vp1 ] ) by @xmath286 yields @xmath287 taking the norm in both sides of ( [ vp2 ] ) and using proposition [ prop1 ] , the fact that @xmath288 and @xmath289 are uniformly bounded , it follows that @xmath290 using the definition of the norm @xmath291 gives the desired result for @xmath292 . to prove @xmath293 , we multiply ( [ vp1a ] ) by @xmath294 and obtain @xmath295 for all @xmath68 and all @xmath282 . taking the norm in both sides of ( [ vp3a ] ) and using the stability property of @xmath145 , @xmath273 with the uniformly boundedness of @xmath289 gives @xmath296 from ( [ vp3b ] ) it holds that @xmath297 applying the continuous gronwall s lemma to ( [ vp3c ] ) completes the proof of @xmath293 . to prove @xmath298 , we recall that the perturbed semigroup satisfies the following variation of parameter formula ( see ( * ? ? ? * page 161 ) ) @xmath299 replacing @xmath111 in ( [ vp3 ] ) by @xmath300 and pre - multiplying both left hand sides of ( [ vp3 ] ) by @xmath286 we obtain @xmath301 taking the norm in both sides of ( [ vp4 ] ) and using the stability property of the semigroup @xmath145 yields @xmath302 this ends the proof of @xmath298 . the following lemma is similar to ( * ? ? ? * lemma 4 ) , but its proof is easier than that of ( * ? ? ? * lemma 6 ) since we do nt use any further lemmas in its proof . [ lemma9 ] under assumption [ assumption1 ] and assumption [ assumption3 ] , the perturbed semigroup @xmath303 satisfies the following stability property @xmath304 where @xmath55 is a positive constant independent of @xmath215 , @xmath216 , @xmath135 , @xmath217 and the sample @xmath218 . as in @xcite we set @xmath305 using the telecospic sum , we can rewrite the perturbed semigroup @xmath306 as follow @xmath307 multiplying both sides of ( [ pertur1 ] ) by @xmath308 yields @xmath309 taking the norm in both sides of ( [ pertur2 ] ) and using the stability property of @xmath145 together with the inequality @xmath310 ( see ( * ? ? ? * ( 55 ) ) ) gives @xmath311 from the variation of parameter formula ( see ( * ? ? ? * corollary 1.7 , chapter 3 ) ) we have @xmath312 multiplying both side of ( [ pertur4 ] ) by @xmath308 gives @xmath313 taking the norm in both sides of ( [ pertur5 ] ) , using the stability property of @xmath314 , the unifomrly boundness of @xmath206 and lemma [ lemma8 ] ( ii ) with @xmath315 gives @xmath316 substituting ( [ pertur6 ] ) in ( [ pertur3 ] ) yields @xmath317 note that @xmath318 . applying the discrete gronwall s inequality to ( [ pertur7 ] ) completes the proof of lemma [ lemma9 ] . gathering our preparatory results , we are now ready to proof our main result in theorem [ mainresult1 ] . using the standard technique in the error analysis , we split the fully discrete error in two terms @xmath319 note that the space error @xmath320 is estimated by lemma [ lemma7 ] . it remains to estimate the time error @xmath321 . note also that in the case of additive noise the proof may be straightforward . we estimate the time error @xmath321 for both @xmath269 and @xmath272 separately in the following two subsections . we recall that the exact solution at @xmath220 of the semidiscrete problem is given by @xmath322 we also recall that the numerical solution at @xmath220 given by ( [ erem ] ) can be rewritten as @xmath323 if @xmath324 then it follows from ( [ num1 ] ) and ( [ num2 ] ) that @xmath325ds\right\vert_{l^2(\omega , h)}\nonumber\\ & + & \left\vert\int_0^{\delta t}\left[e^{(a_h+j^h_0)(\delta t - s)}p_hb(x^h(s))-e^{(a_h+j^h_0)\delta t}p_hb(x^h_0)\right]dw(s)\right\vert_{l^2(\omega , h)}\nonumber\\ & = : & i+ii.\end{aligned}\ ] ] using lemma [ lemma5 ] and lemma [ lemma6 ] it holds that @xmath326ds\right\vert_{l^2(\omega , h)}\nonumber\\ & \leq&\int_0^{\delta t}\left\vert e^{(a_h+j^h_0)(\delta t - s)}[g^h_0(x^h(s))-g^h_0(x^h_0)]\right\vert_{l^2(\omega , h)}ds\nonumber\\ & \leq&\int_0^{\delta t}\left(\mathbb{e}\left[\left\vert e^{(a_h+j^h_0)(\delta t - s)}\right\vert^2_{l(h)}\vert g^h_0(x^h(s))-g^h_0(x^h_0)\vert^2\right]\right)^{1/2}ds\nonumber\\ & \leq & c\int_0^{\delta t}\vert x^h(s)-x^h_0\vert_{l^2(\omega , h)}ds.\end{aligned}\ ] ] using inequality ( [ regular1 ] ) in ( [ addition2 ] ) and taking in account the fact that @xmath327 , we obtain @xmath328\right)^{1/2}ds\nonumber\\ & \leq & c\int_0^{\delta t}\left(\mathbb{e}(1+\vert x_0\vert^2)\right)^{1/2}ds\leq c\delta t.\end{aligned}\ ] ] using the it s isometry property , we have @xmath329dw(s)\right\vert_{l^2(\omega , h)}\nonumber\\ & = & \left(\mathbb{e}\left\vert \int_0^{\delta t}\left[e^{(a_h+j^h_0)(\delta t - s)}p_hb(x^h(s))-e^{(a_h+j^h_0)\delta t}p_hb(x^h_0)\right]dw(s)\right\vert^2\right)^{1/2}\nonumber\\ & = & \left(\int_0^{\delta t}\mathbb{e}\left\vert e^{(a_h+j^h_0)(\delta t - s)}p_hb(x^h(s))-e^{(a_h+j^h_0)\delta t}p_hb(x^h_0)\right\vert^2_{l^0_2}ds\right)^{1/2}.\end{aligned}\ ] ] using triangle inequality , lemma [ lemma5 ] , assumption [ assumption4 ] , ( [ regular1 ] ) and the fact that @xmath330 for all @xmath331 , @xmath332 for all positive real numbers @xmath333 and @xmath111 , it follows from ( [ addition4 ] ) that inserting ( [ addition5 ] ) and ( [ addition3 ] ) in ( [ addition1 ] ) yields @xmath334 for @xmath335 , we iterate the exact solution ( [ num1 ] ) at @xmath220 by substituting @xmath336 in ( [ num1 ] ) by their mild forms similarly , for @xmath335 , we iterate the numerical solution ( [ num2 ] ) at @xmath220 by substituting @xmath337 only in the first term of ( [ num2 ] ) by their expressions therefore , it follows from ( [ num1a ] ) and ( [ num2a ] ) and the triangle inequality that @xmath338 where and using holder s inequality , the stability property of @xmath339 , lemma [ lemma6 ] , the triangle inequality and the fact that @xmath330 yields using lemma [ lemma4 ] , it follows from ( [ num2aaa ] ) that @xmath340 the estimation of @xmath341 follows the same lines as the one of @xmath342 , let us sketch its estimation at the end . we use inequality @xmath330 to split @xmath104 into two terms and have @xmath343 using triangle inequality gives @xmath344\right)^{1/2}ds\right)^2.\nonumber\\\end{aligned}\ ] ] the smoothing properties of the semigroup combining with lemma [ lemma5 ] and lemma [ lemma6 ] yields @xmath345\right)^{1/2}ds\right)^2\nonumber\\ & \leq&cm\sum_{k=0}^{m-2}\left(\int_{t_{m - k-2}}^{t_{m - k-1}}\left(\mathbb{e}\vert x^h(s)-x^h(t_{m - k-2})\vert^2]\right)^{1/2}ds\right)^2\nonumber\\ & = & cm\sum_{k=0}^{m-2}\left(\int_{t_{m - k-2}}^{t_{m - k-1}}\vert x^h(s)-x^h(t_{m - k-2})\vert_{l^2(\omega , h)}ds\right)^2.\end{aligned}\ ] ] using holder s inequality together with lemma [ lemma4 ] , it follows from ( [ num2ab ] ) that @xmath346 again using triangle inequality , the smoothing properties of the semigroup , lemma [ lemma5 ] , lemma [ lemma6 ] and holder s inequality yields @xmath347 substituting ( [ use1a ] ) and ( [ use1 ] ) in ( [ num2aa ] ) yields @xmath348 to estimate @xmath341 , we use the triangle inequality to split in two parts @xmath349dw(s)\vert^2_{l^2(\omega , h)}\nonumber\\ & + & \vert\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}\nonumber\\ & & [ e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}p_hb(x^h(t_{m - k-2}))-e^{(a_h+j^h_{m - k-2})\delta t}p_hb(x^h_{m - k-2})]dw(s)\vert^2_{l^2(\omega , h)}\nonumber\\ & = : & vi_1+vi_2.\end{aligned}\ ] ] using the it isometry property , lemma [ lemma5 ] , assumption [ assumption4 ] and the fact that @xmath350 is uniformly bounded independently of @xmath135 , @xmath216 and the sample @xmath218 yields @xmath351dw(s)\vert^2\nonumber\\ & = & \sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\vert e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}\nonumber\\ & & p_h\left[b(x^h(s))-b(x^h(t_{m - k-2}))\right]\vert^2_{l^0_2}ds\nonumber\\ & \leq&\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\left[\left\vert e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}\right\vert^2_{l(h)}\right.\nonumber\\ & \times&\left.\left\vert p_h\left[b(x^h(s))-b(x^h(t_{m - k-2}))\right]\right\vert^2_{l^0_2}\right]ds\nonumber\\ & \leq&c\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\vert p_hb(x^h(s))-p_hb(x^h(x^h(t_{m - k-2}))\vert^2_{l^0_2}ds\nonumber\\ & \leq&c\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}\vert x^h(s)-x^h(t_{m - k-2})\vert^2_{l^2(\omega , h)}ds.\end{aligned}\ ] ] applying lemma [ lemma4 ] , it follows from ( [ usemidle ] ) that @xmath352 using the inequality @xmath353 , we split @xmath354 in two terms @xmath355dw(s)\vert^2_{l^2(\omega , h)}\nonumber\\ & \leq&2\vert \sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}\nonumber\\ & & \left[e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}-e^{(a_h+j^h_{m - k-2})\delta t}\right]p_hb(x^h(t_{m - k-2}))dw(s)\vert^2_{l^2(\omega , h)}\nonumber\\ & + & 2\vert\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t } e^{(a_h+j^h_{m - k-2})\delta t}\nonumber\\ & & \left[p_hb(x^h(t_{m - k-2}))-p_hb(x^h_{m - k-2})\right]dw(s)\vert^2_{l^2(\omega , h)}\nonumber\\ & = : & 2vi_{21}+2vi_{22}.\end{aligned}\ ] ] using it isometry property and using the triangle inequality yields @xmath356 p_hb(x^h(t_{m - k-2}))\right\vert^2_{l^0_2}ds.\end{aligned}\ ] ] as @xmath357 is a semigroup , we obviously have @xmath358 using relation ( [ smoothnew ] ) , lemma [ lemma8 ] ( i ) with @xmath359 and @xmath360 and lemma [ lemma9 ] with @xmath361 in ( [ use2b ] ) allows to have @xmath362ds\\ & \leq & c\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}t_{k+1}^{-\beta}(s - t_{m - k-2})^{\beta}\vert s_{m - k-2}(t_{m - k-1}-s)\vert^2_{l(h)}\\ & & \vert b(x^h(t_{m - k-2}))\vert^2_{l^0_2}ds\nonumber.\end{aligned}\ ] ] using assumption [ assumption4 ] and the fact that the random perturbed semigroup @xmath363 is uniformly bounded independently of @xmath216 , @xmath135 and the sample @xmath218 , it follows from ( [ case1a ] ) that @xmath364 let us estimate @xmath365 @xmath366dw(s)\right\vert^2_{l^2(\omega . h)}.\nonumber\\\end{aligned}\ ] ] indeed as in the estimate of @xmath367 , the following estimate holds for @xmath365 @xmath368 inserting ( [ case1c ] ) and ( [ case1b ] ) in ( [ esti4 ] ) gives @xmath369 substituting estimates of @xmath367 ( [ use2 ] ) and @xmath354 ( [ case1d ] ) in ( [ esti4 ] ) yields @xmath370 following the same lines as for the estimate of @xmath342 , we obtain @xmath371dw(s)\right\vert^2_{l^2(\omega , h)}\nonumber\\ & \leq & c\delta t^{\beta}+c\delta t\vert x^h(t_{m-1})-x^h_{m-1}\vert^2_{l^2(\omega , h)}.\end{aligned}\ ] ] gathering estimates of @xmath372 , @xmath341 , @xmath104 and @xmath342 in ( [ somme1 ] ) yields @xmath373 applying the discrete gronwall s lemma to ( [ above1 ] ) yields @xmath374 using lemma [ lemma7 ] together with inequality ( [ final ] ) completes the proof of theorem [ mainresult1 ] for @xmath269 . note that the estimates of @xmath372 and @xmath104 in subsection [ firstestimate ] hold for @xmath246 . we only need to re - estimate @xmath341 and @xmath342 . we will only estimate @xmath342 in detail since they are term similar . @xmath375 where @xmath367 and @xmath354 are defined by ( [ esti4 ] ) in subsection [ firstestimate ] . applying lemma [ lemma4 ] , it follows from ( [ usemidle ] ) that @xmath376 using the inequality @xmath353 , we split @xmath354 in two terms as @xmath377 where @xmath378 and @xmath365 are given by ( [ use2a ] ) in subsection [ firstestimate ] . we recall that from ( [ case1c ] ) the following estimate holds for @xmath365 @xmath379 using it isometry property and the triangle inequality , we split @xmath378 in two parts as @xmath380 p_hb(x^h(t_{m - k-2}))\right\vert^2_{l^0_2}ds.\nonumber\\ & \leq & 2\sum_{k=0}^{m-1}\left(\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\left\vert e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}\left[e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}\right.\right.\right.\nonumber\\ & & \left.\left.\left .- e^{(a_h+j^h_{m - k-2})\delta t}\right]\left[p_hb(x^h(t_{m - k-2}))-p_hb(x(t_{m - k-2}))\right]\right\vert^2_{l^0_2}ds\right)\nonumber\\ & + & 2\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1 } } \mathbb{e}\left\vert e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}\left[e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}\right.\right.\nonumber\\ & & \left.\left .- e^{(a_h+j^h_{m - k-2})\delta t}\right]p_hb(x(t_{m - k-2}))\right\vert^2_{l^0_2}ds\nonumber\\ & : = & 2vi_{211}+2vi_{212}.\end{aligned}\ ] ] using lemma [ lemma5 ] yields using assumption [ assumption4 ] and lemma [ lemma7 ] , we have @xmath381 using it isometry property and inserting an appropriated power of @xmath382 yields the following estimate @xmath383p_hb(x(t_{m - k-2}))\right\vert^2_{l^0_2}ds\nonumber\\ & \leq&\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\left[\vert e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}\vert^2_{l(h)}\vert\left(e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}\right.\right.\nonumber\\ & & \left.\left .- e^{(a_h+j^h_{m - k-2})\delta t}\right)(-a_h)^{-\frac{\gamma}{2}}\vert^2_{l(h)}\left(\vert(-a_h)^{\frac{\gamma}{2}}p_hb(x(t_{m - k-2}))\vert^2_{l^0_2}\right)\right]ds\nonumber\\ & \leq&c\sum_{k=0}^{m-2}\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\left[\vert e^{(a_h+j^h_{m-1})\delta t}\cdots e^{(a_h+j^h_{m - k-1})\delta t}(-a_h)^{\frac{1-\gamma}{2}}\vert^2_{l(h)}\right.\nonumber\\ & & \vert(-a_h)^{\frac{-1+\gamma}{2}}\left(e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}-e^{(a_h+j^h_{m - k-2})\delta t}\right)(-a_h)^{\frac{-\gamma}{2}}\vert^2_{l(h)}\nonumber\\ & & \times\left.\vert ( -a_h)^{\frac{\gamma}{2}}p_hb(x(t_{m - k-2}))\vert^2_{l^0_2}\right]ds.\end{aligned}\ ] ] using relation ( [ smoothnew ] ) , lemma [ lemma8 ] ( i ) with @xmath384 and @xmath385 in ( [ case2d ] ) gives @xmath386ds\nonumber\\ & \leq & c\sum_{k=0}^{m-2}t_{k+1}^{-1+\gamma}\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\left[\left\vert(-a_h)^{\frac{-1+\gamma}{2}}\left[e^{(a_h+j^h_{m - k-2})(t_{m - k-1}-s)}\right.\right.\right.\nonumber\\ & & \left.\left.\left .- e^{(a_h+j^h_{m - k-2})\delta t}\right](-a_h)^{-\frac{\gamma}{2}}\right\vert^2_{l(h ) } \vert(-a_h)^{\frac{\gamma}{2}}p_hb(x(t_{m - k-2}))\vert^2_{l^0_2}\right]ds.\nonumber\\\end{aligned}\ ] ] using lemma [ lemma8 ] ( iii ) with @xmath387 , and lemma [ lemma9 ] with @xmath388 yields @xmath389 ds\nonumber\\ & \leq & c\sum_{k=0}^{m-2}\left(t_{k+1}^{-1+\gamma}\int_{t_{m - k-2}}^{t_{m - k-1}}\mathbb{e}\left[\left\vert ( -a_h)^{\frac{-1+\gamma}{2 } } s^h_{m - k-2}(t_{m - k-1}-s)(-a_h)^{\frac{1-\gamma}{2}}\right\vert^2_{l(h)}\right.\right.\nonumber\\ & \times&\left\vert(-a_h)^{\frac{-1+\gamma}{2}}(s^h_{m - k-2}(s - t_{m - k-2})-\mathbf{i})(-a_h)^{-\frac{\gamma}{2}}\right\vert^2_{l(h)}\nonumber\\ & & \left.\left.\vert(-a_h)^{\frac{\gamma}{2}}p_hb(x(t_{m - k-2}))\vert^2_{l^0_2}\right ] ds\right)\nonumber\\ & \leq&c\sum_{k=0}^{m-2}t^{-1+\gamma}_{k+1}\int_{t_{m - k-2}}^{t_{m - k-1}}(s - t_{m - k-2})\mathbb{e}\vert(-a_h)^{\frac{\gamma}{2}}p_hb(x(t_{m - k-2}))\vert^2_{l^0_2 } ds\nonumber\\ & \leq&c\delta t^{2}\sum_{k=0}^{m-2}t_{k+1}^{-1+\gamma}\mathbb{e}\vert(-a_h)^{\frac{\gamma}{2}}p_hb(x(t_{m - k-2}))\vert^2_{l^0_2}. \end{aligned}\ ] ] using the definition of the @xmath181 norm , assumption [ assumption5 ] , lemma [ lemma1 ] , theorem [ theorem1 ] and inequality ( [ regular1a ] ) , we have @xmath390\nonumber\\ & \leq&c\mathbb{e}\left[\sum_{i=0}^{\infty}\vert ( -a)^{\frac{\gamma}{2}}b(x(t_{m - k-2}))q^{1/2}e_i\vert^2\right]\nonumber\\ & = & c\mathbb{e}\vert ( -a)^{\frac{\gamma}{2}}b(x(t_{m - k-2})\vert^2_{l^0_2}\nonumber\\ & \leq & c\mathbb{e}(1+\vert ( -a)^{\frac{\gamma}{2}}x(t_{m - k-2})\vert^2)\nonumber\\ & \leq & c(1+\mathbb{e}(\vert x_0\vert^2_{\gamma}))<\infty.\end{aligned}\ ] ] substituting ( [ case2f ] ) in ( [ case2ea ] ) yields @xmath391 inserting ( [ case2 g ] ) and ( [ case2c ] ) in ( [ case2b ] ) gives @xmath392 inserting ( [ case2h ] ) and ( [ case2aa ] ) in ( [ case2a ] ) gives @xmath393 substituting estimates of @xmath354 ( [ case2i ] ) and @xmath367 ( [ new2 ] ) in ( [ new1 ] ) yields @xmath394 following the same lines as for @xmath342 , we obtain @xmath395dw(s)\right\vert^2_{l^2(\omega , h)}\nonumber\\ & \leq & ch^{2\beta}+c\delta t+c\delta t\vert x^h(t_{m-1})-x^h_{m-1}\vert^2_{l^2(\omega , h)}.\end{aligned}\ ] ] therefore gathering the current estimates of @xmath341 and @xmath342 , and the estimates of @xmath372 and @xmath104 from subsection [ firstestimate ] in the inequality ( [ somme1 ] ) yields @xmath396 applying the discrete gronwall s lemma to ( [ case2ka ] ) yields @xmath397 here we provide two examples to sustain our theoretical results . the first example has exact solution . the reference solution or the exact solution using in the errors computation for our second example is taken to be the numerical solution with small time step . in the legends of our graphs , we use the following notations 1 . sers denotes the strong errors from our sers scheme . 2 . setd1 denotes the strong errors from the stochastic exponential scheme @xcite given by ( [ setd1 ] ) . the exponential matrix function @xmath398 is computed by krylov subspace technique with fixed dimension @xmath399 and tolerance @xmath400 @xcite . note that we compute at every time step the action on the exponential matrix function on a vector and not the whole exponential matrix function . our code was implemented in matlab 8.1 . we first consider the following stochastic reaction diffusion equation with stiff reaction driven by additive noise in two dimensions with neumann boundary conditions @xmath401dt+dw(t ) , \quad x(0)=x_0 , \quad t\in[0,t],\end{aligned}\ ] ] on the domain @xmath402\times[0,l_2]$ ] , @xmath403 a simple computation shows that the eigenfunctions @xmath404 with the corresponding eigenvalues @xmath405 of @xmath406 are given by @xmath407 where @xmath408 , @xmath409 and @xmath410 . in the abstract form ( [ model ] ) our linear operator @xmath9 is taken to be @xmath411 and @xmath412 which obviously satisfies assumption [ assumption3 ] . we take @xmath413 and the triangulation @xmath414 has be contructed from uniform cartesien grid of sizes @xmath415 . we assume that the covariance operator @xmath18 and @xmath9 have the same eigenfunctions . we take the following values for @xmath416 in the representation ( [ noise ] ) @xmath417 to have trace class noise , it is enough to consider @xmath418 . in this example @xmath419 , therefore @xmath7 defined in ( [ nemystskii ] ) obviously satisfies assumption [ assumption4 ] and assumption [ assumption5 ] . the exact solution of ( [ example1 ] ) is constructed in @xcite . and @xmath420 for sers and setd1 schemes respectively . the noise regularity parameter used is @xmath421 and @xmath422 samples are used in the errors computation.,scaledwidth=75.0% ] figure [ fig1 ] shows the strong convergence of sers and setd1 schemes . this figure also shows that setd1 is unstable for large time steps . we can observe the good stability property of the new sers scheme even for large time steps . we can also observe that the two schemes have the same order of accuracy . indeed although setd1 seems be more accurate , the two graphs become very close for small time step . the orders of convergence of the two methods are @xmath423 and @xmath420 for sers and setd1 schemes respectively , which are very close to theoretical results . note that we only use the stable part of the data in the computation of the order of convergence of setd1 scheme . as a more challenging example we consider the the stochastic advection - diffusion - reaction spde with multiplicative noise in two dimensions on the domain @xmath424\times[0,1]$ ] . @xmath425dt+xdw.\\ \mathbf{d}&=&\left ( \begin{array}{cc } 10^{-2}&0\\ 0 & 10^{-3 } \end{array}\right)\end{aligned}\ ] ] with mixed neumann - dirichlet boundary conditions . the dirichlet boundary condition is @xmath426 at @xmath427 and we use the homogeneous neumann boundary conditions elsewhere . the darcy velocity @xmath428 is obtained as in @xcite and to deal with high pclet flows we discretize in space using finite volume method ( viewed as the finite element method as in @xcite ) in rectangular grid of sizes @xmath429 . the reference solution or the exact solution using in the errors computation is the numerical solution with the time step @xmath430 . relatively small time steps are used to stabilize the scheme setd1 . the noise used is the same as in the first example with ( [ eigennoise ] ) . our linear operator @xmath9 is given by @xmath431 and the functions @xmath101 and @xmath102 are given by @xmath432 therefore , from ( * ? ? ? * section 4 ) it follows that the operators @xmath6 and @xmath7 defined by ( [ nemystskii ] ) fulfil obviously assumption [ assumption3 ] , assumption assumption [ assumption4 ] and assumption [ assumption5 ] . and @xmath433 for sers and setd1 schemes respectively . the noise regularity parameter used is @xmath434 and @xmath422 samples are used in the errors computation.,scaledwidth=75.0% ] figure [ fig2 ] shows the strong convergence of sers scheme and setd1 scheme presented in @xcite . we can also observe that the two schemes have the same order of accuracy . indeed although sers seems be more accurate , the difference between the two errors is small . the orders of convergence of the two methods are @xmath435 and @xmath433 for sers and setd1 schemes respectively , which are very close to theoretical results . this work was supported by the german academic exchange service ( daad ) ( daad - project 57142917 ) and the robert bosch stiftung through the aims arete chair programme ( grant no 11.5.8040.0033.0 ) . part of this work was done when antoine tambue visited tu chemnitz . the visit was supported by twas - dfg cooperation visits programme . tambue , a. , berre , i. , and nordbotten , j. , m. , efficient simulation of geothermal processes in heterogeneous porous media based on the exponential rosenbrock euler and rosenbrock - type methods . . * 53 * , 250262 ( 2013 )

in ultracold fermi gases , the effect of spin - imbalance on superfluidity has been the subject of intense study . one of the reasons for this is that spin - imbalance frustrates the bardeen - cooper - schrieffer ( bcs ) superfluid pairing mechanism , in which fermions in different spin states combine into cooper pairs with zero momentum . in 1964 , it was proposed that an exotic superfluid state called the fulde - ferrell - larkin - ovchinnikov ( fflo ) state , in which the cooper pairs have nonzero momentum , could exist in a spin - imbalanced fermi gas . at the saddle - point ( mean field ) level , it has been shown that the fflo state only occupies a very small sliver in the ground state phase diagram of a 3d fermi gas . however , a question that remains to be investigated is : what is the influence of phase fluctuations around the saddle point on the fflo state ? in this work we show that phase fluctuations only lead to relatively small quantitative corrections to the presence of the fflo state in the saddle - point phase diagram of a 3d spin - imbalanced fermi gas . starting from the partition function of the system , we calculate the effective action within the path - integral adiabatic approximation . the action is then expanded up to second order in the fluctuation field around the saddle point , leading to the fluctuation free energy . using this free energy , we calculate corrections due to phase fluctuations to the bcs - fflo transition in the saddle - point phase diagram . at temperatures at which the fflo state exists , we find only small corrections to the size of the fflo area . our results suggest that fluctuations of the phase of the fflo order parameter , which can be interpreted as an oscillation of its momentum vector , do not cause an instability of the fflo state with respect to the bcs state .

1310.3840

in the last two decades , ultracold quantum gases have been the subject of many theoretical and experimental investigations @xcite . among the many systems that have been studied , ultracold fermi gases have received wide attention @xcite . due to the experimental controllability achieved with ultracold gases , quantum many - body phenomena such as fermionic superfluidity can be studied in great detail in these systems . by controlling and tuning the interaction strength between fermions in different states using feshbach resonances @xcite , it has become possible to study the crossover from a bardeen - cooper - schrieffer ( bcs ) superfluid state of weakly interacting cooper pairs , to a bose - einstein condensate ( bec ) of strongly coupled molecules @xcite . aside from the interaction strength , another important parameter that can be tuned is the population imbalance between fermions in different states . this parameter is of importance because spin - imbalance frustrates the bcs superfluid pairing mechanism . in the bcs state , pairing between fermions occurs at the fermi surface . however , a population imbalance will create a gap between the fermi surfaces of the two spin states , making the bcs state energetically less favorable . theoretically it was predicted that at a certain critical spin - imbalance , known as the clogston - chandrasekhar limit @xcite , a first order phase transition from the bcs state to the normal state would occur . by preparing a fermi gas in one hyperfine state , and using a radio - frequency sweep to create a mixture of two hyperfine states ( labeled spin - up and spin - down ) , the transition from a superfluid to a normal gas , induced by spin - imbalance , was demonstrated experimentally @xcite . at this point , the question remains whether a non - uniform superfluid can exist in a spin - imbalanced 3d fermi gas . the most prominent example of non - uniform superfluidity is the fulde - ferrell - larkin - ovchinnikov ( fflo ) state , which was proposed independently by fulde and ferrell ( ff ) @xcite and by larkin and ovchinnikov ( lo ) @xcite in 1964 . the fflo state differs from the bcs state in that it has cooper pairs with non - zero momentum , which in position space results in an oscillating superfluid order parameter . it was suggested that this exotic superfluid state could exist at non - zero polarization . in part of the literature , a further distinction is made between the ff state and the lo state : the former has one momentum component , whereas the latter is the superposition of two momentum components of equal magnitude but opposite sign . in this paper , we will focus on the ff state but we will henceforth call this the fflo state , bearing in mind that we mean the superfluid state with one momentum component . following the success of creating a spin - imbalanced fermi gas , the theoretical investigation of the fflo superfluid state was intensified . the first studies focused on the three - dimensional ( 3d ) fermi gas , at the saddle - point ( mean - field ) level @xcite , and found that the fflo state is only present in a very small sliver of the ground - state phase diagram @xcite . the 1d case has also received wide theoretical attention , and has proven to be a promising setup for detecting the fflo state . in 1d , the presence of the fflo state in the ground - state phase diagram is much larger compared to the 3d case @xcite . following these theoretical predictions , the first indirect experimental evidence for the fflo state was found in a 1d fermi gas by the hulet group at rice university @xcite . inspired by this success , several new experimental detection techniques have been proposed @xcite , both for the 1d and for the 3d case . however , in the latter case , the fflo state still eludes experimental observation . to acquire a better understanding of the fflo state in a 3d fermi gas , it is necessary to go beyond the mean - field level , which , while resulting in quantitatively correct results in the limit of weak interaction ( bcs limit ) at temperature zero , offers a qualitative description at best for temperatures above zero or for stronger interactions . up till now , little attention has been devoted to this subject and to the effect of fluctuations on the fflo state in general . one important exception is the work by radzihovsky , @xcite in which a low - energy model for the fulde - ferrell state and for the larkin - ovchinnikov state is developed , with an in - depth focus on the nature of the emerging goldstone modes for the latter state . in this paper , we contribute to this subject by explicitly studying the effect of phase fluctuations on the presence of the fflo state in the phase diagram of a 3d fermi gas . our main motivation is the following : the fflo state is characterized by a momentum component @xmath0 , which means that the rotational symmetry of the system is spontaneously broken by this state . the momentum component @xmath0 results in an oscillating phase of the order parameter in position space . because of this , fluctuations of the phase of the order parameter are equivalent to fluctuations in the direction of the momentum @xmath0 . since a 3d fermi gas exhibits spherical symmetry , these fluctuations cost zero energy . hence , one would expect these fluctuations to proliferate and destabilize the fflo state . our main point of interest is to see whether the region of fflo in the phase diagram of a 3d fermi gas vanishes due to phase fluctuations , which would help to understand why this state has not been observed experimentally in 3d . to the best of our knowledge , this specific problem has not yet been studied in literature . the rest of this paper is organized as follows . in section [ hydrodynamic effective action ] we derive a hydrodynamic effective action , starting from the partition function of a 3d fermi gas with spin - imbalance , within the path - integral adiabatic approximation . in section [ fluctuation action ] we perform an expansion of the action up to second order in the fluctuation field , which leads to the fluctuation part of the action . from the fluctuation action , the fluctuation free energy is readily derived . subsequently , in section [ phase diagram ] , using this free energy , we calculate the phase diagram of the system and determine whether corrections emerge by taking into account phase fluctuations . finally in section [ conclusions ] we draw conclusions . in this section , the effective action describing the fflo state in a spin - imbalanced 3d fermi gas , with the inclusion of phase fluctuations around the saddle point , is calculated within a hydrodynamic approach . the starting point of this derivation is the partition function of the system , written as a path integral over fermionic grassmann fields @xmath1 and @xmath2 : @xmath3 where the action consists of a single - particle term and an interaction term:@xmath4 the action ( [ action ] ) is written in imaginary time @xmath5 , @xmath6 and @xmath7 represent 3d position vectors , @xmath8 is the inverse temperature and @xmath9 is the chemical potential of the fermions in the spin state @xmath10 . furthermore , we use @xmath11 as units . in the interaction term , @xmath12 represents a general interparticle potential . in this paper , only s - wave scattering will be considered at ultracold temperatures , therefore the interaction potential will be replaced by a pseudo - potential:@xmath13 where the interaction strength @xmath14 is related to the s - wave scattering length @xmath15 through @xcite:@xmath16 an alternative , elegant derivation of expression ( [ 1/g(as ) ( 2 ) ] ) can be found in @xcite . using the standard hubbard - stratonovich transformation , the fourth - degree interaction term in ( [ action ] ) is rewritten as a sum of two second - degree terms . the cost of this transformation is that an additional path integral , over auxiliary bosonic fields @xmath17 and @xmath18 , is introduced . these fields are physically relevant , as they are interpreted as the fields of the fermion pairs . following the hubbard - stratonovich transformation , the partition function becomes@xmath19 . \label{partition function}\ ] ] one way to introduce fluctuations into the partition function is to write the bosonic fields as the sum of a saddle - point contribution and a fluctuation contribution : @xmath20 and @xmath21 . in this way , amplitude- and phase fluctuations are fully intertwined , and both are automatically taken into account @xcite . an alternative way , which will be used in this paper since it allows ( at a later stage in this calculation ) to focus solely on phase fluctuations , is to write the bosonic fields in terms of an amplitude and a phase @xmath22{l}\delta_{\mathbf{x},\tau}=\left \vert \delta_{\mathbf{x},\tau}\right \vert e^{i\theta_{\mathbf{x},\tau}}\\ \bar{\delta}_{\mathbf{x},\tau}=\left \vert \delta_{\mathbf{x},\tau}\right \vert e^{-i\theta_{\mathbf{x},\tau}}\end{array } \right . , \label{amplitude and phase}\ ] ] where both @xmath23 and @xmath24 are real fields . upon substituting ( [ amplitude and phase ] ) into the partition function ( [ partition function ] ) , it is convenient to apply the following gauge transformation to the fermionic fields : @xmath25 . after working out the derivatives in ( [ partition function ] ) , the partition function becomes@xmath26 where the following nambu spinors were used:@xmath27{cc}\bar{\psi}_{\mathbf{x},\tau,\uparrow } & \psi_{\mathbf{x},\tau,\downarrow}\end{array } \right ) \text { and } \eta_{\mathbf{x},\tau}=\left ( \begin{array } [ c]{c}\psi_{\mathbf{x},\tau,\uparrow}\\ \bar{\psi}_{\mathbf{x},\tau,\downarrow}\end{array } \right ) , \label{nambu spinor}\ ] ] in order to write the inverse green s function in block - diagonal form , where the diagonal elements are given by a set of @xmath28 matrices given by@xmath29 ^{2}\right ) \sigma_{3}+\left \vert \delta_{\mathbf{x},\tau}\right \vert \sigma_{1}. \label{-g^-1 x , t}\ ] ] here @xmath30,@xmath31 and @xmath32 are pauli matrices , and furthermore the definitions of the total chemical potential @xmath33 and the imbalance chemical potential @xmath34 have been introduced . the path integral over the bose field @xmath35 in ( [ partition function ( 1 ) ] ) can not be calculated exactly and hence an approximation has to be made . the most basic approximation is to replace the field @xmath23 by a constant @xmath36 : this is the saddle - point approximation . since in this article we want to describe phase fluctuations , we improve on this approximation by also considering fluctuations of the field @xmath37 around its saddle - point value . the specific choice of the saddle point is such that the fflo state is included in the current formalism . the fflo state is defined by fermionic pairs that have a non - zero momentum @xmath0 . in position space , this is equivalent with an oscillating phase of the order parameter @xmath38 . hence the choice of the saddle point , including phase fluctuations , is the following:@xmath22{l}\left \vert \delta_{\mathbf{x},\tau}\right \vert = \delta \\ \theta_{\mathbf{x},\tau}=\mathbf{q.x+}\delta \theta_{\mathbf{x},\tau}\end{array } \right . , \label{saddle point + phase fluctuations}\ ] ] where @xmath39 is the fluctuation field . at this point , we explicitly choose to neglect amplitude fluctuations and to focus solely on phase fluctuations . after substitution of ( [ saddle point + phase fluctuations ] ) in the partition function ( [ partition function ( 1 ) ] ) , we find:@xmath40 the inverse green s function in ( [ partitiesom fl ] ) is given by : @xmath41 ^{2}\right ) \sigma_{3}+\delta \sigma_{1}.\nonumber\end{aligned}\ ] ] the partition function ( [ partitiesom fl ] ) still contains two path integrals : one over the phase fluctuation field and one over the fermionic fields . to calculate the fermionic path integral , a transformation to reciprocal space is necessary , because of the derivatives that are present in the inverse green s function however , since the field @xmath42 is a general function of space and time , this will lead to an infinite number of non - diagonal terms in the action , making the calculation intractable . as a remedy , the path - integral adiabatic approximation will be used . this approximation assumes that the bosonic fluctuation field @xmath43varies slowly in time and space compared to the fermionic fields @xmath1 and @xmath44 . as a result , for a given configuration of the fluctuation field , the configuration of fermionic fields can be coarse - grained by averaging over the ` fast ' degrees of freedom:@xmath45 here @xmath46 are the space - time points for the ` slow ' subsystem , and @xmath47 are the space - time points for the ` fast ' subsystem . the appearance of additional degrees of freedom is due to the assumption that for boson configurations @xmath39 the averaging over fermion configurations can be performed while keeping the bosonic field constant . this method was used successfully for a hydrodynamic description of the berezinskii - kosterlitz - thouless transition in a 2d fermi gas @xcite . using expression ( [ piaa ] ) , the fourier transformation of the fermionic fields can be performed independently of the fluctuation field ( see appendix [ fourier transform ] ) . after fourier transformation , the path integral over fermionic fields is quadratic and can be calculated exactly . the partition function then becomes:@xmath48 \right \ } \right ) + \beta v\frac{\delta^{2}}{g}\right ) , \label{partition function ( 2)}\ ] ] with @xmath49 the momentum of the fermionic fields and @xmath50 the fermionic matsubara frequencies . in ( [ partition function ( 2 ) ] ) the inverse green s function is given by@xmath51 = \left ( \begin{array } [ c]{cc}-i\omega_{n}-\tilde{\zeta}_{\mathbf{k , q}}^{\left ( \theta \right ) } + \tilde { \xi}_{\mathbf{k , q}}^{\left ( \theta \right ) } & \delta \\ \delta & -i\omega_{n}-\tilde{\zeta}_{\mathbf{k , q}}^{\left ( \theta \right ) } -\tilde{\xi}_{\mathbf{k , q}}^{\left ( \theta \right ) } \end{array } \right ) . \label{green 's function}\ ] ] to write ( [ green s function ] ) in a compact form , the following notations were introduced:@xmath22{l}\tilde{\zeta}_{\mathbf{k , q}}^{\left ( \theta \right ) } = \zeta_{\mathbf{k , q}}+\zeta_{\mathbf{k}}^{\left ( \theta \right ) } \\ \zeta_{\mathbf{k , q}}=\zeta~-\mathbf{k.q}\\ \zeta_{\mathbf{k}}^{\left ( \theta \right ) } = \mathbf{-~}\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\mathbf{k}+\dfrac{i}{2}\nabla_{\mathbf{x}}^{2}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \end{array } \right . , \label{zeta}\ ] ] and@xmath22{l}\tilde{\xi}_{\mathbf{k , q}}^{\left ( \theta \right ) } = k^{2}-\tilde{\mu } _ { \mathbf{q}}^{\left ( \theta \right ) } ~~\\ \tilde{\mu}_{\mathbf{q}}^{\left ( \theta \right ) } = \mu_{\mathbf{q}}+\mu_{\mathbf{q}}^{\left ( \theta \right ) } \\ \mu_{\mathbf{q}}=\mu-\dfrac{q^{2}}{4}\\ \mu_{\mathbf{q}}^{\left ( \theta \right ) } = -\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\dfrac{\mathbf{q}}{2}-\dfrac{i}{2}\dfrac{\partial \delta \theta_{\mathbf{x},\tau}}{\partial \tau}-\dfrac{1}{4}\left [ \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \right ] ^{2}\end{array } \right . . \label{mu}\ ] ] these notations are divided in a part that depends explicitly on the phase field @xmath39 ( indicated by a superscript @xmath52 ) , and a part that does not . when @xmath39 is set to zero in ( [ partition function ( 2 ) ] ) , the saddle - point result is obtained @xcite . in ( [ zeta ] ) and ( [ mu ] ) the two competing effects of the fflo state are visible . the first effect is that the imbalance chemical potential @xmath53 can be lowered by the term @xmath54 , which depends on the fflo momentum @xmath0 . this shows that the fflo state is able to cope with spin - imbalance . the second effect is that the momentum @xmath0 increases the total chemical potential , i.e. the term @xmath55 in @xmath56 . this shows that forming cooper pairs with non - zero momentum costs more energy than forming cooper pairs with zero momentum . the fflo state is a trade - off between these two competing effects . finally , the fermionic matsubara summation in ( [ partition function ( 2 ) ] ) can be performed . the resulting action is ultraviolet divergent , which is an artifact of the contact potential ( [ contact potential ] ) . this divergence can be removed by writing the interaction strength @xmath14 as a function of the s - wave scattering length @xmath15 , as in expression ( [ 1/g(as ) ( 2 ) ] ) . the integral over momentum @xmath49 in this expression cancels out the divergence of the action in ( [ partition function ( 2 ) ] ) . the latter then becomes:@xmath57 -\tilde{\xi}_{\mathbf{k , q}}^{\left ( \theta \right ) } -\frac{\delta^{2}}{2k^{2}}\right \ } \label{action ( 2)}\\ & -\beta v\frac{\delta^{2}}{8\pi a_{s}},\nonumber\end{aligned}\ ] ] with @xmath58 . this is the effective action for the fflo state in a 3d fermi gas with spin - imbalance . in the partition function ( [ partition function ( 2 ) ] ) , only the path integral over the fluctuation field @xmath39 remains . however , at this point , the action is still a complicated functional of this field , for which the path integral can not be calculated analytically . therefore , the action will be expanded up to quadratic order in @xmath59 and its derivatives . at this point we have made two approximations regarding the fluctuation field : ( 1 ) the fluctuations are assumed to be small compared to the saddle - point value , which justifies an expansion up to quadratic order in the fluctuation field and ( 2 ) the fluctuations are assumed to be slowly varying in time and space , compared to the fermionic fields , which forms the basis for the path - integral adiabatic approximation . the latter approximation implies that the momentum @xmath60 , associated with the fluctuation field , is also small . for this reason , we will take into account terms only up to order @xmath61 in the action when fourier transforming the fluctuation field , which will be done at a later stage in this work . by expanding the action ( [ action ( 2 ) ] ) up to quadratic order in @xmath39 and its derivatives , the action can be written as the sum of two parts : @xmath62 . the zeroth order term yields the saddle - point action @xmath63 and the second order term leads to the fluctuation action @xmath64 . the saddle - point action corresponds to the result obtained in @xcite . the fluctuation action is given by@xmath65 ^{2}\right ) \right . \nonumber \\ & + \frac{1}{2}y\left ( e_{\mathbf{k , q}}\right ) \left ( \left [ \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\mathbf{k}\right ] ^{2}-i\nabla_{\mathbf{x}}^{2}\left ( \delta \theta _ { \mathbf{x},\tau}\right ) \nabla_{\mathbf{x}}\left ( \delta \theta _ { \mathbf{x},\tau}\right ) .\mathbf{k-}\frac{1}{4}\left [ \nabla_{\mathbf{x}}^{2}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \right ] ^{2}\right ) \nonumber \\ & + \frac{1}{2}\left [ \left ( \dfrac{\xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) ^{2}y\left ( e_{\mathbf{k , q}}\right ) + x\left ( e_{\mathbf{k , q}}\right ) \dfrac{\delta^{2}}{e_{\mathbf{k , q}}^{3}}\right ] \nonumber \\ & \times \left [ \left ( \nabla_{\mathbf{x}}\left ( \delta \theta _ { \mathbf{x},\tau}\right ) .\dfrac{\mathbf{q}}{2}\right ) ^{2}+\dfrac{i}{2}\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\mathbf{q}\dfrac{\partial \delta \theta_{\mathbf{x},\tau}}{\partial \tau}-\frac{1}{4}\left ( \dfrac{\partial \delta \theta_{\mathbf{x},\tau}}{\partial \tau}\right ) ^{2}\right ] \nonumber \\ & + \tilde{y}\left ( e_{\mathbf{k , q}}\right ) \left ( \dfrac{\xi _ { \mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) \left [ \left ( \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\dfrac{\mathbf{q}}{2}\right ) \left [ \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\mathbf{k}\right ] -\dfrac{i}{4}\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\mathbf{q~}\nabla_{\mathbf{x}}^{2}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \right . \nonumber \\ & \left . + \dfrac{i}{2}\dfrac{\partial \delta \theta_{\mathbf{x},\tau}}{\partial \tau}\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau } \right ) .\mathbf{k}\right ] \right \ } , \label{fluctuation action ( 2)}\ ] ] where the following notations were used:@xmath22{l}\smallskip x\left ( a\right ) = \dfrac{\sinh \left ( \beta a\right ) } { \cosh \left ( \beta e_{\mathbf{k , q}}\right ) + \cosh \left ( \beta \zeta_{\mathbf{k , q}}\right ) } \\ \smallskip y\left ( e_{\mathbf{k , q}}\right ) = \beta \dfrac{1+\cosh \left ( \beta e_{\mathbf{k , q}}\right ) \cosh \left ( \beta \zeta_{\mathbf{k , q}}\right ) } { \left [ \cosh \left ( \beta e_{\mathbf{k , q}}\right ) + \cosh \left ( \beta \zeta_{\mathbf{k , q}}\right ) \right ] ^{2}}\\ \smallskip \tilde{y}\left ( e_{\mathbf{k , q}}\right ) = \dfrac{\beta \sinh \left ( \beta e_{\mathbf{k , q}}\right ) \sinh \left ( \beta \zeta_{\mathbf{k , q}}\right ) } { \left [ \cosh \left ( \beta e_{\mathbf{k , q}}\right ) + \cosh \left ( \beta \zeta_{\mathbf{k , q}}\right ) \right ] ^{2}}\end{array } \right . . \label{x , y , y_tilde}\ ] ] more details about the derivation of ( [ fluctuation action ( 2 ) ] ) are given in appendix [ obtaining fluctuation action ] . as a last step before calculating the path integral over the fluctuation field @xmath59 , we have to perform a fourier transformation of this field , again to eliminate the derivatives in the action ( [ fluctuation action ( 2 ) ] ) . the result is:@xmath66 \medskip \right ] \medskip \right \ } \delta \theta_{\mathbf{q},m}\delta \theta_{\mathbf{q},m}^{\ast}. \label{action - fluctuation ( 2)}\ ] ] in expression ( [ action - fluctuation ( 2 ) ] ) , @xmath67 is the momentum of the fluctuation field and @xmath68 are bosonic matsubara frequencies . the action in ( [ action - fluctuation ( 2 ) ] ) contains three scalar products : @xmath69 , @xmath70 and @xmath54 . when calculating the integrals over momentum @xmath49 and @xmath67 , one has to be careful when integrating over the polar and azimuthal angles in spherical coordinates . here , for the benefit of clarity , we explicitly mention our conventions in the definition of these various angles . within the integration over @xmath49 , @xmath0 points along the z - axis , such that @xmath71cos@xmath72 with @xmath73 the polar angle of the spherical coordinate system . analogously , the scalar product of @xmath67 and @xmath0 is written as : @xmath74 . the vector @xmath67 , however , lies in an arbitrary direction with respect to the vector @xmath49 . as a consequence , the scalar product of these two vectors is given by @xmath75 , where the angle @xmath76 can be written in terms of both the polar and the azimuthal angle between @xmath49 and @xmath0 on the one hand and between @xmath67 and @xmath0 on the other hand , by using:@xmath77 where @xmath78 and @xmath79 . substituting ( [ theta_k , q ] ) in the action ( [ action - fluctuation ( 2 ) ] ) and integrating out @xmath80 , the fluctuation partition function , defined by @xmath81 becomes@xmath82 here , @xmath83 equals@xmath84 where the following coefficients have been introduced@xmath22{l}\smallskip a_{\mathbf{q},x}=\dfrac{1}{2}{\displaystyle \int } \dfrac{d\mathbf{k}}{\left ( 2\pi \right ) ^{3}}\left ( 1-\dfrac{\xi _ { \mathbf{k , q}}}{e_{\mathbf{k , q}}}x\left ( e_{\mathbf{k , q}}\right ) -k^{2}\left ( 1-x^{2}\right ) y\left ( e_{\mathbf{k , q}}\right ) \right ) \\ \smallskip b_{\mathbf{q},x}={\displaystyle \int } \dfrac{d\mathbf{k}}{\left ( 2\pi \right ) ^{3}}\left \ { \dfrac{1}{2}k^{2}\left ( 1 - 3x^{2}\right ) y\left ( e_{\mathbf{k , q}}\right ) -kqx\tilde { y}\left ( e_{\mathbf{k , q}}\right ) \left ( \dfrac{\xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) \right . \\ \smallskip \hspace{0.5in}\left . -\dfrac{q^{2}}{4}\left [ \left ( \dfrac { \xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) ^{2}y\left ( e_{\mathbf{k , q}}\right ) + x\left ( e_{\mathbf{k , q}}\right ) \dfrac{\delta^{2}}{e_{\mathbf{k , q}}^{3}}\right ] \right \ } \\ \smallskip c_{\mathbf{q},x}=\dfrac{1}{4}{\displaystyle \int } \dfrac{d\mathbf{k}}{\left ( 2\pi \right ) ^{3}}y\left ( e_{\mathbf{k , q}}\right ) \\ \smallskip d_{\mathbf{q},x}={\displaystyle \int } \dfrac{d\mathbf{k}}{\left ( 2\pi \right ) ^{3}}\left \ { \dfrac{q}{2}\left [ \left ( \dfrac{\xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) ^{2}y\left ( e_{\mathbf{k , q}}\right ) + x\left ( e_{\mathbf{k , q}}\right ) \dfrac{\delta^{2}}{e_{\mathbf{k , q}}^{3}}\right ] \right . \\ \smallskip \hspace{0.5in}\left . + kx\tilde{y}\left ( e_{\mathbf{k , q}}\right ) \left ( \dfrac{\xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) \right \ } \\ \smallskip e_{\mathbf{q},x}=\dfrac{1}{4}{\displaystyle \int } \dfrac{d\mathbf{k}}{\left ( 2\pi \right ) ^{3}}\left [ \left ( \dfrac { \xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) ^{2}y\left ( e_{\mathbf{k , q}}\right ) + x\left ( e_{\mathbf{k , q}}\right ) \dfrac{\delta^{2}}{e_{\mathbf{k , q}}^{3}}\right ] \end{array } \right . .\ ] ] at this point only the path integral over the fluctuation field remains in the partition function given by expression ( [ partition function ( 3 ) ] ) . when calculating this path integral , one has to be careful not to double - count the fields . the reason is that @xmath85 because @xmath39 is a real function . to circumvent this problem , only half the total momentum domain is taken into account : @xmath86 where the symmetry property@xmath87 was used . now , the standard expression for a quadratic bosonic path integral can be used , which leads to @xmath88 \right ) .\ ] ] as a final step , the bosonic matsubara summation can be calculated , which eventually results in the fluctuation free energy:@xmath89 -\beta \sqrt{\psi}\right \ } , \label{free energy}\ ] ] with@xmath22{l}\sqrt{\psi}=\dfrac{1}{2e_{\mathbf{q},x}}\sqrt{d_{\mathbf{q},x}^{2}q^{2}y^{2}+4e_{\mathbf{q},x}\left ( a_{\mathbf{q},x}~q^{2}+b_{\mathbf{q},x}~q^{2}y^{2}+c_{\mathbf{q},x}~q^{4}\right ) } \\ \xi=\dfrac{d_{\mathbf{q},x}}{2e_{\mathbf{q},x}}qy \end{array } \right . .\ in this subsection we briefly explain the method for constructing the phase diagram of the system , including phase fluctuations , starting from the fluctuation free energy ( [ free energy ] ) . in our previous work @xcite , the saddle - point phase diagram for a 3d spin - imbalanced fermi gas at zero temperature was calculated , as a function of the chemical potentials @xmath91 and @xmath53 . for a given value of @xmath91 and @xmath53 , the saddle - point free energy was minimized with respect to the variational parameters @xmath36 ( the superfluid band gap ) and @xmath92 ( the fflo momentum ) . the values of @xmath36 and @xmath92 at a given minimum then determined which state this minimum corresponds to : the bcs state ( @xmath93 ) , the fflo state ( @xmath94 ) or the normal state ( @xmath95 ) . now , phase fluctuations around the saddle point can be included by using the fluctuation free energy ( [ free energy ] ) . for each value of @xmath91 and @xmath53 , the contribution of the fluctuation free energy to the different minima of the saddle - point free energy can be calculated . this will result in a shift of these minima relative to each other , leading to corrections to the phase diagram . however , an important caveat has to be kept in mind . in this paper , phase fluctuations are calculated by using an expansion of the action around the saddle point up to quadratic order in the fluctuation field . this expansion only makes sense when the saddle point is a local or a global minimum . if this is not the case , the quadratic terms will result in negative contributions to the action , which leads to an exponential divergence of the partition function because @xmath96 . because of this fact , only corrections to the bcs - fflo transition and to the bcs - normal transition can be calculated within the current formalism , whereas this is not possible for the fflo - normal transition . the reason for this is that the former two transitions are both of first order , which means there is a competition between two local minima . for given values of @xmath91 and @xmath53 , the fluctuation corrections to both these minima can be calculated and thus their relative shift can be determined . in contrast , the fflo - normal transition is of second order , so that there is a continuous transition between the fflo minimum and the normal minimum . from this it follows that for given values of @xmath91 and @xmath53 , either the fflo minimum or the normal minimum is present , but never both at the same time . hence , fluctuation corrections can only be calculated for one of the two states , for given values of @xmath91 and @xmath53 . as a result , it is not possible to calculate corrections to the fflo - normal transition . finally it should be mentioned that in order to calculate fluctuation corrections to the normal state , both amplitude- and phase fluctuations should be taken into account . this is because when @xmath36 goes to zero , both types of fluctuations become indistinguishable . because of this reason , we will focus solely on fluctuation corrections to the bcs - fflo transition in this paper . figure [ fdmuzeta_fl_3d_english.eps ] shows the saddle - point phase diagram of a 3d fermi gas with spin - imbalance , as a function of the chemical potentials @xmath91 and @xmath53 , at different temperatures . [ h ] fdmuzeta_fl_3d.eps three different states are indicated : the bcs state ( light blue area ) , the fflo state ( light green area ) , and the normal state ( light red area ) . as is well known , at the saddle - point level , the presence of the fflo state in the phase diagram of a spin - imbalanced 3d fermi gas is minimal . with increasing temperature , the area of the fflo state decreases in size , because the free energy of the normal state lowers relatively to the free energy of the fflo state ( as well as the bcs state ) when temperature increases . we find that at @xmath99 , the fflo state vanishes for all values of @xmath91 . when adding fluctuation corrections to these phase diagrams , we find , surprisingly , that these corrections are very small so that their influence is not visible at the scale of fig . [ fdmuzeta_fl_3d_english.eps ] . to give an idea of the order of magnitude of the fluctuation corrections , we ` zoom in ' on the fflo area at @xmath100 by re - scaling the ordinate of the phase diagram in fig . [ fdmuzeta_fl_3d_english.eps](c ) . more specifically , we use @xmath101 instead of @xmath53 , where @xmath102 is the value of @xmath53 ( for a given value of @xmath91 ) at which the bcs - fflo transition occurs within the saddle - point approximation . using this rescaling method , the phase diagram at @xmath100 , now including fluctuation corrections , is given by fig . [ fdmuzeta_fl_t003_details.eps ] . [ h ] fdmuzeta_fl_t003_details.eps we find that the fflo area is slightly enlarged when fluctuations are taken into account ( green + orange area ) compared to the saddle - point result ( green area ) , i.e. the bcs - fflo transition line lies at lower values of @xmath103 . the orange ( fluctuation ) area is relatively small , even compared to the fflo area , which in itself is only a small sliver on the phase diagram , as can be seen in fig . [ fdmuzeta_fl_3d_english.eps ] . thus we find that phase fluctuations , at least within a hydrodynamic approach , have only a relatively small quantitative effect on the fflo state , when considering the bcs - fflo transition . in fig . [ fdmuzeta_fl_t003_details.eps ] we have omitted the values of @xmath91 at which the fflo state is not present , since we only calculated corrections for the bcs - fflo transition and not for the bcs - normal transition . the somewhat irregular shape of the fluctuation area is due to the resolution of our numerical grid . at higher values than @xmath104 , the contribution of fluctuations is negligible , even at the scale of this figure . this may seem counterintuitive , because the interaction strength increases when the density and hence the total chemical potential @xmath91 is increased , and fluctuations are expected to contribute more at higher values of the interaction strength . the reason , however , is that because we use the scattering length as a unit of length ( instead of the fermi vector @xmath105 ) the temperature relative to the fermi temperature @xmath106 is not a constant in figure [ fdmuzeta_fl_t003_details.eps ] : @xmath106 decreases with increasing @xmath91 , because the density and hence @xmath107 ( in units @xmath11 ) increases when @xmath91 is increased . this explains why the fluctuation corrections , which decrease in magnitude at lower temperatures , become smaller for increasing values of @xmath91 in fig . [ fdmuzeta_fl_t003_details.eps](c ) . when converting the temperature in fig . [ fdmuzeta_fl_3d_english.eps ] to units of the fermi temperature , we find that the tri - critical point ( indicated by a black dot in this figure ) lies at approximately @xmath108 . in this paper , we have studied the effect of phase fluctuations on the fflo state in a 3d fermi gas with spin - imbalance . starting from the partition function of the system , the complex fields of the cooper pairs were written in terms of an amplitude field and a phase field . by choosing a suitable saddle point , in which cooper pairs were allowed to have a non - zero momentum @xmath0 , the fflo state was included into the mathematical description . we then considered fluctuations of the phase field around the saddle point . because the resulting path integrals over the fermionic fields and the fluctuation field could not be calculated exactly , we used the path - integral adiabatic approximation . subsequently , the action was expanded up to second order in the fluctuation field , after which the path integral over this field could be calculated . this led to the fluctuation free energy . this free energy was used to calculate corrections to the saddle - point phase diagram of the system . for each value of the chemical potentials @xmath91 and @xmath53 , the contribution of the fluctuation free energy to each minimum of the saddle - point free energy was calculated . this method allows to calculate corrections to the bcs - fflo transition , but not to the fflo - normal transition . we have found that phase fluctuations only lead to relatively small quantitative corrections to the bcs - fflo transition . our results suggest that the fluctuation of the phase of the fflo order parameter , which can be interpreted as an oscillation of the fflo momentum vector around its saddle - point value , does not cause an instability of the fflo state with respect to the bcs state . acknowledgements : : we wish to thank carlos s de melo , nick proukakis , michiel wouters , fons brosens and serghei klimin for interesting and stimulating discussions . jpad gratefully acknowledges a post - doctoral fellowship of the research foundation - flanders ( fwo - vlaanderen ) . this work was supported by fwo - vlaanderen projects g.0119.12.n , g.0115.12.n , and g.0180.09.n . when applying the averaging given by ( [ piaa ] ) to the partition function ( [ partitiesom fl ] ) , the latter becomes@xmath109 \right \ } \eta _ { \mathbf{x}^{\prime},\tau^{\prime}}^{\left ( \mathbf{x},\tau \right ) } + \beta v\frac{\delta^{2}}{g}\right ) , \label{a - partition function ( 2)}\ ] ] where the nambu spinors are given by ( [ nambu spinor ] ) . let us first transform the diagonal terms of the inverse green s function , given by expression ( [ -g^-1 x , t ( 2 ) ] ) . our definitions for the fourier transform are as follows:@xmath22{c}\bar{\psi}_{\mathbf{x}^{\prime},\tau^{\prime},\sigma}^{\left ( \mathbf{x},\tau \right ) } = \left ( \beta v\right ) ^{-\frac{1}{2}}\sum_{\mathbf{k},\omega_{n}}e^{i\omega_{n}\tau^{\prime}-i\mathbf{k.x}^{\prime}}\bar{\psi } _ { \mathbf{k},\omega_{n},\sigma}^{\left ( \mathbf{x},\tau \right ) } \\ \psi_{\mathbf{x}^{\prime},\tau^{\prime},\sigma}^{\left ( \mathbf{x},\tau \right ) } = \left ( \beta v\right ) ^{-\frac{1}{2}}\sum_{\mathbf{k},\omega_{n}}e^{-i\omega_{n}\tau^{\prime}+i\mathbf{k.x}^{\prime}}\psi_{\mathbf{k},\omega_{n},\sigma}^{\left ( \mathbf{x},\tau \right ) } \end{array } \right . .\ ] ] the term proportional to @xmath110 can be transformed as @xmath111{c}\dfrac{\partial}{\partial \tau}-\nabla_{\mathbf{x}}^{2}-\mu+\dfrac{q^{2}}{4}-\zeta - i\mathbf{q}.\nabla_{\mathbf{x}}\end{array } \right . \nonumber \\ & \left . + \dfrac{i}{2}\dfrac{\partial \delta \theta_{\mathbf{x},\tau}}{\partial \tau}-i\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau } \right ) \nabla_{\mathbf{x}}-\dfrac{i}{2}\nabla_{\mathbf{x}}^{2}\left ( \delta \theta_{\mathbf{x},\tau}\right ) + \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\dfrac{\mathbf{q}}{2}+\dfrac{1}{4}\left [ \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \right ] ^{2}\right ) \psi_{\mathbf{x}^{\prime},\tau^{\prime},\uparrow } ^{\left ( \mathbf{x},\tau \right ) } \nonumber \\ & = -\frac{1}{\beta v}\int_{0}^{\beta}d\tau \int d\mathbf{x}\sum_{\mathbf{k},\omega_{n}}\left ( -i\omega_{n}+k^{2}-\mu+\dfrac{q^{2}}{4}-\zeta + \mathbf{q}.\mathbf{k}\right . \nonumber \\ & \left . + \dfrac{i}{2}\dfrac{\partial \delta \theta_{\mathbf{x},\tau}}{\partial \tau}+\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau } \right ) .\mathbf{k}-\dfrac{i}{2}\nabla_{\mathbf{x}}^{2}\left ( \delta \theta_{\mathbf{x},\tau}\right ) + \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\dfrac{\mathbf{q}}{2}+\dfrac{1}{4}\left [ \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \right ] ^{2}\right ) \bar{\psi}_{\mathbf{k},\omega_{n},\uparrow}^{\left ( \mathbf{x},\tau \right ) } \psi_{\mathbf{k},\omega_{n},\uparrow}^{\left ( \mathbf{x},\tau \right ) } , \end{aligned}\ ] ] and analogously for the term proportional to @xmath112 . the off - diagonal term becomes@xmath113 now the partition function has become@xmath114 \right \ } \eta_{\mathbf{k},\omega_{n}}^{\left ( \mathbf{x},\tau \right ) } + \beta v\frac{\delta^{2}}{g}\right ) , \label{a -partition function}\ ] ] where @xmath115 $ ] is given by@xmath116 & = \left ( -i\omega_{n}-\zeta + \mathbf{q}.\mathbf{k}+\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\mathbf{k}-\dfrac{i}{2}\nabla_{\mathbf{x}}^{2}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \right ) \sigma_{0}\nonumber \\ & + \left ( k^{2}-\mu+\dfrac{q^{2}}{4}+\dfrac{i}{2}\dfrac{\partial \delta \theta_{\mathbf{x},\tau}}{\partial \tau}+\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) .\dfrac{\mathbf{q}}{2}+\dfrac{1}{4}\left [ \nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau}\right ) \right ] ^{2}\right ) \sigma_{3}+\delta \sigma_{1},\end{aligned}\ ] ] with the nambu - spinors@xmath117{cc}\bar{\psi}_{\mathbf{k},\omega_{n},\uparrow}^{\left ( \mathbf{x},\tau \right ) } & \psi_{-\mathbf{k},-\omega_{n},\downarrow}^{\left ( \mathbf{x},\tau \right ) } \end{array } \right ) \text { en } \eta_{\mathbf{k}^{\prime},\omega_{n}^{\prime}}^{\left ( \mathbf{x},\tau \right ) } = \left ( \begin{array } [ c]{c}\psi_{\mathbf{k},\omega_{n},\uparrow}^{\left ( \mathbf{x},\tau \right ) } \\ \bar{\psi}_{-\mathbf{k},-\omega_{n},\downarrow}^{\left ( \mathbf{x},\tau \right ) } \end{array } \right ) .\ ] ] note that in ( [ a -partition function ] ) the term proportional to @xmath118 has had its sum over momentum re - indexed as follows : @xmath119 . the expansion of the action @xmath120 -\tilde{\xi}_{\mathbf{k , q}}^{\left ( \theta \right ) } \right \ } -\beta v\frac{\delta^{2}}{g } , \label{a - actie}\ ] ] up to order @xmath121 is done by first expanding the action to second order in @xmath122 and @xmath123 , which are given by ( [ mu ] ) and ( [ zeta ] ) respectively . after expansion of the integrand of ( [ a - actie ] ) , the action can be divided into a saddle - point contribution ( zeroth order in @xmath39):@xmath124 -\xi_{\mathbf{k , q}}\right ) -\beta v\frac{\left \vert \delta \right \vert ^{2}}{g},\ ] ] and a fluctuation contribution ( terms of higher order in @xmath59 and its derivatives):@xmath125 \zeta_{\mathbf{k}}^{\left ( \theta \right ) } + \left ( \left ( \dfrac{\xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) ^{2}y\left ( e_{\mathbf{k , q}}\right ) + x\left ( e_{\mathbf{k , q}}\right ) \dfrac{\delta^{2}}{e_{\mathbf{k , q}}^{3}}\right ) \frac{\left ( \mu_{\mathbf{q}}^{\left ( \theta \right ) } \right ) ^{2}}{2}\nonumber \\ & \left . + y\left ( e_{\mathbf{k , q}}\right ) \frac{\left ( \zeta_{\mathbf{k}}^{\left ( \theta \right ) } \right ) ^{2}}{2}+2\tilde{y}\left ( e_{\mathbf{k , q}}\right ) \left ( \dfrac{\xi_{\mathbf{k , q}}}{e_{\mathbf{k , q}}}\right ) \frac{\mu_{\mathbf{q}}^{\left ( \theta \right ) } \zeta_{\mathbf{k}}^{\left ( \theta \right ) } } { 2}\right ] , \end{aligned}\ ] ] where the definitions in ( [ x , y , y_tilde ] ) were used . the saddle - point action corresponds to the result of @xcite . now the fluctuation action can be simplified further . firstly , only the terms up to order @xmath61 are taken into account . secondly , some terms may vanish due to e.g. boundary conditions . now we briefly point out the main simplifications which can be performed . in the integral over @xmath126 , the first two terms ( we use the order of terms as given in ( [ x , y , y_tilde ] ) ) vanish because of@xmath127 and the boundary condition for bosons@xmath128 respectively . in the integral over @xmath129 , three terms vanish because they are of other @xmath130 or higher . in the integral over @xmath123 , the first term ( again respecting the order as given in ( [ x , y , y_tilde ] ) ) vanishes for the same reason as ( [ boundary condition infinity and zero ] ) , while the second term is zero because of another boundary condition@xmath131 furthermore , in the integral over @xmath132 , two terms again vanish because they are of order @xmath130 or higher . finally , the term@xmath133 equals zero . using green s first identity , the term ( [ term ( 1 ) ] ) becomes@xmath134 } { \partial \tau}.\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau } \right ) , \end{aligned}\ ] ] because @xmath135 . subsequently , we can write@xmath136 } { \partial \tau}.\nabla_{\mathbf{x}}\left ( \delta \theta_{\mathbf{x},\tau } \right ) = -\int_{0}^{\beta}d\tau \int d\mathbf{x}~\frac{1}{2}\dfrac { \partial \left \ { \left [ \nabla_{\mathbf{x}}\left ( \delta \theta _ { \mathbf{x},\tau}\right ) \right ] ^{2}\right \ } } { \partial \tau}=0 , \label{3}\ ] ] again because of boundary conditions . putting it all together , the fluctuation action becomes equal to ( [ fluctuation action ( 2 ) ] ) . p. nozires and s. schmitt - rink , j. low temp . phys . * 59 * , 195 ( 1985 ) ; m. drechsler and w. zwerger , ann . phys . * 1 * , 15 ( 1992 ) ; c.a.r . s de melo , m. randeria , and j.r . engelbrecht , phys . . lett . * 71 * , 3202 ( 1993 ) ; e. babaev and h. kleinert , phys . b * 59 * , 12083 ( 1999 ) ; c.a.r . s de melo , phys . today * 61 * , no . 10 , 45 ( 2008 ) strecker , g.b . partridge , and r.g . hulet , phys . _ _ _ _ * * 91 * * _ , _ 080406 ( 2003 ) ; j. cubizolles , t. bourdel , s.j.j.m.f . kokkelmans , g.v . shlyapnikov , and c. salomon , _ ibid . _ * 91 * , 240401 ( 2003 ) ; s. jochim , m. bartenstein , a. altmeyer , g. hendl , c. chin , j.h . denschlag , and r. grimm , _ ibid . _ * 91 * , 240402 ( 2003 ) ; s. jochim , m. bartenstein , a. altmeyer , g. hendl , s. riedl , c. chin , j.h . denschlag , and r. grimm , science * 302 * , 2101 ( 2003 ) ; m. greiner , c.a . regal , and d.s . jin , nature * 426 * , 537 ( 2003 ) ; m.w . zwierlein , c.a . stan , c.h . schunck , s.m.f . raupach , s. gupta , z. hadzibabic , and w. ketterle , phys . * 91 , * 250401 ( 2003 ) ; c.a . regal , m. greiner , and d.s . jin , _ ibid . _ * 92 * , 040403 ( 2004 ) ; m.w . zwierlein , c.a . stan , c.h . schunck , s.m.f . raupach , a.j . kerman , and w. ketterle , _ ibid . _ * 92 * , 120403 ( 2004 ) ; m. bartenstein , a. altmeyer , s. riedl , s. jochim , c. chin , j.h . denschlag , and r. grimm , _ ibid . _ * 92 * , 120401 ( 2004 ) ; t. bourdel , l. khaykovich , j. cubizolles , j. zhang , f. chevy , m. teichmann , l. tarruell , s.j.j.m.f . kokkelmans , and c. salomon , _ ibid . _ * 93 * , 050401 ( 2004 ) ; c. chin , m. bartenstein , a. altmeyer , s. riedl , s. jochim , j.h . denschlag , r. grimm , science * 305 * , 1128 ( 2004 ) ; j. kinnunen , m. rodrguez , p. trm , _ ibid . * 305 * , 1131 ( 2004 ) ; g.b . partridge , k.e . strecker , r.i . kamar , m.w . jack , and r.g . hulet , phys . lett__. _ _ * 95 * , 020404 ( 2005 ) ; m.w . zwierlein , j.r . abo - shaeer , a. schirotzek , c.h . schunck , and w. ketterle , nature * 435 * , 1047 ( 2005 ) . t. mizushima , k. machida , and m. ichioka , phys . * 94 * , 060404 ( 2005 ) ; d.e . sheehy and l. radzihovsky , _ ibid . _ * 96 * , 060401 ( 2006 ) ; j. kinnunen , l.m . jensen , and p. trm , _ ibid . _ * 96 * , 110403 ( 2006 ) ; k. machida , t. mizushima , and m. ichioka , _ ibid . _ * 97 * , 120407 ( 2006 ) ; p. castorina , m. grasso , m. oertel , m. urban , and d. zappal , phys . a * 72 * , 025601 ( 2005 ) ; n. yoshida and s .- k . yip , _ _ * 75 * , 063601 ( 2007 ) ; w. zhang and l .- m . duan , _ ibid . _ * 76 * , 042710 ( 2007 ) ; t.k . koponen , t. paananen , j .- martikainen , m.r . bakhtiari , and p. trm , new j. phys . * 10 * , 045014 ( 2008 ) . t. mizushima , k. machida , and m. ichioka , phys . rev . lett . * 94 * , 060404 ( 2005 ) ; d.e . sheehy and l. radzihovsky , _ ibid . _ * 96 * , 060401 ( 2006 ) ; j. kinnunen , l.m . jensen , and p. trm , _ ibid . _ * 96 * , 110403 ( 2006 ) ; k. machida , t. mizushima , and m. ichioka , _ ibid . _ * 97 * , 120407 ( 2006 ) ; p. castorina , m. grasso , m. oertel , m. urban , and d. zappal , phys . rev . a * 72 * , 025601 ( 2005 ) ; n. yoshida and s .- k . yip , _ * 75 * , 063601 ( 2007 ) ; w. zhang and l .- m . duan , _ ibid . _ * 76 * , 042710 ( 2007 ) ; t.k . koponen , t. paananen , j .- martikainen , m.r . bakhtiari , and p. trm , new j. phys . * 10 * , 045014 ( 2008 ) . j.m . edge and n.r . cooper , phys . lett . * 103 * , 065301 ( 2009 ) ; a. korolyuk , f. massel , and p. trm , _ ibid . _ * 104 * , 236402 ( 2010 ) ; j. kajala , f. massel , and p. trm , phys . a * 84 * , 041601(r ) ( 2011 ) ; h. lu , l.o . baksmaty , c.j . bolech , and h. pu , phys . . lett . * 108 * , 225302 ( 2012 ) ; i. zapata , f. sols , and e. demler , _ ibid . _ * 109 * , 155304 ( 2012 ) .

in phys . rev . a * 71 * , 060312(r ) ( 2005 ) the robustness of the local adiabatic quantum search to decoherence in the instantaneous eigenbasis of the search hamiltonian was examined . we expand this analysis to include the case of the global adiabatic quantum search . as in the case of the local search the asymptotic time complexity for the global search is the same as for the ideal closed case , as long as the hamiltonian dynamics is present . in the case of pure decoherence , where the environment monitors the search hamiltonian , we find that the time complexity of the global quantum adiabatic search scales like , where is the list length . we moreover extend the analysis to include success probabilities and prove bounds on the run time with the same scaling as in the conditions for the limit . we supplement the analytical results by numerical simulations of the global and local search .

quant-ph0507010

quantum adiabatic algorithms , first conceived in ref . @xcite , have been developed as an alternative to the traditional circuit model of quantum computation . an important aspect of the adiabatic approach is its expected resilience to various kinds of open system effects , which makes it a promising candidate for robust quantum computation . the basic reason for this appealing feature is that the adiabatic quantum computer operates near the instantaneous ground state of a time - dependent hamiltonian and can therefore be expected to be insensitive to relaxation and open system effects among the excited states . the robustness of adiabatic quantum computation has been discussed in the literature for different kinds of perturbations , such as relaxation @xcite , noise @xcite , decoherence @xcite , and unitary control errors @xcite . in this paper , we elaborate on the robustness of the quantum adiabatic search @xcite in the presence of decoherence in the instantaneous eigenbasis of the search hamiltonian . in particular , we extend the analysis of the local search @xcite in ref . @xcite to global adiabatic search @xcite . we further address the dependence of the asymptotic time complexity of the search procedure on the success probability for finding the searched item . to model adiabatic search in the presence of decoherence we consider master equations of the form ( @xmath4 from now on ) @xmath5 -b\bm{[}w,[w,\varrho]\bm{]},\ ] ] where @xmath6 is the density operator , and @xmath7 are constants such that @xmath8 . for convenience , we also assume that @xmath9 . the operator @xmath10 is taken to be the time - dependent hamiltonian of the adiabatic search @xcite , and @xmath11 is a time - dependent hermitian operator such that @xmath12 = 0 $ ] , which gives decoherence in the instantaneous eigenbasis of @xmath10 . decoherence with respect to the energy eigenbasis of a system has been shown to occur in certain regimes of system - environment interaction @xcite . apart from this general regime of weak interaction and dominating self - hamiltonians one can find other settings where this type of decoherence occurs . one example is control errors in terms of fluctuations in the energy levels of the system @xcite . note also that decoherence in the energy eigenbasis does not necessarily imply the particular form of eq . ( [ enkel ] ) . however , this choice provides a sufficiently simple model to be treated analytically , and yet results in non - trivial behavior . moreover , the form of eq . ( [ enkel ] ) quite generally guarantees the solution to be a proper density operator at all times . our results show that for @xmath13 and list length @xmath1 , it is sufficient with a run time of the order @xmath1 in the global case and @xmath14 in the local case , as is the case for the ideal closed evolution . these results suggest that the adiabatic quantum computer is protected against this type of decoherence , which is promising for physical realizations . in the wide - open case where @xmath15 @xcite , with the special choice @xmath16 , the scaling changes to @xmath0 and @xmath1 , for the global and local search , respectively . although the concept of adiabaticity has a well defined meaning for closed systems , there is no obviously unique generalization to the case of open systems . one may consider several different generalizations focusing on various aspects of the standard adiabaticity . for example , in refs . @xcite the concept of adiabaticity is based upon jordan block decompositions of the superoperator that describe the evolution of the open system . a different approach , designed for systems that are weakly open , has been put forward in ref . this latter approach is based upon taking decoupling of the energy subspaces of the ideal hamiltonian as the starting point . here , we use a third method based upon a feature of the adiabatic quantum computer , namely that the evolution takes place near an eigenstate of the ideal hamiltonian . this property is in contrast with , e.g. , implementations of holonomic quantum gates @xcite , where it is essential that the gate can operate on arbitrary superpositions , without too large errors . for the functioning of the adiabatic quantum computer , however , it is sufficient that the state of the system remains close to the correct eigenstate . in the present interpretation of the word `` adiabaticity '' we only require that the probabilities of finding the system in the instantaneous eigenstates of @xmath17 are conserved , but we allow superpositions of the instantaneous eigenstates to decay into mixtures . the structure of the article is as follows . in sec . [ sec : twolev ] we derive an integral equation from the original master equation . section [ sec : search ] gives a brief overview of the adiabatic search algorithm , and we calculate some quantities which are used later . in sec . [ sec : semiopen ] we prove a sufficient condition for the success probability of the adiabatic search to approach unity in the semi - open case . we consider both global and local search . furthermore , a condition on the decoherence term of the master equation , which is used in the proofs in sec . [ sec : semiopen ] , is examined . in sec . [ sec : wopen ] we demonstrate that the wide - open case is essentially different from the semi - open case . we prove sufficient conditions for the success probability to approach unity in both the global , as well as the local case . we further prove that the sufficient conditions are also necessary . in sec . [ sec : success ] , bounds on the run time for fixed success probabilities less that unity , are given . we end with the conclusions . in this section we develop the basic equation used in our analysis . since it is our intention to use adiabaticity we consider a fixed family of hamiltonians @xmath17 , parametrized , in a sufficiently smooth way , by the parameter @xmath18 . we assume that the parameter @xmath18 is proportional to the time @xmath19 , such that @xmath20 at time @xmath21 and @xmath22 at time @xmath23 . the time @xmath24 is referred to as the `` run time '' . we also assume a family of hermitian operators @xmath25 , to be used in the master equation @xmath26 -b\bm{[}w(t / t),[w(t / t),\varrho(t)]\bm{]}.\ ] ] we assume that both @xmath27 and @xmath28 are hermitian and non - degenerate at each @xmath19 . moreover , we assume @xmath29 = 0 $ ] for all @xmath19 . by a change of variables @xmath30 in eq . ( [ startekv ] ) one obtains @xmath31 -tb\bm{[}w(s),[w(s),\rho(s)]\bm{]},\ ] ] where @xmath32 . let @xmath33 be instantaneous orthonormal eigenvectors of @xmath17 , i.e. , @xmath34 . since @xmath17 is assumed to be non - degenerate the arbitrariness in the choice of eigenvectors is limited to phase factors . let @xmath35 denote the eigenvalues of @xmath25 . if we insert @xmath36 into eq . ( [ sparam ] ) , the equations for the components @xmath37 become @xmath38_{nm},\end{aligned}\ ] ] where @xmath39 is an hermitian matrix with elements @xmath40 and @xmath41 is the matrix with elements @xmath37 . consider an ordinary differential equation @xmath42 where @xmath43 and @xmath44 are s - dependent linear operators acting on the vector @xmath45 . if the space is finite - dimensional , @xmath43 and @xmath44 are sufficiently well behaved as functions of @xmath18 , and if @xmath43 fulfills @xmath46 = 0 $ ] for all @xmath18 , @xmath47 , then eq . ( [ sftjf ] ) can be rewritten as the following integral equation @xmath48 in the present case we have @xmath49 and @xmath50.\end{aligned}\ ] ] hence , eq . ( [ grundekv ] ) can be rewritten as @xmath51_{nm}ds',\nonumber\\\end{aligned}\ ] ] where @xmath52^{2}ds',\ ] ] @xmath53ds'.\ ] ] our task is to find a single marked item in a collection of @xmath1 items . the @xmath1-element search problem is associated with an @xmath1-dimensional hilbert space with orthonormal basis @xmath54 , where the marked item corresponds to @xmath55 . we consider the following family of hamiltonians @xcite , @xmath56 where @xmath57 of the initial hamiltonian @xmath58 is easy to prepare , while the ground state of @xmath59 gives the solution of the search problem . hence , if we assume the evolution is adiabatic , this family of hamiltonians takes us from the initial state @xmath60 to the marked state @xmath61 , and thus solves the search problem . a crucial observation is that the problem is essentially two - dimensional , since the space @xmath62 spanned by @xmath60 and @xmath61 is the only relevant subspace . we may gram - schmidt orthogonalize @xmath63 , yielding the orthonormal basis @xmath64 we represent the hamiltonian family in eq . ( [ family ] ) , restricted to the space @xmath62 , as matrices in the basis defined in eq . ( [ basis ] ) . after the substitution @xmath65 , we get @xmath66,\ ] ] which has the eigenvalues @xmath67 and orthonormal eigenvectors @xmath68 = \left[\begin{matrix}e^{(0)}_{1}(s ) \\ e^{(0)}_{2}(s)\end{matrix}\right ] , \nonumber\\ \bm{e}^{(1 ) } & = & \frac{\sqrt{2}}{\sqrt{\delta^{2 } + \delta d } } \left[\begin{matrix } \frac{1}{2}(\delta + d)\\ -s\frac{\sqrt{n-1}}{n } \end{matrix}\right ] = \left[\begin{matrix } e^{(1)}_{1}(s)\\ e^{(1)}_{2}(s)\end{matrix}\right],\end{aligned}\ ] ] where @xmath69 the two relevant eigenvectors of @xmath17 are @xmath70 where @xmath71 is the ground state . we next calculate the matrix @xmath39 , defined in eq . ( [ zdef ] ) . in order to save some efforts we first notice that @xmath72 as well as @xmath73 are identically zero , since @xmath74 , @xmath75 , @xmath76 , and @xmath77 are real valued . secondly @xmath39 has to be an hermitian matrix , which follows from differentiation of the relation @xmath78 . by combining eq . ( [ zdef ] ) with eqs . ( [ sdfh ] ) , ( [ deltadef ] ) , and ( [ ddef ] ) , one obtains the following result @xmath79 define @xmath80 note that the maximum @xmath81 of @xmath82 is obtained at @xmath83 . a useful property of @xmath84 is @xmath85 for all @xmath86 . concerning the reduction to the two - dimensional problem , one should note that , due to the assumption @xmath87 = 0 $ ] , the operator @xmath25 block - diagonalizes with respect to the eigenspaces of @xmath17 . this implies that the solution of eq . ( [ startekv ] ) remains in the subspace @xmath62 once started there . thus , we only need to consider the master equation in eq . ( [ startekv ] ) on this particular subspace . define @xmath88 from eq . ( [ grundint ] ) one can derive the following equation for @xmath89 , @xmath90ds ' \nonumber\\ * & & - 4 \int_{0}^{s } e^{-tbq(s')}\int_{0}^{s'}e^{tbq(s '' ) } { { \textrm{re}}}\left[e^{-itar(s')}e^{itar(s'')}\right]z(s')z(s'')y(s'')ds''ds'.\end{aligned}\ ] ] as seen from eq . ( [ ydef ] ) , the value of @xmath89 determines to what degree the system is in the ground state , i.e. , @xmath91 when the system is in the ground state , and @xmath92 when the system is in the first excited state . we begin with a comment on our terminology regarding the `` openness '' of quantum systems . with a `` wide - open '' @xcite system we intend a system governed by eq . ( [ enkel ] ) , with @xmath15 . as a contrast to the wide - open case we define `` semi - open '' as @xmath93 . finally , a `` closed '' system denotes the case @xmath94 . we define @xmath95}\delta(s ) = \frac{1}{\sqrt{n}},\\ \label{ligammadef } \gamma & = & \min_{s\in [ 0,1]}\gamma(s).\end{aligned}\ ] ] in the proofs we use the following restriction on the decoherence term @xmath96 where @xmath97 is a constant independent of @xmath1 . this assumption essentially means that the fluctuations of @xmath98 in @xmath18 are not allowed to grow with the list - length @xmath1 . the restriction to an @xmath1-independent @xmath97 can be relaxed . we discuss this issue , as well as the condition in general , in sec . [ sec : condition ] . from eq . ( [ tvaanivaa ] ) it follows that @xmath99 where @xmath100}z(s')u_{\pm}(s')ds',\\ \label{gidef } u_{\pm}(s ' ) & = & \int_{0}^{s'}e^{t[bq(s'')\pm iar(s'')]}z(s'')y(s'')ds'',\nonumber\\\end{aligned}\ ] ] and @xmath101ds'.\ ] ] for the sake of simplicity we separate the treatment of the terms @xmath102 and @xmath103 . to begin with we define the following integral and perform a partial integration @xmath104(s ) & = & \int_{0}^{s}e^{- t[bq(s')\pm iar(s')]}z(s')g(s')ds'\nonumber\\ & = & \int_{0}^{s}e^{- t[bq(s')\pm iar(s')]}\left[b\frac{dq}{ds'}(s')\pm ia\frac{dr}{ds'}(s')\right]\frac{z(s')g(s')}{b\frac{dq}{ds'}(s')\pm ia\frac{dr}{ds'}(s')}ds'\nonumber\\ * & = & -\frac{1}{t}e^{- t[bq(s)\pm iar(s)]}\frac{z(s)g(s)}{b\frac{dq}{ds}(s)\pm ia\frac{dr}{ds}(s)}\nonumber\\ & & + \frac{1}{t}\frac{\sqrt{n-1}}{n}\frac{g(s)}{b\frac{dq}{ds}(s)\pm ia\frac{dr}{ds}(s)}\bigg|_{s = 0}\nonumber\\ * & & + \frac{1}{t}\int_{0}^{s}e^{- t[bq(s')\pm iar(s')]}\frac{d}{ds'}\left(\frac{z(s')}{b\frac{dq}{ds'}(s')\pm ia\frac{dr}{ds'}(s')}\right)g(s')ds'\nonumber\\ * & & + \frac{1}{t}\int_{0}^{s}e^{- t[bq(s')\pm iar(s')]}\frac{z(s')}{b\frac{dq}{ds'}(s')\pm ia\frac{dr}{ds'}(s')}\frac{d}{ds'}g(s')ds'.\end{aligned}\ ] ] if we put @xmath105 into eq . ( [ gdef ] ) the result is @xmath106(s ) \nonumber\\ & = & -\frac{1}{t}e^{-t[bq(s)\pm iar(s)]}\frac{z(s)}{b\frac{dq}{ds}(s)\pm ia\frac{dr}{ds}(s ) } \int_{0}^{s}e^{t[bq(s')\pm iar(s')]}z(s')y(s')ds'\nonumber\\ * & & + \frac{1}{t}\int_{0}^{s}e^{-t[bq(s')\pm iar(s ' ) ] } \frac{d}{ds'}\left(\frac{z(s')}{b\frac{dq}{ds'}(s')\pm ia\frac{dr}{ds'}(s')}\right)\int_{0}^{s'}e^{t[bq(s'')\pm iar(s'')]}z(s'')y(s'')ds '' ds'\nonumber\\ * & & + \frac{1}{t}\int_{0}^{s}\frac{z(s')}{b\frac{dq}{ds'}(s')\pm ia\frac{dr}{ds'}(s')}z(s')y(s')ds'.\end{aligned}\ ] ] in the next step we calculate an upper bound to @xmath107 . to do this we use that @xmath108 . moreover , since @xmath109 it follows that @xmath110 , which implies @xmath111\}\leq1 $ ] . we also use that @xmath112 , @xmath113 , and eq . ( [ zegensk ] ) to obtain @xmath114 now we use the fact that @xmath115 , as well as eqs . ( [ lideltadef ] ) and ( [ ligammadef ] ) , which result in @xmath116 to treat the integrand in eq . ( [ idelsteg ] ) , we define @xmath117 which can be estimated as @xmath118 the function @xmath82 is symmetric around @xmath83 and increasing on the interval @xmath119 $ ] . it follows that the function @xmath120 is also symmetric around @xmath83 , and that @xmath121 on @xmath119 $ ] . thus @xmath122 since both @xmath82 and @xmath123 are symmetric around @xmath83 , @xmath124 has the same symmetry . moreover , @xmath123 is increasing on the interval @xmath125 $ ] . hence , @xmath126 on @xmath125 $ ] , which leads to @xmath127 by combining eq . ( [ condition ] ) with eqs . ( [ xolikhet ] ) , ( [ absderint ] ) , and ( [ zderd ] ) , one obtains @xmath128 by combining eqs . ( [ lideltadef ] ) , ( [ iuppd ] ) , ( [ idelsteg ] ) , and ( [ absd ] ) it follows that @xmath129 equation ( [ jdef ] ) can be rewritten as @xmath130}z(s')ds ' \right).\ ] ] we write @xmath131(s)|$ ] , with @xmath132 as in eq . ( [ gdef ] ) with @xmath133 . by a line of reasoning very similar to the one in the previous section one obtains @xmath134 which results in @xmath135 by combining eqs . ( [ diff ] ) , ( [ iresultat ] ) , and ( [ jresultat ] ) , one obtains @xmath136 note that @xmath137 which , as long as @xmath93 , implies that @xmath138 where we have used eq . ( [ ydef ] ) . the leading term in eq . ( [ sogl ] ) is proportional to @xmath139 . in the @xmath140 limit the probability to find the system in the groundstate is conserved . thus , it is a sufficient condition for adiabaticity that @xmath141 . as is well known @xcite , the standard global adiabatic search algorithm performs as the classical search , but the local adiabatic search @xcite outperforms the classical algorithms . here we further elaborate and extend the analysis of ref . @xcite concerning the effect of decoherence on the local adiabatic search . the idea behind the local search is to adjust the speed of parameter change in such a way that more time is spent near the minimum energy gap . this is obtained by a reparametrization of the functions @xmath17 and @xmath25 to @xmath142 and @xmath143 , where @xmath144 is a strictly increasing , sufficiently smooth function , such that @xmath145 and @xmath146 . the new master equation becomes @xmath147\nonumber\\ & & -bt \bm { [ } w \bm { ( } f(s ) \bm { ) } , [ w \bm { ( } f(s ) \bm { ) } , \rho(s ) ] \bm{]}.\end{aligned}\ ] ] by a change of variables @xmath148 and by defining @xmath149 , eq . ( [ startlocal ] ) becomes @xmath150\nonumber\\ & & -bt\frac{df^{-1}}{dr}(r)\bm{[}w(r),[w(r),\widetilde{\rho}(r)]\bm{]}.\end{aligned}\ ] ] note that @xmath151 and @xmath152 . the equation for @xmath153 is fully analogous to eq . ( [ tvaanivaa ] ) with @xmath154 and @xmath155 replaced by @xmath156 and @xmath157 , where @xmath158 the choice @xmath159 gives the optimal efficiency for the quantum search problem @xcite . one obtains @xmath160 @xmath161 define @xmath162}\frac{\gamma^{2}(s)}{\delta(s)}.\ ] ] as in the global case we let @xmath163 , and as before we separate the treatment of the terms @xmath164 and @xmath165 . the terms @xmath166 and @xmath167 are defined as in eqs . ( [ iuppd])-([gidef ] ) and ( [ jekv ] ) , but with @xmath154 and @xmath155 replaced by @xmath156 and @xmath157 , respectively . the analogue of eq . ( [ ipmutr ] ) , is obtained by replacing @xmath154 and @xmath155 by @xmath156 and @xmath157 , respectively . we use @xmath168 $ ] , @xmath169 $ ] , @xmath170\}\leq 1 $ ] if @xmath171 , as well as eq . ( [ zegensk ] ) , which result in @xmath172 by using eq . ( [ zetadef ] ) it follows that @xmath173 one may use @xmath174 and eq . ( [ zegensk ] ) to obtain @xmath175 the derivative contained in the integrand on the right - hand side of the above equation is performed by first multiplying both the numerator and the denominator by @xmath176 and then using @xmath177 . moreover , we use eq . ( [ zetadef ] ) . the result is @xmath178 note that in the inequality above we have extended the integration interval from @xmath179 $ ] to @xmath180 $ ] . by combining eqs . ( [ condition ] ) , ( [ zderd ] ) , ( [ ldef ] ) , ( [ tiiabs ] ) , and ( [ absderiv ] ) one obtains @xmath181 as in the case of global search one obtains @xmath182 as in eq . ( [ gdef ] ) with @xmath183 , and @xmath154 and @xmath155 replaced by @xmath156 and @xmath157 . note that @xmath184 and thus @xmath185 . using similar arguments as in the previous sections , one obtains @xmath186 next we use @xmath174 together with eqs . ( [ condition ] ) , ( [ zderd ] ) , ( [ ldef ] ) , and ( [ absderiv ] ) , yielding @xmath187 we combine eqs . ( [ wti ] ) and ( [ wtj ] ) together with @xmath188 , @xmath93 , @xmath189 , and @xmath190 . furthermore , we assume @xmath191 , which gives @xmath192 , and results in @xcite @xmath193 the leading term of this equation is proportional to @xmath194 . hence , a sufficient condition for adiabaticity is @xmath195 . equation ( [ soloc ] ) can be expressed in terms of the solution @xmath196 of eq . ( [ startlocal ] ) by the substitution @xmath197 . however , this change of variables does not affect the condition for adiabaticity . both for the global and local search we have restricted the decoherence term of the master equation to fulfill the condition in eq . ( [ condition ] ) . this condition puts a limit on the fluctuations of @xmath98 with respect to @xmath18 , as a function of the list - length @xmath1 . however , one may note that the condition can be relaxed , in the sense that we may allow @xmath97 to be dependent on @xmath1 . by inserting @xmath198 , where @xmath199 and @xmath200 are constants , into eqs . ( [ sogl ] ) and ( [ soloc ] ) , one finds that the new scaling becomes @xmath201 and @xmath202 for the global and local search , respectively . thus , the scalings get worse , but one finds that there is a `` window '' @xmath203 where the local search still outperforms the classical search , although not with the optimal efficiency of the grover search @xcite . in the following we attempt to obtain an indication of to what extent the condition in eq . ( [ condition ] ) is restrictive or not . for the sake of simplicity we do so for the more restrictive case where @xmath97 is independent of @xmath1 . we assume @xmath204 , where @xmath205 , is an increasing and sufficiently smooth function . now , a larger value of @xmath206 implies stronger decoherence . thus , we have chosen to consider a decoherence model for which the strength of the decoherence increases with the energy difference between the two instantaneous eigenstates , but we do not specify any details on how it grows . we intend to derive a sufficient condition on @xmath207 for @xmath98 to fulfill the condition in eq . ( [ condition ] ) . by the assumptions on @xmath207 and the form of @xmath208 , it follows that @xmath98 is decreasing on the interval @xmath119 $ ] and increasing on @xmath125 $ ] . partial integration implies @xmath209 we use @xmath210 and @xmath211 . moreover , we use @xmath212 , which leads to @xmath213 only the integral on the right - hand side of the above expression may potentially grow when @xmath1 increases . if we treat this term separately we find @xmath214 this implies that a sufficient condition for the inequality in eq . ( [ condition ] ) to hold , is that the expression on the right - hand side of the above equation , is bounded . from the assumptions on @xmath207 it follows that the limiting behavior of this integral is determined by the asymptotic behavior of the function @xmath207 as @xmath215 . assume that @xmath216 for some constants @xmath217 and @xmath218 . then it follows that @xmath219 in the case where @xmath220 , the function @xmath98 satisfies the inequality in eq . ( [ condition ] ) , since @xmath221 is a bounded function on the nonzero natural numbers . in the case where @xmath222 , one can see that the right - hand side of eq . ( [ uppd ] ) contains two terms . none of these terms increase in magnitude , with increasing @xmath1 , if @xmath223 . hence , we may conclude that for all @xmath224 the condition in eq . ( [ condition ] ) is fulfilled . in other words , for a sufficiently smooth , positive , and increasing function @xmath207 , the only condition we have required is @xmath225 with @xmath224 . the wide - open case is obtained if we let @xmath227 ( and @xmath228 for convenience ) which in the global search case yields the master equation @xmath229\bm{]}\ ] ] and in the local search case yields @xmath230\bm{]},\ ] ] where we have changed variables to @xmath197 as described in sec . [ sec : semiopenlocal ] . even with the assumption that @xmath25 should be simultaneously diagonalizable with @xmath17 , we still have a rather large freedom in the choice of @xmath11 , in terms of how @xmath98 depends on the parameter @xmath18 and the list - length @xmath1 . in the previous sections it was only necessary to require that @xmath98 should fulfill the condition in eq . ( [ condition ] ) . in this case , however , we need to specify more about @xmath11 . our main object of study in this section is the case @xmath231 , which corresponds to @xmath232 . however , in order to prove that in the wide - open case the asymptotic behavior is dependent on the choice of @xmath11 , we consider the generalization @xmath233 . note in particular that if @xmath234 , where @xmath235 is an odd natural number , then @xmath236 can be obtained with @xmath237 . here we prove that in the wide - open case , with @xmath238 , the global and local conditions for adiabaticity are @xmath239 and @xmath240 , respectively . we first prove that these are sufficient conditions . thereafter follows a proof of necessity . let @xmath227 and @xmath228 in eq . ( [ tvaanivaa ] ) and assume that @xmath226 , which gives @xmath241 assume @xmath242 , and use @xmath243 . it follows that @xmath244 , for all @xmath245 . moreover , since @xmath109 , it follows that @xmath246 . this can be used , together with @xmath108 , to show that eq . ( [ tvaanivaa ] ) leads to @xmath247 one may use @xmath115 , integrate , and use eq . ( [ zegensk ] ) to obtain @xmath248 thus , we have proved that a sufficient condition for @xmath249 to approach @xmath250 is @xmath251 , if @xmath218 . here we require @xmath252 to fulfill @xmath253 . as before we let @xmath227 and @xmath228 in eq . ( [ tvaanivaa ] ) , but we replace @xmath154 with @xmath156 defined as @xmath254 with @xmath255 as in eq . ( [ ldef ] ) , and where we in the inequality have used @xmath256 and the assumption @xmath253 . inserting eq . ( [ wqwh ] ) into eq . ( [ tvaanivaa ] ) yields @xmath257 by a reasoning very similar to the derivation leading from eq . ( [ sdghn ] ) to eq . ( [ globwo ] ) it follows that @xmath258 which , together with eq . ( [ ldef ] ) , gives @xmath259 thus , a sufficient condition for adiabaticity is @xmath260 . in secs . [ sec : globsuff ] and [ sec : locsuff ] we proved sufficient conditions for conservation of the probability to find the system in the ground state , throughout the evolution . here we prove that these sufficient conditions are also necessary , but under the simplifying assumption that we start in the ground state of the system . moreover , we focus on the end point @xmath22 , i.e. , we consider the success probability @xmath261 , and necessary conditions for @xmath262 to approach unity . due to the effectively two - dimensional hilbert space of the problem we may use a bloch vector representation in terms of which the state of the system and the hermitian operator @xmath11 read @xmath263,\\ w(s ) & = & \frac{1}{2}\{[w_{0}(s)+w_{1}(s)]\hat{1 } + \bm{u}(s)\cdot{\bm\sigma}\},\end{aligned}\ ] ] where @xmath264 and @xmath265 are elements in @xmath266 , @xmath267 , and @xmath268 are the pauli spin operators . it follows that @xmath269 , where @xmath270 is the bloch unit - vector representing the ground state of @xmath17 . if we use the assumption @xmath226 , then eq . ( [ widoppen ] ) can be rewritten in the bloch representation as @xmath271\},\ ] ] where @xmath272 in the local case eq . ( [ widoppenlocal ] ) again becomes eq . ( [ blochekv ] ) , but with the parameter @xmath18 replaced by @xmath273 and @xmath274 most of the material presented in this subsection is independent of whether we consider the global or the local search . the difference is exclusively carried by the function @xmath275 . due to this we will in the rest of this subsection use the variable @xmath18 , although we should , to be correct , use @xmath273 in the local search case . in sec . [ sec : locnecess ] we properly use the variable @xmath273 again . let @xmath60 and @xmath276 defined in eq . ( [ basis ] ) correspond to the north and south pole of the bloch sphere , respectively . the bloch vector @xmath277 , corresponding to the instantaneous ground state , can then be expressed as @xmath278 note that @xmath277 moves in the @xmath45-@xmath279 plane only , and that @xmath280 . let @xmath281 be the angle between @xmath270 and @xmath282 . this angle can be expressed as @xmath283 for @xmath284 the right - hand side of the above equation is strictly decreasing on the interval @xmath285 , beginning at the value @xmath286 and ending at the value @xmath287 . it follows that we can choose @xmath281 as an increasing function with range in @xmath288 $ ] . since the system is started in the ground state @xmath289 . since both @xmath270 and @xmath290 are fixed to the @xmath45-@xmath279 plane it follows , from the form of eq . ( [ blochekv ] ) , that the motion of @xmath264 is fixed to the @xmath45-@xmath279 plane . thus , we may restrict our analysis to this plane only . [ lemins ] let @xmath291 , @xmath292 , and @xmath293 be three unit vectors in @xmath294 , and let @xmath264 be such that @xmath267 . suppose that the angle between @xmath291 and @xmath293 is greater than or equal to the angle between @xmath291 and @xmath292 . moreover , let both these angles be in the interval @xmath288 $ ] . then @xmath295 implies @xmath296 _ proof . _ consider a coordinate system such that @xmath291 corresponds to @xmath297 , @xmath292 to @xmath298 , and @xmath293 to @xmath299 , with @xmath300 . in the cases when @xmath301 assumes possible combinations of @xmath302 , the lemma is trivially true . hence , we assume @xmath303 . if we let @xmath304 , then eq . ( [ inputas ] ) becomes @xmath305 . by using @xmath306 , this can be rewritten as @xmath307 since cotangent is a decreasing function on the interval @xmath308 , and since @xmath309 , it follows that the right - hand side of eq . ( [ sfgn ] ) is larger than or equal to @xmath310 . the above procedure can be reversed to obtain eq . ( [ outputas ] ) . @xmath311 [ lemmabloch ] if @xmath312 is the solution of eq . ( [ blochekv ] ) with initial condition @xmath313 , and with @xmath270 as in eq . ( [ qutr ] ) , then @xmath314.\ ] ] _ proof . _ the strategy of this proof is to make a cauchy - euler polygon approximation of eq . ( [ blochekv ] ) , and show an inequality similar to eq . ( [ basolikhet ] ) for every step size . since the solution @xmath312 is obtained in the limit of small step sizes , the solution will satisfy the inequality in eq . ( [ basolikhet ] ) . let @xmath315 be a partition of the interval @xmath180 $ ] such that @xmath316 . the partition has the step size @xmath317 . let @xmath318 and let @xmath319.\end{aligned}\ ] ] let @xmath320 be the linear interpolation of @xmath321 , meaning that on the interval @xmath322 $ ] the function @xmath320 is defined by @xmath323 now we prove that the approximation @xmath320 converges uniformly on the interval @xmath180 $ ] to the solution @xmath312 , as @xmath324 . define @xmath325\}.\ ] ] in both the global and the local case the function @xmath326 can be shown to be continuous on @xmath327 , and to have a continuous partial derivative with respect to @xmath18 . if one uses the assumption @xmath218 in the global case , and @xmath328 as well as the additional assumption @xmath284 in the local case , one finds @xmath329 it follows that @xmath326 fulfills the lipschitz condition @xmath330\times\mathbb{r}^{3}.\ ] ] if we assume that @xmath331 , then eq . ( [ xicond ] ) implies @xmath332 ) and ( [ instangn ] ) one can show that @xmath333 . this in turn implies that the linear interpolation @xmath320 fulfills latexmath:[\[\label{spifub } @xmath333 , together with eq . ( [ xgn ] ) , can be used to show that @xmath335 this gives @xmath336.\ ] ] the fact that the partial derivative of @xmath326 , with respect to @xmath18 , is continuous on @xmath327 , implies that there exists a constant @xmath217 such that @xmath337 for all @xmath338 $ ] , and all @xmath339 . by combining eqs . ( [ lipschitz])-([mdfs ] ) one obtains @xmath340 for all @xmath341 $ ] . this can be used to show that @xmath342 for all @xmath338 $ ] , and where @xmath343 . since both @xmath326 and @xmath344 are continuous , eq . ( [ eapproximativ ] ) implies that @xmath344 is an @xmath345-approximative solution @xcite to eq . ( [ blochekv ] ) . since @xmath326 is continuous and fulfills the lipschitz condition in eq . ( [ lipschitz ] ) it follows @xcite from eq . ( [ eapproximativ ] ) that @xmath346 for all @xmath347 $ ] . since @xmath348 it follows from eq . ( [ nldsvk ] ) that @xmath320 converges uniformly on @xmath180 $ ] to the solution @xmath312 , as @xmath324 . ( one may note that the solution @xmath312 exists and is unique , since @xmath326 is continuous and lipschitz on @xmath180\times\mathbb{r}^{3}$ ] . ) the solution @xmath312 is continuous as it is the limit function of the continuous and uniformly converging functions @xmath320 on the subset @xmath180 $ ] of @xmath349 . ( see theorem 7.12 in ref . @xcite . ) in the following we focus on the proof of the inequality in eq . ( [ basolikhet ] ) . we use the fact that the vectors @xmath350 and @xmath351 are restricted to the @xmath45-@xmath279 plane of the bloch sphere . by induction we prove that @xmath352 . by using eq . ( [ xgn ] ) one finds @xmath353 if this is combined with the induction hypothesis @xmath352 we obtain @xmath354 the angle between @xmath355 and @xmath291 is larger than the angle between @xmath351 and @xmath291 , and these angles are in the range @xmath288 $ ] . according to lemma [ lemins ] with @xmath356 and @xmath357 , it follows from eq . ( [ enbn ] ) that @xmath358 since the induction hypothesis is true for @xmath359 , we obtain @xmath360 for all @xmath361 . this is true regardless of the choice of step size @xmath362 . suppose there exists a point @xmath363 for which the inequality @xmath364 does not hold . then there exists a number @xmath365 , such that @xmath366\cdot \bm{q}(\widetilde{s } ) = -a < 0 $ ] . by continuity of @xmath264 and @xmath277 , there must exist an interval @xmath367 such that @xmath368\cdot \bm{q}(s)\leq -a/2,\quad \forall s\in[\widetilde{s}_{1},\widetilde{s}_{2}].\ ] ] now consider approximations @xmath320 in this interval . @xmath369\cdot \bm{q}(s ) & \leq & [ \bm{v}^{(d\!s)}(s)-\bm{v}(s)]\cdot \bm{q}(s ) -a/2 \nonumber\\ & \leq & \sup_{s\in [ \widetilde{s}_{1},\widetilde{s}_{2}]}\big|\big|\bm{v}^{(d\!s)}(s ) -\bm{v}(s)\big|\big|\nonumber\\ & & -a/2.\end{aligned}\ ] ] since @xmath320 converges uniformly to @xmath312 , as @xmath324 , there exists a sufficiently small @xmath362 such that @xmath370 . thus , for a sufficiently small @xmath362 it follows that @xmath371\cdot \bm{q}(s)<0 $ ] , for all @xmath372 $ ] . this contradicts eq . ( [ olik ] ) , since we may choose @xmath18 to be some element of the partition @xmath315 inside the interval @xmath373 $ ] . ( such an element exists if @xmath362 is sufficiently small . ) hence , by contradiction we have proved the statement of the lemma . @xmath311 in the following we let @xmath374 denote the solution of eq . ( [ tvaanivaa ] ) for a given run time @xmath24 , with @xmath375 and @xmath228 . similarly we let @xmath376 denote the solution of eq . ( [ widoppen ] ) , for the run time @xmath24 . [ lemmatre ] @xmath377 _ proof . _ if @xmath378 , then , by construction , we have @xmath379 where @xmath380 is the bloch vector representing @xmath376 . similarly @xmath381 , where @xmath382 is the bloch vector corresponding to @xmath383 which is the solution of eq . ( [ widoppen ] ) with @xmath384 , and initial condition @xmath385 . the solution is @xmath386 . we may conclude that @xmath387 . hence , @xmath388 is only eq . ( [ basolikhet ] ) in disguise . @xmath311 if we combine eq . ( [ tvaanivaa ] ) with eq . ( [ yineq ] ) and let @xmath389 , @xmath390 , we find the following inequality @xmath391 where @xmath392}z(s')z(s'')y_{0}(s'')ds''ds'\ ] ] and where we have used that @xmath393 . this implies that if @xmath394 is to approach @xmath286 , then @xmath395 has to go to zero , or be negative , since @xmath396 . however , @xmath397 , as is shown below . equation ( [ ynoll ] ) yields @xmath398 by inserting eq . ( [ ynolikhet ] ) into eq . ( [ iintegdef ] ) and by making the change of variables @xmath399 and @xmath400 one obtains @xmath401}}{(1+x^{2})(1+y^{2})^{3/2}}dydx.\ ] ] in the above expression we have introduced the function @xmath402 . furthermore , @xmath403^{\sigma}ds ' \nonumber\\ & = & \alpha [ \phi(x)-\phi(-\sqrt{n-1})],\end{aligned}\ ] ] where @xmath404 and @xmath405 note that @xmath406 . by using eq . ( [ qdefeni ] ) one can write @xmath407.\ ] ] define @xmath408 @xmath409 from eq . ( [ mvbdiu ] ) it follows that @xmath395 fulfills the inequality @xmath410 the definition in eq . ( [ defiab ] ) yields @xmath411 the second equality in the expression above is obtained by separating the integrals `` @xmath412 '' into `` @xmath413 '' and make the change of variables @xmath414 ( and @xmath415 ) in the @xmath416 integrals . if one uses the inequality @xmath417.\ ] ] in eq . ( [ didb ] ) , one obtains @xmath418 since the integrand of the above integral is strictly positive , one can conclude that @xmath419 hence , for a fixed @xmath420 the function @xmath421 is strictly increasing . similarly , @xmath422 increases with @xmath1 . if we assume @xmath284 it follows that @xmath423 we can conclude that if @xmath424 , then @xmath425 , necessarily . this , in turn , implies @xmath426 . we now investigate how the last expression depends on @xmath420 . @xmath427 @xmath428 both terms on the right - hand side in the above expression are strictly less than zero , for @xmath429 . it follows that @xmath430 is a strictly decreasing function of @xmath420 . thus , a necessary condition for the integral to go to zero is that @xmath420 goes to infinity . due to the form of @xmath420 in eq . ( [ defalphaglobal ] ) it follows , for large @xmath1 , that @xmath431 is a necessary condition for @xmath432 . since @xmath432 is equivalent to @xmath3 , we have shown that the sufficient condition @xmath433 , proved in sec . [ sec : globsuff ] , is also necessary . the proof of the necessary condition in the local case resembles rather closely the proof of the global search case . however , we assume that @xmath253 . we let @xmath434 denote the solution of eq . ( [ tvaanivaa ] ) for a given run time @xmath24 , with @xmath375 and @xmath228 , and with @xmath154 replaced by @xmath156 defined in eq . ( [ wqwh ] ) . similarly we let @xmath435 denote the solution of eq . ( [ widoppenlocal ] ) , for the given run time @xmath24 . it is straightforward to prove the counterpart of lemma [ lemmatre ] @xmath436 similarly we can obtain @xmath437 with @xmath438 as in eq . ( [ iintegdef ] ) , but with @xmath154 replaced by @xmath156 . moreover , @xmath439 due to eq . ( [ ynolikhetlocal ] ) , eq . ( [ mvbdiu ] ) remains true if we replace @xmath395 with @xmath438 and @xmath440 with @xmath441 . we also obtain @xmath442,\ ] ] where @xmath443 and @xmath444 now we define @xmath445 as in eq . ( [ defiab ] ) but with @xmath420 replaced by @xmath446 , and @xmath447 , @xmath448 replaced by @xmath449 the line of reasoning from eq . ( [ iiineq ] ) to eq . ( [ sistaglobal ] ) is unaltered up to a replacement of @xmath450 , and @xmath421 with @xmath451 , and @xmath445 . we define @xmath452 in analogy with @xmath453 in eq . ( [ fdefineq ] ) . analogously to the global search case we find that a necessary condition for the limit @xmath454 is that @xmath455 , since @xmath456 is a strictly decreasing function in @xmath446 . for large @xmath1 , it follows , due to the form of @xmath446 in eq . ( [ defalphalokal ] ) that @xmath457 is a necessary condition for @xmath3 . thus , we have proved that the sufficient condition for adiabaticity , proved in sec . [ sec : locsuff ] , is also necessary . we have so far investigated the conditions for the success probability @xmath262 to approach unity , which echoes the formulation of the standard adiabatic theorem . however , for an actual implementation of the adiabatic quantum computer we have to be satisfied with a success probability less than unity in order to have a finite run time . it thus seems reasonable to ask how much of the results obtained for the @xmath3 limit survives when @xmath2 . consider the set of pairs of list lengths and run times @xmath458 that gives precisely the success probability @xmath262 . one may note that @xmath24 does not necessarily have to be a function of @xmath1 , as there may accidentally be several values of @xmath24 , for a fixed @xmath1 , which give the same @xmath262 . we show that the run time @xmath24 is bounded by curves with the same asymptotic behavior as found in the @xmath3 cases . we assume that the system initially is in the ground state , i.e. , @xmath459 and @xmath460 . we denote @xmath461 . the strategy is to obtain expressions of the form @xmath462 or @xmath463 . if @xmath464 and @xmath465 are continuous and strictly decreasing functions , we may invert them , on suitable domains , and obtain bounds @xmath466 . we derive both upper and lower bounds in the wide - open case . in the semi - open case we obtain upper bounds only . one may note that the upper bounds can be seen as sufficient conditions on the run time to achieve a success probability which is at least @xmath262 . an upper bound for the global wide - open case can be obtained directly from eq . ( [ globwo ] ) . the result is @xmath467 we obtain a lower bound by using results from sec . [ sec : globnecess ] . by combining eqs . ( [ grdlolik ] ) and ( [ samlat ] ) one obtains @xmath468 where @xmath469 is defined in eq . ( [ fdefineq ] ) , and @xmath420 in eq . ( [ defalphaglobal ] ) . we know from sec . [ sec : globnecess ] that @xmath217 is strictly decreasing in @xmath420 . moreover , @xmath217 is continuous . hence , @xmath217 is invertible and @xmath470 the bounds in eqs . ( [ wogluppett ] ) and ( [ wogllow ] ) fulfill the same @xmath471 rate as we found in the @xmath3 investigation . for the wide - open local case we may use eq . ( [ wioloc ] ) , and @xmath472 to obtain @xmath473 analogously to the global search case we can obtain a lower bound to the wide - open local search from sec . [ sec : locnecess ] , as @xmath474 with @xmath446 as in eq . ( [ defalphalokal ] ) . this results in the bound @xmath475 which , for large @xmath1 , increases as @xmath476 . we have found that for a fixed success probability @xmath262 the run time is locked between two bounds . these bounds have the same scaling with the list length @xmath1 as found in the @xmath3 limit . in the semi - open global search case we use eq . ( [ sogl ] ) , with @xmath460 and @xmath477 to obtain @xmath478 in the semi - open local search case one may use eq . ( [ soloc ] ) to obtain @xmath479 the bounds in eqs . ( [ boundsoglob ] ) and ( [ boundsoloc ] ) are sufficient conditions on the run time to obtain at least the success probability @xmath262 , for a given list length @xmath1 @xcite . we have numerically calculated the actual run times of the wide - open global and local search . let us choose @xmath480 and @xmath481 . this choice of @xmath482 and @xmath483 is merely a convenient way to obtain an interpolation between the closed case @xmath484 and the wide - open case @xmath485 , without effectively changing @xmath24 . moreover we have chosen @xmath231 . thus , we consider an environment that monitors the energy of the system . the initial condition at @xmath20 is @xmath486 with @xmath60 as in eq . ( [ eqalsuperpos ] ) . in other words , we begin at the initial ground state of the hamiltonian in eq . ( [ family ] ) . we should also comment upon the units . as seen in sec . [ sec : search ] we have put the energy gap between the ground state and the degenerate first excited state to @xmath286 in the initial hamiltonian @xmath58 . depending on the actual system used to realize the search procedure , the energy gap is actually @xmath487 in some appropriate energy unit . to put the energy gap to @xmath286 corresponds to measuring energy in units of @xmath487 . furthermore , as we have chosen @xmath4 , it follows that the run time @xmath24 , used throughout this paper , can be seen as the quotient @xmath488 , of the physical run time @xmath489 and @xmath490 defined by @xmath491 in fig . [ fig : globalpvar ] , the dotted curves correspond to the wide - open global search with success probability ranging from @xmath492 to @xmath493 . the curves seem to follow closely to a linear asymptote with slope @xmath494 , independent of the choice of @xmath2 . the dotted curves in fig . [ fig : localpvar ] similarly correspond the wide - open local search case . the asymptotic slope of these curves seems to be @xmath495 for every choice of @xmath2 . we have also calculated the actual run times of the semi - open global and local search numerically . again we use @xmath231 , @xmath496 , and @xmath497 . the results for the global case are shown in fig . [ fig : globalpvar ] , in form of the dashed curves corresponding to @xmath498 , and the solid curves corresponding to @xmath499 . all these curves seem to tend to the asymptotic slope @xmath495 , which is in line with our analytical results . in fig . [ fig : localpvar ] we similarly have created plots for the semi - open local search case . these again correspond to the dashed curves with @xmath498 , and the solid curves with @xmath500 . the results are qualitatively similar to the semi - open global search case , but with asymptotic slope @xmath501 . figures [ fig : globalw ] and [ fig : localw ] show the global and local search case for a fixed success probability @xmath502 but for different degrees of openness @xmath503 . figure [ fig : globalw ] suggests that in the semi - open cases the asymptotic slope is @xmath504 , while in the wide - open case the asymptotic slope is @xmath494 , which is consistent with our analytical results . in fig . [ fig : localw ] the curves in the local search case appears to have a qualitatively similar behavior to the global search case , except that the asymptotic slope is @xmath495 in the wide - open case , and @xmath501 in the semi - open case . from figs . [ fig : globalw ] and [ fig : localw ] one may obtain an intuitive understanding of the `` discontinuity '' of the scaling of the run times , when going from the semi - open to the wide - open case . the closer we get to the wide - open case ( i.e. , the closer @xmath503 gets to @xmath286 ) , the higher up in list - lengths the curve `` follows '' the curve of the wide - open case , before it turns to its asymptotic slope . in this context one should also note that for every fixed list - length @xmath1 , the system is continuous with respect to @xmath503 . in other words , the discontinuity only appears for the asymptotes in the @xmath505 limit , not in any system corresponding to a fixed list . the list - length @xmath1 corresponds to the dimension of the total hilbert space , but in the effective two - dimensional hilbert space the list - length only enters as a parameter in the hamiltonian . if we are to understand the discontinuity of the asymptotic scaling as a function of the degree of openness , in the case @xmath231 , we may consider how the parameter @xmath1 enters the terms @xmath506 $ ] and @xmath507 $ ] in the master equation . the minimum energy gap of the hamiltonian is proportional to @xmath508 . thus , at the minimum gap the list length @xmath1 enters as @xmath509 in the hamiltonian part of the master equation , but as @xmath510 in the decohering part . no matter how small @xmath482 is , as long as it is nonzero , @xmath509 always dominates @xmath510 in the limit of large @xmath1 . although this , strictly speaking , only holds at @xmath511 , it seems intuitively reasonable that the hamiltonian term dominates the evolution , for large @xmath1 , as long as @xmath512 . intuitively it also seems reasonable that if the hamiltonian dominates the evolution , then the hamiltonian determines the asymptotic scaling of the run time , and thus the scaling becomes the one of the ideal closed case . however , if @xmath375 then the decohering term determines the scaling . we investigate the effect of decoherence on the adiabatic search of a marked element in a disordered list of @xmath1 elements . more specifically we consider decoherence with respect to the instantaneous eigenbasis of the adiabatic search hamiltonian . the decoherence is modeled by using a master equation of the form @xmath513 -b\bm{[}w,[w,\rho]\bm{]}$ ] ( with @xmath514 ) , where @xmath10 is the time - dependent search hamiltonian and @xmath11 is a time - dependent hermitian operator that commutes with @xmath10 . the operator @xmath11 is only subjected to the condition in eq . ( [ condition ] ) , which , in some sense , states that the fluctuations of @xmath11 ( with respect to time ) should not grow with the list - length @xmath1 . we prove sufficient conditions for the @xmath3 limit of the success probability @xmath262 . in the semi - open case ( @xmath13 ) we prove that these sufficient conditions are @xmath515 and @xmath516 in the global and local search case , respectively , where @xmath24 is the run time of the adiabatic search . in other words , the results for the ideal adiabatic search ( @xmath94 ) remains valid even in the presence of decoherence with respect to the instantaneous eigenbasis . in the wide - open case ( @xmath227 ) , sufficient and necessary conditions for the @xmath3 limit , are deduced . in the case where @xmath16 we find that the conditions for @xmath3 are @xmath517 and @xmath141 for the global and local search , respectively . we finally investigate how the run time @xmath24 depends on @xmath1 for a fixed success probability @xmath262 less than unity . we find that it is sufficient with run times that scales like @xmath1 and @xmath14 , in the semi - open case , as for the ideal global and local adiabatic searches . in the wide - open case we instead find the scalings @xmath0 and @xmath1 , for the global and local search , respectively . in view of these results the following picture emerges . in the semi - open cases the hamiltonian dynamics dominates the behavior of the system for sufficiently large list lengths . to be more precise , an increased degree of eigenbasis decoherence does increase the run time of the adiabatic search , but only through the constant @xmath217 of the asymptote @xmath518 . in the wide - open case , however , the protective effect of the hamiltonian dynamics is not present . consequently , the asymptotic behavior depends directly on the choice of @xmath11 . as a consequence the asymptotic behavior of the wide - open case may be distinctly different from that of the semi - open case , although we have the same choice of @xmath11 . one may note that the abrupt change in asymptotic behavior with respect to the transition from semi - open to wide - open system , raises a warning sign concerning the use of the wide - open evolution as an approximation to an `` almost '' wide - open evolution . it should be noted , however , that for any fixed list length , the system is continuous with respect to the degree of openness , and that the discontinuity rather concerns the asymptotic scaling of the run time in the limit of large list - lengths . the decoherence model provided by eq . ( [ enkel ] ) implies that we disregard the memory effects in the environment and the corresponding fluctuations , which seems justifiable if the relevant time scales of the evolution is much larger than the dominating time scales of the memory of the environment . the modeling of memory effects generally leads to integro - differential equations @xcite , or to time - local non - markovian equations @xcite . to include such effects into the analysis of the influence of the environment on the efficiency of the adiabatic quantum computer would be interesting but challenging . finally we note that , in spite of the apparent difference between the adiabatic quantum computing scheme and the traditional circuit model , it has been shown that these two models are , in a certain sense , equivalent @xcite . however , this equivalence does not concern the robustness to noise , relaxation , or decoherence . previous findings suggest that the adiabatic quantum computer to a certain extent should be resilient against various kinds of open system effects . the results of this paper provide further evidence for this . if we consider hamiltonians with a time - independent eigenbasis , then randomness in the eigenenergies causes decoherence with respect to the eigenstates , when averaged over several instances . although this gives rise to decoherence in the eigenbasis , it is not necessarily of a form that can be expressed as in eq . ( [ enkel ] ) ; this depends on the statistics of the fluctuations . a time - independent two - level system with an energy gap that is drawn from an ensemble with lorentzian distribution , results in an averaged evolution that can be described by eq . ( [ enkel ] ) with time - independent @xmath10 and @xmath11 . ( [ enkel ] ) the eigenbasis of the hamiltonian depends on time . however , in the adiabatic limit many aspects of the above time - independent case seems likely to be retained . d. aharonov , w. van dam , j. kempe , z. landau , s. lloyd , and o. regev , _ proceedings of the 45th symposium on foundations of computer science _ ( ieee computer society press , new york , 2004 ) , p.42 ; e - print quant - ph/0405098 .

we investigate for the first time the effects of a warm dark matter ( wdm ) power spectrum on the statistical properties of galaxies using a semi - analytic model of galaxy formation . the wdm spectrum we adopt as a reference case is suppressed - compared to the standard cold dark matter ( cdm ) case - below a cut - off scale mpc corresponding ( for thermal relic wdm particles ) to a mass kev . this ensures consistency with present bounds provided by the microwave background wmap data and by the comparison of hydrodynamical n - body simulations with observed lyman- forest . we run our fiducial semi - analytic model with such a wdm spectrum to derive galaxy luminosity functions ( in b , uv , and k bands ) and the stellar mass distributions over a wide range of cosmic epochs , to compare with recent observations and with the results in the cdm case . the predicted color distribution of galaxies in the wdm model is also checked against the data . when compared with the standard cdm case , the luminosity and stellar mass distributions we obtain assuming a wdm spectrum are characterized by : i ) a flattening of the faint end slope and ii ) a sharpening of the cutoff at the bright end for . we discuss how the former result is directly related to the smaller number of low - mass haloes collapsing in the wdm scenario , while the latter is related to the smaller number of satellite galaxies accumulating in massive haloes at low redshift , thus suppressing the accretion of small lumps on the central , massive galaxies . these results shows how a adopting a wdm power spectrum may contribute to solve two major problems of cdm galaxy formation scenarios , namely , the excess of predicted faint ( low mass ) galaxies at low and - most of all - high redshifts , and the excess of bright ( massive ) galaxies at low redshifts . cosmology : theory cosmology : dark matter galaxies : formation

1201.1617

the present understanding of galaxy formation in a cosmological context is based upon the collapse via gravitational instability of tiny dark matter ( dm ) perturbations dominating the matter contribution @xmath4 to the energy density of the universe , with baryons contributing only a fraction @xmath5 . in such a context , the number density and the properties of galaxies with different mass ultimately arise from the power spectrum of dm perturbations at the epoch of recombination . for example , in a universe dominated by cold dark matter ( cdm ) the low thermal speed of dm particles makes the density perturbations gravitationally unstable down to mass scales negligible for galaxy formation , and structure formation proceeds bottom - up with the progressive collapse of larger and larger regions ; massive galaxies , groups and clusters form by subsequent aggregations . when coupled with the evidences for a positive vacuum energy @xmath6 dominating the energy - density of the universe ( with @xmath7 see spergel et al . 2007 ) , such a @xmath6cdm scenario appears to be consistent with a wide range of observables ; from the temperature anisotropies in the microwave background radiation ( komatsu et al . 2011 ) to the clustering of galaxies on large scales ( cole et al . . however , such observables directly probe scales from@xmath8 gpc down to @xmath9 mpc . at galactic and sub - galactic scales @xmath10 mpc , the computations based on n - body simulations pose several problems to the @xmath6cdm scenario . for example , the cdm haloes are predicted to have inner density profiles much steeper than those inferred from the rotation curves of real galaxies ( moore et al . 1999 , abadi et al . 2003 , reed et al . 2005 ; madau , diemand , kuhlen 2008 ; for observations see , e.g. , gentile et al . 2007 , mcgaugh et al . 2007 ) , the discrepancy being larger for dwarf galaxies ( see governato et al . 2007 ; abadi et al . 2003 ; de blok et al . 2008 ) . in addition , cdm haloes are predicted to contain a wealth of substructure , which we expect to observe as satellite galaxies within the haloes of galaxies and groups , in contrast with observations ( see klypin et al . 1999 , moore et al . 1999 ; see also mateo et al . 1998 ) . since at such small scales the model predictions depend on the complex physics involved in the process of galaxy formation , part of the solutions to the above problems may be of astrophysical nature . feedback processes such as heating or winds caused by supernovae explosions and heating due to the uv background may be effective in suppressing star formation in the low - mass haloes of satellite galaxies ( bullock et al . 2000 ; benson et al . 2002 ; somerville et al . 2002 ; kravtsov , gnedin & klypin 2004 ; governato et al . 2007 ; mashchenko , wadsley & couchman 2008 ) , and in flattening the the inner density profiles in both spirals and dwarfs ( see governato et al . 2007 ; 2010 ) . however , the origin of the above problems may also lie , at least partially , in the shape of the cdm power spectrum , which could over predict the amplitude of small - scale ( @xmath11 mpc ) perturbations . indeed , a strong indication that this may be the case comes from the statistical properties of galaxies and active galactic nuclei ( agn ) , usually investigated in the framework of @xmath6cdm through semi - analytic models of galaxy formation ( sams , see kauffman , white & guiderdoni 1993 , cole et al . 1994 ; monaco , salucci & danese 2000 , kauffmann & haenhelt 2000 ; granato et al . 2004 ; menci et al . 2006 ; croton et al . 2006 ; bower et al . 2006 ; marulli et al . 2008 ) , which connect the physical processes involving gas and star formation to the merging histories ( and hence to the power spectrum ) of dm haloes . all such models ( including different astrophysical feedback process ) concur in predicting an excess of small - mass galaxies compared to observations . in particular : i ) the predicted galaxy luminosity functions at low redshift are much steeper than observed unless a strong feedback is assumed ( see , e.g. , somerville , primack 1999 ; cole et al . 2000 ; menci et al . 2002 ) , while at high redshifts the larger escape velocities of galaxies ( due to their larger densities ) make the feedback inefficient in suppressing star formation in low - mass objects . this results into an over - prediction of faint galaxies increasing with redshift when compared , e.g. , with the evolution of the k - band luminosity function for @xmath12 ( see cirasuolo et al . 2010 ) or with the faint lyman - break galaxies ( lo faro et al . 2009 ) at @xmath13 ; the excess of star - forming low - mass galaxies at @xmath13 reflects into an excess of red low - mass galaxies at @xmath14 ( see , e.g. , croton et al . 2006 ; salimbeni et al . ii ) the number of galaxies with stellar mass @xmath15 m@xmath16 in the universe is systematically over predicted by all theoretical cdm models ( fontana et al . 2006 ; fontanot et al . 2009 ; marchesini et al . 2009 ) in the whole redshift range @xmath17 where the small - mass end of the mass function has been measured ( see also santini et al . 2012 ) . thus , the baryonic processes ( and the feedback in particular ) implemented so far in all galaxy formation models do not seem able to provide an explanation for the _ predicted _ excess . while it is still possible that the problem stems from a poor understanding of the laws governing star formation especially at high redshifts @xmath18 ( see , e.g. , krumholz & dekel 2011 ) , it is interesting to note that the over - predictions of low - mass objects in the cdm scenario concerns also the agn population ; recent estimates of the x - ray luminosity functions of agn extending down to log @xmath19(2 -10 kev)@xmath20 ( fiore et al . 2011 ) are in sharp contrast with the large density of faint agn predicted by analytic ( shankar 2009 ) or semi - analytic ( menci et al . 2006 , 2008 ) models for the co - evolution of galaxies and agns ( se also shankar & mathur 2007 ; wyithe & loeb 2003 ; lapi et al . 2006 , shen 2009 ; shankar 2010 and references therein ) , even when the observed luminosity functions are corrected for the estimated fraction of obscured objects ( e.g. , la franca et al . 2005 ) . although , individually , the above problems do not definitely rule out the cdm model due to the complexity of the baryonic physics relating dm haloes with observable properties , the concurring evidences listed above call for a more radical approach , to investigate the possibility that the conflicts are rooted in an excess of power of the cdm spectrum at small scales , and that density perturbations could be dumped below same characteristic scale @xmath21 . a natural mechanism for such a suppression is the free - streaming of dm particles on scales @xmath22 , where @xmath23 is the dm thermal speed at the time of matter - radiation equality @xmath24 . for large dm velocities @xmath25 ( corresponding to a neutrino with mass @xmath26 ev ) this leads to the well - known hot dm scenario ( for a review see primack & gross 2000 ) which would yield @xmath27 mpc , well above the most recent observational limits @xmath28 set by the comparison of the lyman-@xmath2 forest spectra at @xmath29 with the n - body simulations ( viel et al . 2005 ; 2008 ) . however , for dm particles with larger mass @xmath30 the streaming speed could be reduced so as to bring @xmath21 close to @xmath31 mpc ( corresponding to a mass scale @xmath32 ) , consistent with present bounds but still large enough to affect galaxy formation . if constituted by thermal relic particles , such warm dm ( wdm ) would have streaming velocities @xmath33 , where @xmath34 is the effective temperature determining the value of the wdm density @xmath35 , so that @xmath36 . thus , to lower the smoothing scale @xmath37 of a factor @xmath38 compared to the above hot dm case , the wdm particles should have @xmath39 kev if their density has to match the observed dm density parameter @xmath40 . this type of relic wdm was discussed by peebles ( 1982 ) , bond , szalay & turner ( 1982 ) , and could have a physical counterpart in the gravitino , the supersymmetric partner of the graviton , in gauge - mediated supersymmetry breaking models ( pagels , & primack 1982 ; see steffen 2006 for a recent review ) . another possibility is that wdm particles have never been in thermal equilibrium , as is the case , e.g. , for sterile neutrinos created from oscillations with active neutrinos ( olive & turner 1982 ; dodelson & widrow 1994 ; shi & fuller 1999 ; abazajian , fuller & patel , 2001 ; dolgov & hansen 2002 ) . a discussion on the wdm candidates can be found in colombi , dodelson , lawrence ( 1996 , and references therein ) . for scenarios based on composite dm see khlopov et al . ( 2008 ) . previous studies based on n - body simulations have shown that wdm can resolve some of the tension between theoretical predictions and observations ( bode et al . 2001 ; avila - reese et al . some simulations suggest that the density profiles of wdm haloes are shallower in the inner regions ( colin , valenzuela , avila - reese 2008 ; see also de vega , salucci , sanchez 2010 ) , and that wdm can provide an explanation for the observed kinematics of satellites of massive galaxies ( lovell et al . simulations focussed on the halo properties and mass distribution in wdm scenario ( see knebe et al . 2002 ; zavala et al . 2009 ; polisensky & ricotti 2011 , smith & markovic 2011 ) have shown that there is a sensible suppression in the number of haloes with @xmath41 . however , the effects of a wdm spectrum on the properties of galaxies at such mass scales would be best investigated with sams , since not only they allow to probe in detail the effects on the halo statistics for @xmath41 , but also they can investigate the chain of effects that such a suppression has on the baryonic processes involved in galaxy formation : a modification of the power spectrum changes the number and the formation epoch ( and hence the internal densities ) of the of high - redshift progenitors of local galaxies , an effect which propagates to gas cooling , to the disk sizes , and to star formation ; this also affects both dynamical friction and binary merging , the processes determining the fate of sub - haloes associated to satellite galaxies ( see , e.g. , somerville & primack 1999 ; menci 2002 ) , and consequently the stellar mass growth of large galaxies due to accretion of smaller objects ; finally , starbursts and agn accretion and feedback ( included in almost all state - of - the - art sams ) induced by galaxy interactions are affected by the change in the number and in the properties of sub - haloes . in the present paper we investigate for the first time the effects of a wdm power spectrum on the statistical properties of galaxy using a sam of galaxy formation . we adopt a wdm power spectrum with a power suppression ( compared to the cdm spectrum ) below a cut - off scale @xmath42 mpc , to retain consistency with present bounds provided by the microwave background wmap data and by the comparison of hydrodynamical n - body simulations with observed lyman-@xmath2 forest ( see viel et al . 2005 ) . in terms of the properties of the wdm particles , such a cut off scale @xmath42 mpc is achieved for a mass of dm particles @xmath39 kev ( see sect . 2 ) whose free streaming length is @xmath43 mpc , corresponding to the wavelength of the mode for which the linear perturbation amplitude is suppressed by a factor of 2 . we run our fiducial semi - analytic model ( menci et al . 2006 , 2008 ) with such a wdm power spectrum to derive galaxy luminosity functions ( in b , uv , and k bands ) and the stellar mass distributions over a wide range of cosmic epochs , to compare with recent observations ( sect . 3 ) ; the predicted color distribution of galaxies in the wdm model is also checked against the data . our results are discussed in sect . 4 , while sect . 5 is devoted to conclusions . the sam we use connects the physical processes involving baryons ( physics of gas , star formation , feedback , growth of supermassive black holes ) to the merging histories of dm haloes , ultimately determined by the dm initial power spectrum . in the present paper we use the same model developed by menci et al . ( 2005 , 2006 , 2008 ) with the same choice of free parameters ; the only change is that merging trees are computed in a wdm cosmology . this allows to single out the effects of changing the dm spectrum with the same baryon physics . note that in the present paper the tree resolution as been increased with respect to menci et al . ( 2005 ) to resolve dm masses down to @xmath44 m@xmath16 for both the wdm and the cdm case , to better investigate the effect of changing the spectrum on the low - mass end of the mass distribution . here we briefly recall the basic features of the model ( for a complete description see menci et al . 2005 , 2006 , 2008 ) , and we describe the power spectrum we use to compute the merging trees in the wdm cosmology . galaxy formation and evolution is driven by the collapse and growth of dm haloes , which originate from the gravitational instability of overdense regions in the primordial density field . this is taken to be a random , gaussian density field within the concordance cosmology " ( spergel et al . 2007 ) , for which we adopt round parameters @xmath45 , total matter density parameter @xmath46 ( with baryons contribution corresponding to @xmath47 and hubble constant ( in units of 100 km / s / mpc ) @xmath48 . the normalization of the spectrum is taken to be @xmath49 in terms of the variance of the field smoothed over regions of 8 @xmath50 mpc . adopting the exact best fit wmap7 values for the above parameters does not change our results appreciably . as cosmic time increases , larger and larger regions of the density field collapse , and eventually lead to the formation of groups and clusters of galaxies ; previously formed , galactic size condensations are enclosed . the corresponding merging rates of the dm haloes are provided by the extended press & schechter formalism ( eps , see bond et al . 1991 ; lacey & cole 1993 ) ; this depends on the variance of the power spectrum on a mass scale @xmath51 @xmath52 where @xmath53 is the linear power spectrum of dm perturbations at a wavelengths @xmath54 , @xmath55 is the total mass in a sphere of radius @xmath56 , and @xmath57 is the top - hat window function ( see peebles 1993 ) . the clumps included into larger dm haloes may survive as satellites , or merge to form larger galaxies due to binary aggregations , or coalesce into the central dominant galaxy due to dynamical friction ( see menci et al . 2005 , 2006 ) ; in addition satellite halos are partially disrupted as the density in their outer parts becomes lower than the density of the host halo within the pericentre of its orbit ( see menci et al . 2002 for details ) . since the above galaxy coalescence processes take place over time scales that grow longer over cosmic time , the number of satellite galaxies increases as the dm host haloes grow from groups to clusters . all the above processes are implemented in our model through a monte carlo realization of merging trees ( generated with eps probabilities depending on the spectrum through eq . 1 ) and of the dynamical processes affecting the sub - haloes once they are included into larger haloes , following the canonical prescriptions of sams . we do not expect our results to be appreciably affected by uncertainties in the description of the above processes . although some variance exists between the galaxy merging times predicted by n - body simulations and the different sams ( see de lucia et al . 2010 for a discussion ) , there is an overall agreement in terms of the mass build up . indeed , henriques et al . ( 2011 ) compare the evolution of the k - band luminosity distribution predicted by three different sams , including the one used here and the de lucia & blaizot ( 2007 ) model . the latter model , which adopts galaxy merging histories extracted from high - resolution n - body simulations , differs at most by a factor two in the luminosity of the brightest galaxies from the present sam ; since the major fraction of such effect is due to the inclusion of a radio mode agn feedback in the de lucia & blaizot ( 2007 ) model ( contributing to suppress the k - band luminosities by about 1 magnitude , see croton et al . 2006 ) , the difference in the growth of massive galaxies due to uncertainties in the treatment of dynamical processes of sub - haloes is negligible . similar conclusions have been presented by fontanot et al . ( 2009 ) . while the above dynamical sector of the model is entirely determined by the initial power spectrum , the observable properties of galaxies are determined by the physics of baryons , which is related to the dm merging trees through the canonical approach of sams . the radiative gas cooling , the ensuing star formation and the supernova events with the associated feedback occurring in the growing dm haloes are computed for each sub - halo hosting the galaxy . the cooled gas with mass @xmath58 settles into a rotationally supported disk with radius @xmath59 ( typically ranging from @xmath31 to @xmath60 kpc ) , rotation velocity @xmath61 and dynamical time @xmath62 , all related to the dm sub - halo mass . the gas gradually condenses into stars at a rate @xmath63 ; the stellar ensuing feedback returns part of the cooled gas to the hot gas phase at the virial temperature of the halo . an additional channel for star formation implemented in the model is provided by interaction - driven starbursts , triggered not only by merging but also by fly - by events between galaxies ; such a star formation mode provides an important contribution to the early formation of stars in massive galaxies , as described in detail in menci et al . ( 2003 , 2005 ) . the model also describes the growth of black holes ( bhs ) inside each dm halo from primordial seeds with mass @xmath64 through merging between galaxies and accretion of a fraction of the cold galactic gas @xmath58 . the accretion is triggered by galaxy encounters , and the fraction of cold gas funnelled to the central bh is derived from the model by cavaliere & vittorini ( 2000 ) . the active accretion phase of bhs corresponds to the agn , which release part of their radiation energy in the surrounding interstellar medium , thus heating and expelling part of the galactic gas ( lapi , cavaliere , menci 2005 ) ; the implementation of such a form of agn feedback in the semi - analytic model is described in detail in menci et al . ( 2008 ) . detailed calculations of the free - streaming scale @xmath21 yield ( see smith & markovic and references therein ) @xmath65^{1/3}\,\big[{m_x\over { \rm kev}}\big]^{-4/3}\,{\rm mpc}\ ] ] the lighter is the dm particle the larger is the free - streaming length @xmath21 , corresponding to the scale below which the perturbations are wiped out . in the case wdm is composed by relic thermalized particles , the suppression ( with respect to the cdm case ) of density perturbations in wdm scenarios is quantified through the ratio of the two linear power spectra ( transfer function ) which can be parametrized as ( bode , ostriker & turok 2001 ) @xmath66^{-5\,\mu},\ ] ] where the cdm power spectrum @xmath67 is computed following bardeen et al . fitting integrations of the full einstein - boltzmann integration ( see viel et al . 2005 ) yields @xmath68 and @xmath69^{0.11}\ , \big[{m_x\over { \rm kev}}\big]^{-1.11}\ , \big[{h\over 0.7}\big]^{1.22}\,h^{-1}\,{\rm mpc}.\ ] ] the above relation between transfer function and dm particle mass @xmath30 is valid for thermal relics ; a similar relation holds for sterile neutrinos provided one substitutes the mass @xmath30 with a mass @xmath70 ( but see kusenko 2009 for variations due to their mode of production ) . in both cases , the smaller the wdm mass @xmath30 ( or @xmath71 ) the larger the suppression with respect to the standard cdm spectrum however , observational upper limits on the power suppression ( corresponding to lower limits on @xmath30 ) have been derived by viel et al . ( 2005 ) by comparing the observed lyman-@xmath2 forest in absorption spectra of quasars at @xmath72 with the results of n - body simulations run assuming different wdm power spectra of perturbations . as a result , for thermal relics a lower limit @xmath73 kev has been obtained , corresponding to @xmath74 kev . since our aim is to estimate the possible impact of wdm on galaxy formation , we shall assume the largest suppression of the dm power spectrum which is still safely above the constraints set by the lyman-@xmath2 forest ; thus we adopt as a reference case a wdm particle mass @xmath1 kev , and compute the transfer function after eq . the corresponding power spectrum , which constitutes the basic input for the merging trees at the basis of our semi - analytic model , is shown in fig . 1 . and compared with the standard cdm case . the power spectrum determines the merging histories and the mass distribution of dm haloes as follows ; starting from a grid of parent dm halo masses at redshift @xmath75 ( whose abundance is determined from the standard press & schechter 1974 mass function ) , we fragment them into progenitor masses at earlier cosmic times through a monte carlo simulation , with fragmentation probabilities given by the eps theory ( again dependent on the power spectrum through the variance in eq . 1 ) . by construction , the eps probabilities lead to merging trees where - at each time level - the average number of generated dm haloes of given mass reproduces the press & schechter mass distribution at the corresponding cosmic time . variants in this process associated with a modified press & schechter approach are discussed in sec . 4 . on the basis of such dm merging trees , we run our sam . we start by populating with baryons the dm haloes at the highest redshifts ( @xmath76 ) in each dm merging tree ; then we compute all the physical processes involving baryons as described in sect . 2.1 , thus associating a galaxy ( with disk gas , stars and a star formation rate ) to each dm halo ; when the dm haloes merge , the galaxies associated with the progenitor dm haloes may coalesce into a larger galaxy or survive as satellites according to the processes described in sect . 2.1 . [ 0.55 ] we now proceed to investigate the effect of implementing the wdm power spectrum on galaxy formation . we show in fig . 2 the low redshift b - band galaxy luminosity function , and its uv counterpart at high redshift @xmath77 ; the luminosity functions in wdm cosmology are compared with data and with the corresponding cdm case . a first effect of adopting the wdm spectrum is a decrease in the number of faint objects at both low and high redshifts . this is expected due to the suppression in the amplitude of initial perturbations ( see fig . quantitatively , the suppression of a factor @xmath78 for @xmath79 with respect to the cdm case is similar to that obtained for the halo mass function by smith & markovic ( 2011 ) for dm halo masses @xmath80 m@xmath16 for the same value of @xmath30 , and directly derives from the smaller number of low mass dm halos that collapsed at high redshift and survived to form faint galaxies . however , a second interesting effect of adopting a wdm power spectrum appears at low redshifts ( fig . 2 , left panel ) ; at bright magnitudes , the luminosity function shows a steeper cut off compared with the cdm case . this is due to the fact that in large dm haloes at low redshift the growth of massive galaxies due to the accretion of low mass objects is suppressed due to the smaller number of satellite galaxies which accumulate in the host dm halo . [ 0.5 ] [ 0.5 ] thus , changing the power spectrum in a way consistent with present observational limits seems not only to provide a viable solution for the over prediction of low - mass objects typical of cdm cosmology , but also to contribute to solve a long - standing problem of cdm models , namely , the over prediction of bright galaxies at low redshifts . the latter problem has been only recently alleviated by the inclusion of the radio mode feedback ( see , e.g. , bower et al . 2006 , cattaneo et al . 2006 ; croton et al . 2006 ; kang , jing , silk 2006 ) ; however , at present such a feedback , associated with a low - accretion state of bhs in massive galaxies ( presumably at the origin the radio activity ) , does not have a clear observational counterpart , since the observed number of radio galaxies is much smaller than that predicted by sams ( which need to associate such a radio mode accretion to each massive galaxy with stellar mass @xmath81 @xmath82 , see fontanot et al . 2011 ) ; in addition , at present the implementation of a radio mode still leads to predict a blue fraction of central galaxies that is too high and with an inverted luminosity dependence compared with what observed ( see weinmann et al . 2006 ) . [ 0.6 ] [ 0.5 ] note that the proposed approach leaves unchanged the baryonic mechanisms affecting the faint end ( the supernovae feedback ) and the bright end ( the agn feedback ) of the galaxy luminosity distribution . indeed , we have checked that adopting a wdm spectrum leaves unchanged the properties of the predicted color - magnitude diagram , which is still characterized by a bimodal distribution in agreement with observations ( see strateva et al . 2001 ; baldry et al . 2004 ) which start to appear already at @xmath83 as observed ( see bell et al . 2004 , giallongo et al . 2005 ) ; a quantitative comparison with detailed observed color distribution for different magnitudes ( fig . 3 bottom panel ) shows an overall agreement with existing measurements , although a moderate excess of bright ( @xmath84 ) galaxies with @xmath85 is still present . note that such excess would be larger ( to include a fraction @xmath86 of bright galaxies ) in the absence of agn feedback . to investigate how the two effects of adopting wdm power spectrum evolve with redshift , and thus to establish how the different number of satellites in common dm haloes affect the mass growth of galaxies in the wdm vs. the cdm case , we compare with statistical observables more directly related to the stellar mass . the evolution of the k - band luminosity function in the wdm case is compared with that derived for the cdm cosmology and with the observational data in the top panel of fig . 4 . ; the same comparison is performed for the evolution of the stellar mass function in the bottom panel . these two observables are related in the model since the first is computed by convolving the stellar spectral energy distributions ( seds ) with the star formation history of the galaxy progenitors , while the latter is the time integral over the same history . the stochastic nature of the merging and star formation histories results in a whole distribution of computed galaxies in the @xmath87 plane that we have checked to be consisted with the observed distribution . on the observational side , the two quantities are measured in different ways . the k - band luminosity functions are taken from the ultra deep survey ( uds ) , the deepest survey from the ukirt infra - red deep sky survey ( ukidss ) , containing imaging in the j - and k - bands , with deep multi - wavelength coverage in @xmath88 @xmath89 @xmath90 @xmath91 @xmath92 filters in most of the field . the sample contains 50,000 objects with high completeness down to @xmath93 ( cirasuolo et al . 2010 ) . as for the observed stellar mass functions , we compare mainly with the data from santini et al . ( 2012 ) ; stellar masses were estimated by fitting a 14 bands photometry ( up to 5.5 @xmath94 m rest - frame ) to the bruzual & charlot synthetic models of stellar populations ; the analysis of observations taken with the hubble wide field camera 3 in the goods - s field allows to achieve an excellent determination of the low - mass - end of the distribution down to small stellar masses @xmath95 m@xmath16 even at the highest redshifts ( @xmath96 ) . note that , while the stellar mass function constitutes a direct probe of the effect of wdm spectrum on the growth of stellar mass of galaxies , observational determinations of such a quantity are prone to several observational uncertainties connected , e.g. , to the estimate of metallicities or extinction curves necessary to derive the stellar masses , to the treatment of the tp - agb phase , or to the reconstruction of the star formation history of each galaxy , that is necessary to estimate the appropriate @xmath97 ratio and that may be poorly described by simplistic models like those adopted in the stellar synthesis codes ( maraston et al . 2010 ; lee et al . although the above systematic uncertainties ( not included in the error bars in the bottom panel of fig . 4 ) do not allow to definitely rule out any of the models on the basis of the stellar mass distribution , its flat logarithmic slope at small masses ( less prone to the above systematics , see marchesini et al . 2009 ) is better matched by the wdm model . [ 0.67 ] fig . 4 . - top panel . the evolution of the k - band luminosity functions in the wdm cosmoligy ( solid line ) is compared with data from cirasuolo et al . dashed line refer to the standard cdm case . + bottom panel the evolution of the stellar mass function in the wdm cosmology ( solid line ) is compared with the standard cdm case ( dashed ) ; data in the leftmost panel are from drory et al . ( 2004 , squares ) and fontana et al . ( 2006 , circles ) ; data in all remaining panels are from santini et al . + in both the upper and the lower left panels , the arrows show the range of magnitudes and stellar masses corresponding to the free streaming mass ; the dispersion characterizing the above relations are related to the stochastic nature of the merging trees . thus , also in this case both effects ( flattening of the faint end slope and sharpening of the cutoff at the bright end ) are effective at low redshift , providing a substantial improvement of the fit to the observations . at higher redshifts , the flattening at the faint end remains an approximatively constant feature , being related to the smaller number of low mass dm halo collapsed in wdm cosmology ; although models still slightly overestimates the number of small - mass galaxies ( with @xmath98 ) at @xmath99 , the agreement is appreciably improved by the adoption of a wdm spectrum , a result difficult to achieve adopting different feedback or star formation recipes within the cdm framework ( see the comparison with other sam in cdm cosmology in fontanot et al . 2009 ; santini et al . 2012 ) . interestingly , the effect of adopting a wdm spectrum on the bright end appears at redshift @xmath100 ; this indicates that it is indeed related to the later phase of stellar mass growth in massive galaxies , that associated with the accretion of small lumps onto a central dominant galaxy . in fact , earlier comparison of different sam predictions with observed k - band luminosity functions ( see henriques et al . 2011 ) have shown that the cdm over prediction of massive galaxies is caused by excessive mass growth in this redshift range . to investigate this point in detail , we first show in fig . 5 ( left panel ) the number of satellites ( with mass @xmath101 m@xmath16 ) predicted by our sam in the wdm case ( normalized to that predicted in the cdm case ) as a function of the host dm halo mass ; we also compare our results with n - body simulations with the same lower mass limit for satellites ( bode et al . note that the distribution of the satellite suppression factor in fig . 5 ( left panel ) is close to that computed from n - body simulations , although the comparison can only be approximate since the distribution of the suppression factor from the n - body has been compute by bode et al . ( 2001 ) for wdm particles with mass @xmath102 kev , while the recent simulations by kamada ( 2011 ) and polisensky & ricotti ( 2011 ) - both with wdm particle mass @xmath103 kev- have been computed only for milky way - size haloes . a suppression factor @xmath104 for milky way - size has been found also for @xmath105 kev by colin et al . ( 2000 ; see also knebe et al . 2002 ) . [ 0.5 ] [ 0.6 ] thus , our sam seems to provide a reasonable description of the effect of wdm on the the sub - halo population . the predicted suppression of the satellite number in the wdm case impacts the growth of massive galaxies since : i ) they are hosted in larger and larger haloes for decreasing redshifts , ii ) their merging histories is dominated by the accretion of much smaller , satellite galaxies for @xmath106 ( see baugh 2006 ) , and iii ) their low star formation at low redshift affects their stellar mass growth only mildly . such an impact , and why it becomes appreciable for @xmath106 , is better shown in the right panel of fig . 5 , where we show the fraction of stellar mass gained by accretion of satellites by galaxies with different stellar mass found at different redshift ; the color code shows such a fraction in the wdm case normalized to that in the cdm case . note that at high redshifts , when galaxies have small stellar masses , the accretion of satellites in wdm cosmology is similar to that predicted in cdm scenarios , since at high redshifts merging between comparable clumps is more frequent , the number of satellites is however small , and thus the impact of minor merging on the accretion of stellar mass is small . however , as cosmic time proceeds ( for @xmath106 ) , satellite galaxies progressively accumulate in larger and larger dm haloes , while the merging histories of massive objects are dominated by the infall of small mass clumps ; at this stage , the lower number of satellites predicted in the wdm scenario effectively suppresses the accreted stellar mass with respect to the cdm case . to better illustrate the epoch when such a suppression is effective , we also show in fig . 5 ( right panel ) the growth history of three typical massive galaxies ( with mass @xmath107 at @xmath75 ) ; note how all such paths cross the transition region ( where the suppression of accretion in the wdm becomes appreciable , see the color scale ) at redshifts @xmath108 . this explains why for @xmath3 the population of massive galaxies shows a slower increase in the wm d case compared to cdm , as shown by fig . we turn now to discuss the robustness of our results and conclusions when the uncertainties in the theoretical modelling of the abundance of low - mass dm haloes in wdm are considered . in the previous section , we adopted a straightforward extended press & schechter approach to compute the abundance of haloes for a wdm power spectrum , and to generate the dm merging trees through a monte carlo sam . however , at present , uncertainties exist in the determination of the abundance of halos with masses smaller than the free streaming mass @xmath109 ( where @xmath21 is given by eq . 2 and @xmath110 is the average wdm density ) . there is some debate as to whether haloes with such masses can form at all ( bode et al . 2001 , wang & white 2007 ) ; recent results from numerical simulations ( see wang & white 2007 ; zavala et al . 2009 ) show that , when the effects of spurious clumps associated with the limited resolution of simulations are accounted for , a strong suppression in the number of haloes with @xmath111 occurs ( see also schneider et al . 2011 ) . thus , it is important to explore the maximal effect that such an uncertainty can have on our results . to this aim , we assumed a cut off in the halo abundance for @xmath111 . following the approach in smith & markovic ( 2011 ) , we adopted an _ ad hoc _ suppression factor in the form of an error function , so that the mass function @xmath112 in wdm becomes @xmath113 \big\}\,{dn \over d log m}~,\ ] ] where @xmath114 controls the logarithmic width of the step ( see left panel of fig . thus , we run our monte carlo computations and the whole semi - analytic model after suppressing the abundance of dm haloes with masses @xmath111 after the above equation , assuming @xmath115 ( as in smith & markovic 2011 ) or @xmath116 ( corresponding in practice to a complete suppression of the abundance of dm haloes with mass below the free - streaming mass ) . to implement such a suppression in our monte carlo realizations of merging trees , we first run a monte carlo simulation with probabilities given by the eps theory with the wdm power spectrum ( see sect . 2.1 and . 2.2 ) . this generates a set of merging trees where - at each time level - the average number of generated dm haloes as a function of mass reproduces the standard press & schechter mass distribution @xmath117 at the corresponding cosmic time ; then , before running the full sam on such merging trees , we suppress the number of low - mass halos in the trees by randomly eliminating haloes of mass @xmath51 with a suppression probability @xmath118 . by construction , we end up with dm merging trees where the abundance of dm halos follows eq . note that , when @xmath116 is adopted in eq . ( 5 ) , @xmath119 for all haloes with mass @xmath41 , leading in practice to a complete removal of sub-@xmath120 dm haloes from our merging trees . [ 0.67 ] the effects on the low - redshift luminosity function and on the stellar mass functions at @xmath121 are shown in the fig . the overall result is a further flattening of the luminosity / stellar mass functions , caused by the lack of low - mass dm haloes . however , when passing from the abundance of dm halos to galaxy counting ( in luminosity or stellar mass bins ) , the effect of suppressing the abundance of sub-@xmath120 haloes is smeared out over a range of luminosities / stellar masses due the concurrence of two effects . on the one hand , the effect of suppressing the abundance of sub-@xmath120 haloes is not concentrated on lowest - luminosity objects , since - even when sub-@xmath120 haloes are allowed to form at some redshifts - only a fraction of the galaxies formed in them remain as an isolated clump , the remaining fraction being included into a larger galaxy ( with a larger luminosity / stellar mass ) at some lower redshifts . on the other hand , the stochastic nature of the star formation histories in sams results a ) in a relevant fraction of the faintest galaxies to form in halos with masses @xmath122 ( whose abundance is not suppressed ) , and b ) in a relevant fraction of sub-@xmath120 halos ( when they are allowed to form ) to host galaxies with extremely low stellar mass and star formation rate , so that they contribute only to luminosity and stellar masses well below the observational threshold . thus , the theoretical uncertainty on the abundance of dm halos with @xmath41 results in the uncertainty regions for the predicted luminosity and stellar mass functions shown in fig . 6 . a similar degree of uncertainty is obtained for all the luminosity functions shown in sect . 4 , illustrating how our basic conclusions are robust with respect to present uncertainties in the modelling of sub-@xmath120 wdm haloes . in this paper we investigated the effect of assuming a wdm power spectrum on galaxy formation using a sam , leaving unchanged the baryonic mechanisms already implemented in our previous cdm model ( in particular the stellar and the agn feedback ) ; this allows to single out the effects of changing the dm spectrum with the same baryon physics . we focussed on the statistical distribution of luminosity and stellar masses , leaving to next paper a full study of the effects of wdm on other properties of the galaxy population ( e.g. , chemical abundances , ages of stellar populations , effects of the environment ) . within the present study , we find that adopting power spectrum corresponding to wdm particles with mass @xmath123 kev may contribute to solve two major problems of cdm galaxy formation scenarios , namely , the excess of predicted faint ( low mass ) galaxies at low and especially at high redshifts , and the excess of bright ( massive ) galaxies at low redshifts . in fact , adopting a wdm spectrum on the one hand results in a smaller number of collapsed low - mass haloes ; on the other hand it results into a smaller number of satellite galaxies accumulating in massive haloes at low redshifts , thus suppressing the accretion of small lumps on the central , massive galaxies for @xmath3 . such conclusions are robust with respect to the uncertainties in the modelling of the wdm halo abundance around the free streaming mass scale , as discussed in sect . 4 . our results on the luminosity and stellar mass distributions of galaxies concur to indicate that particles with mass @xmath42 kev constitute interesting candidates for dm ; such an indication adds to the existing astrophysical evidences from ( see references in the introduction ) : a ) the observed galaxy structures at small scales ( @xmath124 kpc ) ; b ) the cored profiles of galaxies ; c ) the number of sub haloes and satellites in the haloes of massive galaxies . directly testing whether dm is constituted by such particles , is a realistic possibility for the near future . a cosmological direct probe of wdm could be provided by cosmic shear ( weak gravitational lensing ) , which has the advantage of being weakly dependent on baryonic physics . first results in the simple relic scenario indicate that future lensing surveys like euclid could see wdm signals for @xmath125 kev ( smith & markovic 2011 ) , close to the mass adopted here . if the wdm is constituted by sterile neutrinos , as advocated to account for the non - zero active neutrino mass ( for a review see boyarsky , ruchayskiy & shaposhnikov 2009 ) , astrophysical bounds on its mass can be placed because it would decay in a standard neutrino plus a photon with energy of the order of the kev , thus producing narrow line emission in x - rays . claims for detection of lines hardly identifiable with known transitions have been published using both chandra and suzaku observations of dwarf galaxies and the galactic centre ( loewenstein & kuseno 2010 , prokhorov & silk 2010 ) . both the claimed lines are at energies where the effective area is strongly reducing ( 8.7 kev suzaku ) or near the gold m edge feature of the chandra mirror ( 2.51 kev ) . in the case wdm is constituted by sterile neutrinos direct detection is also possible through triple beta decay experiments like mare and katrin ( see de vega et al . 2011 ) . as for the case of thermal wdm , the favourite candidate is the gravitino . it emerges as a natural candidates for the lightest supersymmetric ( lsp ) particle in gauge - mediated supersymmetry breaking models ( see dine , nelson , nir , shirman 1996 ) with conserved r parity . contrary to the gravity - mediated supersymmetry breaking case , where the lsp ( typically the neutralino ) has masses exceeding the 10 tev scale , the gravitino mass can range from a fraction of ev up to o(gev ) . however , if it constitutes the lsp with mass @xmath123 kev , a large value @xmath126 for the degrees of freedom would be needed at the time of its decoupling to comply with the observed matter density @xmath127 ( pagels & primack 1982 ) ; this is larger than @xmath128 expected in minimal supersymmetric standard model ( mssm ) . of course , scenarios with @xmath129 exceeding the mssm expectations are conceivable , as well as scenarios where entropy - producing processes could dilute the gravitino abundance ( fujiii and yanagida 2002 ) , or where vanishing gauge coupling at high temperature suppresses the gravitino production ( buchmuller , hamaguchi & ratz 2003 ; see also lemoine , moultaka , jedamzik 2006 ) . a gravitino mass @xmath123 kev effective to provide a viable wdm scenario for galaxy formation would correspond to a relatively low energy scale for supersymmetry breaking @xmath130 gev ( here @xmath131 is the reduced plank mass ) . while its extremely weak coupling with matter ( @xmath132 , see lee & wu , 1999 ) makes gravitino dm inaccessible to direct searches , upcoming data from colliders may allow for signals from light gravitinos with mass in the ev to mev range ( see feng , kamionkowski & lee 2010 ) . light gravitinos would be produced primarily in the decays of the next - to - lightest supersymmetric particle , resulting in a variety of signals , including di - photons , delayed and nonpointing photons , kinked charged tracks , and heavy metastable charged particles . if light gravitino constitutes wdm , the 7 tev lhc with 1 fb@xmath133 could see evidence for hundreds of light - gravitino events . our results have straightforward implications concerning the growth of supermassive black holes and the evolution of agns . in cosmological galaxy formation theories , these are tightly related to the evolution of their host galaxies , as indicated also by the observed correlation between the mass of supermassive black holes and global properties of their host galaxies , like the stellar mass or the velocity dispersion ( kormendy & richstone 1995 ; magorrian et al . 1998 ; gebhardt et al . 2000 ; ferrarese & merritt 2000 ) . as recalled in the introduction ( see references therein ) , the over - predictions of low - mass objects in the cdm scenario concerns also the agn population ; assuming a wdm power spectrum could substantially alleviate the problem , analogously to what we showed to happen for the galaxy population . note that , contrary to what we derived for low redshift massive galaxies , we do not expect bright agns to be affected by assuming a wdm spectrum . this is because such objects are thought to be powered by gas accretion triggered by galaxy encounters with partners of comparable mass ( extremely rare at low redshift ) , only mildly affected by the abundance of low mass satellites . we shall investigate this points in a next paper . abadi , m. g. , navarro , j. f. , steinmetz , m. eke , v. r. 2003 , apj 591 , 499 abazajian , k. , fuller , g.m . , patel , m. 2001 , phys . d , 64 , 02351 avila - reese , v. , colin , p. , valenzela , o. , donghia , e. , firmani , c. , 2001 , apj , 559 , 516 baldry , i.k . , glazebrook , k. , brinkmann , j. , zeljko , i. , lupton , r.h . , nichol , r.c . , szalay , a.s . 2004 , apj , 600 , 681 bardeen j. m. , bond j. r. , kaiser n. , szalay a. s. , 1986 , apj , 304 , 15 baugh , c.m . 2006 , rep . prog . phys . 69 , 3101 bell , e. , wolf , c. , meisenheimer , k. , rix , h .- w . , borch , a. , dye , s. , kleineinrich , m. , mcintosh , d. 2004 , apj , 608 , 752 benson , a.j . , frenk , c.s . , lacey , c.g . , baugh , c.m . , cole , s. 2002 , mnras , 333 , 177 blanton , m.r . 2001 , apj , 121 , 2358 bode , p. , ostriker , j.p . , turok , n. 2001 , apj , 556 , 93 bond , j.r . , szalay , a.s . , turner , m. s. 1982 , phys . 48 , 1636 bond , j.r . , cole , s. , efstathiou , g. , & kaiser , n.,1991 , apj , 379 , 440 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 boyarsky , a. , ruchayskiy , o. , shaposhnikov , m. 2009 , annual review of nuclear and particle science , 59 , 191 bouwens , r.j . , illingworth , g.d . , franx , m. , ford , h. 2007 , apj , 670 , 928 buchmuller , w. , hamaguchi , k. & ratz , m. 2003 , phys . b , 574 , 156 bullock , j.s . , kravstov , a.v . , weinberg , d.h . 2000 , apj , 539 , 517 cattaneo , a. , dekel , a. , devriendt , j. , guiderdoni , b. , blaizot , j. , mnras , 370 , 1651 cavaliere , a. , & vittorini , v. 2000 , apj , 543 , 599 cirasuolo m. , mclure r. j. , dunlop j. s. , almaini o. , foucaud s. , simpson c. , 2010 , mnras , 401 , 1166 cole , s. , aragon - salamanca , a. , frenk , c.s . , navarro , j.f . , zepf , s.e . , 1994 , mnras , 271 , 781 cole , s. , lacey , c.g . , baugh , c.m . , frenk , c.s . , 2000 , mnras , 319 , 168 cole , s. et al . 2005 , mnras , 362 , 505 colin , p. , valenzuela , o. , avila - reese , v. , 2000 , apj , 673 , 203 colombi , s. , dodelson , s. , widrow , l.m . 1996 , apj , 458 , 1 croton , d.j , springel , v. , white , s.d.m . , de lucia , g.,frenk , c.s . , gao , l. , jenkins , a. , kauffmann , g. , navarro , j.f . , yoshida , n. , 2006 , mnras , 365 , 11 de blok , w.j.g . ; walter , f. , brinks , e. , trachternach , c. , oh , s .- h . , kennicutt , r.c . , 2008 , aj , 136 , 2648 de vega , h.j . , salucci , p. , sanchez , n.g . 2010 preprint ( arxiv:1004.1908 ) de vega , h.j . 2011 , prepriny [ astro - ph/1109.3452 ] de lucia g. , blaizot j. , 2007 , mnras , 375 , 2 de lucia g. , boylan - kolchin m. , benson a. j. , fontanot , f. , monaco p. , 2010 , mnras , 406 , 1533 dine , m. , nelson , a. , nir , y. & shirman , y. 1996 , phys . rev . d53 , 2658 dodelson , s. , widrow , l.m . 1994 , phys . 72 , 17 dolgov , a.d . , hansen , s.h . 2002 , astropart . , 16 , 339 drory , n. , bender , r. , feulner , g. , hopp , u. , maraston , c. , snigula , j. , hill , g. j. 2004 , apj , 608 , 742 feng , j.l . , kamionkowski , m. , lee , s.k . 2010 , phys . d , 82 , 015012 ferrarese , l. merritt , d. , 2000 , apj , 539 , l9 fiore , f. et al . 2011 , a&a , 537 , 16 fontana a. et al . , 2006 , a&a , 459 , 745 fontanot , f. , de lucia , g. , monaco , p. , somerville , r.s . , santini , p. 2009 , mnras , 397 , 1776 fontanot , f. , pasquali , a. , de lucia , g. , van den bosch , f.c . , somerville , r.s . ; kang , x. 2011 , mnras , 413 , 957 fujii , m. & yanagida , t. 2002 , phys . lett.b , 549 , 273 gebhardt , k. et al . 2000 , apj , 539 , l13 giallongo , e. , salimbeni , s. , menci , n. , zamorani , g. , fontana , a. , dickinson , m. , cristiani , s. , pozzetti , l. 2005 , apj , 622 , 116 gentile , g. tonini , c. salucci , p. 2007 , a&a , 467 . 925 governato , f. , willman , b. , mayer , l. , brooks , a. , stinson , g. , valenzuela , o. , wadsley , j. , quinn , t. 2007 , mnras , 374 , 1479 governato , f. et al . 2010 , nature , 403 , 203 granato , g.l , de zotti , g. , silva , l. , bressan , a. , danese , l. 2004 , apj , 600 , 580 henriques , b. , maraston , c. , monaco , p. , fontanot , f. , menci , n. , de lucia , g. , tonini , c. 2011 , mnras , 415 , 357 kamada , f.a . 2011 , proceedings of warm dark matter in the galaxies : theoretical and observational progresses , preprint [ astro - ph/1109.3187 ] kang , x. , jing , y.p . , silk , j. 2006 , apj , 648 , 820 kauffmann , g. , white , s.d.m . , & guiderdoni , b. , 1993 , mnras , 264 , 201 kauffmann , g. , haehnelt , m. , 2000 , mnras , 311 , 576 khlopov , m.yu . 2008 , proc . of the blois 2008 conference ( arxiv:0806.3581 ) klypin , a. , kravtsov , a.v . , valenzuela , o. , prada , f. 1999 , apj , 522 , 82 knebe , a. , devriendt , j.e.g . , mahmood , a. , silk , j. 2002 , mnras , 329 , 813 komatsu , e. et al . 2011 , apjs , 192 , 18 kormendy , j. , richstone , d. 1995 , ara&a , 33 , 581 kravtsov , a.v . , gnedin , o.y . , klypin , a.a . 2004 , apj , 209 , 482 krumholz , m.r . , dekel , a. 2011 , apj , in press kusenko , a. 2009 , phys.rept . , 481,1 lacey , c. , & cole , s. , 1993 , mnras , 262 , 627 la franca , f. et al . 2005 , apj , 635 , 864 lapi , a. , cavaliere , a. , & menci , n. 2005 , apj , 619 , 60 lapi , a. , shankar , f. , mao , j. , granato , g.l . , silva , l. , de zotti , g. , danese , l. 2006 , apj , 650 , 42 lee , s .- k . , ferguson , h. c. , somerville , r. s. , wiklind , t. , & giavalisco , m. 2010 , apj , 725 , 1644 lee , t. , wu , g.h . 1999 , phys . b 447 , 83 lemoine , m. , moultaka , m. , jedamzik , j. 2006 , phys . lett.b , 645 , 222 lo faro . b , monaco , p. , vanzella , e. , fontanot , f. , silva , l. , cristiani , s. 2009 , mnras , 399 , 827 lovell , m.r . , eke , v.r . , frenk , c.s . , jenkins , a. 2011 , mnras , 413 , 3013 loewenstein , m. & kuseno , a. 2010 , apj , 714 , 652 madau , p. diemand , j. , kuhlen , m. 2008 , apj , 679 , 1260 madgwick , d.s . , et al . , 2002 , mnras , 333 , 133 magorrian , j. et al . 1998 , aj , 115 , 2285 maraston , c. , pforr , j. , renzini , a. , daddi , e. , dickinson , m. , cimatti , a. , tonini , c. 2010 , mnras , 407 , 830 marchesini , d. , van dokkum , p.g . ; frster schreiber , n.m . , franx , m. , labb , i. , wuyts , s. 2009 , apj , 701 , 1765 marulli , f. , bonoli , s. , branchini , e. , moscardini , l. , springel , v. 2008 , mnras , 385 , 1846 mashchenko , s. , wadsley , j. and couchman , h. , m. p 2008 , science , 319 , 174 mateo , m. 1998 , ara&a , 36 , 435 mcgaugh , s.s . , de blok , w.j.g . , schombert , j.m . , kuzio de naray , r. , kim , j. h. 2007 , apj , 659 , 149 menci , n. , cavaliere , a. , fontana , a. , giallongo , e. , poli , f. , 2002 , apj , 578 , 18 menci , n. , cavaliere , a. , fontana , a. , giallongo , e. , poli , f. , vittorini , v. 2003 , apj , 587 , l63 menci , n. , fontana , a. , giallongo , e. , salimbeni , s. 2005 , apj , 632 , 49 menci , n. , fontana , a. , giallongo , e. , grazian , a.,salimbeni , s. 2006 , apj , 647 , 753 menci , n. , fiore , f. , puccetti , s. , cavaliere , a. 2008 , apj , 686 , 219 monaco , p. , salucci , p. , danese , l. , 2000 , 311 , 279 moore , b. , quinn , t. , governato , f. , stadel , j. , lake , g. 1999 , mnras , 310 , 1147 olive , k.a . , turner , m.s . 1982 , phys . d , 25 , 213 pagels , h. , primack , j. 1982 , phys . lett . , 48 , 223 peebles , p.j.e . 1982 , apj , 258 , 415 peebles , p.j.e . 1993 , _ principles of physical cosmology _ ( princeton : princeton univ . press ) polisensky e. , ricotti m. , 2011 , phys . d , 83 , 3506 press , w.h . , & schechter , p. , 1974 , apj , 187 , 425 primack , j.r . & gross , m.a.k . 2000 , in current aspects of neutrino physics , editor d.o . caldwell , springer ( berlin ) , p. 207 prokhorov d. , silk j. , 2010 , apj , 725 , 131 reed , d. , governato , f. , verde , l. , gardner , j.,quinn , t. , stadel , j. , merritt , david ; lake , g. 2005 , mnras , 357 , 82 salimbeni , s. et a. 2008 , a&a , 477 , 763 santini , p. et al . 2012 , a&a , 538 , 33 schneider , a. , smith , r.e . , maccio , a.v . , moore , b. 2011 , preprint ( arxiv:1112.0330 ) shankar f. , & mathur s. 2007 , apj , 660 , 1051 shankar , f. 2009 , new astronomy reviews , 53 , 57 shankar f. , crocce m. , miralda - escude j. , fosalba p. , weinberg d. h. 2010 , apj , 718 , 231 shen , y. 2009 , apj , 704 , 89 shi , x.d . , fuller , g.m . 1999 , phys . 82 , 2832 smith , r. e. , markovic k. , 2011 , phys . d , 84 , 3507 somerville , r.s . , & primack , j.r . , 1999 , mnras , 310 , 1087 somerville , r. 2002 , apj , 572 , 23 spergel , d.n . 2007 , apjs , 170 , 377 steffen , f.d . 2006 , journal of cosmology and astroparticle physics , 9 , 1 strateva , i. et al . 2001 , aj , 122 , 1861 tegmark , m. , zaldarriaga , m. 2002,phys . d , 66 , 3508 viel , m. , legourgues , j. , haenhelt , m.g . , matarrese , s. riotto , a. 2005 , phys . d , 71 , 063534 viel , m. , becker , g.d . , bolton , j.s . , haenhelt , m.g . , rauch , m. , sargent , w.l.w . 2008 , phys . d , 100 , 041304 volonteri , m. , madau , p. , quataert , e. , rees , m.j . 2005 , apj , 620 , 69 wang , j. , white , s.d.m . 2007 , mnras , 380 , 93 weinmann , s.m . , van der bosch , f.c . , yang x. , mo , h.j . 2006 , mnras , 366 , 2 wyithe , j.s.b . , loeb , a. 2003 , apj , 586 , 693 zavala , j. , jing , y.p . , faltenbacher , a. , yepes , g. hoffman , y. , gottlber , s. , catinella , b. 2009 , apj , 700 , 1779

quantum annealing is a promising technique which leverages quantum mechanics to solve hard optimization problems . considerable progress has been made in the development of a physical quantum annealer , motivating the study of methods to enhance the efficiency of such a solver . in this work , we present a quantum annealing approach to measure similarity among molecular structures . implementing real - world problems on a quantum annealer is challenging due to hardware limitations such as sparse connectivity , intrinsic control error , and limited precision . in order to overcome the limited connectivity , a problem must be reformulated using minor - embedding techniques . using a real data set , we investigate the performance of a quantum annealer in solving the molecular similarity problem . we provide experimental evidence that common practices for embedding can be replaced by new alternatives which mitigate some of the hardware limitations and enhance its performance . common practices for embedding include minimizing either the number of qubits or the chain length , and determining the strength of ferromagnetic couplers empirically . we show that current criteria for selecting an embedding do not improve the hardware s performance for the molecular similarity problem . furthermore , we use a theoretical approach to determine the strength of ferromagnetic couplers . such an approach removes the computational burden of the current empirical approaches , and also results in hardware solutions that can benefit from simple local classical improvement . although our results are limited to the problems considered here , they can be generalized to guide future benchmarking studies . this research was supported by 1qbit . the authors would like to thank clemens adolphs , hamed karimi , anna levit , dominic marchand , and arman zaribafiyan for useful discussions , robyn foerster for valuable support , and marko bucyk for editorial help . we thank helmut katzgraber for reviewing the manuscript . we also acknowledge the support of the universities space research association ( usra ) quantum artificial intelligence laboratory research opportunity program .

1701.04433

quantum annealing ( qa ) , the quantum counterpart of simulated annealing , is an approach for harnessing quantum mechanical effects in searching the energy landscapes of classical np - hard optimization problems @xcite . the development of a quantum annealer by _ d - wave systems _ @xcite has initiated a great deal of theoretical and experimental research into the usefulness of the qa approach and its potential supremacy over classical algorithms @xcite . in recent years , the research community has focused mainly on searching for a useful area of application where qa demonstrates a scaling advantage over classical algorithms . although there is evidence that the d - wave quantum annealer exhibits quantum means of energy landscape exploration such as tunnelling @xcite and entanglement @xcite , the efforts to identify an application for which the device is able to outperform classical optimization have not yet been conclusive . the highlight of the search for potential quantum supremacy is the recent study by google @xcite on an artificially crafted weak strong cluster problem showing that quantum approaches using either the d - wave 2x ( dw2x ) quantum annealer or quantum monte carlo simulation scale significantly better than simulated annealing . the primary difficulty in demonstrating a quantum scaling advantage on useful optimization problems can be attributed to the architecture of the quantum annealer . the quantum annealer is designed to find the ground state of ising hamiltonians with pair - wise interactions on a fixed sparse graph called `` chimera '' . although the majority of optimization problems across various disciplines can be translated into ising problems , their formulations usually have connectivity different from that specified by the chimera s structure . minor embedding ( me ) is a technique for mapping such _ non - native _ problems to the hardware , where several physical qubits encode one logical qubit @xcite . due to the me overhead , embedded problems are suboptimal for determining a scaling advantage @xcite . however , the analysis of the quantum annealer s performance on these problems is critical for the design of future hardware architectures and the setting of various programming parameters . examples of studies on parametrized families of hard embedded problems include the sherrington kirkpatrick model with random @xmath0 couplings , which is directly related to the graph partitioning problem @xcite , operational navigation and scheduling problems @xcite , the job - shop scheduling problem @xcite , the multi - period portfolio optimization problem @xcite , and the graph isomorphism problem @xcite . in this paper , we report on the performance of the dw2x and discuss efficient programming guidelines for the problem of measuring similarity among small molecules that are modelled as labelled graphs . to determine the similarity between two molecular graphs while accounting for noise , a relaxation of the maximum weighted independent set problem , known as maximum weighted co-@xmath1-plex problem , is formulated such that it is consistent with the hardware s architecture . previous studies of non - native problems were conducted using randomly generated instances . to the best of our knowledge , our work is the first to examine the performance of a quantum annealer on real instances of problems in the context of molecular similarity . encoding and decoding are challenges specific to solving non - native problems on a quantum annealer . encoding includes two problems : topological embedding and parameter setting . in the former , the mapping between each logical qubit and a set of connected physical qubits is determined . in the latter , the strength of internal couplings ( among physical qubits corresponding to the same logical qubit ) is set , and the logical local fields and coupling values are distributed among the physical qubits and couplers . it is well known that both topological embedding and parameter setting problems have a significant impact on the efficiency of a quantum annealer @xcite . decoding refers to the process of inferring the solution of each logical qubit from the retrieved solutions of the corresponding physical qubits . an important consequence of the limited available precision @xcite and the existence of errors @xcite of the quantum annealer is that the physical qubits representing a logical qubit might be assign to different values . classical post - processing techniques are often used to assign the right value to the logical qubit . in this paper , we review current approaches for encoding and decoding non - native problems and provide alternatives to the current suboptimal practices , which require substantial computational time . the rest of the paper is organized as follows . in section [ sec : molecular_similarity ] , we define the molecular similarity problem . in section [ sec : quantum_annealing ] , a background on qa is presented and a novel formulation of the molecular similarity problem amenable to qa is developed . different challenges of solving an embedded problem on the hardware architecture are then discussed . detailed experimental results are presented in section [ sec : experimental_results ] . we conclude and discuss future work in section [ sec : conclusion ] . supplementary information is presented in the appendix . the measurement of structural similarity among molecules plays an important role in chemical and pharmaceutical research in areas such as drug discovery . the similarity measures proposed in the literature can be categorized into two classes . the first class uses a vector - based representation called a fingerprint in which the molecules are compared using distance metrics such as euclidean distance @xcite . although fingerprints are simple and computationally efficient , they can not provide accurate information on the common substructures of molecules . thus , in this paper , we use the other class , which was developed based on graph theoretical concepts . the second class of similarity measures uses the intuitive graph representation of the labelled molecular atom bond structure . formally , a labelled graph of a molecule can be written as @xmath2 , where @xmath3 is the set of vertices , @xmath4 is the set of edges , @xmath5 is the set of labels assigned to each vertex , and @xmath6 is the set of edge labels . graph representations of molecules are often compared based on the property of isomorphism . two graphs @xmath7 and @xmath8 are called isomorphic if there is a bijection between their vertex sets such that there exists a mapping between the adjacent pairs of vertices of @xmath7 and @xmath8 . a more practical variation of isomorphism is the maximum weighted common subgraph ( mwcs ) , which identifies the largest weighted subgraph of @xmath7 that is isomorphic to a subgraph of @xmath8 . there is a correspondence between the mwcs problem and another well - known problem the maximum weighted independent set ( mwis ) of a third graph , which can be induced from the graphs being compared @xcite . the vertices and edges of the third graph , called a _ conflict graph _ , represent possible mappings and the conflicts between them , respectively . the goal of the mwis problem is to find the largest weighted set of vertices such that there is no edge between all selected pairs , forming the largest conflict - free mapping . since molecular data are subject to regular errors , representing the mwcs problem as the mwis problem makes it easier to relax the definition of similarity to account for the effect of noise in the data by considering as similar substructures with conflicts up to a certain threshold . there are different relaxations of the similarity requirement in the literature @xcite . one of the relaxations is known as the maximum weighted co-@xmath1-plex problem , in which the goal is to find the largest weighted set of vertices in the graph such that each vertex has at most @xmath9 edges connecting it to the other vertices . it is clear that the maximum weighted co-@xmath10-plex problem is the mwis problem @xcite . the majority of the similarity methods discussed above , including the mwis and maximum weighted co-@xmath1-plex problems , are in general np - hard , having exponentially increasing computational complexity due to the combinatorial nature of the graphs involved @xcite . an illustration of the molecular similarity problem and its co-@xmath1-plex formulation is shown in figure [ fig : cokplex_example ] . the maximum weighted co-@xmath1-plex solution is shown for two cases , @xmath11 and @xmath12 . ] the details of reducing molecules to graphs and building the corresponding conflict graph are discussed in our previous work @xcite . in the next section , we present an overview on qa and discuss how the problem of measuring molecular similarity can be solved by a quantum annealer . quantum annealing is a heuristic technique that was introduced to solve hard optimization problems by exploiting quantum mechanical effects such as quantum tunnelling . the solution to a combinatorial optimization problem is encoded into the ground state of a classical ising hamiltonian @xmath13 where the local fields and couplers are represented by @xmath14 and @xmath15 , respectively . the variables @xmath16 denote classical ising spin variables and the sums run over the weighted vertices @xmath3 and edges @xmath17 of a graph @xmath18 . specifically , the task is to find the spin configuration @xmath19 which minimizes @xmath20 . a quantum annealing solver attempts to minimize @xmath20 by implementing a time - dependent ising hamiltonian @xmath21,\ ] ] where @xmath22 is the pauli spin operator on qubit @xmath23 and @xmath24 , with @xmath25 referred to as the annealing time . the functions @xmath26 and @xmath27 specify the annealing schedules where typically @xmath26 and @xmath27 are monotonically decreasing and increasing functions , respectively . the d - wave devices accept problems in terms of the hamiltonian described in equation [ eq : ising ] . however , in conventional optimization formulations , binary variables @xmath28 commonly take values from @xmath29 instead of @xmath30 . this mathematical form is usually referred to as a quadratic unconstrained binary optimization ( qubo ) problem . fortunately , the qubo and ising formulations can be related by taking @xmath31 . thus , to solve any optimization problem on a quantum annealer , it is sufficient to reformulate it as an instance of qubo . in the sections to follow , we describe how to formulate the maximum weighted co-@xmath1-plex problem in qubo form , how to map a qubo problem to the quantum hardware , and how to retrieve the logical answers from the physical solutions provided by the quantum annealer . a co-@xmath1-plex of a graph @xmath32 is a subgraph of @xmath32 in which each vertex has a degree of at most @xmath9 . formally , let @xmath33 be the conflict graph of two molecular graphs @xmath7 and @xmath8 , where @xmath34 , and @xmath35 represents both the bijection and the user - defined requirements . the latter allows the enforcement of customized structure restrictions , for example , adding an edge to @xmath35 if there is a mismatch between the edge labels of two molecular graphs . the maximum weighted co-@xmath1-plex of the conflict graph @xmath36 corresponds to the mwcs of @xmath7 and @xmath8 , where each possible pairing can violate at most @xmath9 constraints . before presenting the qubo formulation , let us first define a star graph . + [ stardef ] a graph @xmath37 is a star graph of size @xmath1 if it is a tree with @xmath38 vertices and one vertex of degree @xmath1 . based on the co-@xmath1-plex formulation , we do not penalize the conflict edges . each vertex in the conflict graph can have up to @xmath9 edges . therefore , we only penalize in situations where a subset of the vertices induces a subgraph in which there is one vertex with degree greater than @xmath9 . in other words , we penalize all subsets of vertices whose induced subgraph forms a star graph of size @xmath1 . further , let us define the binary parameter @xmath39 as @xmath40 where @xmath41 are the vertices of the conflict graph @xmath36 . the maximum weighted co-@xmath1-plex problem is formulated as @xmath42 , \end{aligned}\ ] ] where @xmath43 is a binary variable equal to 1 if the vertex @xmath44 is included in the maximum independent set or 0 otherwise , @xmath45 is the weight of vertex @xmath44 , and @xmath46 . the tunable parameter @xmath1 should be determined by the user . the objective function of formulation is a higher - order polynomial . there are several algorithms in the literature that map higher - order polynomials to quadratic polynomials ( * ? ? ? * and references therein ) . the hardware architecture of the d - wave devices consists of a fixed sparse graph called a chimera graph , in which each vertex is a physical flux qubit and each edge is a physical coupler between two qubits . each generation of devices has shown significant progress , including an increase in the number of qubits and reduction of noise . however , implementing a practical problem remains a challenge , mainly because of the physical constraints of the hardware . one of the fundamental limitations of these devices is their sparse connectivity . typically , a real - world problem formulation might require a different connectivity from that defined by the hardware graph . this is true of the molecular similarity problem , which involves finding the maximum weighted co-@xmath1-plex in a conflict graph @xmath36 . therefore , we need to encode the problem into the device s architecture . there are two aspects to this encoding process : the embedding problem and the parameter - setting problem . the embedding problem is commonly addressed by me techniques where each vertex of @xmath36 is replaced by a connected subgraph of the hardware graph , denoted by @xmath47 . each connected subgraph is ferromagnetically coupled and is referred to as a chain in the literature , though it does not necessarily have a linear , acyclic structure . an example of me is illustrated in figure [ fig : embedding_example ] , and more details about me can be found elsewhere @xcite . after embedding , the logical local field @xmath48 and the couplings @xmath49 are respectively distributed among the vertices and edges of the subgraph @xmath50 such that @xmath51 the parameter - setting problem consists of determining the qubit biases @xmath52 and coupler strengths @xmath53 of the embedded problem . the encoding process is of particular importance since the performance of the quantum processor is highly sensitive to the choice of embedding and its corresponding parameters . in what follows , we discuss both aspects of the encoding process , presenting the current approaches and describing the methods used in this work . into the hardware graph . the logical variable @xmath10 is replaced by a connected subgraph with physical vertices @xmath54 which are ferromagnetically coupled . ] the problem of finding an me is in general np - hard . in this work , we use the embedding software provided by d - wave systems . an me found by this algorithm is not necessarily unique or optimal . in fact , different mes representing the same qubo problem might result in different hardware performance . hence , a criterion or method to select an embedding which maximizes the hardware s performance is needed . we refer to this problem as the _ embedding selection problem_. various embedding properties have been suggested as contributing factors affecting the hardware s performance . for instance , cai et al . @xcite advise that it is preferable to select an embedding which minimizes either the total number of physical qubits used or the maximum number of physical qubits representing a logical qubit ( i.e. , it minimizes the length of the longest chain ) . also , pudenz et al . @xcite show that for the problem of antiferromagnetic chains , the performance of the hardware decreases with chain size . however , for some problems with a more complex topology , such as operational planning problems @xcite , the chain length in the embedding did not contribute significantly to the hardware s performance . further evidence that these embedding properties do not necessarily affect the performance of the device has been presented by bian et al . they introduce a locally structured embedding algorithm which considers the local structure of the hardware , in contrast to an me approach which is referred to as a global embedding . in particular , they show that the local embedding , which requires more qubits and longer chains in general , yields better performance compared with the me approach . another embedding property has been considered in ref . @xcite , in which the authors suggest that it is preferable to use embeddings with chains of equal length . considering these embedding properties as criteria in selecting an embedding has been sufficient for some specific problems . however , for general problems , a more sophisticated embedding selection strategy is still lacking . given that none of the current embedding selection criteria seem to predict or optimize the performance of the hardware for general problems , we implement an empirical approach to select an embedding . we refer to the embedding selected by this method as _ empirical embedding_. our empirical selection method consists of generating @xmath55 distinct embeddings for each problem instance and setting their parameters according to the method described in section [ subsubsec : parameter_setting ] . we run @xmath56 anneals at @xmath57 @xmath58 and use majority vote decoding as described in section [ sec : decoding ] for each embedded instance . from this pool of embeddings , we select the @xmath59 embeddings , which results in a higher success probability . the experiment is repeated on the reduced pool of embeddings . we then select the embedding that maximizes the success probability . a similar approach has been used by king et al . @xcite . to empirically select the embedding , we use a priori knowledge of the optimal value . in general , a different approach , which selects the solutions with minimum energy , could be considered instead . for example , perdomo - ortiz et al . @xcite introduce a performance estimator based on minimum - energy solutions to guide the selection of the best programming specifications . we use the optimal known solution to better demonstrate that the current embedding selection criteria do not necessarily optimize the performance of the hardware . a problem graph describing the structure of an optimization problem , such as the conflict graph @xmath36 , has been reduced accurately to its minor - embedding @xmath60 as long as there is a one - to - one correspondence between the ground states of the optimization problems defined on both graphs . it is well known that regardless of how the logical parameters , including local fields and couplers , are distributed among the physical qubits , a large negative value of the ferromagnetic couplers , @xmath61 , ensures the correspondence between the ground states such that there is no broken connected subgraphs ( i.e. , the physical qubits encoding one logical qubit result in an identical state ) . however , since a quantum annealer is an analogue machine with finite available precision , it is important to find a sufficiently small value for these ferromagnetic couplers . in this case , the values of the ferromagnetic couplers are also dependent on how the logical parameters are distributed . an additional restriction of the dw2x architecture is the limited range of the @xmath62 and @xmath63 parameters , i.e. , @xmath64 $ ] and @xmath65 $ ] , respectively . the values of @xmath62 and @xmath63 can be scaled so that they lie within their respective ranges by multiplying them with a positive constant factor @xmath66 . this parameter is called the `` scaling factor '' in the rest of the paper . the literature on setting the parameters of an embedded graph can be divided , in general , into empirical and theoretical schemes . the former are the most widely used approaches , and include the work of venturelli et al . @xcite , vinci et al . @xcite , and perdomo - ortiz et al . @xcite . in the empirical approaches , the logical field and coupler values are evenly distributed among the physical qubits . to determine the strengths of the ferromagnetic couplers , the sample data from the device is used to experimentally find the optimal strength of the ferromagnetic couplers such that the probability of observing the ground state is maximized . the drawbacks of empirical approaches are twofold . first , inferring the strength of the ferromagnetic couplers based on the samples from a device that is subject to various types of errors does not guarantee that the ground state of the input graph @xmath36 is the same as the ground state of the embedded graph @xmath60 . second , the empirical approaches are essentially trial - and - error approaches requiring significant computation time that increases as the size of the input graph increases . to the best of our knowledge , there is only one paper that discusses a theoretical approach to setting the parameters of an embedded graph @xcite . in the approach proposed by choi @xcite , the distribution of the logical coupler values is similar to the empirical approaches , though the logical local field values are differently distributed . furthermore , it is assumed that the subgraphs representing each logical qubit are subtrees of the hardware graph . however , this assumption does not necessarily hold for all embedded real - world problems , including some molecular similarity problems . in this work , we use a generalization of choi s theoretical method and compare it with an empirical approach to investigate the potential benefit of a theoretical parameter - setting approach . it is worth mentioning that the empirical approach used in this paper is different from the approaches used in the literature . both are explained below and the results of their performance are discussed in section [ subsection : emp_theo_parameter_setting ] . * empirical parameter - setting * we empirically set the strengths of ferromagnetic couplers using the hamze de freitas selby ( hfs ) algorithm @xcite . initially , we set @xmath67 , where @xmath68 refers to the strength of all ferromagnetic couplers tying together the physical qubits representing the logical qubit @xmath23 . we distribute the logical local field @xmath48 and couplings @xmath49 evenly among the corresponding physical qubits and couplers , and scale their values to be in the available range . we then solve the scaled problem with the hfs algorithm @xmath69 times . if there is no broken connected subgraph in at least one of the solutions , we stop and set the ferromagnetic coupler strengths to @xmath70 . otherwise , we decrease @xmath68 by @xmath71 and repeat the above procedure . the empirical strength of ferromagnetic couplers used is the first value found for which none of the connected subgraphs are broken . we use @xmath72 and @xmath73 in our experiment . * theoretical parameter - setting * to theoretically set the @xmath68 values , we generalize the approach proposed by choi @xcite , where the logical couplings are distributed evenly among the physical couplers and the logical local fields are distributed following the procedure in algorithm [ parameter ] . in our generalization , there is no need to reduce the subgraphs to trees . the details of our generalization are explained in appendix [ subsec : appendix_theoretical_parameter_setting ] . a decoding strategy refers to the process of inferring the value of each logical qubit given the values of the physical qubits obtained from the quantum annealer . if the values of the physical qubits encoding one logical qubit agree , the common value of the physical qubits is assigned to the logical qubit . otherwise , the logical qubits are considered broken and should be repaired . there are two general decoding strategies known as local and global @xcite . the former approach decodes the broken logical qubits individually . examples of local algorithms include heuristic algorithms with polynomial time complexity such as majority vote , coin tossing , and greedy descent . the latter approach decodes the broken qubits simultaneously . this process involves solving an optimization problem induced by the broken logical qubits , which is referred to as a `` decoding '' hamiltonian . the disadvantage of a global decoding technique is that it requires solving an ising problem belonging to the class of np - complete problems . examples of such algorithms are exhaustive search and simulated annealing . majority vote ( mv ) , which assigns the most repeated value within an encoded subgraph to its corresponding logical qubit , is the simplest and most widely used approach in the literature @xcite . however , vinci et al . @xcite point out that mv is not appropriate for the minor - embedded problems since physical excitations are more likely to occur during the evolution of embedded problems and the probability that physical qubits break independently ( the required criterion for the success of mv ) can not be considered valid . furthermore , they mention that mv can not be more effective than coin tossing in cases where the breakage of an encoded subgraph in various ways results in the same energy . instead , they use simulated annealing to globally minimize the energy of the broken qubits . the main drawback of the global decoding strategies is that minimizing the energy of the decoding hamiltonian can be as hard as solving the original hamiltonian . to justify the computational cost , vinci et al . @xcite use the critical probability , @xmath74 , as a decoding threshold . the critical probability is the probability of an infinite cluster appearing for the first time in an infinite lattice . if the probability for an encoded ( logical ) qubit to be broken , @xmath75 , is greater than @xmath74 , global approaches can not be performed efficiently since the probability of having infinitely large broken clusters approaches 1 as the total number of logical qubits goes to infinity ( @xmath76 ) . however , if @xmath77 , the probability of having infinitely large broken clusters approaches 0 if the largest size of the decodable connected logical qubits is close to @xmath78 . vinci et al . @xcite show that the percolation threshold of their encoded logical graph is well beyond the experimental probability that a logical qubit breaks @xcite . in this paper , we use mv to decode the broken hardware answer , map it back to the logical space , and then apply a greedy descent method to the logical answer as a post - processing technique to further refine the retrieved solution . justification for the selection of mv as a local decoding strategy can be found in section [ subsec : greedy_descent ] . the quantum annealer used in this work is a dw2x processor located at nasa s ames research center . it consists of 1097 working flux qubits and 3060 working couplers arranged in a chimera graph architecture and operates at 15 mk . to empirically evaluate the performance of a stochastic solver such as the dw2x , the standard metric is time - to - solution ( tts ) , the total time required by the solver to find the ground state at least once with a probability of @xmath79 . defining @xmath80 as the number of runs required by the solver to find the ground state at least once with a probability of @xmath79 , we have @xmath81 , where @xmath82 is the duration of each annealing run . the calculation of tts is explained in detail in appendix [ subsec : appendix_tts ] . in this section , we first discuss the experimental setup . we then present the tts results and investigate the effect of different embedding criteria , parameter setting approaches , and annealing times on the performance of the dw2x . finally , we justify the use of mv as a decoding strategy and report on the improvement made on the performance of the dw2x using a simple post - processing technique . we generate a parametrized family of the molecular similarity problems by using molecules from an available data set that consists of @xmath83 small molecules @xcite . the molecular similarity problem corresponds to finding the maximum weighted co-@xmath1-plex in a conflict graph @xmath36 as defined in section [ sec : molecular_similarity ] . the vertices @xmath3 and edges @xmath17 of @xmath36 represent the matchings and conflicts between the molecules being compared , respectively . the size of @xmath36 , defined in terms of its number of vertices @xmath84 , depends on the size of the molecules and the similarity conditions considered . it is worth mentioning that the number of logical variables in the molecular similarity problem corresponds to the size of @xmath36 , not the size of the molecules being compared . the size of @xmath36 grows as the value of @xmath1 in formulation increases . for example , the maximum size of @xmath36 that we can successfully embed into a dw2x processor is @xmath85 and @xmath86 for @xmath11 and @xmath87 , respectively . in this paper , we consider @xmath11 in order to study the performance of the dw2x on larger non - native problems . the conflict graphs are then parametrized by @xmath88 , where the number of vertices @xmath84 is selected from the set @xmath89 , and the density @xmath90 is selected from the range in the set @xmath91 | k=0 , \ldots , 2 \ } $ ] . for each combination of @xmath84 and @xmath90 , 100 different instances are generated . 0.5 , @xmath92 , and @xmath93 for @xmath94 $ ] , @xmath95 $ ] , and @xmath96 $ ] , respectively.,title="fig : " ] 0.5 , @xmath92 , and @xmath93 for @xmath94 $ ] , @xmath95 $ ] , and @xmath96 $ ] , respectively.,title="fig : " ] * embeddability * to study the embeddability of the molecular similarity instances , we use the ` find_embedding ( ) ` function from sapi 2.2 @xcite with default parameter values , but with the number of trials per function call set to @xmath97 . figure [ fig : phy_qubits_estimation : a ] shows the mean success rate of the embeddability across 100 instances . the embeddability success rate is defined as the fraction of the number of embeddings found over the total number of trials . as shown in figure [ fig : phy_qubits_estimation : a ] , embeddings are found for all instances up to size @xmath98 , regardless of their densities . the number of embeddings generated decreases for size @xmath85 and eventually reaches 0 for size @xmath99 . we also estimate the mean number of physical qubits required to embed the given instances . as illustrated in figure [ fig : phy_qubits_estimation : b ] , the mean number of physical qubits scales similarly across different densities . the scaling is @xmath100 , @xmath101 , and @xmath102 for density @xmath94 $ ] , @xmath95 $ ] , and @xmath96 $ ] , respectively . figure [ fig : tts_main ] shows the 25th , 50th , 75th , and 99th percentiles of the tts for three different densities over 100 instances for each problem size . for every instance , the true ground state is found using the gurobi solver ( version 6.5.1 ) , the best empirical embedding is selected , the parameters are theoretically set , the annealing time is set at @xmath103 @xmath104 ( see section [ subsec : annealing_time ] ) , the retrieved solution from the hardware is decoded using mv . a common strategy to average out systematic errors is to perform gauge transformations between each call to the quantum annealer @xcite . since we are plotting the tts for a class of instances of similar size , not an individual instance , and the tts has a higher variance over different instances than different gauges , the timing data is not averaged over multiple gauges @xcite . the 99th percentile of the tts is not shown in figure [ fig : tts_main ] if all 100 instances are not solved given 5 calls to the dw2x and 10,000 anneals per call . the lower limit of the standard deviation of the 99th percentile is also not shown for several sizes in case it is bigger than the mean of the 99th percentile . as illustrated , the tts is higher for problems with lower densities . as the density of the qubo graph reaches the extreme values of 0 and 100 , we intuitively expect that the problem of identifying the maximum independent set becomes easier . furthermore , the large differences between the median and 99th percentiles of the tts for several sizes and densities are indications of heavy tails in the distributions of the tts . in other words , this result conveys that the hardness of our problems does not depend solely on the number of logical variables . it significantly differs across instances with a similar number of logical variables . [ .4][c ] $ ] , ( b ) @xmath95 $ ] , and ( c ) @xmath96 $ ] . the number of calls to the dw2x and the number of anneals per call are 5 and 10,000 , respectively.,title="fig : " ] [ .4][c ] $ ] , ( b ) @xmath95 $ ] , and ( c ) @xmath96 $ ] . the number of calls to the dw2x and the number of anneals per call are 5 and 10,000 , respectively.,title="fig : " ] [ .4][c ] $ ] , ( b ) @xmath95 $ ] , and ( c ) @xmath96 $ ] . the number of calls to the dw2x and the number of anneals per call are 5 and 10,000 , respectively.,title="fig : " ] following the discussion in section [ subsec : embedding_selection ] , different embedding representations of the same ising hamiltonian might result in different performance of the qa hardware . in the literature , it has been suggested that optimizing some embedding properties could be an important factor that affects the performance of the hardware , for example , minimizing the total number of qubits used in an embedding . figure [ fig : emb_motivation ] depicts a counterexample to this argument for an instance of the molecular similarity problem . in particular , we do not observe any type of correlation between the mean success probability and the number of physical qubits used in the embedding . additionally , we note that if the embedding yielding a low success probability is selected , such an instance will be misclassified as a hard instance and might mask a heavy tail in the final tts results . these observations motivate us to perform a more comprehensive study to address the embedding selection problem . and density @xmath95 $ ] . the results are shown for @xmath55 different embedding representations of the same instance . ] to test the overall impact of the embedding selection problem on the quantum hardware s performance , we experimentally compare three criteria previously proposed in the literature . we generated @xmath97 instances for each problem size and @xmath55 embeddings for each instance , and selected @xmath10 embedding according to each of the criteria . specifically , we consider the following criteria : 1 . selecting the embedding with the smallest number of physical qubits ( pq ) ; 2 . selecting the embedding with the shortest longest chain ( lch ) ; 3 . selecting the embedding with chains of equal length ( std ) whereas the first two criteria are straightforward , heuristic me algorithms do not necessarily return embeddings with chains of equal length as required for the third criterion . to account for the selection of an embedding with chains of equal size , we calculate the standard deviation of the chain size for each embedding and select the embedding with the minimum standard deviation . further , we consider the empirical selection criteria for which we select the embedding that yield the highest success probability . each point indicates the mean across @xmath97 instances of each problem size . a ) shows the comparison in terms of the mean success probability for the dw2x device performing @xmath56 annealing runs . additional results show comparison in terms of ( b ) the mean number of physical qubits , ( c ) the mean size of the largest chain , and ( d ) the mean std of chain size.,title="fig : " ] 0.5 $ ] . each point indicates the mean across @xmath97 instances of each problem size . a ) shows the comparison in terms of the mean success probability for the dw2x device performing @xmath56 annealing runs . additional results show comparison in terms of ( b ) the mean number of physical qubits , ( c ) the mean size of the largest chain , and ( d ) the mean std of chain size.,title="fig : " ] 0.5 $ ] . each point indicates the mean across @xmath97 instances of each problem size . a ) shows the comparison in terms of the mean success probability for the dw2x device performing @xmath56 annealing runs . additional results show comparison in terms of ( b ) the mean number of physical qubits , ( c ) the mean size of the largest chain , and ( d ) the mean std of chain size.,title="fig : " ] 0.5 $ ] . each point indicates the mean across @xmath97 instances of each problem size . a ) shows the comparison in terms of the mean success probability for the dw2x device performing @xmath56 annealing runs . additional results show comparison in terms of ( b ) the mean number of physical qubits , ( c ) the mean size of the largest chain , and ( d ) the mean std of chain size.,title="fig : " ] figure [ fig : emb_selec : a ] shows the embedding selection comparison for the three criteria and the empirical embedding in terms of the mean success probability . we observe that these three selection criteria perform similarly poorly against the best embedding found empirically . further , the empirical embedding uses on average more physical qubits , its longest chain is longer , and its chains have higher std as shown in figures [ fig : emb_selec : b ] , [ fig : emb_selec : c ] , and [ fig : emb_selec : d ] , respectively . therefore , the use of these criteria has no effect on the success probability . an additional observation can be made from figure [ fig : emb_selec : a ] . as the logical problem size increases , the difference between the embedding selection methods decreases . a possible reason could be that for larger problems , most of the embeddings found with the heuristic algorithm reach the limit size of the chip . thus , we have less freedom to generate different embeddings . this observation is in agreement with the argument provided by cai et al . @xcite for the success of a heuristic me algorithm . specifically , they state that if the problem to be embedded is significantly smaller than the hardware graph s size , there are probably more distinct embeddings . to generate the results shown in figures [ fig : emb_motivation ] and [ fig : emb_selec ] , we perform @xmath56 annealing runs for each problem instance . in order to verify if performing gauge averaging or a larger number of anneals will affect our conclusions regarding the embedding selection problem , we perform additional experiments . specifically , we repeat the test shown in figure [ fig : emb_motivation ] for two cases : @xmath56 anneals and @xmath59 random gauges ; and 50,000 anneals and the default gauge . we also repeat the test for the first criterion presented in figure [ fig : emb_selec : a ] using @xmath56 anneals and 5 random gauges . in all cases , the new results verify that there is no correlation between the number of physical qubits used and the success probability . these results suggest that further investigation is needed in order to understand which properties of embeddings have an influence on the hardware s performance when solving problems with a complex topology , such as the conflict graphs in the molecular similarity problems . meanwhile , an empirical method like the one performed here can be applied . figure [ fig : theor_empirical ] shows the comparison of the performance of the dw2x using two different parameter - setting schemes with respect to four measures : success probability , size of the biggest broken cluster , residual energy , and computation time . the second measure represents the largest size of a decodable domain in which all broken logical qubits are connected . the third measure is the relative energy difference between the quantum annealer s lowest - energy solution and the actual ground state found by the gurobi solver . 0.5 $ ] . ( a ) shows the success probability ( log ) when the parameters of the embedded graphs are set theoretically or empirically . ( b ) shows the size of the biggest broken cluster size for theoretical and empirical parameter settings . ( c ) illustrates the residual energy for theoretical and empirical parameter settings . ( d ) illustrates the computation time ( seconds ) to set the parameters theoretically and empirically . all four measures represent the average value over 100 instances . these figures are generated by setting @xmath105 @xmath104 , and the number of calls and anneals per call are set to 5 and 10,000 , respectively.,title="fig : " ] 0.5 $ ] . ( a ) shows the success probability ( log ) when the parameters of the embedded graphs are set theoretically or empirically . ( b ) shows the size of the biggest broken cluster size for theoretical and empirical parameter settings . ( c ) illustrates the residual energy for theoretical and empirical parameter settings . ( d ) illustrates the computation time ( seconds ) to set the parameters theoretically and empirically . all four measures represent the average value over 100 instances . these figures are generated by setting @xmath105 @xmath104 , and the number of calls and anneals per call are set to 5 and 10,000 , respectively.,title="fig : " ] 0.5 $ ] . ( a ) shows the success probability ( log ) when the parameters of the embedded graphs are set theoretically or empirically . ( b ) shows the size of the biggest broken cluster size for theoretical and empirical parameter settings . ( c ) illustrates the residual energy for theoretical and empirical parameter settings . ( d ) illustrates the computation time ( seconds ) to set the parameters theoretically and empirically . all four measures represent the average value over 100 instances . these figures are generated by setting @xmath105 @xmath104 , and the number of calls and anneals per call are set to 5 and 10,000 , respectively.,title="fig : " ] 0.5 $ ] . ( a ) shows the success probability ( log ) when the parameters of the embedded graphs are set theoretically or empirically . ( b ) shows the size of the biggest broken cluster size for theoretical and empirical parameter settings . ( c ) illustrates the residual energy for theoretical and empirical parameter settings . ( d ) illustrates the computation time ( seconds ) to set the parameters theoretically and empirically . all four measures represent the average value over 100 instances . these figures are generated by setting @xmath105 @xmath104 , and the number of calls and anneals per call are set to 5 and 10,000 , respectively.,title="fig : " ] as illustrated , the theoretical approach is overall superior to the empirical approach . it results in a higher probability of success ( especially as the size of the problem increases ) , smaller broken clusters , lower residual energy , and lower computation time . these results indicate that the theoretical approach can be a strong alternative to replace current empirical practices in the literature . the theoretical approach reduces the pre - processing computation time to set the problem parameters and boosts the performance of the quantum annealer . it is worth mentioning that the reported performance for the empirical approach in this paper is most likely an upper bound on the performance of the typical empirical approaches in the literature for two main reasons . first , the value of the ferromagnetic couplers is optimized per instance in our empirical approach , whereas it is usually optimized per class of instances with the same size in the literature . second , in our approach , the analogue , noisy quantum solver is replaced by a classical solver with higher precision , thereby decreasing the probability that the qubits representing one logical qubit resolve into different states . @xmath104 . ] to gain more insight into the nature of the superiority of the theoretical parameter - setting approach , the difference between the theoretical scaling factor and the empirical scaling factor for each instance was found . figure [ fig : theor_empirical_scalefactor ] shows the @xmath106th , @xmath55th , and @xmath107th percentiles of the scaling difference over 100 instances for each problem size . as illustrated , the theoretical approach yields a larger scaling factor as the number of logical variables increases ( the difference is positive ) , which explains its better performance over the empirical approach . solving classical optimization problems by exploiting quantum effects is of central interest in the field of experimental quantum annealing . although the presence of entanglement @xcite and multiqubit tunnelling @xcite in d - wave processors have been experimentally demonstrated , any scaling advantage over classical algorithms remains elusive . when studying scaling , it is important to determine an optimal implementation . in the case of non - native problems , optimality refers to determining an effective encoding , as has been addressed in the previous sections . once an encoding has been determined , it is necessary to find the optimal annealing time . in ref . @xcite , it was suggested that the annealing time must be optimized for each problem size . in this context , various studies have tried to determine an optimal annealing time for different random instances and have arrived at the conclusion that the minimum possible @xmath108 @xmath58 of d - wave processors is longer than the optimal annealing time @xcite . alternatively , hen et al . @xcite have found instances whose optimal annealing time on the same processor is greater than @xmath109 @xmath58 . although it is outside of the scope of this paper to probe for quantum speedup , we are interested in studying the dw2x s performance dependance on the annealing time for our molecular similarity problem instances . a recent upgrade to the dw2x processor introduced a faster minimum annealing time of @xmath110 @xmath58 , allowing us to test the quantum annealer s performance dependency on @xmath25 for shorter annealing times . @xmath58 and @xmath108 @xmath58 . results are shown for different numbers of logical variables , varying from @xmath111 to @xmath112 and coloured accordingly , and density @xmath113 $ ] . ] from the correlation plot in figure [ fig : sp_correlation ] , it seems that a longer annealing time increases the success probability . in order to gain better insights , we performed a _ t_-test to determine whether the difference between these samples is statistically significant . we obtained a @xmath114-value close to @xmath115 , which implies that the higher mean success probability at @xmath116 @xmath58 has not occurred only by chance . this result provided enough evidence to accept the hypothesis that the difference between the mean success probabilities at @xmath117 @xmath58 and @xmath57 @xmath58 is statistically greater than @xmath115 , which can be understood in two different ways . firstly , it has been suggested that the success probability is a monotonically increasing function of @xmath25 @xcite . in this sense , we should expect that longer annealing times will increase the success probability until the asymptotic regime is reached . a counterexample to this assumption has been presented by amin @xcite , who presented a nonmonotonic success probability function of @xmath25 . according to his findings , our results for the success probability at @xmath110 @xmath58 and @xmath108 @xmath58 indicate that our system is in a quasistatic regime and that those annealing times are still too long . figure [ fig : percolation_prob_all_sizes ] shows the mean experimental probability of there being broken qubits ( @xmath118 ) for different sizes of the molecular similarity instances with @xmath119 $ ] , as well as their corresponding mean percolation thresholds ( @xmath74 ) . the calculation of the percolation threshold is explained in detail in appendix [ subsec : appendix_percolation ] . as illustrated , the values of @xmath118 approach @xmath74 as the number of logical variables increases . the trend of the @xmath118 and @xmath74 values shows the experimental probability of there being broken qubits would eventually be larger than the percolation threshold for larger graph similarity instances . therefore , the global decoding schemes will be inefficient for large instances of our graph similarity problem . the probability of having large broken clusters would be high , and solving the remaining problem would be as hard as solving the original . the average is taken over 100 instances . ] to locally decode the physical solutions , we apply mv since it requires the least possible effort to retrieve a solution from the physical space . it also gives a clear picture of the efficiency of the quantum annealer , with minimal contribution from a classical computer . the decoded logical solutions obtained by mv might be local optima . to further improve the quality of the decoded solutions , we use the greedy descent method . figure [ fig : r99_mv_gd ] shows the effect of applying greedy descent on the decoded logical solutions obtained by mv . as shown , there is a noticeable difference between the median @xmath80 before and after applying greedy descent , implying that refining the quantum solutions via classical post - processing techniques would be necessary to efficiently solve non - native problems using a quantum annealer . for three different densities using mv and greedy descent , where the number of calls and anneals per call are set to 5 and 10,000 , respectively . ] in this work , we have studied the performance of a quantum annealer in solving instances of the molecular similarity problem utilizing a real data set . the effective use of a quantum annealer presents many challenges . here , we focused on challenges derived from encoding and decoding real - world problems . we addressed the challenges present in both aspects of encoding , that is , embedding and parameter setting , and demonstrated how a careful encoding strategy helps to improve the performance of the dw2x device . in particular , our results emphasized the importance of the embedding selection problem . for some instances , we observed that two different embedding representations of the same original problem yield very different success probabilities . commonly , this difference is attributed to various properties of embedding , for example , the number of qubits used . however , we have shown that none of the commonly observed embedding properties correlate with the hardware s performance . in this work , we have incorporated an empirical method that selects the embedding that maximizes the success probability . we have also observed that the performance of the quantum annealer is less sensitive to the choice of embedding when increasing the size of the original problem . this result is expected since a minor - embedding heuristic will successfully find a larger number of distinct embeddings if the size of the problem to be embedded is significantly smaller than the size of the chip . thus , we expect that the poor performance of embedded problems with a size as large as the size of the current chip should improve in a next - generation quantum annealer . we have also shown that using a theoretical , rather than empirical , approach to select the parameters in the embedding can have significant advantages . besides eliminating the impractical experimental penalty optimization , a theoretical parameter - setting approach ensures an accurate representation of the logical ising hamiltonian , and as a consequence reduces the probability of broken qubits . of particular importance is the overall boosting of the hardware s performance , which benefits from an improved scaling factor over the empirical approach . another important question we have addressed is the selection of a decoding technique . for our problem set , we have found that a simple local decoding technique is effective . this is mainly a consequence of the reduced connectivity of broken logical qubits . specifically , we have shown that fufrther improvement in the hardware performance can be achieved if we use majority vote to fix the broken solutions and subsequently apply a greedy descent post - processing technique . whereas our experimental conclusions are restricted to the specific type of problem studied here , the results provide useful insight for future non - native benchmarking studies . in future work , we will address some of the questions still open regarding the effective implementation of a problem embedded into a quantum annealer . of practical importance is gaining a better understanding of which properties correlate to the performance of the quantum annealer . one possible research direction could be to study the quality of the physical qubits and couplers in the chip . an alternative direction could be to study whether the properties defining the hardness of spin glass problems have an impact on the hardware s performance . after having determined an optimal encoding , we will address the problem of determining the optimal annealing time . 10 e.s . giuseppe , t. erio , j. phys . gen . * 39 * , r393 ( 2006 ) e. farhi , j. goldstone , s. gutmann , j. lapan , a. lundgren , d. preda , science * 292 * , 472 ( 2001 ) m.w . johnson , m.h.s . amin , s. gildert , t. lanting , f. hamze , n. dickson , r. harris , a.j . berkley , j. johansson , p. bunyk , e.m . chapple , c. enderud , j.p . hilton , k. karimi , e. ladizinsky , n. ladizinsky , t. oh , i. perminov , c. rich , m.c . thom , e. tolkacheva , c.j.s . truncik , s. uchaikin , j. wang , b. wilson , g. rose , nature * 473 * , 194 ( 2011 ) v.s . denchev , s. boixo , s.v . isakov , n. ding , r. babbush , v. smelyanskiy , j. martinis , h. neven , phys . x * 6 * , 031015 ( 2016 ) i. hen , j. job , t. albash , t.f . rnnow , m. troyer , d.a . lidar , phys . a * 92 * , 042325 ( 2015 ) d. venturelli , s. mandr , s. knysh , b. ogorman , r. biswas , v. smelyanskiy , phys . x * 5 * , 031040 ( 2015 ) w. vinci , t. albash , g. paz - silva , i. hen , d.a . lidar , phys . a * 92 * , 042310 ( 2015 ) h.g . katzgraber , f. hamze , z. zhu , a.j . ochoa , h. munoz - bauza , phys . x * 5 * , 031026 ( 2015 ) z. zhu , a.j . ochoa , s. schnabel , f. hamze , h.g . katzgraber , phys . a * 93 * , 012317 ( 2016 ) a. perdomo - ortiz , b. ogorman , j. fluegemann , r. biswas , v.n . smelyanskiy , sci . rep . * 6 * ( 2016 ) . : 10.1038/srep18628 s. boixo , v.n . smelyanskiy , a. shabani , s.v . isakov , m. dykman , v.s . denchev , m.h . amin , a.y . smirnov , m. mohseni , h. neven , nat . commun . * 7 * ( 2016 ) . : 10.1038/ncomms10327 t. lanting , a.j . przybysz , a.y . smirnov , f.m . spedalieri , m.h . amin , a.j . berkley , r. harris , f. altomare , s. boixo , p. bunyk , n. dickson , c. enderud , j.p . hilton , e. hoskinson , m.w . johnson , e. ladizinsky , n. ladizinsky , r. neufeld , t. oh , i. perminov , c. rich , m.c . thom , e. tolkacheva , s. uchaikin , a.b . wilson , g. rose , phys . x * 4 * , 021041 ( 2014 ) j. cai , w.g . macready , a. roy , arxiv preprint arxiv:1406.2741 ( 2014 ) v. choi , quantum . inf . process . * 10 * , 343 ( 2011 ) s. mandr , z. zhu , w. wang , a. perdomo - ortiz , h.g . katzgraber , arxiv preprint arxiv:1604.01746 ( 2016 ) e.g. rieffel , d. venturelli , b. ogorman , m.b . prystay , v.n . smelyanskiy , quantum . inf . process . * 14 * , 1 ( 2015 ) d. venturelli , d.j.j . marchand , g. rojo , arxiv preprint arxiv:1506.08479 ( 2015 ) g. rosenberg , p. haghnegahdar , p. goddard , p. carr , j. wu , m. de prado , ieee journal of selected topics in signal processing * 10 * , 1053 ( 2016 ) k.m . zick , o. shehab , m. french , sci . * 5 * ( 2015 ) . : 10.1038/srep11168 v. choi , quantum . inf . process . * 7 * , 193 ( 2008 ) d. baum , a point - based algorithm for multiple 3d surface alignment of drug - sized molecules . ph.d . thesis , free university of berlin ( 2007 ) f.n . abu - khzam , n.f . samatova , m.a . rizk , m.a . langston , in _ ieee / acs international conference on computer systems and applications _ ( 2007 ) , pp . 367373 b. balasundaram , f. mahdavi pajouh , in _ handbook of combinatorial optimization _ , ed . by p.m. pardalos , graham ( springer new york , 2013 ) , pp . 15591598 b. balasundaram , s. butenko , i.v . hicks , oper . 59 * , 133 ( 2011 ) m.r . garey , d.s . johnson , _ computers and intractability : a guide to the theory of np - completeness _ ( w. h. freeman and company , new york , 1979 ) r.g . downey , m.r . fellows , theor . sci * 141 * , 109 ( 1995 ) m. hernandez , a. zaribafiyan , m. aramon , m. naghibi , arxiv preprint arxiv:1601.06693 ( 2016 ) e. boros , a. gruber , in _ international symposium on artificial intelligence and mathematics _ ( 2012 ) k.l . pudenz , t. albash , d.a . lidar , nat . commun . * 5 * ( 2014 ) . : 10.1038/ncomms4243 z. bian , f. chudak , r. israel , b. lackey , w.g . macready , a. roy , arxiv preprint arxiv:1603.03111 ( 2016 ) t. boothby , a.d . king , a. roy , quantum . process * 15 * , 495 ( 2016 ) a.d . king , c.c . mcgeoch , arxiv preprint arxiv:1410.2628 ( 2014 ) a. perdomo - ortiz , j. fluegemann , r. biswas , v.n . smelyanskiy , arxiv preprint arxiv:1503.01083 ( 2015 ) f. hamze , n.d . freitas , in _ proceedings of the 20th conference on uncertainty in artificial intelligence _ ( 2004 ) , pp . 243250 a. selby , arxiv preprint arxiv:1409.3934 ( 2014 ) c. xu , f. cheng , l. chen , z. du , w. li , g. liu , p.w . lee , y. tang , j. chem . model . * 52 * , 2840 ( 2012 ) s. boixo , t.f . ronnow , s.v . isakov , z. wang , d. wecker , d.a . lidar , j.m . martinis , m. troyer , nat . * 10 * , 218 ( 2014 ) t.f . rnnow , z. wang , j. job , s. boixo , s.v . isakov , d. wecker , j.m . martinis , d.a . lidar , m. troyer , science * 345 * , 420 ( 2014 ) m.h . amin , phys . a * 92 * , 052323 ( 2015 ) b.s . clarke , a.r . barron , j. stat . plan . inference * 41 * , 37 ( 1994 ) m.e.j . newman , s.h . strogatz , d.j . watts , phys . e. * 64 * , 026118 ( 2001 ) d.s . callaway , m.e.j . newman , s.h . strogatz , d.j . watts , phys . rev . lett . * 85 * , 5468 ( 2000 ) as mentioned in section [ subsubsec : parameter_setting ] , we generalize the approach presented by choi @xcite to theoretically set the parameters of the embedded graph , in which the connected subgraphs representing the logical qubits are not necessarily subtrees of the hardware graph . the procedure used in this paper is explained in algorithm [ parameter ] . it is worth mentioning that we do not discuss the proof for the validity of the theoretical approach . the key idea as presented by choi @xcite is to ensure that the ground state of the input graph before and after embedding matches . let us consider a conflict graph @xmath120 with logical local fields and couplings denoted by @xmath62 and @xmath63 , respectively . assume that logical qubit @xmath121 is represented by @xmath122 physical qubits forming the physical subgraph @xmath123 . further assume that @xmath124 is the set of neighbouring vertices of the logical qubit @xmath23 in the logical graph @xmath36 and @xmath125 is the set of neighbouring vertices of the physical qubit @xmath126 , @xmath127 , excluding the vertices representing the same logical qubit . the theoretical approach detailed below has an iterative pre - processing step in which several logical qubits might be removed from the logical graph , since their optimal values can be inferred in advance . the parameters are then set on the reduced logical graph . find the scaling factor to bring the distributed @xmath128 to the range specified by the user ( we use the range @xmath129 for both @xmath62 and @xmath63 in this paper ) multiply the distributed @xmath130 by the scaling factor set all ferromagnetic coupler values ( @xmath61 ) to @xmath70 since the quantum annealer is a stochastic solver , we consider the successive annealing runs as a sequence of binary experiments that might succeed in returning the ground state with some probability . let us formally define @xmath131 as a sequence of random independent outcomes of @xmath132 annealing runs , where @xmath133 denotes the probability of observing the ground state at the @xmath23-th anneal . defining @xmath134 as the number of successes observed in @xmath132 anneals ( @xmath135 ) , we have @xmath136 ( @xmath137 ) . that is , @xmath138 has a binomial distribution with parameters @xmath132 and @xmath139 . the @xmath80 then equals @xmath132 such that @xmath140 . it is easy to verify that @xmath141 . since the probability of success @xmath139 is unknown , the challenge is to estimate @xmath139 . we follow the bayesian inference technique to estimate the probability of success for each instance @xmath23 @xcite . in the bayesian inference framework , we start with a guess on the distribution of @xmath139 known as prior and update it based on the observations from the device in order to get the posterior . since the observations from the device have a binomial distribution , the proper choice of prior is a beta distribution which is the conjugate prior of the binomial distribution . this choice guarantees that the posterior also has a beta distribution . the beta distribution with parameters @xmath142 and @xmath143 ( the jeffery prior ) is chosen as prior since it is invariant under reparameterization of the space and it learns the most from the data @xcite . updating the jeffery prior based on the data from the device , the posterior distribution denoted by @xmath144 is then @xmath145 where @xmath146 is the number of calls to the quantum annealer , @xmath147 is the number of anneals in each call , and @xmath148 is the number of times that the ground state of instance @xmath23 is observed at the @xmath149-th call . to estimate the tts ( or @xmath150 ) for the entire population of instances with similar parameters , let us assume that there are @xmath151 instances with similar properties , for example , with the same number of variables . we are interested in using the data on these @xmath151 instances to estimate the tts for the entire population of instances with the same number of variables . after finding the posterior distribution @xmath144 for all instances in set @xmath152 , we use the bootstrap methodology to estimate the distribution of the @xmath153-th percentile of the tts . the procedure is described in algorithm [ tts ] . fix the number of bootstrap re - samples @xmath154 ( we set it at @xmath56 ) consider the empirical distribution of @xmath155 as an approximation of the true @xmath156 distribution ( we have plotted the mean and the standard deviation of this empirical distribution in figure [ fig : tts_main ] ) the qubo graphs of molecular similarity problem instances can be considered random graphs . a random graph is a collection of vertices with edges connecting pairs of them randomly . newman et al . @xcite and callaway et al . @xcite developed an approach based on generating functions to determine the statistical properties of random graphs with arbitrary degree distribution . here we review their approach to determine the percolation threshold . let us denote the probability that a randomly chosen vertex has degree @xmath1 by @xmath157 . let us further define @xmath158 where @xmath159 and @xmath160 are the generating functions for the probability distribution of vertex degrees and outgoing edges , respectively , and @xmath161 is the average vertex degree . callaway et al . @xcite have shown that the percolation threshold , or critical probability , can be calculated as follows : @xmath162 the details of the derivation can be found in @xcite . the key idea is that the critical probability is the point at which the mean cluster size goes to infinity . the context in which we apply this criterion is on random graphs whose degree distribution is known . it is known because the degree distribution can be measured directly @xcite . below , we provide an example with a known degree distribution for which we calculate the critical probability . * example * consider a graph similarity problem instance with 18 vertices and @xmath94 $ ] . the number of vertices with degree 11 , 12 , and 13 are , in respective terms , 8 , 8 , and 2 . the distribution of vertex degrees can be generated by @xmath163 applying the formula above , the critical probability is @xmath164 .

the paper demonstrates that the w uma - type systems have a potential of playing an important role in studies of the galactic structure due to their high spatial frequency of occurrence , ease of detection and reasonably tight absolute - magnitude ( period , color ) calibration . an algorithm , based on fourier decomposition of light curves , permitted to define a sample of 388 contact binaries with well observed light curves , periods shorter one day and with available colors ( the r - sample ) , from among 933 eclipsing binary systems in the ogle variable - star catalog for 9 fields of baade s window . the sample of such systems which was visually classified by the ogle project as ew - type binaries ( the o - sample ) is by 55% larger and consists of 604 stars . the algorithm prevents inclusion of rr lyr and sx phe stars which in visual classification might be mistakenly taken for contact binaries with periods equal to twice their pulsation periods . determinations of distances for the contact systems , utilizing the absolute - magnitude calibration and the map of reddening and extinction of stanek ( 1996 ) , indicate an approximately uniform distribution of contact binaries almost all the way to the galactic bulge ( there is a hint of a gap at 6.5 to 8 kpc ) , with 9 well observed systems identified in the bulge . the distances have been derived assuming two hypotheses : ( 1 ) extinction extends uniformly all the way to the bulge at kpc or ( 2 ) extinction is truncated at kpc . analysis of the period color diagram favors the latter hypothesis which has been assumed throughout the paper . the uniform distribution of the contact systems with distance , implying heights up to pc , as well as a tendency for their colors to be concentrated in the region normally occupied by old turn - off - point stars , confirm the currently held opinion that contact binary systems belong to the old stellar population of the galaxy . a first attempt to construct the luminosity function for contact binaries has been made for samples defined by distances of 2 and 3 kpc . the _ apparent _ density of contact systems is about systems per pc , with the main uncertainty coming from the definitions of the r- and o - samples . if the _ spatial _ density ( corrected for undetected low - inclination systems ) is two times higher than the lower limit of the above range , and equals systems per pc , then one star among 400 main sequence stars is a contact system ; however , this number includes m - dwarfs among which contact binaries do not occur . an independent estimate of the _ apparent _ frequency , relative to nearby dwarfs with colors similar to those of the contact binaries , obtained in the volume - limited sense to 2 and 3 kpc , is one contact system per about 250 300 main sequence stars , which agrees well with the recent estimates for old open clusters and their background / foreground fields .

astro-ph9607009

the optical gravitational lensing experiment ( ogle ) is a long term observing project aimed at detecting invisible matter in our galaxy by detection of gravitational microlensing events in the direction of the galactic bulge . it has been giving a rich spectrum of by - product results , among them excellent color magnitude diagrams ( udalski et al . 1993 ) which have provided interesting insight into the structure of the galactic disk ( paczyski et al . 1994a = p94 ) , non - axiality of the bulge ( stanek et al . 1994 , paczyski et al . 1994b ) , and maps of the interstellar extinction in baade s window ( stanek 1996 = s96 ) . the first utilization of another advantage of this dataset , of its systematic temporal coverage permitting detection of very large numbers of periodic variable stars , was presented by rucinski ( 1995a = bwc ) in a preliminary analysis of 77 contact binaries , from the first instalment of the periodic variable - star catalog for the central baade s window field ( udalski et al . 1994 ) . since then , the ogle project released the catalog of periodic variable stars in the remaining 8 fields of baade s window ( udalski et al . 1995a , 1995b ) , with data now available for the whole @xmath8 field . this database will be from now on called the catalog . the present paper is basically a full version of the bwc study for whole baade s window and addresses the same issue , which is the utility of contact binaries ( or w uma - type or ew - type systems , as we will call them below ) as distance tracers in galactic structure studies . from among all 933 eclipsing systems , 676 systems with periods shorter than one day were first selected , but this number was further substantially reduced to 388 by acceptance as a contact binary through a strict criterion based on the fourier decomposition of the light curves ( section [ alg ] ) and availability of @xmath0 colors . this selection was entirely independent of the original classification as ew - type systems in the catalog . we note , that among 662 ew systems in the catalog , 604 systems have @xmath0 colors and periods shorter than one day , so that our ( impersonal ) selection of 388 systems imposed more severe restrictions on the acceptance as a contact binary than the ogle classification . the present paper utilizes the fact that sizes of contact binaries can be estimated from their periods so that with color information absolute magnitudes can be derived permitting usage of these systems as distance tracers ( bwc , rucinski 1994a = r94 , rucinski 1996 = r96 ) . although functional dependencies of dynamical time scales are underlying absolute - magnitude calibrations for both , the rr lyrae - type stars and the contact binaries , the latter are probably less good standard candles than rr lyrae stars , mostly because of the relatively strong color dependence ( but infra - red observations might help here ) . depending on the combination of the color and period , the absolute magnitudes for contact binaries can be estimated to 0.2 0.5 mag . the spread for the rr lyrae stars of a given metallicity is some 2 4 times smaller ( cf . recent results for m5 by reid 1996 ) , although for large , inhomogeneous samples it might be substantially larger . we note that the small spread in luminosities of the rr lyrae stars is simply due to their occurrence in a very small region of the stellar parameter space . in this respect , the contact binaries have a great advantage : they occur with high frequency ( as we show later , in the solar neighborhood , they are some 24,000 more common than rr lyrae stars ! ) and , by being slightly evolved , they trace the most common main sequence population to greater depths than do ordinary single stars . also , apparently their luminosities are relatively insensitive to @xmath9 $ ] ( rucinski 1995b = r95 ) . besides , the rr lyr stars are present only in very old stellar systems , older than about 10 gyr , whereas contact binaries seem to start appearing in open clusters with ages around 1 gyr ( kauny & rucinski 1993 , rucinski & kauny 1994 ) . the luminosity calibrations for the contact binaries do have weaknesses related to occasional occurrence of spots on some systems and lack of good calibrators over the whole period color sequences ( rucinski 1996 ) , but hopefully the large numbers of these systems in the micro - lens databases might help compensate for these deficiencies or reveal trends which could be explored in more in - depth studies . origins and evolution of contact binary stars are still poorly understood although their low angular momenta , presence in old open and globular clusters ( kauny & rucinski 1993 , rucinski & kauny 1994 ) and space motions ( guinan & bradstreet 1988 ) suggest formation by a relatively slow process of coalescence of close , detached binaries through magnetic - wind induced loss of angular momentum . advanced evolution is particularly strongly suggested by occurrence of contact binaries among blue straggler stars in old open and globular clusters ( mateo 1993 , 1996 ) . there have been several general reviews on the subject of contact binaries , the most recent ones by rucinski ( 1993a ) and eggleton ( 1996 ) . of interest for this work are results of recent systematic searches in stellar clusters which have led to an upward revision of the _ apparent _ frequency , both in clusters and in the galactic field , to about one contact system per 280 stars ( r94 ) , with the implied _ spatial _ frequency ( corrected for undetected systems at low orbital inclinations ) of about 1/140 , in agreement with entirely independent estimates of mateo which were based on numbers of stellar x - ray sources ( hut et al . the contact systems are also recognized as extremely important for dynamical evolution of globular clusters ( hut et al . 1992 ) . there exists an abundance of light curve data for many bright contact binaries in the sky , but the material is highly inhomogeneous in terms of accuracy as well as sampling of the parameter space . perhaps the most troublesome is lack of standard color data even for quite bright w uma - type systems . the new micro - lensing samples contain more uniform data than presently available for field systems , the reason being that the latter in their majority have come from very non - systematic searches of the sky , with equally non - systematic follow - up studies . the general catalogue of variable stars lists 562 contact systems , but for only 510 of them are the periods reliably known , whereas moderately good photometric data exist for some 130 systems , but again only for half of that in standard photometric systems . thus , although the amount of information for each system in the ogle catalogue is limited , its quality is uniform or at least controllable , and can be easily subject to statistical scrutiny . section [ ecl ] of the present paper contains a brief assessment of the database for eclipsing systems in the catalogue . section [ alg ] describes an algorithm used to select a sample of contact binaries used as a primary dataset in the paper ( the r - sample ) . section [ dist ] discusses estimates of observed and absolute magnitudes of the systems and then of their distances . section [ pop ] discusses implications of the present results for our understanding of the contact systems and section [ sum ] gives a summary of the paper . the sample of eclipsing binaries in the ogle catalogue ( udalski et al . 1994 , 1995a , 1995b ) contains 933 entries . the data listed in the catalogue consist of : the @xmath10 and @xmath11 coordinates , @xmath12 , @xmath13 , @xmath14 , the period and zero epoch , and the light curve ( typically 100 190 points ) in @xmath15 . the sample contains systems with @xmath16 , but selection effects against discovery might be expected at fainter magnitudes due to progressively stronger blending of images . a histogram of magnitudes at maximum light , @xmath12 , shown in figure 1 , indicates an increase in numbers of detected eclipsing systems with decrease in brightness which is surprisingly consistent with the uniform - density , empty - space law of `` 4-times - per - magnitude '' increase . a clear deficiency relative to this law shows up for @xmath17 and an abrupt cut - off takes place at @xmath18 , close to the stated limit of the catalog at @xmath19 . possibility of selection effects at fainter magnitudes led us to consider in section [ dist ] the numbers of contact systems in relation to the total numbers stars in baade s window rather than in absolute sense . the histogram of amplitudes shown in figure 2 indicates that selection effects against discovery might be present for amplitudes @xmath14 smaller than about 0.3 , although this distribution can not be obviously separated from the intrinsic distribution of the variability amplitudes . the random errors in the data were estimated by evaluating spread in light curve points at phases close to phase 0.25 for all eclipsing systems in the catalog . this was done in a re - normalization step , while fitting the light curves by fourier cosine series , as described in the next section . the distribution of errors with brightness is shown in figure 3 where it is separated into four magnitude bins . from inspection of this figure , we conclude that typical random errors in @xmath15 ( per observation ) were about 0.02 for @xmath20 and increased somewhat to about 0.03 0.04 for @xmath21 . the ogle project , when releasing the variable - star catalog attempted a visual classification of variability on the basis of morphology of light curves . while separation of contact binaries from other eclipsing systems in most cases does not present problems , it would be useful to have an automatic classifier which would be impersonal , easy to repeat and if need occurs equally easy to modify . we would also like to have a simple way of removing poorly observed systems as these might be stars of other types ( spotted , pulsating ) or might have poor determinations of periods . such a classifier has been constructed here on the basis of the fourier decomposition of light curves . as was discussed in rucinski ( 1993b = r93 ) , light curves of contact systems are very simple in their shape and usually only two cosine coefficients are needed to adequately describe a light curve . although this revealed a basic difficulty of extracting several geometric elements from light curves of such a low information content , this is a convenient circumstance for large databases of variables observed with moderate accuracy , as in the case of the ogle catalog . as was described in r93 , comparison of the cosine coefficients @xmath22 and @xmath23 permits a crude estimate of the degree of contact which is weakly dependent on the inclination and mass - ratio of the system . this property was noted already in rucinski ( 1973 ) and explained through different sensitivity of @xmath22 and @xmath23 to the global distortion of the contact structure and to the more localized eclipse effects . however , it remained unclear whether the fourier coefficients would retain usefulness for detached systems as no single database existed which could be used to compare in large numbers light curves of detached and contact eclipsing system . the usefulness of the approach has been proven positively here by application of the fourier coefficients to the whole sample of the ogle eclipsing systems . the light curves of all 933 systems were analyzed exactly following the prescriptions in r93 and the results in the @xmath24 plane are shown in figure 4 . note that the light curves for the fourier analysis were expressed in light units and each time a new normalization was done for phases around 0.25 ( at least 10 observations ) . this normalization permitted to estimate spread of observations at these phases and gave information of measurement errors , as shown in figure 3 in previous section . in figure 4 , the systems have been marked differently following the ogle classification : ew or contact binaries of the w uma type ( filled circles ) , eb or systems showing beta lyrae light curves with unequal minima ( open circles ) , and ea and e detached binaries and binaries of uncertain type ( crosses ) . a theoretical envelope based on the results in r93 for the inner ( marginal ) contact is shown as a continuous curve ( see below ) . the ew - type systems are indeed confined below the theoretical limit for marginal contact ( with a well - known tendency for weak contact ) , while most of the systems classified as detached eclipsing ( ea , e and other uncertain , 250 systems in total ) are above this limit . the relatively small number of systems classified as beta lyrae - type ( eb , 21 systems ) basically follow the division line between the detached and contact groups . in what follows we simply disregarded this group and included those systems in the contact - binary sample which passed the fourier filter . we note that some genuine contact systems show unequal minima indicating poor thermal contact . this matter will remain open until the @xmath0 color curves become available , as then the correlated behaviors of the @xmath25 coefficients for the light and color curves could be used as a detection criterion for eb systems . once the color - curves become available , they will provide useful criteria to guard against inclusion of pulsating stars with periods equal to twice their pulsation periods . at present , we rely entirely on the fourier coefficients . in particular , negativity of @xmath23 prevents inclusion of the rr lyrae stars , although the rrc - subclass might occasionally create problems for less accurate light curves . as an illustration , we show in figure 4 the @xmath24 coefficient pairs for the pulsating stars in one ( central ) field of the ogle project . the envelope for the inner contact in figure 4 can be approximated very well by a following simple relation ( cf . figure 6 in r93 ) : @xmath26 , where both coefficients are negative . in this paper , position relative to the inner - contact curve is the main criterion for classifying a star as of the ew - type ( below the curve ) or other eclipsing ( above the curve ) . this criterion was supplemented by one measuring the over - all quality of fit by a simplified light curve consisting only of 5 fourier terms : @xmath27 . the fit was usually excellent for contact binaries , whereas for detached binaries it showed large systematic deviations which made the formal errors of the coefficients @xmath28 very large and thus basically useless in the analysis . typical fits are shown in figure 5 for the first ew and ea systems of the ogle sample , v4 and v19 , in the field bwc , where @xmath29 is the field number ( for the central field @xmath30 ) , and @xmath31 is the variable number in the field . ] . the @xmath24 diagrams for different ranges of the mean standard errors of the 5-term fits , in light units , are shown in figure 6 . whereas 525 systems with periods shorter than one day passed the @xmath24 filter , the requirement that the overall fourier fit be better than 0.04 eliminated most of the detached and many poorly - observed contact systems leaving 404 objects . among these , 388 systems have @xmath0 colors at maximum light . the sample of contact systems selected in this way will be , from now on , called the `` restricted '' sample or the r - sample . we feel that although this sample might have been somewhat conservatively selected , it does contain only genuine and well - observed contact systems . most of analyses were done for this sample . the sample of ew systems selected by the ogle project , limited to periods shorter than one day and with available colors consist of 604 systems and , from now on , will be called the o - sample . spot checks of those systems which belong to the o - sample , but were rejected for the r - sample fully confirm that , indeed , these cases should not be considered , either due to large measurement errors ( most probably due to seeing - dependent blending of images ) or to light curves with partial phase coverage . a more extensive discussion of information contained in light curves of the eclipsing binaries in the catalog is planned for a separate paper of this series . several elements are needed to derive distances to the contact systems . in addition to the observed maximum magnitudes and colors , @xmath15 and @xmath0 , one needs values of reddening and extinction as well as absolute - magnitudes derived from a calibration utilizing colors and periods . these details will be described in the subsequent subsections . however , much information is already contained in the observed color magnitude diagram ( cmd ) for the contact system which is shown in figure 7 . this diagram shows systems of the r - sample , together with a schematic representation of the distribution for normal stars . position of one blue system in figure 7 at @xmath32 and @xmath33 , # 6.121 should be noted . although its light curve seems to be well defined , a closer examination of the data by dr . andrzej udalski revealed that its color might be affected by image blending . another blue system in the o - sample , # 4.177 with @xmath34 , has been rejected by the fourier filter because @xmath23 for it was positive . however , its color seems to be genuinely blue . both systems fall outside the range where our absolute - magnitude calibration is strictly applicable so that very large estimates of their distances might be entirely incorrect . besides those two blue systems , we also see unusually red systems or systems with very short orbital periods . in the r - sample , systems # 0.160 , # 3.053 , # 5.143 , # 6.123 , # 7.112 , # 8.072 have @xmath35 . such very red colors are unusual for contact binaries . at this point , we might only hope that these are not indications of a systematic deficiency in the color transformations in the ogle data at the red end of the color distribution . turning to very short periods : the system # 4.040 has the orbital period of 0.228 day with @xmath36 , whereas the system # 3.038 , with the orbital period of only 0.198 day and @xmath37 must remain in the o - sample , as it has not passed ( marginally ) the fourier filter with @xmath38 and @xmath39 ; these coefficients suggest a detached , but very close system of strongly distorted , late - type components . these special cases will be a subject of another paper of this series . here , we concentrate on normal systems and limit ourselves to time - independent quantities , temporary disregarding outliers in the sample . a schematic outline of the location of the majority of stars in the ogle fields in the color magnitude diagram is shown in figure 7 by a shaded area . spatial distributions of these normal , predominantly single stars , were analyzed in p94 in terms of the galactic structure in the direction to the bulge . the slanted sequence of relatively blue stars was interpreted there as main - sequence stars of a young population , but this interpretation has been recently replaced ( kiraga , paczyski & stanek 1996 ; see also bertelli et al . 1995 and ng et al . 1996 ) by postulating that the sequence is formed by old , turn - off - point stars , progressively reddened with distance . the contact binaries scatter around this sequence in figure 7 , so that any inferences that we could establish for the contact binaries should possibly also relate to these stars as well . to visualize the relation of numbers of the w uma systems to normal stars , the next figure 8 shows sections of equal color vertically through the cmd in figure 7 . these sections were defined as bands 0.1-magnitude wide in @xmath0 in this part of the cmd where we see most of the contact binaries , i.e. within @xmath40 and @xmath41 . the sample of normal stars in figures 7 and 8 is the one used by udalski et al . ( 1993 ) and p94 in their discussions of the color magnitude diagrams . these cmd - stars were selected for best photometry and therefore represent a sample of different stars than those which were studied for periodic variability . an arbitrary normalization factor of 100 , used in figure 8 , was apparently approximately a correct one to show both , the cmd stars and the contact binaries , in one figure . in fact , the plots directly show that the frequency of occurrence of contact binaries in our sample is similar to , but exceeds , the apparent frequency for old open clusters and for milky way background / foreground , estimated from the recent ccd surveys ( r94 ) at about one w uma system per 280 stars . although , for the first time , we have here a luxury of being able to obtain the apparent frequency in color magnitude bins , and not just by relating numbers of detected w uma - type systems to numbers of observed stars , we should remember that ( 1 ) the sample of the cmd - stars is biased toward good photometric quality with many faint stars ( estimated here at 45% ) missing from the statistics , and ( 2 ) contact binaries are intrinsically brighter than normal stars , so that the samples are not co - spatial . in spite of these qualifications , the _ shapes _ of the distributions in figure 8 are the same to the accuracy of statistical uncertainties ( the reduced @xmath42 ) when the distributions are simply normalized to the total number of systems in each color band . in the four 0.1-mag wide color bands vertically across the color magnitude diagram in figure 7 , from blue to red we have 13 , 33 , 81 and 97 contact binaries of the r - sample , which are out - numbered by cmd - stars by factors of @xmath43 , @xmath44 , @xmath45 and @xmath46 , where the uncertainties are given by poisson errors . we will return to the matter of relative frequency in sections [ freq ] and [ local ] , in discussions of a volume - limited sample , after presenting determinations of absolute magnitudes and distances . to determine these latter quantities , an iterative scheme was applied which involved several components . these will be discussed in the next sub - sections . the sample of standard w uma - type systems which was used to derive the @xmath47 calibration was basically the same as in the previous related papers ( bwc , r94 , r95 ) , but the data have been somewhat improved . the final calibration ( rucinski 1996 , unpublished ) is only slightly different from the one which was used in bwc . it is based on 19 systems : 3 with known trigonometric parallaxes , 12 in high - latitude open clusters and 4 in visual binaries with fainter companions which can be placed on the main sequence . relative to bwc , the major improvements are in that ( 1 ) three systems in m67 and one in praesepe have now measured @xmath0 colors which confirm those previously estimated from @xmath48 colors , ( 2 ) new observations of the ct eri ( @xmath49 day ) in a visual binary permit extension to slightly longer periods than before , ( 3 ) systems in ngc 188 with colors transformed from @xmath48 are de - emphasized as some important systems there show unexplained deviations : er cep and es cep are too bright by @xmath50 mag , and the important evolved system v5 is too faint by @xmath51 mag , probably reflecting poor contact ( its relatively red color should be even _ redder _ in good contact ) . according to the most recent proper - motion study by dinescu et al . ( 1996 ) , es cep is not a member of ngc 188 . at present , the calibration extends over the following ranges : @xmath52 day , @xmath53 and @xmath54 . applicability of the calibration beyond the above ranges is unknown at this time . because of this limitation , figures subsequent to figure 10 distinguish ( by filled circles ) those cases where , after de - reddening , systems fall within the strict ranges in all three quantities . the equations giving the absolute magnitudes and the coefficients are : @xmath55 with @xmath56 , @xmath57 and @xmath58 , and @xmath59 ( per star ) . the bootstrap sampling as in r94 permitted another set of estimates , in terms of median and @xmath60-@xmath61 ranges for the distinctly non - gaussian distributions of the coefficients : @xmath62 , @xmath63 and @xmath64 . the strong correlations between the coefficients result in smaller uncertainties in @xmath65 than these large errors in the coefficients would imply . monte - carlo simulations show that within these ranges , predictions on @xmath65 should be good to 1-@xmath61 level of about @xmath66 and 2-@xmath61 level of about @xmath67 mag , although systematic errors are still possible , mostly because of spots on some of the late - type calibrating systems . comparison with the previous calibrations ( bwc , r95 ) indicates that the period coefficient is slightly larger in the absolute sense than before and that the color coefficient is smaller . although this makes little difference for the final values of @xmath65 ( since the terms are correlated ) , the new calibration is therefore somewhat less sensitive to the uncertainties in the reddening than the previous versions . we note also that the @xmath68-based calibration is relatively insensitive to variations in @xmath9 $ ] ( r95 ) . the lines of constant @xmath65 are over - plotted in the period color diagram in figure 10 ( section [ pc ] ) . to find individual values of the reddening ( @xmath69 ) , extinction ( @xmath70 ) , distance ( @xmath71 ) and absolute magnitude ( @xmath65 ) , an iterative procedure was used . this procedure will be described in sec.[det ] . it utilized maps of extinction @xmath72 and reddening @xmath69 in baade s window which have been recently published by stanek ( 1996 ; see also woniak & stanek 1996 ) . this excellent tool permits estimates of _ maximum _ values of both quantities in this complex region of strongly variable and patchy interstellar absorption . extinction in the @xmath15-band was found from : @xmath73 . both , @xmath74 and @xmath75 , are indeed the maximum values as they were estimated in s96 from the red clump stars in the bulge at about 8 kpc : we do not know the actual values of @xmath72 and @xmath69 for locations of our systems in space except that they should fall somewhere between zero and these largest values at some distance @xmath76 . this distance would measure the effective thickness of the absorption layer and , by definition , should be smaller than the distance to the bulge , here set at 8 kpc . although , following s96 , independent values of reddening and extinction were used in most analyses , for some simplified estimates that will follow , @xmath77 was sometimes assumed . an approximate average for the central portion of baade s window is @xmath78 ( paczyski et al . 1994a ) , hence @xmath79 and @xmath80 . these median values are a bit smaller than the modal ones of about 0.65 and 1.0 , respectively . however , as demonstrated in s96 , the range in extinction over the whole area is much wider , @xmath81 , so that allowance for positional dependence should substantially improve the quality of the data over an assumption of an average extinction . the input data for each star were : the position ( @xmath82 , @xmath83 ) , the observed magnitude ( @xmath15 ) , the observed color ( @xmath0 ) and the orbital period @xmath84 . two independent quantities were taken from the s96 maps : the maximum reddening @xmath74 and the maximum extinction @xmath85 . both have been derived by two - dimensional interpolation in the extinction maps for all systems of the sample . for 135 cases of stars outside the maps , from among 933 of the ogle sample , all falling within 2 arc minutes of the borders , the values at edges were assumed . the relation between the period and color ( pc ) , observationally established by eggen ( 1961 , 1967 ) , plays a special role in studies of w uma - type systems . this relation is only moderately tight as it reflects evolution of some contact systems leading to longer periods and redder colors for more evolved systems . however , the short - period / blue envelope of the pc relation is particularly important as it is expected to be well defined , being delineated by the least - evolved systems . differences in metallicities enter here and might affect the colors and location of the envelope , but as was shown in rucinski ( 1995b ) , the @xmath0 color is relatively insensitive to variations in metallicity , @xmath86 $ ] . for the galactic disk stars , the range in metallicities should be confined to @xmath87 < + 0.5 $ ] so that larger effects might be expected only for strongly metal - deficient systems ( rucinski 1994b ) . since we have no independent information on metallicities of individual systems , this matter will be entirely disregarded here . the shape of the short - period / blue envelope ( called from now on spbe ) has been determined on the basis of data for 59 bright , nearby systems , tabulated by mochnacki ( 1985 ) and transformed to @xmath0 using the main - sequence relations of bessell ( 1979 , 1990 ) . the recent improvements of the absolute - magnitude calibration ( sec.[cal ] ) have led to a confirmation that the the main sequence @xmath48 to @xmath0 transformations are basically valid for contact binary systems . it was found that the following simple , two - parameter approximation describes the spbe very well : @xmath88 ; the period @xmath84 is in days . this expression has no astrophysical significance and is used here only for convenience . the envelope and the data are shown in figure 9 . the period color relation for the r - sample is shown graphically in figure 10 , with vertical reddening vectors determined through iteration described in the next sub - section . if we concentrate our attention on lower ends of these vectors , we can see that practically all systems are located substantially below the spbe . obviously , a large fraction of this shift is due to the reddening , but part of the shift might be also due to the evolution . we also note that the systems occupy a much wider range of periods than currently in the standard - system sample ( which becomes unreliable above about 0.6 day ) , practically up to the ( conventional ) limit of one day . the long - period systems consist of relatively massive stars and are expected to evolve much more quickly than their late - type analogues so that the un - evolved envelope is particularly difficult to define for longer orbital periods . obviously , we have no information on evolution of individual systems , which is actually unimportant for distance determinations ( except that it actually helps seeing distant systems ) , but we can approximately correct for the reddening , using the data on maximum extinction and making an assumption on the extent of the dust layer . two assumptions on the effective line - of - sight thickness of the dust layer , @xmath76 , were used : 8 and 2 kpc . the former assumed that interstellar absorption increases linearly up to the distance of the bulge at @xmath2 kpc whereas the latter assumed , following arp ( 1965 ) and p94 , that beyond 2 kpc there is no further absorption . this second assumption is related to the small thickness of the dust layer in the galaxy . it is equivalent to postulating that at the distance @xmath89 kpc and the galactic latitude of the bw region of @xmath90 , the line of sight leaves the dust layer whose half - thickness at this point is about 150 pc . note , that the sun is supposed to be located about 20 30 pc above the plane ( reid & majewski 1993 , humphreys & larsen 1995 ) so that the constraint on the dust layer thickness is even more stringent . the iteration leading to a simultaneous determination of a distance and extinction / reddening started with an assumption of @xmath91 for each star and consisted of up to 4 cycles of the following steps ( always leading to stabilization of @xmath65 to 0.01 ) : @xmath92 the de - reddened period - color diagrams for both values of @xmath76 are shown in figure 10 , where they are identified as r@xmath93 and r@xmath94 . on the average , the results strongly favor @xmath3 kpc ( or perhaps slightly more ) as for @xmath2 kpc all systems fall far below the upper envelope . to explain this behavior in another way , we would have to assume that practically all systems are intrinsically different than in the solar vicinity , i.e. that they are strongly evolved with down- and right - ward shifts , away from the envelope of un - evolved systems . however , assumption of @xmath3 kpc carries also a danger of over - correcting the reddening and extinction is some individual cases . at this time , we feel that we must resort to treating the two values of @xmath76 as two extremes for the real spatial distribution of the interstellar dust . we note however that ng & bertelli ( 1996 ) , have recently presented arguments ( using the ogle data ) that @xmath95 kpc . as we will be shown below , the results of distance determinations for both assumptions on @xmath76 do differ for intermediate distances , but the differences are not very large . also , it should be stressed that many systems fall outside the strict applicability of the absolute - magnitude calibration which is marked in figure 10 by a broken - line box . this is perhaps most evident for periods longer than 0.63 days , where not only the calibration is uncertain , but where the upper envelope for nearby systems is also practically undefined . inspection of figure 10 shows that even for @xmath3 kpc only few systems fall above and to the left of the spbe . this can be explained by absence of very metal - poor systems ( with @xmath9 < -2 $ ] ) which are expected to have their own spbe , shifted to bluer colors ( rucinski 1994b , 1995b ) . the cmd for the contact binaries for the r - sample and @xmath89 kpc , in terms of absolute magnitudes @xmath65 and de - reddened colors @xmath96 , is shown in figure 11 . this figure contains also the simplified main sequence fits for pleiades , as given by paczyski et al . ( 1994 ) : @xmath97 , and for nearby stars , as given by reid & majewski ( 1993 ) : @xmath98 for @xmath99 and @xmath100 for @xmath101 . the contact binaries form a band above the main sequence , which is about 1.5 magnitude wide . the band is somewhat wider for systems which fall outside the applicability range of the calibration ( open circles ) . we encounter here , for the first time , an area of concern : a large fraction of the contact systems in baade s window are located in figure 11 above the calibrating systems from the solar neighborhood , the latter tracing well the main sequence dependences . the location above the main sequence is expected for contact binaries on two accounts : ( 1 ) they are binaries hence they might show a spread within 0 0.75 mag , ( 2 ) at least a fraction among them should show indications of advanced evolution , as the best explanation for their origin is through the relatively slow process of magnetic - wind braking with typical time scales of several gyr . the system v5 in ngc 188 ( v371 cep ) , discovered by kaluzny & shara ( 1987 ) ( see also kaluzny 1990 ) , might be taken as a prototype for such evolved systems . they should be , on the average , brighter than other systems and thus should be visible deeper in space leading to their over - representation in the ogle sample . we note that the numerous faint systems with @xmath102 tend to be rather uniformly spread within the whole length of the colors . an obvious over - population of the upper part of the diagram is definitely due to numerous distant systems , which are in fact very rare in space . the systematic deviations of absolute magnitudes could be also due to application of wrong reddening corrections , as the individual values of @xmath65 depend on the reddening - corrected color . however , the sense of the discrepancy is wrong . figure 11 shows the results for the preferred by us , the shorter dust length scale of @xmath89 kpc . for the longer scale of @xmath2 kpc , the discrepancy becomes larger and a gap opens up between the baade s window data and the standard - system data . since for the larger @xmath76 the reddening corrections are on the average smaller ( the reddening climbs slower with distance ) , these results would imply that even for @xmath89 kpc , the reddening corrections are systematically slightly too small . this is inconsistent with the assumption that that data in s96 gave maximum values of the reddening . unfortunately , we have no way of disentangling color shifts due to evolution from those due to reddening . in particular , we can not attribute the whole deviations in @xmath65 to incorrect reddening , as solutions of such an inverse problem are unstable and lead to a very large spread in @xmath69 . in this respect , the maps in s96 give us at least a way to confine the values of reddening within reasonable limits . the other possibility is that the @xmath0 colors contain a small systematic error ( in the sense of being too red ) for faint stars . we discuss such a possibility in section [ problem ] here we note that a simple allowance for such an error of about 0.2 mag in @xmath0 does not change our conclusions , in that the deviations in @xmath65 between baade s window and calibrating systems become only slightly smaller . the data in the cmd are subject to truncation from below and from above in absolute magnitudes . the upper boundary at about @xmath103 ( but dependent on color ) is related to the conventional limit on the orbital period of one day . it is shown by a dotted line in the figure which was obtained by inserting @xmath104 day into the absolute - magnitude calibration . the lower boundary is obviously related to the magnitude limit of the sample at @xmath105 . from the discussion of distances ( the next sub - section ) we have indications that the sample is probably complete to about 3 kpc . with the average extinction in the field , the corresponding absolute - magnitude limit would be then at @xmath106 . the luminosity functions in @xmath65 , based on systems to 2 and 3 kpc are shown in figure 12 . the faint end is by one magnitude deeper for the smaller volume , but the statistics is obviously more limited . the same functions , transformed from the @xmath65 to @xmath107 magnitudes are shown also in the lower panels of the same figure . the @xmath107 functions are blurred by a spread in @xmath0 , but have an advantage that they can be compared directly with the lf s for other objects . in particular , when we compare them with the luminosity function for blue stragglers in globular clusters ( fusi peci et al . 1993 , sarajedini 1993 ) , which peaks at @xmath108 and extends between @xmath109 with strongly tapered ends , we see that , on the average , the contact binaries are fainter than the globular - cluster blue stragglers so that only the relatively weak bright tail of the lf might contain a bs contribution . three directly observed quantities , @xmath84 , @xmath12 and @xmath0 , two values found by interpolations in the maps , @xmath70 and @xmath69 , and one derived value , the distance @xmath71 , can be related by combining our calibration with the standard expression linking observed and absolute magnitudes : @xmath110 in simplified relations below we normally set @xmath111 , with @xmath112 , but for individual systems the iterative process described above was used . we should note that accuracies of distance determinations depend practically only on uncertainties in magnitudes , colors and values of reddening as , for the present purpose , the periods are known practically without errors . thus , the respective contributions to the distance errors @xmath113 , from errors in magnitudes , colors and reddening would be : @xmath114 , @xmath115 , and @xmath116 . because of the size of the color coefficient @xmath117 in the absolute - magnitude calibration , the dependence on any possible systematic color deficiency in the calibration of the ogle photometric system is about two times smaller than it would have been for the standard technique of absolute magnitudes estimated from the main - sequence fitting . in particular , for a systematic error of @xmath118 , the distances would be systematically wrong by @xmath119% . results of the determinations of the distances for the r - sample and @xmath89 kpc are shown in figure 13 . the upper panel shows the data in the period distance plane . in principle , the two quantities should not correlate as barring really unusual and then extremely - interesting astrophysical causes the detection rate should be the same at all distances for a given period . in fact , we clearly see lack of short period , low luminosity systems beyond about 2.5 3 kpc , as expected given the magnitude - limited nature of the sample . this way , numerous short - period systems which dominate the sky - field sample , with a peak at about 0.35 day and a sharp cutoff at 0.22 day ( rucinski 1992 ) , are entirely eliminated for distances beyond about 3 kpc . at larger distances , rarer , long - period systems dominate in numbers . they do not show any tendency for a decrease in numbers beyond about 2.5 3 kpc , and seem to be visible all the way to the bulge , with a few systems admittedly outside the calibration applicability by about 1.5 mag located in the very bulge at 8 kpc . it is interesting to note a hint of a small gap between 6.5 and 8 kpc . up to about 6.5 kpc , the distribution in logarithmic units of the distance climbs approximately linearly without showing any structure , indicating approximately flat distribution in the distance . all 9 systems which seem to be located at the distance of the bulge have well defined light curves and are prime candidates for a difficult , but extremely important , radial - velocity follow - up study ) so that its distance might be entirely erroneous . ] . the systems in the bulge have designations : # 1.121 , # 1.138 , # 1.199 , # 2.132 , # 3.094 , # 3.142 , # 3.171 , # 3.178 , # 7.163 . except # 1.121 and # 3.094 which have @xmath120 , all are fainter than @xmath121 . this location is very close to the turn - off point of the bulge , which is now placed by kiraga et el . ( 1996 ) at about @xmath122 . in the cmd in figure 7 , all nine appear in the `` corner '' just below the disk ms , with @xmath123 . some of these nine systems show unequal minima . the curved , short - period cutoff in the distribution of distant systems can be explained entirely by the existence of the spbe on the period color relation . in order to be visible from large distances , a system must be blue and must have a long period . both quantities are constrained by the envelope . the upper panel of figure 13 shows the spbe , as in figure 10 , transformed to the distance period plane using the average extinction @xmath124 , for two values of @xmath15 , 17.9 and 17.0 . the un - evolved envelope of the period color relation does indeed set a strong limit on observability of distant contact binaries . our determinations of distances have been based on a more likely assumption on the spatial distribution of reddening , with an abrupt stop in its increase at @xmath89 kpc . has this assumption influenced the distance determinations to a large degree ? the results for the opposite extreme of @xmath125 kpc , designated as r@xmath93 , are shown in figure 14 . the distance distribution is generally flatter than for r@xmath94 , with largest changes taking place for intermediate distances . a weak depression in increase of numbers with distance at about 3 kpc seems to be somewhat stronger than before , but in general the distributions of distances are similar . the results presented in figures 13 and 14 seemed to the author so important that he could not resist a temptation to turn back to the whole o - sample and repeat the whole procedure again ( remembering that some systems of this sample might be less carefully screened for quality of data and classification than those of the r - sample ) . the results are shown in figure 15 , for @xmath89 kpc . they do not differ much from those for the r - sample . the small subset of systems in the bulge seems to be even better defined , forming a vertically clustered grouping at @xmath126 kpc in the upper panel . the relation between the distances and intrinsic colors for the r - sample is presented in figure 16 . we see a large range of intrinsic colors reaching quite red values for a few systems , but most are confined to a moderately narrow range between @xmath127 . the three systems which are intrinsically very red with @xmath128 are : # 3.053 , # 7.112 and # 8.072 . at large distances , there exists an obvious limit at red colors related to faintness of red , short - period systems . on the blue side of the diagram , we do not see any influence of the selection effects related to the spbe , since colors as blue as approaching @xmath129 should be permitted by the envelope . thus , there exist no contact systems in the ogle sample bluer than @xmath130 . having the determinations of distances , we can return to the color magnitude diagram for the whole sample and consider positions of systems in the 3-d space of observed @xmath0 , @xmath15 and distance . this is shown in figure 17 where distances are coded by sizes of symbols . as expected , distant systems cluster in the lower part of the diagram , with a striking excess in the `` corner '' below @xmath131 and within @xmath132 or @xmath133 . the selection effects operating here are related to two limitations of the sample . one of these is due to the cut - off of the sample at one day . for any distance , slanted straight lines ( parallel to the two straight ones for 4 and 8 kpc ) , define two different volumes : above the line , we see a mixture of systems at different distances , but below this line , systems must have periods _ shorter _ than 1 day to be detectable between us and the assumed distance . there exists also a limit on the period on the short side , which results from the period color relation , as not all combinations of periods and colors are permitted , both quantities being constrained by the spbe of un - evolved systems in the period color relation ( cf . section [ pc ] ) . these two limitations produce for each distance a region on the cmd diagram where contact binaries might be observed . two such regions are shown in figure 17 for the distances of 4 kpc and 8 kpc . similar `` catchment areas '' can be defined for other distances , by simple vertical shifts on the cmd . the areas do not impose limits on the presence of blue systems in the sample so that such systems do not apparently exist at large distances . finally , the distribution of the systems in the period magnitude diagram ( figure 18 ) shows a tendency for clustering at faint magnitudes and long periods . this tendency can be identified with the increase in numbers of systems at large distances , close to the limits of our sample , as the observed clustering tendency approximately follows lines of equal distance and approximately equal color . it is gratifying to see this link to the distances as it indirectly proves that our interpretations are basically correct . as was shown in the previous section , the sample of contact systems within the `` pencil - beam '' view of the ogle search is probably complete to 3 kpc . to see the sensitivity to this assumption , we will consider two distances , 2 kpc and 3 kpc , as limits for completness . for @xmath18 and assuming for simplicity ( but inaccurately ) the same absorption of @xmath124 , the limiting absolute magnitudes would be @xmath134 and @xmath135 , respectively . within the volumes defined by the two limiting distances , there are 27 and 98 contact systems in the r@xmath94-sample and 39 and 141 systems in the o@xmath94-sample . the _ apparent _ density of contact systems can be simply evaluated by dividing these numbers by the volume of the spatial cone defined by the ogle field . for the two limiting distances , one obtains then @xmath136 and @xmath137 stars / pc@xmath6 for the r - sample , and @xmath138 and @xmath139 stars / pc@xmath6 for the o - sample , where the errors come from the poisson statistics . we note that the estimates for 2 kpc and 3 kpc are very similar indicating completness of the sample . this should not be suprising as the contact binaries show a short - period , red - color cut - off so that faint systems practically do not exist . while the densities are well defined for both distances , the essential uncertainty here comes from the initial definitions of the r- and o - samples . the density of the contact systems is apparently high . in particular , if we multiply the above estimates of the apparent density by 2 times , to correct for undetected systems of low orbital inclinations , and compare the resulting _ spatial _ density with the local density of rr lyrae stars , we see that the contact binaries outnumber the latter by a huge factor . the local density of all rr lyr stars is @xmath140 stars / pc@xmath6 ( suntzeff , kinman & kraft 1991 ) , so that for a round number of @xmath141 contact systems per cubic parsec , we obtain the ratio of 24,000 times . obviously , this number is applicable only to the solar neighborhood as the rr lyrae stars show a strong galacto - centric concentration . while the density of the w uma - type systems in the current sample can be relatively easily evaluated , the relative frequency requires assumptions on the sample of normal stars to use as reference . it should be remembered that contact systems with spectral types later than about k2k5 do not exist , a limit apparently related to the existence of the sharp cutoff in the period distribution . this range of the spectral types coincides with the region where the luminosity function for normal stars climbs steeply , so that the results on the relative frequency must depend on the limiting distance or limiting absolute magnitude assumed for normal stars . by integrating the stellar density function of wielen , jahreiss & krger ( 1983 ) to limiting magnitudes of our sample ( which gives about 0.01 stars / pc@xmath6 ) we obtain a very high spatial frequency of of w uma systems of about 0.015 or one such a system per about 67 main sequence stars . however , this number critically depends on how far is the luminosity function integrated in its rising part , and carries uncertainty of about two times . it is actually easier to relate the number of w uma systems to the _ total _ number of stars by integrating the stellar density function all the way to very small stars . here we used the extension of the luminosity function to faint m - dwarfs by gould , bahcall & flynn ( 1996 ) which drops sharply for faint stars . such integrated total number density of 0.06 stars / pc@xmath6 is well defined . the absolute spatial frequency of the w uma systems is then 0.0025 , or one such a system per 400 stars , including m - dwarfs . the relative frequency of the contact systems can be also evaluated using the ogle data for normal stars . distributions for single stars in figures 7 , 8 and 16 were based on what we called in section [ cmd ] the `` cmd - stars '' , i.e. stars with high - quality photometry which could best define the color magnitude diagrams . the databases for these stars were easily accessible from the _ ftp _ sources , as described in p94 , so that these stars were the main basis for relating the contact binaries to normal stars . however , these cmd - stars are not useful for statistical analyses , because their numbers were biased in progressively fainter magnitude bins in a different way than stars which were analyzed for variability . for statistical applications , the full count of all stars observed and analyzed in the ogle project would be necessary . such a full catalog is not available at this moment . however , the first installment of the full catalog for the baade s window central field has just been published ( szymaski et al . 1996 ) . as described in this paper , statistics of faint stars are strongly biased in the cmd sample , with the ratio of all stars to the cmd stars reaching a large factor of 1.90 for the cululative count of stars with @xmath142 . using the new central - field data available over the _ ftp _ , we determined this ratio in magnitude intervals for the range of @xmath143 . in steps of one magnitude , it changes approximately as : 1.05 , 1.35 , 1.8 , 2.15 , 1.6 , and peaks at @xmath144 reaching 2.2 . the smoothed version of this ratio was applied as a correction factor for all 9 fields , in an estimate of the total number of stars that had undergone scrutiny for variability by the ogle project . to estimate the total number of main sequence stars to the distances of @xmath145 and 3 kpc , we counted all cmd - stars in all nine fields with @xmath146 ( to avoid contamination by the red clump giants ) , which fell above the main - sequence lines given by p94 , @xmath147 , for the assumed values @xmath124 , @xmath112 . these numbers were corrected by the magnitude - dependent factors ( as described above ) , giving the expected numbers of the main sequence stars with @xmath146 equal to 10065 and 33055 , to 2 and 3 kpc , respectively . using the numbers of the contact systems , as given at the beginning of this section , we obtain the apparent reciprocal frequencies ( in the sense of numbers of ordinary dwarfs per one w uma - type binary ) , and for both distances : @xmath148 and @xmath149 for the r - sample and @xmath150 and @xmath151 for the o - sample . remembering that the o - sample might contain mis - classified systems of other variability types and the r - sample might be overly restrictive , we can say that in round numbers the apparent reciprocal frequency is 250 300 . this frequency compares very well with similar estimates for the old open clusters and their background / foreground fields of @xmath152 and @xmath153 given in r94 . our current estimates of the apparent frequency of the w uma systems are substantially higher than estimated by duerbeck ( 1984 ) for nearby systems at about 1000 , confirming our earlier supposition ( kauny & rucinski 1993 , rucinski & kauny 1994 ) that the sample of bright , field systems is incomplete , being biased against low - amplitude systems . the implied _ spatial _ reciprocal frequency based on the ogle data is approximately 125 150 , assuming that undetected systems with low inclinations contribute one half of all systems . this result is consistent with the direct estimate based on stellar densities , at the beginning of this section , of about 67 main sequence stars per one contact system which can be evaluated only to a factor of about two times . the luminosity functions based on the volume - limited samples to 2 and 3 kpc have been discussed in section [ abs ] . here , we look at the statistics of the two most important , time - independent quantities , the orbital periods and intrinsic colors . the histograms of both quantities for the r@xmath94 sample , limited to @xmath154 kpc , are shown in figure 19 . the upper panel of figure 19 shows the distribution of orbital periods . it is very similar to such a distribution for field contact binaries ( rucinski 1992 ) , which shows a peak at short periods , with a sharp cut - off close to 0.25 day and a gentle slope for longer periods . apparently , fears expressed by this author that the field - system distribution might be affected by observational selection effects were premature . potentially , this histogram should contain very important information on the formation and structure of the contact system . we note that most of its features do not have good and un - equivocal explanations . the intrinsic - color distribution in the lower panel of figure 19 shows a broad maximum between @xmath155 with the total extension between @xmath156 . there exist also a few systems with very red colors , reaching @xmath157 . as we comment in section [ pop ] , the concentration of colors is reminiscent to that observed for turn - off point stars of old population , although the distribution for contact binaries is wider and extends further to the red . in section [ abs ] concerns have been expressed about a small but systematic difference in average absolute magnitudes between the baade s window systems and nearby systems used for our calibrations . this difference might indicate a problem with the color scale in the ogle data for faint stars . in fact , kiraga , paczyski and stanek ( 1996 ) have recently pointed out some evidence for lack of consistency between the color - magnitude diagrams in baade s window as available from ogle ( udalski et al . 1993 , paczyski et al . 1994a ) , from the hst ( holtzman et al . 1993 ) , and the color magnitude diagrams for the globular clusters 47 tucanae ( kauny 1996 ) and the open cluster ngc 6791 ( kauny & udalski 1992 ) . a possible reason for the inconsistency is a systematic shift in the ogle @xmath0 colors for the stars fainter than the `` red clump '' . the inconsistency would manifest itself for @xmath158 where the listed @xmath0 colors in the ogle data seem to become too red , with the discrepancy reaching about 0.1 in @xmath0 for @xmath121 , and stabilizing at this value for fainter stars . this is just a possibility at this time , and it may be caused by a non - linearity of the ccd detector ( cf . udalski et al . 1993 ) or , partly , by blending , with the red stars below the detection threshold affecting in a small , but systematic way the stars above the detection threshold . this issue shall be clarified in the near future using independent photometry based on a different ccd detector ( kauny & thompson , in preparation ) . to estimate the influence of such an error , the data have been analyzed with colors artificially changed following the above suspected dependence of the error on brightness , by assuming that the discrepancy increases linearly within @xmath159 to 0.2 and then stays constant . even for such an exaggerated case , the changes are small . the period color diagram in figure 10 is almost un - affected , although the gap in the distribution of points and the spbe less evident , especially for the r@xmath93 case . this is as expected , as faint systems , which would be mostly affected by the color error , have on the average longer periods ( cf . figure 18 ) . further experiments show that the discrepancy between the absolute magnitudes of the ogle systems and nearby systems in figure 11 remains for artificial data , although the deviations do become smaller . the relatively small change can be explained by our scheme of the coupled absolute - magnitude , distance , reddening determination which distributes color differences in a more complex way than just a simple shift @xmath160 . the largest differences appear in the distance period relations ( figures 13 15 ) : the data points tend to be more spread at large distances , there is no concentration of points at 8 kpc , and several systems are moved to distances as large as 10 kpc . the gap between 6.5 8 kpc is filled . the distance color relation in figure 16 obviously shows the same features in terms of distances , but the concentration in intrinsic colors around the top colors , which was discussed in section [ distr ] remains . thus , even if an error in the color is present , we can uphold our main conclusion of the paper on the global distribution of contact binaries in space , but some details would have to be modified . also , since the error may appear only at faint magnitudes , the results on the local frequency of contact binaries should remain entirely un - affected . two mechanisms of formation of contact binaries from detached binaries have been proposed . one involves the magnetic - wind evolution of the orbital angular momentum ( vant veer 1979 , vilhu 1982 , guinan & bradstreet 1988 ) and seems to be the main route to produce such systems in open clusters and in the field . the second mechanisms utilizes dynamical interactions in dense cluster cores , possibly involving binary systems ( leonard & fahlman 1991 , leonard & linnell 1992 ) , and is expected to contribute mostly in dense stellar clusters . of importance for distinguishing between these two mechanisms would be data on population characteristics of the contact binaries . although we have scant data on specific ages of contact binaries and we have no idea about the distribution of the ages , there are indications that they are quite advanced . this is suggested by space motions of the nearby systems ( guinan & bradstreet 1988 ) , which correlate weakly with @xmath161 metallicity indices ( rucinski , unpublished ) , as well as by the recent detections in old open clusters . there exist single contact binaries in praesepe and be 33 with ages of 0.7 0.9 gyr , but then their numbers increase for older clusters with several systems in the old clusters ngc 6791 , ngc 188 , be 39 and cr 261 ( kauny & rucinski 1993 , rucinski & kauny 1994 , mazur et al . 1995 ) . of particular importance is the fact that contact binaries seem to exist in old open and globular clusters on both sides of the turn - off point ( top ) . first w uma systems in globular clusters were found with surprisingly high frequency among blue stragglers ( mateo 1993 , 1996 ) , but recently yan & mateo ( 1994 ) discovered them also below the top ( but above and along the ms ) in m71 . observational selection effects are formidable below the top in globular clusters in most cases so that it would be premature to conclude that the ms contact systems do not exist there . however , the first data from the hubble telescope ( rubenstein & bailyn 1996 , edmonds et al . 1996 ) indicate that globular clusters might , indeed , be deficient in main sequence contact systems . an open cluster with systems on both sides of the top is the populous old open cluster cr 261 ( mazur et al . 1995 ) : for some 12 certain and 6 uncertain members of the cluster ( among 28 contact systems in the field ) , 4 are blue stragglers . not only this is the only cluster unambiguously showing w uma systems on both sides of the top , but its contact blue stragglers seem to have another property : their amplitudes are systematically smaller than for the rest of the systems . if this is not a statistical coincidence due to the small sample , it excellently links with the suggestion of eggen & iben ( 1989 ) that the extremely small mass - ratio system , aw uma , is a blue straggler in the field ( for references to other possible cases , cf . mateo 1993 ) . although we have no radial velocity or abundance information for the contact binaries in baade s window , their spatial distribution and relative frequencies , as well as preference of colors to group within a range around @xmath162 ( section [ local ] ) shed much light on their population characteristics . apparently , we see contact binaries at all distances without any indication of any structure in this distribution ; they do not seem to be concentrated to the galactic plane . these characteristics are consistent with an old galactic population . the concentration of colors to a relatively narrow interval is reminiscent of the enhancement observed as a broad vertical band on color magnitude diagrams of the stellar field ( gilmore 1990 , reid & majewski 1993 ) , in the range of `` vertical '' evolution in the top region . in the old population , there are practically no bluer stars than about @xmath163 or , equivalently , @xmath164 , but then , within @xmath165 or @xmath166 , the density is substantially enhanced by stars of the top region . obviously , the actual location of the enhancement depends on the reddening in the particular direction . expressed in the de - reddened colors , the enhancement for the contact binaries seems to be wider than the normal width of @xmath167 ; in figure 19 it extends to the red to @xmath168 or even further . while this extension can be perhaps explained through changes in stellar structure during the interaction of components when establishing the contact , one of the main goals for the future would be to establish whether the red cutoff is real and does not result from a selection effect or another bias in the photometric material . in figures 16 and 19 , some systems have surprisingly red colors , so that we directly see that contact binaries are not confined exclusively to the top region . having established that contact binaries belong to an old galactic population , we must pose the question : which old population is it ? it seems unlikely that contact binaries in their majority are as old as stars of the halo or the thick disk . according to gilmore 1990 and reid & majewski 1993 , the extended / thick disk population has scale heights of about 1400 1600 pc . the line of sight to baade s window passes the galactic center at the distance of about 560 pc , well within these scale heights , so that uniform density of the w uma systems in space would agree with characteristics of this population . the main problem is with the low local density of such very old stars of only about 2% . the apparent frequency of the contact systems found here suggests a spatial density of about 1/150 or more , or 0.7% , which would require that about 1/3 of these very old stars were contact binaries . this is difficult to accept . a much more likely parent population of the contact systems is the old disk of solar - age stars , with the scale height of about 325 pc . these stars dominate in numbers in the solar vicinity and are the most likely donor of close binaries for formation of contact systems . a hint of a break in the continuity of the distance distribution at about 6.5 kpc might be actually due to the line of sight leaving that disk at @xmath169 pc . the current data can not unfortunately shed light on the contribution of blue stragglers to the contact binary sample . possibly , a careful search in our database for low - amplitude , totally eclipsing systems similar to aw uma could lead to isolating a sample likely to consist of blue stragglers . we should note that if the nine systems which we see in the bulge do belong to the oldest galactic population , some of them might be actually blue stragglers as they are all relatively blue . however , even the bluest among them would be then relatively `` mild '' bs s , as their colors are only moderately different from those expected for the bulge top : the observed colors , @xmath170 , and the intrinsic colors , @xmath171 . the paper presents discussion of properties of contact systems discovered in 9 fields monitored during the micro - lensing project ogle . a simple automatic classifier based on the fourier analysis of light curves permitted isolation of 388 well observed contact systems with available @xmath0 colors from among 933 eclipsing systems in the ogle periodic variable star catalog . utilization of the @xmath172 calibration permitted determination of distances in an iterative scheme involving maximum values of reddening and extinction taken from the maps of stanek ( 1996 ) . analysis of the distance distribution reveals no discontinuities or unexpected features , except for those which can be at least partly explained by an interplay of known selection effects . an apparent lack of contact systems bluer than @xmath173 at large distances seems to be a genuine feature of the color magnitude diagram . most of the systems appear in the color range @xmath127 , i.e. the blue edge of the distribution is the same as for stars of the old population which concentrate in the turn - off point ( top ) region , but the red edge appears at redder colors . contact binaries with periods longer than one day , which are intrinsically brighter than the w uma - type variables , but which normally are not considered together with the w uma - type systems will be the subject of a separate investigation . no absolute calibration is at present available for those long - period systems . we note , however , that the calibration for the w uma - type systems within @xmath174 day is really an extrapolation and should be properly established . also , due to its particular importance , the short - period / blue envelope ( spbe ) for the un - evolved systems , in the period color diagram , urgently needs a better definition . these progresses can come only with an increase in quality and availability of photometric data for contact binaries , which are still lagging behind availability of light curves . a first attempt to construct luminosity functions for volumes limited by distances of 2 and 3 kpc , where selection effects should be small was made . these functions are truncated at @xmath175 and @xmath176 , respectively by the magnitude limit of the sample . this truncation makes it impossible to firmly establish whether there exists excess of contact binaries in the top region , whose presence is indicated by the distribution of intrinsic colors for the whole sample . extension of the luminosity functions to fainter magnitudes should be one of the primary goals for the future . the numbers of stars contained in the volumes used for the luminosity functions permitted an estimate of the frequency of contact binaries . the apparent frequency equals to one contact system per about 250 300 main sequence stars . this estimate , with an approximate correction for undetected , low - inclination systems , leads to the spatial frequency of about 1/125 1/150 . a direct evaluation of the spatial density of the contact systems of about @xmath177 per cubic parsec leads to an even higher frequency when comparison is made with main sequence stars of similar absolute magnitudes . while the density estimate is quite robust , the relative frequency estimate obtained that way is uncertain as it depends strongly on the assumed magnitude limit in the rising part of the luminosity function for normal stars . the high spatial frequency of the w uma - type systems , together with the uniform spatial distribution all the way to the bulge and with the preference for intrinsic colors to cluster within the @xmath96 interval of 0.4 to 1.0 suggest that the contact binaries belong to the old stellar galactic population . considering predominance of the old disk of solar - age stars in the solar neighborhood over the extended / thick disk or halo populations , it is most likely that the old disk stars are the main donor of contact binaries . the relative contribution of the contact blue stragglers to the sample can not be presently evaluated . however , analyses of individual light curves might help in isolating a sub - sample of very low mass - ratio systems , possibly dominating among the contact blue stragglers . the author would like to express thanks to the ogle project for access to their database and to hilmar duerbeck and yuen ng for useful comments . special thanks are due to bohdan paczyski , andrzej udalski and krzysztof stanek for their expert help with various technical and interpretative details of this work . udalski , a. , kubiak , m. , szymaski , m. , kauny , j. , mateo , m. & krzeminski , w. 1994 , aca , 44 , 317 udalski , a. , szymaski , m. , kauny , j. , kubiak , m. , mateo , m. & krzeminski , w. 1995a , aca , 45 , 1 udalski , a. , olech , a. , szymaski , m. , kauny , j. , kubiak , m. , mateo , m. & krzeminski , w. 1995b , aca , 45 , 433 vant veer , f. 1979 , , 80 , 287

improved resolution in both , energy and momentum in arpes - data has lead to the establishment of a definite energy scale in the dressed quasiparticle dispersion relations . the observed structure aroundmev has been taken as evidence for coupling to phonons and has re - focused the debate about the mechanism of superconductivity in the cuprates . here we address the relative merits of phonon as opposed to spin fluctuation mechanisms . both possibilities are consistent with arpes . on the other hand , when the considerations are extended to infrared optical data , a spin fluctuation mechanism provides a more natural interpretation of the combined sets of data in bisrca cuo ( bi2212 ) .

cond-mat0304029

with the advent of improved resolution , both in energy and momentum , several angular resolved photo emission spectroscopy ( arpes ) groups have measured the renormalized quasiparticle energies and/or lifetimes in the cuprates . the data shows that a definite energy scale exists for the renormalization . this energy scale has tentatively been assigned by some@xcite to interactions of the charge carriers with phonons . this provocative possibility has again brought to the forefront the open question of mechanism in the cuprates and in particular the possibility that the electron - phonon interaction plays an important role . while the electron - phonon interaction is widely believed to cause superconductivity in conventional materials , the gap symmetry in the cuprates is @xmath3-wave@xcite rather than @xmath4-wave which means that it is the projection of the electron phonon interaction on the @xmath3-channel which enters the equation for the critical temperature @xmath5 . this would imply that some of the details of the electron - phonon interaction would be drastically different in the oxides compared to conventional metals . while conventional superconductors do exhibit anisotropy@xcite in their superconducting gap as a function of momentum * k * on the fermi surface , the @xmath4-channel always overcomes the @xmath3-channel interaction . also in dirty @xmath4-wave materials with the mean free path @xmath6 smaller than the coherence length @xmath7 , the anisotropy is washed out and the gap becomes isotropic . on the other hand , in the pure limit there are many ways in which this gap anisotropy manifests itself in a conventional superconductor.@xcite one example is the low temperature @xmath8 electronic specific heat which exhibits the expected exponential dependence @xmath9 only for temperatures much lower than the lowest gap @xmath10 in the system . arguments for why the form of the electron - phonon interaction can be very different in the oxides as compared with conventional metals have been presented in the literature . some of the critical ideas have been reviewed by kuli.@xcite the main argument for our purpose here is that the electron - phonon interaction becomes strongly peaked in the forward direction when corrections for strong correlations are properly accounted for through charge vertices.@xcite another related idea is that because of the large static dielectric function in the oxides the screening is greatly reduced as compared with ordinary metals . this also leads to enhanced forward scattering as can be seen@xcite from the following simple argument . the fourier transform of the bare coulomb potential diverges like @xmath11 for small * q * , here * q * is the momentum transfer . screening eliminates this singularity and the screened potential goes , instead , like @xmath12 , where @xmath13 is an inverse screening length . for small @xmath13 the increased forward scattering has the general tendency of increasing the weighting of the higher harmonics in its expansion . for example the expansion of a delta function @xmath14 has equal weight in each spherical harmonics.@xcite strong forward scattering can in fact lead to @xmath3-wave superconductivity which is not suppressed by the strong coulomb repulsion in contrast to the @xmath4-wave channel.@xcite while the possibility of phonon induced @xmath3-wave superconductivity can not be eliminated on general grounds , it has not been favored in much of the literature . while there is no consensus on mechanism , many workers believe instead in an electronically driven scenario : for instance , the @xmath15 model can exhibit superconductivity.@xcite a spin fluctuation mechanism , such as is envisaged in the nearly antiferromagnetic fermi liquid ( naffl ) model of pines and coworkers@xcite is also electronic in nature with the exchange of spin fluctuations rather than phonons . another electronic mechanism is the marginal fermi liquid ( mfl ) model@xcite which in its original form had @xmath4-wave symmetry . in this paper we attempt to understand the renormalization effect observed in arpes within a boson exchange mechanism . we will consider explicitly both , phonon and spin fluctuation exchange . we will also try to reconcile within the same model , both , arpes data on equilibrium quasiparticle properties and the optical conductivity . infrared measurements in the cuprates have produced a wealth of information on charge dynamics in the oxides . they have played a key role in our present understanding of the microscopic nature of the cuprates . in particular , @xmath16-axis infrared conductivity data have given detailed spectroscopic information on the pseudogap@xcite which has been widely viewed as a direct manifestation of strong correlation effects . recently , improvements in resolution have also been achieved in @xmath17-plane measurements.@xcite from reflectance measurements as a function of frequency it is possible to extract separately the real and the imaginary part of the conductivity @xmath18 as a function of @xmath19 for a fixed temperature @xmath20 . within a generalized drude model , in which the optical scattering rate and optical effective mass acquire a frequency dependence , we can define an isotropic optical scattering time @xmath21 as @xmath22 , where @xmath23 is the plasma frequency . this quantity can be determined from the optical sum rule on the real part of @xmath18 , namely @xmath24 . the optical scattering rate defined above is related to the quasiparticle lifetime which may be considered to be more fundamental as they directly define quasiparticle motion . while these two lifetimes are related , they are by no means identical , however , except for a simplified case of isotropic elastic impurity scattering . even in this case , if the electronic density of states has an important dependence on energy or the elastic scattering is anisotropic , quasiparticle and transport scattering rates are no longer the same . note that , in as much as the in - plane conductivity is isotropic , the optical scattering rate is also isotropic and represents a weighted average of the more fundamental momentum dependent quasiparticle scattering . while there have been some attempts in the literature@xcite to compare both sets of data , i.e. : arpes and optical directly , we will emphasize here that , because they are not so simply related , they can not easily be compared . on the other hand , any viable microscopic model needs to be able to provide a unifying description of both sets of experiments . in this regard , we will emphasize that in addition to the differences between quasiparticle and transport scattering rate which we have already discussed , they differ in another fundamental way . in a boson exchange mechanism the charge carrier - exchange boson spectral density ( denoted by @xmath25 in the explicit case of phonons ) which describes the equilibrium properties can be quite different from the corresponding transport spectral density , usually denoted by @xmath26.@xcite the origin of this fundamental difference lies in a well - known factor @xmath27 , where @xmath28 is the angle between initial and final electron momentum undergoing a scattering process . for quasiparticle properties such as its lifetime , only the probability that an electron of momentum * k * leaves the state @xmath29 is relevant , while for transport , for instance for the d.c . resistivity , it is also important where it ends up since backward scattering depletes the current more than forward scattering . this basic physics is accounted for by the @xmath27 factor which eliminates forward scattering @xmath30 and weights with a factor two backward scattering @xmath31 . serious estimates@xcite of these two spectral densities for a model electron - phonon interaction in the poor screening regime with explicit inclusion of correlation effects , have lead to an estimate that @xmath26 may be smaller by about a factor of @xmath32 compared with @xmath25 , the corresponding quasiparticle quantity . on the other hand , in the naffl model the interaction with spin fluctuations is believed to strongly peak around @xmath33 in momentum . this means that the scattering potential itself weights more strongly the backward scattering processes , and so we expect that the transport spectral density will be larger than its quasiparticle counterpart . the above arguments immediately suggest that phonons effects could very well show up prominently in quasiparticle properties while at the same time play only a minor role in transport , i.e. : in the infrared conductivity and vice versa for spin fluctuations . in section ii we make general remarks about the renormalization effects due to boson exchange mechanisms based on perturbation theory . in particular we contrast the electron - phonon case with spin fluctuations . in section iii we deal with fits to the arpes data in bi2212 and in section iv we consider , in addition , optical conductivity data . we describe the constraints on microscopic models that a fit to both sets of data imposes in the specific case of bi2212 . in section v we draw conclusions . for an electron - phonon system , the renormalization of the electronic quasiparticles follows from a knowledge of the electron - phonon spectral density @xmath25 . this function which depends only on frequency , and which is limited in range to the maximum phonon frequency , contains all of the complicated information about electronic band wave functions , dispersion relation , phonon dynamics , and electron - lattice vibration coupling which is needed to compute self energy effects . in lowest order perturbation theory , the quasiparticle scattering rate @xmath34 at energy @xmath19 is given by@xcite @xmath35 for a @xmath36-function einstein phonon at @xmath37 of the form @xmath38 @xmath39 and we see that @xmath37 sets the energy at which @xmath40 jumps from zero to a finite value . for an @xmath25 distributed in energy , the rise will be more gradual but for @xmath41 , with the debye energy @xmath42 , the scattering rate will , again , become constant . the expression for the optical scattering rate , on the other hand , which is also obtained in lowest order perturbation theory is@xcite @xmath43 which gives @xmath44 for @xmath45 in an einstein model , but now for @xmath46 @xmath47 which starts at zero with @xmath48 and gradually increases towards @xmath49 with increasing @xmath19 . this is quite distinct from the abrupt jump to @xmath50 at @xmath48 found for the quasiparticle scattering rate . here the subscript ` @xmath51 ' denotes transport as opposed to equilibrium properties . it is clear that the singularity in @xmath52 associated with the boson energy scale @xmath37 is more significant in the quasiparticle ( equilibrium case ) than in the transport scattering rate . there is yet another difference between quasiparticle and transport rates . if we denote the electron - phonon interaction matrix element by @xmath53 for electrons scattering from * k * to @xmath54 with a phonon of energy @xmath55 ( @xmath56 is a branch index ) the quasiparticle spectral density is @xmath57 with the double average over the fermi surface denoted by @xmath58 @xmath59 where @xmath60 is the chemical potential , so all integrals are pinned at the fermi surface . the optical spectral weight is related@xcite instead to @xmath61 the extra factor of @xmath62 with @xmath63 the velocity of the electron * k * weights forward scattering by zero and emphasizes most backward collisions . the functions @xmath25 of eq . ( [ eq:5 ] ) and @xmath26 of eq . ( [ eq:7 ] ) are isotropic , momentum averages , given by the right hand side of each of these equations . both involve a double integral over initial and final states @xmath29 and @xmath64 . the full complicated momentum dependence of the electron - phonon coupling @xmath53 and of the phonon dispersion @xmath55 is included with no approximations , although after the double summation over momentum , indicated in eq . ( [ eq:6 ] ) , @xmath25 is now isotropic , dependent only on frequency . this is the function that enters isotropic eliashberg theory . an anisotropic formulation of the eliashberg equations@xcite also exists and in this case a directional electron - phonon spectral density distinct for each electron state @xmath29 on the fermi surface , which we denote by @xmath65 , would replace its fermi surface average : @xmath66 to calculate @xmath25 from first principles is difficult and complex.@xcite it can , fortunately , be measured directly from tunneling data through consideration of current voltage characteristics.@xcite the resulting function of @xmath19 often , but not always , looks qualitatively very much like the phonon frequency distribution @xmath67 but this does not imply that momentum dependence of coupling and/or of phonons is not included in eq . ( [ eq:5 ] ) . any nesting effects such as are present in the nested fermi liquid theory of virosztek and ruvalds@xcite and in the work of savrasov and andersen@xcite are fully incorporated on the right hand side of eq . ( [ eq:5 ] ) . it is only because we take the left hand side from experiment that these important issues do not become prominent in our work . a further complication arises in the case of coupling to spin fluctuations.@xcite in this case the magnetic spin susceptibility @xmath68 is involved as well as its coupling to charge . the susceptibility in the cuprates is sharply peaked in momentum space at the antiferromagnetic wave vector @xmath69 ( momentum transfer ) for electron scattering from @xmath29 to @xmath64 . in this case , what enters eliashberg theory@xcite is the complex function @xmath68 weighted by electron coupling . for instance , in the naffl model@xcite @xmath70 with @xmath71 , @xmath72 the static susceptibility , @xmath73 the magnetic coherence length , @xmath74 the commensurate antiferromagnetic momentum , and @xmath75 the characteristic energy of the spin fluctuations at the point * q*. a fermi surface to fermi surface approximation does not naturally reveal itself since this may not include transitions with @xmath76 for which the susceptibility is the largest . however , one can still introduce , as an approximation , some average effective electron - spin fluctuation spectral function denoted by @xmath77 which can be used in the isotropic version of the eliashberg equations . it represents the appropriate average of the susceptibility that enters superconductivity . its value is to be determined from consideration of experimental data . as described by carbotte _ et al._@xcite this isotropic spectral density is given by the second derivative of @xmath19 times the optical scattering rate @xmath78 . recall , @xmath79 , where @xmath80 is the in - plane optical conductivity . this conductivity is isotropic in tetragonal systems . for the orthorhombic case , @xmath80 with @xmath81 and @xmath82 can be different . defining @xmath83 the approximate relation @xmath84 holds.@xcite we emphasize that in terms of microscopic theory @xmath85 is related to an appropriate average of the full momentum and energy spin susceptibility @xmath68 ( eq . ) . recently , infrared optical data has been used very effectively to obtain @xmath86 in the oxides and we will use these in our work . in this way we circumvent all of the complications that would arise in a first principle calculation of the average effective function @xmath86 . we will return to this important point later in equation ( [ eq:10 ] ) . we note in passing that @xmath85 very often exhibits a pronounced resonance like structure particularly if @xmath87 data at low temperatures in the superconducting state is analyzed . in the following we use the expression ` optical resonance ' in reference to such a structure . in most materials analyzed so far , this optical resonance has a spin resonance equivalent measured using inelastic neutron scattering . we will discuss this in more detail in sec . [ sec:4 ] . in some sense the opposite extreme to the naffl model of pines and coworkers@xcite in which the interaction peaks at @xmath33 is the mfl model of varma _ et al._@xcite in which the interaction with the fluctuation spectrum is thought of as momentum independent in a first approximation . as originally conceived , this model leads to an @xmath4-wave superconducting gap rather than @xmath3-wave as is now generally believed to be the case . this represents a real limitation for the model . nevertheless , it has been very useful in correlating much data on the normal state transport properties in the cuprates . in as much as it is only the magnitude and energy dependence of the resulting average spectral function @xmath25 ( @xmath77 ) that matter , these may not be very different between the mfl and naffl models . both have an average spectral density which is reasonably flat as a function of energy and which extends to some high frequency of order several hundred mev . of course , if anisotropies on the fermi surface were to be considered , the directional @xmath65 ( @xmath88 ) , eq . , would be expected to vary strongly in the naffl model and not in the mfl model , but here , for simplicity , we have assigned to each electron the same average interaction . for phonons in conventional metals the mass renormalization parameter @xmath89 can vary by several tens of percentage points over the fermi surface.@xcite for the naffl model based on the susceptibility , the variations are much larger , and can be of the order of a factor of two or three.@xcite it is clear that arpes data in a particular direction could considerably under or overestimate the average renormalizations . in this sense optics is better . we should make one further remark . in conventional materials for which @xmath25 and @xmath26 have been calculated from band structure and phonon information usually obtained by inelastic neutron scattering , they have been found to differ in shape as a function of @xmath19 and in size . but these differences have often been overlooked in the literature . in the work of kuli and zeyher @xcite the corresponding differences are large . these authors considered directly the cuprates and try to account seriously for the ionicity and the increased effect of correlations . as described in the introduction , they find that the electron - phonon interaction in this case is strongly peaked in the forward direction and this leads to large differences between @xmath25 and @xmath26 . they estimate @xmath25 to be about a factor of three larger than @xmath26 . the consequence of this is that phonons would show up much more prominently in quasiparticle properties ( equilibrium ) than in the corresponding transport property . in particular , at large @xmath19 , @xmath34 and @xmath87 ( eqs . ( [ eq:2 ] ) and ( [ eq:4 ] ) , respectively ) would approach each other if @xmath25 and @xmath26 were identical , but in fact @xmath87 is expected to be larger than @xmath34 . the opposite would hold for the naffl model which emphasizes backward rather than forward scattering because the interaction peaks at @xmath33 . in this case spin fluctuations should show up more prominently in transport than in equilibrium properties . even though these differences are hard to quantify , the general qualitative features of transport as compared with quasiparticle spectral weight just described must be kept in mind when analyzing data . in the mfl model we expect smaller differences between quasiparticle and transport spectral densities because the underlying interaction is assumed to be approximately momentum independent . so far we have only discussed scattering rates and have emphasized the differences and similarities between infrared optical absorption and quasiparticle lifetimes . in arpes the dressed electronic dispersion relation is measured and denoted by @xmath90 . it is related to the bare band dispersion @xmath91 through the electron self energy . in the electron - phonon case with an einstein phonon spectrum and at @xmath92 ( zero temperature ) the self energy reduces to a simple form @xmath93.\ ] ] in eq . ( [ eq:8 ] ) @xmath94 is the electron mass enhancement parameter defined as @xmath95 where @xmath96 is the renormalized electron mass at the fermi surface . in terms of @xmath25 , @xmath97 for the simplified einstein case . the imaginary part of eq . ( [ eq:8 ] ) just gives back @xmath98 of eq . ( [ eq:3 ] ) . the real part gives the renormalized energies as solutions of equation @xmath99 and for @xmath100 ( on the fermi surface ) this gives @xmath101 : the quasiparticle mass is simply renormalized to @xmath96 . note that in eq . ( [ eq:8 ] ) the self energy has a logarithmic divergence at @xmath102 and this leads to a singular structure in @xmath90 at that energy as has been investigated in detail by verga _ et al._@xcite to which the reader is referred for details . here we have chosen to emphasize instead the nature of the structure in the imaginary part . it is sufficient to remark that for a finite distribution of phonon energies in @xmath25 the logarithmic singularity is moderated as it also is when temperature @xmath8 is included . in the general case , @xmath103 takes the form @xmath104 , \label{eq:9}\end{aligned}\ ] ] where @xmath105 is the digamma function . in lead , the electron - phonon spectral density @xmath25 is well known from tunneling spectroscopy@xcite as well as from direct calculations . it is shown as an insert in the top frame of fig . [ fig:1 ] . in lead as a function of the bare band energy @xmath91 ; the renormalization is due to the electron phonon interaction characterized by the electron - phonon spectral density @xmath25 shown in the inset . the bare dispersion is shown as the dotted line . the other curves are for the normal state at @xmath106k ( solid ) , @xmath107k ( dashed ) , and @xmath108k ( dash - dotted ) . bottom frame : same as the top frame but for the superconducting state at @xmath106k ( solid ) and @xmath109k ( dashed ) just below @xmath110k . the normal state results are shown as gray lines for comparison . , width=340 ] the phonons extend to @xmath111mev and show a characteristic peak for transverse and for longitudinal branches . in the top frame of this figure we show numerical results in the normal state for @xmath90 on the fermi surface as a function of the bare band electronic energy @xmath91 up to @xmath112mev at which point bare ( dotted curve ) and renormalized energies are pretty well parallel to each other . the solid curve applies to the normal state at @xmath113 . we see structure in the renormalized quasiparticle energy @xmath90 ranging up to roughly @xmath114mev . the structure clearly corresponds to the structure in the phonon spectrum ( inset ) . beyond this range multi - phonon processes are operative and the renormalization effects are less . further results in the normal state are for @xmath107k ( dashed ) and @xmath108k ( dash - dotted ) . we see that by the time this last temperature is reached , which is of the order of the maximum phonon energy , the thermal effects have smeared out much of the structure . note also , that in all three cases the curves go through zero at @xmath115 and the slope of each line out of the origin gives the renormalized effective mass parameter @xmath116 at the temperature @xmath20 . for the convenience of the reader we included in the bottom frame of fig . [ fig:1 ] results for the superconducting state at @xmath106k ( black solid ) and at @xmath109k ( black dashed ) just below the critical temperature of lead which is @xmath110k . we note that in each case the phonon structure is somewhat more pronounced than in the corresponding normal state ( shown as gray lines ) . also at @xmath117 the renormalized energy @xmath90 now goes to a finite value equal to the superconducting gap @xmath118 at that temperature . in an arpes experiment the gap is seen as a shift of the leading edge of the electron spectral density @xmath119 , downward from the chemical potential level . before moving on to other possible models for the renormalization we wish to make an important point about how some of the arpes data has been analyzed in the literature . it is essential to be aware that the bare electron band energies are not known independently in arpes experiments . however , electron momentum change @xmath120 can be measured in the particular direction of interest . the bare energy is then related to this momentum change through the fermi velocity @xmath121 . the bare value of @xmath121 denoted @xmath122 is then determined from @xmath90 under the assumption that at @xmath123mev renormalization effects have become small and are negligible in a first approximation . this assumption can , however , result in a significant underestimate of the true mass renormalization @xmath94 involved as is illustrated as a function of the bare band energy @xmath91 for various models for the electron - boson exchange spectral density . the dotted curve is the bare dispersion relation and is for comparison . the solid curve is for an einstein spectrum with @xmath124ev and a mass renormalization factor @xmath125 . the dashed line which shows a slight curvature is for a model in which the charge carriers are coupled to spin fluctuations described by an mmp - form , eq . ( ) , with a spin fluctuation energy @xmath126mev and a mass renormalization @xmath127 . the dash - dotted line is a straight line construction through @xmath128mev which is used in arpes experiments to determine the `` bare '' dispersion as it is not independently known . , width=340 ] in our fig . [ fig:2 ] . the solid curve gives @xmath90 vs @xmath91 for a case where the einstein frequency defining @xmath25 is taken to be very large , @xmath124ev , compared with the energies of interest in this figure . the curve is almost a perfect straight line ( there is no structure ) with slope giving @xmath116 with @xmath125 by choice . this is to be compared with the dotted line which gives the bare dispersion and has slope 1 . applying the construction that the `` arpes bare dispersion '' is a straight line going through the renormalized energy at @xmath128mev would give the same straight line as the solid curve and , therefore , we would conclude that @xmath129 , i.e. : there is no renormalization . as a second example we use for the spectral density a form appropriate to a spin fluctuation model . the form is obtained directly from consideration of experimental data on the infrared conductivity in the normal state of the cuprates . it is given by the experimental form @xmath85 of eq . which can be adequately represented by a lorentzian @xmath130 ( we will refer to this form in the following as mmp - form . ) in eq . ( [ eq:10 ] ) @xmath131 is a characteristic spin fluctuation frequency taken for illustrative purposes @xmath132mev and @xmath133 with @xmath134mev . with this spectral density we get the dashed line in fig . [ fig:2 ] which is to be compared with both , the dotted line ( real bare dispersion ) and the dash - dotted line ( arpes constructed ` bare ' dispersion ) . it is clear that the arpes construction again drastically underestimates the spin fluctuation mass renormalization giving a value of @xmath135 rather than 1.1 . also , the curve does not show any identifiable sharp structure , instead it changes only rather gradually . in order to get the correct value of @xmath94 one would need data up to energies higher than the maximum energy involved in the fluctuation spectrum that is responsible for the renormalization . for spin fluctuation theories this energy scale is set by the magnitude of @xmath136 of the @xmath15 model and is high of order several hundred mev.@xcite the @xmath15 model is widely believed to provide the appropriate hamiltonian needed to describe the strongly correlated charge carriers in the cuo@xmath1-planes in the cuprates . in this instance arpes experiments at much higher energies than presently sampled would be required to get the full value of @xmath94 involved . in the interpretation of their arpes data some experimentalists have considered the possibility that the renormalizations are due to phonons . we consider here only the specific case of bi2212 for which there also exists a recent set of high accuracy optical conductivity measurements.@xcite the phonon frequency distribution in this material has been measured by incoherent inelastic neutron scattering by renker _ et al._@xcite such experiments give directly @xmath67 and a first attempt at a model for an electron - phonon interaction spectral density @xmath25 can be constructed by multiplying the experimental @xmath67 by a constant adjusted to best fit the arpes data on the renormalized quasiparticle energy @xmath90 as shown in as a function of bare band energy @xmath91 using an electron - phonon spectral density @xmath25 to describe the coupling . the solid squares are the data of johnson _ et . al . _ and the solid curve is our fit . the full phonon spectral density revealed by inelastic neutron scattering was used with the area under @xmath25 adjusted to get a mass renormalization parameter @xmath137 . in the theory at @xmath138 @xmath139 @xmath140mev while the value obtained from the arpes construction is nearly the same , @xmath141mev . the curve is for @xmath142k . the dash - dotted curve is for a @xmath36-function model for @xmath25 at @xmath143mev and @xmath144 . the bottom frame applies to the lower temperature @xmath145k and is in the superconducting state . , width=340 ] fig . [ fig:3 ] . the data@xcite at @xmath142k are shown in the top frame and at @xmath145k in the bottom frame as solid squares . on the top horizontal scale we show momentum differences @xmath146 in @xmath139 while on the bottom horizontal scale we have the value corresponding to our bare band energy @xmath91 which implies a theoretical bare energy of @xmath147mev at @xmath138@xmath139 . the fit in both cases was obtained with a value of @xmath137 . we see that the resulting theoretical results ( solid curves ) fit reasonably well the experimental data . here , no attempt has been made to get a best fit by varying the shape of the underlying spectral density @xmath25 . the reader is referred to the work by verga _ et al._@xcite for an alternative approach to the analysis of arpes data for the material la@xmath148sr@xmath149cuo@xmath150 ( lsco ) . these authors do attempt to get the shape of the spectrum from the data itself and we will return to this issue later . two final comments on fig . [ fig:3 ] . the dotted curve which is the arpes construction for the bare dispersion gives at @xmath138@xmath139 an @xmath151mev which is nearly the same as the theoretical value of @xmath147mev . this is to be expected since phonon renormalization effects are small at @xmath132mev and beyond . secondly , we wish to point out that the data at @xmath145k is in the superconducting state . the formalism needed to treat superconductivity is more complicated than that for the normal state which we have sketched , but the ideas and concepts are the same and no details are included here . the reader is referred to an extensive existing literature.@xcite here it will be sufficient to point out that the value of the critical temperature for superconductivity in a @xmath3-wave superconductor is determined most directly by a different function that which comes into the mass renormalization channel . for an @xmath4-wave isotropic superconductor this issue does not arise . the quasiparticle @xmath25 would also determine superconductivity and one could ask if arpes data on @xmath94 are indeed consistent with the observed superconductivity . but in the cuprates the superconducting gap has @xmath3-wave symmetry . in this case it is the projection of the quantity @xmath152 onto the @xmath3-wave channel which determines the spectral function which is most directly involved in superconductivity ( rather than its projection on the @xmath4-channel ) . arpes does not measure this projection separately and so this technique , stricktly speaking , remains silent on the issue of mechanism for superconductivity even if phonons should be the main cause of renormalization of the electron dispersion curves . returning to the top frame of fig . [ fig:3 ] the dash - dotted curve which does not agree well with the data is shown for comparison . it is based on a model electron - phonon spectral density @xmath25 in which the coupling is entirely to a single frequency fixed at @xmath143mev@xcite with a mass enhancement factor @xmath144 chosen to get a critical temperature of @xmath153k when @xmath4- and @xmath3-channel spectral densities are taken to be the same . such a model does not agree well with experiment and when it is used , in the superconducting state at @xmath145k , the disagreement is even worse . the @xmath36-function model corresponds to an extreme in which it is assumed that the electron - phonon coupling is to a single mode . this is unlikely to be the case . a model based on the entire phonon spectrum is more realistic although , as we said before , @xmath25 does not need to have the same shape as @xmath67 . special modes could be weighted more heavily in @xmath25 as compared to @xmath67 . we can impose a second constraint on renormalization effects by considering the infrared conductivity . there exists much data on optical conductivity in the cuprates and these have given us considerable insight into the inelastic scattering involved . the conductivity @xmath80 at @xmath92 ( to remain simple ) in the normal state also follows in a first approximation from the knowledge of the electron self energy:@xcite @xmath154 more complicated formulas exist at any finite temperature @xmath20 and/or for the superconducting case and including vertex corrections which we will not repeat here for the sake of brevity . they can be found in many places including our own work . since here we will fit the electron - boson exchange density to optical data directly , the resulting form can be thought of as including higher order corrections . in eq . ( [ eq:11 ] ) @xmath155 is the elastic impurity scattering rate while the imaginary part of @xmath156 deals with the corresponding inelastic scattering . what is most often measured in optical experiments , is the reflectance @xmath157 as a function of @xmath19 . this is mev . the solid curve presents the data of tu _ et al . _ the dashed curve is for the phonon spectrum and a @xmath137 . the temperature is @xmath158k . the bottom frame shows the derived optical scattering rate @xmath87 as a function of @xmath19 . the solid curve presents the data of tu _ et al . _ the dashed curve is a phonon based result which does not give enough variation with increasing @xmath19 . by contrast , the dash - dotted curve fits well and is based on a simple spin fluctuation spectrum of mmp - form with @xmath159mev and @xmath160 . , width=340 ] shown in the top frame of fig . [ fig:4 ] for bi2212 . the solid curve is the data of tu _ et al._@xcite for room temperature @xmath161k . the dashed curve was obtained with the same neutron based electron - phonon interaction spectrum having a @xmath137 as we found from our arpes analysis . it provides a poor fit as is illustrated , even more sharply , in the bottom frame of fig . [ fig:4 ] which deals with optical scattering rates . it has now become standard procedure for experimentalists to extract from their reflectance data the real and imaginary part of the optical conductivity and to construct from this information the optical scattering rate @xmath87 . the solid line in the bottom frame is the data . the two other curves are theoretical and the dashed is based on the neutron @xmath25 with @xmath137 . it is clear that it does not agree with the data , particularly as @xmath19 increases . this defect remains whatever value of @xmath94 is used . the fundamental problem is that the phonon spectrum cuts off too soon as a function of frequency to fit the optics and , consequently , the calculated @xmath19-dependence of @xmath87 is much too flat . any boson mechanism that fits the data will need to have an energy scale that extends to order of a few @xmath162mev or so . this is consistent with a spin naffl@xcite or a mfl model@xcite but is not compatible with phonons . it is important to note that this limitation of a phonon mechanism is independent of any difference there might be in the size of @xmath26 as compared with @xmath25 . both functions cut off at the maximum phonon energy ( @xmath163mev in case of bi2212 ) and this can not give agreement with the optics . on the other hand , a spin fluctuation model provides a natural explanation because it involves a higher boson energy scale of the order @xmath136 of the @xmath15-model as previously noted.@xcite this is demonstrated in the lower frame of fig . [ fig:4 ] by the dash - dotted curve which is for an mmp - form with @xmath159mev and @xmath160 . it fits the experiment almost perfectly with no need for any adjustment of any kind . ( deviations from @xmath159mev within @xmath164mev will change the quality of this fit only marginally . ) it is interesting to note in closing this discussion that in the case of electron - phonon interaction the same value for @xmath94 is required to get a reasonable fit of the arpes data and to get agreement at least in the low energy regime with the optical data . this is in conflict with the findings by kuli@xcite that in the phonon assisted case @xmath26 should be significantly smaller than @xmath25 . in the top frame of fig . [ fig:5 ] we compare theory with experiment at three k , @xmath165k , and @xmath108k . all are in the normal state . the gray solid lines are experiment of tu _ _ , dotted are theory based on an mmp - form for the charge carrier - exchange boson spectral density , and the dashed have in addition coupling to an optical resonance at @xmath166mev . the bottom frame shows the derived spectral density based on our well defined procedure to invert optical data . the @xmath158k spectrum has no resonance and is a pure mmp - form with @xmath159mev ( solid line ) . the other two spectra are modified through the addition of coupling to a resonance mode at @xmath166mev . also shown is the phonon spectrum ( gray solid curve ) . , width=340 ] temperatures , namely @xmath108k , @xmath165k , and @xmath158k . the data on the optical scattering rate was obtained by tu _ et al._@xcite from their infrared measurements in bi2212 and is denoted by gray solid lines ( top to bottom is decreasing temperature ) . the black dotted lines are theory obtained from an mmp - form given by eq . ( [ eq:10 ] ) with @xmath159mev . the @xmath158k curve repeats the fit shown in the bottom frame of fig . [ fig:4 ] . the same spectral density , however , does not fit well the @xmath165k and @xmath108k data ( dotted lines ) . to get agreement with these data sets it is necessary to augment the mmp - form by adding coupling to an additional optical resonance at @xmath166mev ( this gives the dashed curves ) . this is the frequency at which inelastic neutron scattering experiments have revealed a strong peak in the magnetic susceptibility at momentum @xmath33 in the two dimensional brillouin zone of the cuo@xmath1 plane,@xcite in the superconducting state . the first report of such a resonance in neutron scattering was by j. rossat - mignot _ et al . _ for yba@xmath1cu@xmath167o@xmath168.@xcite we have already described our method for obtaining @xmath86 , thus it is not necessary to give details here except to mention that tu _ et al._@xcite also get this peak using a slightly different method . we present final results for our model @xmath169 in the bottom frame of fig . [ fig:5 ] . while no peak at @xmath166mev is observed at @xmath158k ( solid line ) one shows up at lower temperatures , dotted line for @xmath165k and a somewhat bigger peak at @xmath108k ( dashed curve ) . at present , neutron scattering does not reveal a spin resonance at these temperatures . returning to the top frame of fig . [ fig:5 ] we stress that the introduction of a resonance peak in addition to the mmp - form is responsible for the observed rapid rise around @xmath170mev in the optical scattering rate at @xmath108k and also , although less obvious , at @xmath165k . the model spectral density fits the data remarkably well . such optical resonances have been seen before in yba@xmath1cu@xmath167o@xmath171@xcite as well as in tl compounds@xcite in the superconducting state of optimally doped samples , and have their equivalent in spin resonances observed by inelastic neutron scattering.@xcite here the high temperature optical resonances in bi2212 are seen for the first time in the normal state above @xmath5 , as reported by tu _ et al._@xcite their exact microscopic origin , however , can not be deduced from optical data alone . it could be that the @xmath166mev neutron spin resonance forms above @xmath5 in this optimally doped material . alternatively one might argue that this may be a phonon contribution ( see gray solid curve in the bottom frame of fig . [ fig:5 ] ) to the total electron - boson transport spectral density . if this were the case , however , we would not expect the peak of fig . [ fig:5 ] ( bottom frame ) to show significant temperature dependence in this temperature regime . also , the phonon peak in the bottom frame of fig . [ fig:5 ] has more weight at low @xmath19 than is indicated in optics . in terms of the area under the spectral density @xmath86 we note that the peak itself contributes less than about 15% of the total weight ( at @xmath108k ) , the rest comes from the mmp - form . while the issue of the origin of the peak is important , here we concentrate instead on the consequences of its existence for the arpes data . the possibility of coupling to a resonance mode has been discussed in the past in connection with such data.@xcite in fig . [ fig:6 ] we compare our theoretical results for the as a function of the bare band energy @xmath91 in mev ( light dash - dotted ) . the solid curve is based on the spin fluctuation spectral density with resonance peak obtained from conductivity data ( fig . ) scaled down to a mass renormalization of @xmath172 , otherwise left unmodified . the heavy dash - dotted curve is for comparison and has no resonance peak . the light dashed straight line is for comparison and is the arpes `` bare dispersion '' obtained by making the curve go through experiment at @xmath141mev . this corresponds to a theoretical @xmath173mev at @xmath174@xmath139 , much greater than the arpes ` bare ' value . this frame applies to @xmath142k and is in the normal state . the bottom frame is for @xmath145k in the superconducting state . , width=340 ] renormalized quasiparticle energies with the arpes experimental data ( solid squares ) . the theory ( solid curve ) is based on the optics derived spectral density ( see fig . [ fig:5 ] , bottom frame ) . the top frame is for @xmath142k in the normal state and the bottom frame is for @xmath145k in the superconducting state . for the top frame we renormalized downward the @xmath86 to account for the expectation that in a spin fluctuation mechanism the quasiparticle electron - boson interaction spectral density should be smaller than its transport counterpart . to get the fit seen in the figure we used @xmath175 . a smaller value of @xmath176 was used in the bottom frame . we see that we can get an equally good fit to arpes with spin fluctuations including the @xmath166mev resonance as we did with phonons . this new fit ( see fig . [ fig:6 ] ) , however , has the advantage that it can equally well explain the optical data while this is impossible with phonons . it is also clear from the calculations that the transport electron - boson interaction spectral density is larger than the quasiparticle spectral density by a factor of about two as is expected in theories of the naffl model . it should be acknowledged , however , that part of the difference in the value of @xmath94 needed to fit arpes data as compared with optics reflects anisotropies in the quasiparticle renormalizations over the brillouin zone . also included in the two frames of fig . [ fig:6 ] are our results for a @xmath90 ( heavy dash - dotted ) when the resonant peak in the spectral density is left out of the calculations . we see that the pronounced structure at @xmath177@xmath139 seen in the superconducting state is not reproduced in this case . if we used instead the mfl model , i.e. : @xmath178 , the deviation from the data would be worse and even clearly seen in the normal state . the resonance peak , absent in the mmp - form and in the mfl model , is what gives the observed structure . also shown in the figure is the the bare dispersion relation ( light dash - dotted curve ) on which our calculations are based . this is quite different from the arpes derived ` bare ' dispersion , the light dashed curve , obtained by making the curve go through the experimental point at @xmath141mev on the assumption that the renormalization by boson - exchange interaction has ended at this energy . reference to the difference between the theoretical bare dispersion ( light dash - dotted ) and the dressed curve ( solid line ) shows clearly that this is not the case in a spin fluctuation model . it is interesting to note the similarities and differences between our work and that of verga _ et al._@xcite an important difference is that these authors invert the arpes data directly from which they obtain a model electron - boson spectral density . instead , we have inverted the optical data which serves as a second constraint on the microscopics not considered by verga _ _ while they invert data in lsco at three different doping levels @xmath179 , they , nevertheless , find a spectrum which has a peak which is , however , not as prominent as either the spin resonance peak or the phonon peak in the bottom frame of our fig . [ fig:5 ] . their peak could simply be a peak of the mmp - form as in the solid black curve in the bottom frame of fig . [ fig:5 ] . nothing definite can be concluded . however , we point out that they do find tails at higher energies ( @xmath180mev ) , as we have , which rather support spin fluctuations and not phonons . in this paper we consider in addition to arpes optical data and from this we conclude more forcefully that high energy tails extending up to several hundred mev exist in the boson assisted spectral functions describing the electron boson interaction in bi2212 . whatever the mechanism by which the quasiparticles are renormalized the @xmath181-component of the @xmath182 nambu green s function in the superconducting state can be expressed in the general form@xcite @xmath183 where the renormalized frequency @xmath184 . the inverse quasiparticle lifetime is defined as twice the imaginary part of the poles of @xmath185 for a given momentum label * k*. for @xmath186 we obtain@xcite @xmath187 where the indices 1 and 2 refer to the real and imaginary part respectively , and the corresponding renormalized energy is a solution of @xmath188 for the normal state @xmath189 these expressions reduce to those previously given in the limit of weak scattering with the modification that @xmath190 is renormalized by a mass enhancement factor @xmath116 . in fig . [ fig:7 ] we emphasize once again the difference obtained obtained from optics and from arpes . the gray solid line is the experimental optical scattering rate at @xmath191k obtained from the work of tu _ et al . _ while the black solid curve is our theoretical result based on the electron - spin fluctuation spectral density @xmath86 shown in the bottom frame of fig . ( but using data for @xmath191k ) . the other two curves are theoretical results for the imaginary part of the quasiparticle self energy . , width=340 ] between quasiparticle scattering rates and its optical counterpart . it applies to @xmath191k for bi2212 in the superconducting state . the gray solid curve represents the data of tu _ et al._@xcite for the optical scattering rate and the solid line our theoretical fit to this data which also determines the transport spectral density @xmath86 . the corresponding value of @xmath192 is 2.18 . consideration of the arpes data on quasiparticle renormalizations shows us that the equilibrium spectral density @xmath77 must be considerably smaller . to calculate the imaginary part of @xmath193 of eq . ( [ eq:13 ] ) we have reduced @xmath86 by a factor of two but without changing its shape . this is done to get reasonable agreement with arpes shown in fig . [ fig:6 ] . the results are shown as dashed and dotted lines for the antinodal and nodal directions respectively . not only is the shape of the resulting scattering rate different from the frequency variation obtained in the optical case but its magnitude is also different reflecting the difference in magnitude of @xmath86 and @xmath77 . we have reconsidered the interpretation of the quasiparticle renormalizations observed in high resolution arpes data for the specific case of bi2212 . we confirm that the observed energy scale is compatible with coupling to phonons . we find , however , that present data and , in particular , the way they are analyzed might well significantly underestimate renormalization due to other boson exchange mechanisms distinct from phonon exchange and for which the energy scale is larger . as an example this would be the case in a spin fluctuation exchange mechanism as is envisaged in the naffl model where the energy scale for the spin fluctuations is set by @xmath136 of the @xmath15 model and is consequently very high . if in addition to arpes , optical data is also considered , it is found that an explanation in terms of phonons becomes less tenable . of course , there are many complications and these make a definitive interpretation impossible . what is clear , however , is that transport properties and in particular the observed infrared conductivity can not be understood as due to the electron - phonon interaction . this holds even if modifications are introduced to account for the reduced screening present in the cuprates because of the low electron density and also the increased effects of correlations . the main point is that the underlying charge carrier - phonon spectral density is cut off at the maximum phonon frequency and this feature is in conflict with optical data . by contrast , spin fluctuations give a natural explanation of the data . our analysis of the data of tu _ et al._@xcite also provides a natural explanation of the arpes data . what we find is that an optical resonance is present in the infrared data even in the normal state above @xmath5 and this resonance can explain the energy scale seen in arpes . ( below @xmath5 this optical resonance has a spin resonance equivalent observed by fong _ et al._@xcite using inelastic neutron scattering . ) the microscopic nature of the optical resonance is not entirely known . it could have its origin in spin fluctuations theories and be the spin one resonance observed in inelastic polarized neutron scattering experiments , or could even contain a phonon contribution over and beyond the spin fluctuation contribution . our analysis of both arpes and optical data together has served to emphasize once more the differences between the equilibrium electron - boson exchange spectral density @xmath25 and its transport counterpart @xmath26 be it for phonon exchange or spin fluctuations . the data indicates that these two quantities differ in magnitude by roughly a factor of two or so . this is understood to arise from the fact that transport scattering rates emphasize more strongly backward collisions than do equilibrium rates . while detailed and elaborate calculations beyond the scope of this paper would be required to pin down the expected differences quantitatively , present ideas about reduced screening in the cuprates lead directly to the expectation that phonons will be much less important in transport than for equilibrium properties while ideas about the naffl model lead to opposite expectations . spin fluctuations dominate the transport while at the same time could be less important in determining quasiparticle properties . this agrees well with the available combined set of optical and arpes data in bi2212 . research supported by the natural sciences and engineering research council of canada ( nserc ) and by the canadian institute for advanced research ( ciar ) . we thank dr . t. valla for discussions and for making his data available to us . 99 a. kaminski , j. mesot , h. fretwell , j.c . campuzano , m.r . norman , m. randeria , h. ding , t. sato , t. takahashi , t. mochiku , k. kadowaki , and h. hoechst , , 1788 ( 2000 ) . t. valla , a.v . fedorov , p.d . johnson , j. xue , k.e . smith , and f.j . disalvo , , 4759 ( 2000 ) . p.d . johnson , t. valla , a.v . fedorov , z. yusof , b.o . wells , q. li , a.r . moodenbaugh , g.d . gu , n. koshizuka , c. kendziora , sha jian , and d.g . hinks , , 177007 ( 2001 ) . a. kaminski , m. randeria , j.c . campuzano , m.r . norman , h. fretwell , j. mesot , t. sato , t. takahashi , and k. kadowaki , , 1070 ( 2001 ) . a. lanzara , p.v . bogdanov , x.j . zhou , s.a . kellar , d.l . feng , e.d . lu , t. yoshida , h. eisaki , a. fujimori , k. kishio , j .- shimoyama , t. noda , s. uchida , z. hussain , and z .- x . shen , nature ( london ) * 412 * , 510 ( 2001 ) . bogdanov , a. lanzara , s.a . kellar , x.j . zhou , e.d . lu , w.j . zheng , g. gu , j .- shimoyama , k. kishio , h. ikeda , r. yoshizaki , z. hussain , and z .- x . shen , , 2581 ( 2000 ) . t. valla , a.v . fedorov , p.d . johnson , b.o . wells , s.l . hulbert , q. li , g.d . gu , n. koshizuka , science * 285 * , 2110 ( 1999 ) . t. valla , a.v . fedorov , p.d . johnson , q. li , g.d . gu , and n. koshizuka , , 828 ( 2000 ) . f. marsiglio and j.p . carbotte , in handbook on superconductivity : conventional and unconventional , eds . bennemann and j.b . ketterson ( springer , berlin , in print ) . hardy , d.a . bonn , d.c . morgan , r. liang , and k. zhang , , 3999 ( 1993 ) . z .- x . shen , d.s . dessau , b.o . wells , d.m . king , w.e . spicer , a.j . arko , d. marshall , l.w . lombardo , a. kapitulnik , p. dickinson , s. doniach , j. dicarlo , a.g . loeser , and c.h . park , , 1553 ( 1993 ) . wollman , d.a . van harlingen , w.c . lee , d.m . ginsberg , and a.j . leggett , , 2134 ( 1993 ) . tsuei , j.r . kirtley , c.c . chi , lockseeyu - jahnes , a. gubta , t. shaw , j.z . sun , and m.b . ketchen , , 593 ( 1994 ) . p. tomlinson and j.p . carbotte , , 4738 ( 1976 ) . carbotte , in _ anisotropy effects in superconductors _ weber , plenum ( new york , 1977 ) , p. 183 . kuli , phys . rep . * 338 * , 1 ( 2000 ) . r. zeyher and m.l . kuli , , 2850 ( 1996 ) ; * 54 * , 8985 ( 1996 ) . kuli and r. zeyher , , 4395 ( 1994 ) ; physica c * 199 - 200 * , 358 ( 1994 ) ; * 235 - 240 * , 358 ( 1994 ) . m. weger , b. barbelini , and m. peter , z. phys . b * 94 * , 387 ( 1994 ) . m. weger , m. peter , and l.p . pitaevskii , z. phys . b * 101 * , 573 ( 1996 ) . danylenko , o.v . dolgov , m.l . kuli , and v. oudovenko , euro . b - cond . matter * 9 * , 201 ( 1999 ) . kuli and o.v . dolgov , in _ high temperature superconductivity _ , ed . : s. barnes , j. ashkenazi , j. cohn , and f. zuo , aip conference proceedings * 483 * , 63 ( 1999 ) . s. sorella , g.b . martins , f. becca , c. gazza , l. capriotti , a. parola , and e. dagotto , , 117002 ( 2002 ) . millis , h. monien , and d. pines , , 167 ( 1990 ) . p. monthoux and d. pines , , 6069 ( 1993 ) ; , 4261 ( 1994 ) ; , 16015 ( 1994 ) . varma , int . j. mod . phys . * 3 * , 2083 ( 1989 ) . littlewood , c.m . varma , s. schmitt - rink , and e. abrahams , , 12371 ( 1989 ) . varma , p.b . littlewood , s. schmitt - rink , e. abrahams , and a.e . ruckenstein , , 1996 ( 1989 ) ; _ ibid . _ * 64 * , 497 ( 1990 ) . puchkov , d.n . basov , and t. timusk , j. phys . : condens . matter * 8 * , 10049 ( 1996 ) . t. timusk and b. statt , rep . . phys . * 62 * , 61 ( 1999 ) . tu , c.c . homes , g.d . gu , d.n . basov , and m. strongin , , 144514 ( 2002 ) . p. tomlinson and j.p . carbotte , can . j. phys . * 55 * , 751 ( 1977 ) leung , f.w . kus , n. mckay , and j.p . carbotte , , 4358 ( 1977 ) . allen , 305 ( 1971 ) . g. grimvall , the electron - phonon interaction in metals , ( north - holland , new york , 1981 ) . carbotte , rev . phys . * 62 * , 1027 ( 1990 ) . daams and j.p . carbotte , j. low temp . phys . * 43 * , 263 ( 1981 ) . w.l . mcmillan and j.m . rowell , , 108 ( 1965 ) . mcmillan and j.m . rowell , in superconductivity , ed . parks ( marcel dekker inc . , new york , 1969 ) , p. 561 . a. virosztek and j. ruvalds , , 4064 ( 1990 ) . j. ruvalds and a. virosztek , , 5498 ( 1991 ) . savrasov and o.k . andersen , , 4430 ( 1996 ) . d. branch and j.p . carbotte , , 603 ( 1995 ) ; j. supercond . * 12 * , 667 ( 1999 ) ; can . j. phys . * 77 * , 531 ( 1999 ) ; j. supercond . * 13 * , 535 ( 2000 ) . carbotte , e. schachinger , and d.n . basov , nature ( london ) * 401 * , 354 ( 1999 ) . f. marsiglio , t. startseva , and j.p . carbotte , phys . a * 245 * , 172 ( 1998 ) . e. schachinger , j.p . carbotte , and d.n . basov , europhys . lett . * 54 * , 380 ( 2001 ) . e. schachinger and j.p . carbotte , , 9054 ( 2000 ) . e. schachinger and j.p . carbotte , , 094501 ( 2001 ) . s. verga , a. knigavko , and f. marsiglio , cond - mat/0207145 ( unpublished ) . b. renker , f. gompf , d. ewert , p. adelmann , h. schmidt , e. gering , and h. mutka , z. phys . b * 77 * , 65 ( 1989 ) . f. marsiglio and j.p . carbotte , aust . j. phys . * 50 * , 975 ( 1997 ) ; aust . j. phys . * 50 * , 1011 ( 1997 ) . fong , p. bourges , y. sidis , l.p . regnault , a. ivanov , g.d . gull , n. koshizuka , and b. keimer , nature ( london ) * 389 * , 588 ( 1999 ) . j. rossat - mignot , l.p . regnault , c. vettier , p. bourges , p. burlet , j. bossy , j.y . henry , and g. lapertot , physica c * 185 - 189 * , 86 ( 1991 ) . f. marsiglio , molecular physics reports * 24 * , 73 ( 1999 ) . e. schachinger and j.p . carbotte , physica c * 364 * , 13 ( 2001 ) . dai , h.a . mook , s.m . hayden , g. aeppli , t.g . perring , r.d . hunt , and f. doan , science * 284 * , 1344 ( 1999 ) . h. he , p. bourges , y. sidis , c. ulrich , l.p . regnault , s. pailhs , n.s . berzigiarova , n.n . kolesnikov , and b. keimer , science * 295 * , 1045 ( 2002 ) . norman and h. ding , , r11 089 ( 1998 ) . norman , m. eschrig , a. kaminski , and j.c . campuzano , , 184508 ( 2001 ) . m. eschrig and m.r . norman , , 144503 ( 2003 ) .

wightman function and vacuum expectation value of the field square are evaluated for a massive scalar field with general curvature coupling parameter subject to robin boundary conditions on two codimension one parallel branes located on-dimensional background spacetime with a warped internal space . the general case of different robin coefficients on separate branes is considered . the application of the generalized abel - plana formula for the series over zeros of combinations of cylinder functions allows us to extract manifestly the part due to the bulk without boundaries . unlike to the purely ads bulk , the vacuum expectation value of the field square induced by a single brane , in addition to the distance from the brane , depends also on the position of the brane in the bulk . the brane induced part in this expectation value vanishes when the brane position tends to the ads horizon or ads boundary . the asymptotic behavior of the vacuum densities near the branes and at large distances is investigated . the contribution of kaluza - klein modes along is discussed in various limiting cases . in the limit when the curvature radius for the ads spacetime tends to infinity , we derive the results for two parallel robin plates on background spacetime . for strong gravitational fields corresponding to large values of the ads energy scale , the both single brane and interference parts of the expectation values integrated over the internal space are exponentially suppressed . as an example the case is considered , corresponding to the bulk with one compactified dimension . an application to the higher dimensional generalization of the randall - sundrum brane model with arbitrary mass terms on the branes is discussed . pacs numbers : 04.62.+v , 11.10.kk , 04.50.+h

hep-th0508038

anti - de sitter ( ads ) spacetime is remarkable from different points of view . the early interest to this spacetime was motivated by the questions of principal nature related to the quantization of fields propagating on curved backgrounds . the investigation of the dynamics of fields on ads is of interest not only because ads space is a space of high symmetry , and hence exact solutions for the free field theory can be written down , but also because it is a space of constant nonzero curvature , and thus field theory in ads background is not a trivial rewriting of minkowski spacetime field theory . the presence of the both regular and irregular modes and the possibility of interesting causal structure lead to many new phenomena . the importance of this theoretical work increased when it has been discovered that ads spacetime generically arises as ground state in extended supergravity and in string theories . further interest in this subject was generated by the appearance of two models where ads geometry plays a special role . the first model , the so called ads / cft correspondence ( for a review see @xcite ) , represents a realization of the holographic principle and relates string theories or supergravity in the ads bulk with a conformal field theory living on its boundary . it has many interesting consequences and provides a powerful tool to investigate gauge field theories . the second model is a realization of a braneworld scenario with large extra dimensions and provides a solution to the hierarchy problem between the gravitational and electroweak mass scales ( for reviews in braneworld gravity and cosmology see ref . @xcite ) . in this model the main idea to resolve the large hierarchy is that the small coupling of four dimensional gravity is generated by the large physical volume of extra dimensions . braneworlds naturally appear in string / m - theory context and provide a novel setting for discussing phenomenological and cosmological issues related to extra dimensions . the model introduced by randall and sundrum @xcite is particularly attractive . the corresponding background solution consists of two parallel flat branes , one with positive tension and another with negative tension embedded in a five dimensional ads bulk . the fifth coordinate is compactified on @xmath6 , and the branes are on the two fixed points . in the original version of the model it is assumed that all matter fields are confined on the branes and only the gravity propagates freely in the five dimensional bulk . more recently , alternatives to confining particles on the brane have been investigated and scenarios with additional bulk fields have been considered . apart from the hierarchy problem it has been also tried to solve the cosmological constant problem within the braneworld scenario . the braneworld theories may give some alternative discussion of the cosmological constant . the basic new ingredient is that the vacuum energy generated by quantum fluctuations of fields living on the brane may not curve the brane itself but instead the space transverse to it . randall - sundrum scenario is just the simplest possibility within a more general class of higher dimensional warped geometries . such a generalization is of importance from the viewpoint of a underlying fundamental theory in higher dimensions such as ten dimensional superstring theory . from a phenomenological point of view , higher dimensional theories with curved internal manifolds offer a richer geometrical and topological structure . the consideration of more general spacetimes may provide interesting extensions of the randall - sundrum mechanism for the geometric origin of the hierarchy . spacetimes with more than one extra dimension can allow for solutions with more appealing features , particularly in spacetimes where the curvature of the internal space is non - zero . more extra dimensions also relax the fine - tunings of the fundamental parameters . these models can provide a framework in the context of which the stabilization of the radion field naturally takes place . several variants of the randall sundrum scenario involving cosmic strings and other global defects of various codimensions has been investigated in higher dimensions ( see , for instance , @xcite-@xcite and references therein ) . in particular , much work has been devoted to warped geometries in six dimensions ( see references in kofi05 ) . in braneworld models the investigation of quantum effects is of considerable phenomenological interest , both in particle physics and in cosmology . the braneworld corresponds to a manifold with boundaries and all fields which propagate in the bulk will give casimir - type contributions to the vacuum energy ( for reviews of the casimir effect see ref . @xcite ) , and as a result to the vacuum forces acting on the branes . casimir forces provide a natural alternative to the goldberger - wise mechanism for stabilizing the radion field in the randall - sundrum model , as required for a complete solution of the hierarchy problem . in addition , the casimir energy gives a contribution to both the brane and bulk cosmological constants and , hence , has to be taken into account in the self - consistent formulation of the braneworld dynamics . motivated by these , the role of quantum effects in braneworld scenarios has received a great deal of attention @xcite-@xcite . however , in the most part of these papers the authors consider the global quantities such as the total casimir energy , effective action or conformally invariant fields . the investigation of local physical characteristics in the casimir effect , such as expectation value of the energy - momentum tensor and the field square , is of considerable interest . local quantities contain more information on the vacuum fluctuations than the global ones . in addition to describing the physical structure of the quantum field at a given point , the energy - momentum tensor acts as the source in the einstein equations and therefore plays an important role in modelling a self - consistent dynamics involving the gravitational field . quantum fluctuations play also an important role in inflationary cosmology as they are related to the power spectrum . as it has been shown in ref . pujo05 , the quantum fluctuations of a bulk scalar field coupled to a brane located scalar field with a bi - quadratic interaction generate an effective potential for the field on the brane with a true vacuum at the nonzero values of the field . in particular , these calculations are relevant to the bulk inflaton model @xcite , where the inflation on the brane is driven by the bulk scalar field . in the case of two parallel branes on ads background , the vacuum expectation value of the bulk energy - momentum tensor for a scalar field with an arbitrary curvature coupling is investigated in refs . @xcite . in particular , in ref . @xcite the application of the generalized abel - plana formula @xcite to the corresponding mode sums allowed us to extract manifestly the parts due to the ads spacetime without boundaries and to present the boundary induced parts in terms of exponentially convergent integrals for the points away the boundaries . the interaction forces between the branes are investigated as well . depending on the coefficients in the boundary conditions , these forces can be either attractive or repulsive . the local vacuum effects for a bulk scalar field in brane models with ds branes are discussed in refs . pujo04,nayl05 . on background of manifolds with boundaries , the physical quantities , in general , will receive both volume and surface contributions . for scalar fields with general curvature coupling , in ref . @xcite it has been shown that in the discussion of the relation between the mode sum energy , evaluated as the sum of the zero - point energies for each normal mode of frequency , and the volume integral of the renormalized energy density for the robin parallel plates geometry it is necessary to include in the energy a surface term located on the boundary . an expression for the surface energy - momentum tensor for a scalar field with general curvature coupling parameter in the general case of bulk and boundary geometries is derived in ref . the corresponding vacuum expectation values are investigated in ref . @xcite for the model with two flat parallel branes on ads bulk and in @xcite for a ds brane in the flat bulk . in particular , for the first case it has been shown that in the randall - sundrum model the cosmological constant induced on the visible brane by the presence of the hidden brane is of the right order of magnitude with the value suggested by the cosmological observations without an additional fine tuning of the parameters . the purpose of the present paper is to study the wightman function and the vacuum expectation value of the field square for a scalar field with an arbitrary curvature coupling parameter obeying robin boundary conditions on two codimension one parallel branes embedded in the background spacetime @xmath1 with a warped internal space @xmath2 . the quantum effective potential and the problem of moduli stabilization in the orbifolded version of this model with zero mass parameters on the branes are discussed recently in ref . in particular , it has been shown that one loop - effects induced by bulk scalar fields generate a suitable effective potential which can stabilize the hierarchy without fine tuning . the corresponding results are extended to the type of models with unwarped internal space @xmath2 in ref . we organize the present paper as follows . in the next section we evaluate the wightman function in the region between the branes . by using the generalized abel - plana formula , we present this function in the form of a sum of the wightman function for the bulk without boundaries and boundary induced parts . the vacuum expectation value of the field square for a general case of the internal space @xmath2 is discussed in section [ sec : vevphi2 ] for the case of a single brane geometry and in section [ sec : phi2twopl ] in the region between two branes . various limiting cases are discussed when the general formulae are simplified . a simple example with the internal space @xmath7 is then considered in section [ sec : example ] . the last section contains a summary of the work . the vacuum expectation values of the energy - momentum tensor and the interaction forces between the branes will be discussed in the forthcoming paper . consider a scalar field @xmath8 with curvature coupling parameter @xmath9 satisfying the equation of motion @xmath10@xmath11 , with @xmath12 being the scalar curvature for a @xmath0-dimensional background spacetime , @xmath13 is the covariant derivative operator associated with the metric tensor @xmath14 ( we adopt the conventions of ref . @xcite for the metric signature and the curvature tensor ) . for minimally and conformally coupled scalars one has @xmath15 and @xmath16 correspondingly . we will assume that the background spacetime has a topology @xmath17 , where @xmath2 is a @xmath18-dimensional manifold . first let us consider a general class of spacetimes described by the line element ( see also the discussion in ref . @xcite ) @xmath19with @xmath20 being the metric for the @xmath21-dimensional minkowski spacetime @xmath22 and the coordinates @xmath23 cover the manifold @xmath2 , @xmath24 . here and below @xmath25 and @xmath26 . the scalar curvature for the metric tensor from ( [ metric ] ) is given by the expression @xmath27 where we have introduced the notation @xmath28 with @xmath29 and @xmath30 is the scalar curvature for the metric tensor @xmath31 . the background metric for many braneworld models falls into the general class corresponding to eq . ( [ metric ] ) . in particular , the metric with @xmath32 and @xmath33 is the randall - sundrum case . the metric solutions with @xmath34 , @xmath35 , @xmath36 and @xmath34 , @xmath37 correspond to global @xcite and local @xcite strings respectively . in the case of more extra dimensions generalizations of these models with @xmath38 were considered in olas00,gher00b , oda01,roes02 . the general case of a symmetric homogeneous space @xmath2 with topologically non - trivial yang - mills fields is investigated in @xcite . in particular , when the bulk cosmological constant is negative , a simple solution exists with @xmath39 . for the case of @xmath2 with nonzero ricci scalar one can have a regular solution in the bulk with @xmath36 and randall - sundrum type warp in the minkowski direction . the phenomenology of this type of scenario has been considered in @xcite . our main interest in this paper will be the wightman function and the vacuum expectation value ( vev ) of the field square brought about by the presence of two parallel infinite branes of codimension one , located at @xmath40 and @xmath41 , @xmath42 . we will assume that on this boundaries the scalar field obeys the mixed boundary conditions @xmath43 with constant coefficients @xmath44 , @xmath45 . in the orbifolded version of the model which corresponds to a higher dimensional randall - sundrum braneworld these coefficients are expressed in terms of the surface mass parameters and the curvature coupling of the scalar field ( see below ) . the imposition of boundary conditions on the quantum field modifies the spectrum for the zero point fluctuations and leads to the modification of the vevs for physical quantities compared with the case without boundaries . the resulting quantum effects can either stabilize or destabilize the branewolds and have to be taken into account in the self - consistent formulation of the braneworld dynamics . as a first stage in the investigations of local quantum effects , in this section we will consider the positive frequency wightman function defined as the expectation value @xmath46 where @xmath47 is the amplitude for the vacuum state . the vevs of the field square and the energy - momentum tensor can be obtained from the wightman function in the coincidence limit with an additional renormalization procedure . instead of the wightman function we could take any other two - point function . the reason for our choice of the wightman function is related to the fact that this function also determines the response of particle detectors in a given state of motion . let @xmath48 be a complete set of positive frequency solutions to the field equation ( [ fieldeq ] ) satisfying boundary conditions ( boundcond ) and @xmath49 denotes a set of quantum numbers specifying the solution . expanding the field operator over this set of eigenfunctions and using the commutation relations , the wightman function is presented as the mode sum : @xmath50 the symmetry of the bulk and boundary geometries under consideration allows to present the corresponding eigenfunctions in the decomposed form @xmath51with the standard minkowskian modes in @xmath22 : @xmath52the separation constants @xmath53 are determined by the geometry of the internal space @xmath2 and by the boundary conditions imposed on the branes and will be given below . the modes @xmath54 are eigenfunctions for the operator @xmath55 : @xmath56 \psi _ { \beta } ( x)=-\lambda _ { \beta } ^{2}\psi _ { \beta } ( x ) , \label{eqint1}\]]with eigenvalues @xmath57 and normalization condition @xmath58 @xmath59 is the laplace - beltrami operator for the metric @xmath60 . in the consideration below we will assume that @xmath61 . substituting eigenfunctions ( [ eigfunc1 ] ) into the field equation ( [ fieldeq ] ) and using equation ( [ eqint1 ] ) , for the function @xmath62 one obtains the following equation@xmath63 f_{n}=m_{\beta , n}^{2}e^{2\sigma } f_{n } . \label{eqforfn}\]]equation ( [ eqforfn ] ) is valid for any warp factors @xmath64 and @xmath65 . in this paper we consider the case of two equal warp factors , with @xmath66 in the case when @xmath2 is a torus , this corresponds to a toroidal compactification of @xmath5 spacetime . for warp factors ( [ rosigma ] ) equation ( [ eqforfn ] ) takes the form@xmath67 f_{n}=\left ( m_{\beta , n}^{2}-\lambda _ { \beta } ^{2}\right ) e^{2\sigma } f_{n}. \label{eqforfn1}\]]the operator in the left hand side does not depend on the internal quantum numbers @xmath68 . thus in this case the eigenfunctions @xmath69 and the combination @xmath70 do not depend on @xmath68 . as a result , the dependence on @xmath68 and @xmath71 of the masses is factorized ( see also @xcite):@xmath72for the region between the branes , @xmath73 , the solution to equation ( eqforfn1 ) satisfying the boundary condition on the brane @xmath40 is given by expression @xmath74where @xmath75@xmath76 , @xmath77 are the bessel and neumann functions , @xmath78 and @xmath79 in formula ( [ genu ] ) for a given function @xmath80 we use the notation @xmath81with the coefficients @xmath82 in the discussion below we will assume values of the curvature coupling parameter for which @xmath83 is real . for imaginary @xmath83 the ground state becomes unstable @xcite . for a conformally coupled massless scalar one has @xmath84 and the cylinder functions in eq . ( [ genu ] ) are expressed via the elementary functions . for a minimally coupled massless scalar @xmath85 and the same is the case in odd spatial dimensions . note that in terms of the coordinate @xmath86 introduced by relation ( [ zy ] ) , the metric tensor from ( [ metric ] ) with warp factors ( [ rosigma ] ) is conformally related to the metric of the direct product space @xmath87 by the conformal factor @xmath88 . from the boundary condition on the brane @xmath41 we receive that the eigenvalues @xmath89 have to be solutions to the equation @xmath90this equation determines the tower of radial kaluza - klein ( kk ) masses . we denote by @xmath91 , @xmath92 , the zeros of the function @xmath93 in the right half - plane of the complex variable @xmath86 , arranged in the ascending order , @xmath94 . the eigenvalues for @xmath89 are related to these zeros as @xmath95from the orthonormality condition for the functions @xmath96 for the coefficient @xmath97 in eq . ( [ fny ] ) one finds @xmath98where we have introduced the notation @xmath99 -a_{a}^{2}+b_{a}^{2}(u^{2}-\nu ^{2})\right\ } ^{-1}. \label{tnu}\]]note that , as we consider the quantization in the region between the branes , @xmath100 , the modes defined by ( [ fny ] ) are normalizable for all real values of @xmath83 from eq . ( [ nu ] ) . substituting the eigenfunctions ( [ eigfunc1 ] ) into the mode sum ( wightvev ) , for the expectation value of the field product one finds@xmath101where @xmath102 represents the spatial coordinates in @xmath22 , and@xmath103as the expressions for the zeros @xmath104 are not explicitly known , the form ( [ w11 ] ) of the wightman function is inconvenient . in addition , the terms in the sum over @xmath71 are highly oscillatory functions for large values @xmath71 . it is possible to overcome both these difficulties by applying to the sum over @xmath71 the summation formula derived in refs . sahreview , sahsph by making use of the generalized abel - plana formula . for a function @xmath105 analytic in the right half - plane @xmath106 this formula has the form @xmath107 , \label{cor3form}\end{aligned}\]]where @xmath108 @xmath109 and @xmath110 are the bessel modified functions . formula ( [ cor3form ] ) is valid for functions @xmath105 satisfying the conditions @xmath111and @xmath112 , @xmath113 , where @xmath114 , @xmath115 for @xmath116 . using the asymptotic formulae for the bessel functions for large arguments when @xmath83 is fixed ( see , e.g. , @xcite ) , we can see that for the function @xmath117 from eq . ( [ hab ] ) the condition ( [ cond31 ] ) is satisfied if @xmath118 . in particular , this is the case in the coincidence limit @xmath119 for the region under consideration , @xmath120 . applying to the sum over @xmath121 in eq . ( [ w11 ] ) formula ( [ cor3form ] ) , one obtains @xmath122 % \bigg\ } . \label{w13}\end{aligned}\]]where we have introduced notations @xmath123(the function with @xmath124 will be used below ) . note that we have assumed values of the coefficients @xmath125 and @xmath126 for which all zeros for eq . ( [ cnu ] ) are real and have omitted the residue terms in the original formula in refs . @xcite . in the following we will consider this case only . by the way similar to that used in ref . @xcite , it can be seen that under the condition @xmath127 , the first term in the figure braces in eq . ( [ w13 ] ) is presented in the form @xmath128 . \label{rel1term}\end{aligned}\]]substituting this into formula ( [ w13 ] ) , for the wightman function one finds @xmath129 . \label{w15}\end{aligned}\]]here the term @xmath130does not depend on the boundary conditions and is the wightman function for the @xmath131 spacetime without branes . the second term on the right of eq . ( [ w15 ] ) , @xmath132 , \label{1bounda}\end{aligned}\]]does not depend on the parameters of the brane at @xmath133 and is induced in the region @xmath134 by a single brane at @xmath135 when the boundary @xmath136 is absent . thus the last term on the right of formula ( [ w15 ] ) is the part in the wightman function induced by the presence of the second brane . hence , the application of the summation formula based on the generalized abel - plana formula allowed us ( i ) to escape the necessity to know the explicit expressions for the zeros @xmath104 , ( ii ) to extract from the two - point function the boundary - free and single brane parts , ( iii ) to present the remained part in terms of integrals with the exponential convergence in the coincidence limit . by the same way described above for the wightman function , any other two - point function can be evaluated . note that expression ( [ wads ] ) for the boundary - free wightman function can also be written in the form @xmath137 where @xmath138 is the wightman function for a scalar field with mass @xmath139 in the @xmath140-dimensional minkowski spacetime . by using the identity @xmath141where @xmath142 for the region @xmath143 and @xmath144 for the region @xmath145 ( hence , in ( [ ident11 ] ) one has @xmath146 , @xmath147 as we consider the region @xmath73 ) , and @xmath148the wightman function in the region @xmath100 can also be presented in the form @xmath149 . \label{w17}\end{aligned}\]]in this formula @xmath150 \label{1boundb}\end{aligned}\]]is the boundary part induced in the region @xmath151 by a single brane at @xmath136 when the brane @xmath135 is absent . note that in the formulae given above the integration over angular part can be done by using the formula @xmath152for a given function @xmath153 . combining two forms , formulae ( [ w15 ] ) and ( [ w17 ] ) , we see that the expressions for the wightman function in the region @xmath100 is symmetric under the interchange @xmath154 and @xmath155 . note that the expression for the wightman function is not symmetric with respect to the interchange of the brane indices . the reason for this is that the boundaries have nonzero extrinsic curvature tensors and two sides of the boundaries are not equivalent . in particular , for the geometry of a single brane the vevs are different for the regions on the left and on the right of the brane . for the case @xmath156 one has @xmath157 , where @xmath158 is the @xmath159-dimensional wave vector and replacing in the formulae above @xmath160 we obtain the results for the @xmath5 bulk investigated in ref . examining ( 1boundb ) tells one that the brane induced part in the wightman function vanishes if one or both its arguments lie on the ads boundary corresponding to @xmath161 . note that this does not mean that the presence of the brane in the bulk has no consequences in the conformal field theory on the boundary , translated from the bulk physics by using the ads / cft correspondence . here one should bear in mind that in the ads / cft correspondence the asymptotic behavior of the bulk modes @xmath162 is related to the corresponding boundary fields @xmath163 by the relation @xmath164 . it follows from here that the presence of the brane in the bulk leads to the additional contribution in the boundary two point function which is obtained from ( [ 1boundb ] ) as the limit @xmath165 . this limit is nonzero and gives nontrivial consequences in the boundary conformal field theory . it would be interesting to discuss the holographic interpretation of the brane degrees of freedom in terms of conformal field theory . however , such a discussion is beyond the scope of the present paper . now let us consider the application of the results given above to the higher dimensional generalization of the randall - sundrum braneworld based on the bulk @xmath131 . in the corresponding model @xmath166 coordinate is compactified on an orbifold , @xmath6 of length @xmath167 , with @xmath168 . the orbifold fixed points at @xmath169 and @xmath170 are the locations of two @xmath171-dimensional branes . the corresponding line element has the form ( metric ) with @xmath172 . the absolute value sign here leads to @xmath173-type contributions to the corresponding ricci scalar , located on the branes . as a result @xmath173-terms appear in the equation for the eigenfunction @xmath96 . additional @xmath173-terms come from the surface action of the scalar field with mass parameters @xmath174 and @xmath175 on the branes at @xmath169 and @xmath170 respectively . the @xmath173-terms in the equation for the function @xmath96 in combination with the @xmath176 symmetry lead to the boundary conditions on this function . by the way similar to that for the usual randall - sundrum model ( see , for instance , gher00,flac01b , saha04a ) it can be seen that for untwisted scalar the boundary conditions are robin type with the coefficients @xmath177 for twisted scalar field dirichlet boundary conditions are obtained . note that in the orbifolded version the integration in the normalization integral for the function @xmath96 goes over two copies of the bulk manifold . this leads to the additional coefficient @xmath178 in the expression for the normalization coefficient @xmath179 in ( [ cn ] ) . hence , the wightman function in the higher dimensional randall - sundrum braneworld is given by formula ( [ w15 ] ) or equivalently by ( [ w17 ] ) with an additional factor 1/2 and with robin coefficients given by eq . ( [ atilders ] ) . the one - loop effective potential and the problem of moduli stabilization in this model with zero mass parameters @xmath180 are discussed in ref . in particular , a scenario is proposed where supersymmetry is broken near the fundamental planck scale , and the hierarchy between the electroweak and effective planck scales is generated by a combination of redshift and large volume effects . the masses of the corresponding kk modes along @xmath2 are of the order of tev . the phenomenology of this type braneworld models is discussed in refs . in this section we will consider the vev of the field square induced by a single brane located at @xmath40 . as it has been shown in the previous section the wightman function for this geometry is presented in the form @xmath181where the boundary - free part on the left is defined by formula ( [ wads ] ) . the brane induced part @xmath182 is given by formula ( [ 1bounda ] ) in the region @xmath134 and by formula ( [ 1boundb ] ) with replacement @xmath183 in the region @xmath184 . the vev of the field square is obtained from the wightman function taking the coincidence limit of the arguments . similar to eq . ( wigh1b ) this vev is presented as the sum of boundary - free and boundary induced parts : @xmath185 for the part without boundaries the direct substitution of the coincident arguments in the expression of the wightman function leads to the divergent expression . for the regularization of this divergence the combination of the zeta function method for the sum over @xmath68 and the dimensional regularization method for the integral over @xmath186 can be used . by this way , taking the coincidence limit in the expression ( [ wads ] ) and evaluating the @xmath187-integral , for the vev of the field square in the boundary - free case one finds @xmath188 where we have introduced the local spectral zeta function @xmath189 for the operator @xmath190 . this corresponds to a scalar field with the mass @xmath139 propagating on background of the manifold @xmath2 . for @xmath191 from ( [ phi2bfree ] ) we obtain the standard result for the @xmath5 bulk . in this case the vev of the field square does not depend on the coordinate @xmath86 . for the bulk @xmath192 , even in the case of homogeneous compact subspace @xmath193 , this vev depends on @xmath86 . we explicitly illustrate this below in section [ sec : example ] for the simple case @xmath194 . for points away from the brane the boundary induced part in the wightman function is finite in the coincidence limit and we can directly put @xmath195 . introducing a new integration variable @xmath196 , transforming to the polar coordinates in the plane @xmath197 and integrating over angular part , the following formula can be derived @xmath198by using this formula and eq . ( [ 1bounda ] ) , the boundary induced vev for the field square in the region @xmath134 is presented in the form @xmath199 where the contribution of a given kk mode along @xmath2 is determined by formula @xmath200the corresponding formula in the region @xmath184 is obtained from eq . ( 1boundb ) by a similar way and differs from eq . ( [ phi2spl ] ) by the replacements @xmath155 . using the properties of the bessel modified functions , it can be seen that @xmath201 for dirichlet boundary condition ( @xmath202 ) and @xmath203 for the ratio of the coefficients in the interval @xmath204 in both regions @xmath205 and @xmath206 . in the general case , the vev may change the sign as a function of @xmath86 . in the models with a homogeneous internal space @xmath2 , the sum over @xmath68 in eq . ( [ phi2beta ] ) and , hence , the vev of the field square do not depend on the coordinates @xmath23 in this space . for the comparison with the case of the bulk spacetime @xmath207 when the internal space is absent , it is useful in addition to the vev ( phi2beta ) to consider the vev integrated over the subspace @xmath2 : @xmath208 comparing this integrated vev with the corresponding formula from ref . saha04a , we see that the contribution of the zero kk mode ( @xmath209 ) in eq . ( [ phi2integrated ] ) differs from the vev of the field square in the bulk @xmath207 by the order of the modified bessel functions : for the latter case @xmath210 with @xmath211 defined by eq . ( [ nu ] ) with the replacement @xmath212 . note that for @xmath213 one has @xmath214 . in particular , this is the case for minimally and conformally coupled scalar fields . if the internal space @xmath215 is a one - parameter manifold of size @xmath216 then one has @xmath217 . by taking into account that from the normalization condition it follows that the function @xmath54 contains the factor @xmath218 , we see that @xmath219 is a function on @xmath220 and @xmath221 . the first ratio is related to the proper distance from the brane by the equation @xmath222 hence , for a given size @xmath216 of the internal space , the vev of the field square in addition to the proper distance from the brane , also depends on the absolute position of the brane in the bulk . note that in the case of ads bulk the corresponding quantity depends only on the proper distance from the brane @xcite . to discuss the physics from the point of view of an observer residing on the brane @xmath40 , it is convenient to introduce rescaled coordinates @xmath223 . for this observer the physical size of the subspace @xmath224 is @xmath225 and the corresponding kk masses are rescaled by the warp factor , @xmath226 . now we see that the vev induced by a single brane is a function of the proper distance from the brane and on the ratio @xmath227 of the physical size for the internal space @xmath2 for an observer residing on the brane to the ads curvature radius . as a partial check for derived formulae let us consider the limit @xmath228 . this corresponds to a boundary on the bulk @xmath3 . for @xmath228 , from ( [ nu ] ) we see that the order @xmath83 of the cylindrical functions is large . introducing in ( [ phi2spl ] ) the new integration variable @xmath229 , we can replace the bessel modified functions by their uniform asymptotic expansions for large values of the order . to the leading order this gives : @xmath230 here and below we use the notations @xmath231 and @xmath232 is defined after formula ( [ ident11 ] ) . for the points on the brane the vev given by formula ( [ phi2spl ] ) diverges due to the divergence of the @xmath233-integral in the upper limit . the surface divergences in the renormalized vevs of the local physical observables result from the idealization of the boundaries as perfectly smooth surfaces which are perfect reflectors at all frequencies , and are well known in quantum field theory with boundaries ( see , for instance , deut79,kenn80,birrell ) . it seems plausible that such effects as surface roughness , or the microstructure of the boundary on small scales can introduce a physical cutoff needed to produce finite values of surface quantities . in brane models the imperfectness would come from the quantum gravity effects on the planck scale . an alternative mechanism for introducing a cutoff which removes singular behavior on boundaries is to allow the position of the boundary to undergo quantum fluctuations ford98 . such fluctuations smear out the contribution of the high frequency modes without the need to introduce an explicit high frequency cutoff . in order to find the renormalized value of the field square on the brane many regularization techniques are available nowadays . in particular , the generalized zeta function method is in general very powerful to give physical meaning to the divergent quantities . this method has been used in ref . @xcite for the evaluation of the field square on the brane in the model with @xmath5 bulk . another possibility is to remove from ( phi2spl ) the terms diverging in the limit @xmath234 . the relation between various procedures for the evaluation of the renormalized vev of the field square on the brane in the thin wall approximation is recently discussed in ref . in particular , on the specific example it has been shown that different methods give the same result up to finite renormalization of the surface mass and extrinsic curvature counter - terms . from the divergent behavior of the vacuum fluctuations on the brane it follows that for the points near the brane the main contribution into the integral comes from large values of @xmath233 . assuming @xmath235 , @xmath236 and replacing the bessel modified functions by their asymptotic expansions for large values of the argument , to the leading order for the contribution of the mode with a given @xmath68 one finds @xmath237 where the notation @xmath238 is introduced . note that in this limit the proper distance from the brane is much smaller compared with the ads curvature radius and the proper size @xmath239 of the internal manifold @xmath2 . now let us consider the contribution of the large kk masses along the internal subspace @xmath2 , assuming that @xmath240 . this corresponds to the kk masses @xmath241 much larger than the ads energy scale @xmath242 . in particular , for @xmath243 these conditions are satisfied for all modes along @xmath2 with nonzero kk masses . introducing a new integration variable @xmath244 , we can replace the bessel modified functions by their asymptotic expressions for large values of the argument . for the contribution of the mode with a given @xmath68 one finds @xmath245 where we have introduced the notation @xmath246 and @xmath232 is defined after formula ( [ ident11 ] ) . additionally assuming that @xmath247 or @xmath202 , one has @xmath248 from this formula it follows that for @xmath249 ( the observation point is not too close to the brane ) the contribution of kk modes with large @xmath250 is exponentially suppressed . for small values @xmath251 , @xmath252 , using the asymptotic formulae for the bessel modified functions for small values of the argument , we obtain the formula @xmath253 this limit corresponds to large distances from the brane , @xmath254 and small kk masses @xmath241 compared with the ads energy scale , @xmath255 . in particular , from eq . ( phi2splsmallza ) it follows that for fixed values @xmath242 , @xmath166 , @xmath256 the brane induced vev in the region @xmath206 vanishes as @xmath257 when the brane position tends to the ads boundary , @xmath258 . formula ( [ phi2splsmallza ] ) is further simplified for two subcases . in the case @xmath259 or equivalently @xmath260 , the main contribution into the integral comes from the lower limit ant to the leading order we find @xmath261 with the exponential suppression of the boundary induced vev . in the limit @xmath262 ( small kk masses , @xmath263 ) to the leading order we can put 0 in the lower limit of the integral . then the integral is evaluated by the standard formula for the integrals involving the square of the macdonald function and one finds ( see , for instance , @xcite ) @xmath264 in particular , this formula is valid for the zero mode . in the limit @xmath265 , again using the asymptotics of the modified bessel functions for small values of the argument , we obtain the formula @xmath266this limit corresponds to large proper distances from the brane @xmath267 in the region @xmath268 . as we see for fixed values @xmath269 , @xmath270 , @xmath250 the brane induced vev vanishes as @xmath271 when the observation point tends to the ads boundary . as it has been explained in the previous section [ see paragraph after formula ( [ angint ] ) ] , this does not mean that the presence of the brane in the bulk does not contribute to the corresponding vev in the boundary conformal field theory within the framework of ads / cft duality . to further simplify the formula ( [ phi2splsmallz ] ) we consider two subcases . for the case @xmath272 corresponding to @xmath273 , to the leading order we obtain the following formula @xmath274 in the opposite limit for @xmath275 , @xmath276 , which corresponds to small kk masses , @xmath277 , the lower limit in the integral can be replaced by 0 and to the leading order the contribution of the mode with a given @xmath68 does not depend on @xmath278 : @xmath279 next we consider the limit @xmath280 when @xmath281 is fixed , corresponding to @xmath282 , @xmath283 . the main contribution to the integral in eq . ( [ phi2spl ] ) comes from the lower limit and to the leading order one finds @xmath284 with the exponentially suppressed vev . in particular , the contribution of the nonzero kk modes along @xmath2 exponentially vanishes when the observation point tends to the ads horizon , @xmath285 . note that for the zero mode ( @xmath286 ) the brane induced vev near the ads horizon behaves as @xmath287 ( see formula ( [ phi2splsmallza2 ] ) ) . in the purely ads bulk ( @xmath32 ) this vev vanishes on the horizon for @xmath288 . for an internal spaces with @xmath289 the vev diverges on the horizon . note that for a conformally coupled massless scalar and @xmath34 the boundary induced vev takes nonzero finite value on the horizon . in the limit when the brane position tends to the ads horizon , @xmath290 , for massive kk modes along @xmath2 the main contribution into the vev of the field square in the region @xmath291 comes from the lower limit of the @xmath233-integral . to the leading order we find @xmath292 and the vev is exponentially small . as it can be seen from general formula , for the zero mode in the same limit the vev vanishes as @xmath293 . it is also of interest to consider the behavior of the vev for large values of the ads energy scale @xmath242 corresponding to strong gravitational fields . when the values of the other parameters and the coordinates @xmath294 and @xmath166 are fixed , for nonzero kk modes one has @xmath295 . additionally assuming @xmath249 , from eq . ( [ phi2spllargelamb1 ] ) we see that @xmath296 behaves as @xmath297 $ ] and is exponentially small . for the zero kk mode from the general formulae we can see that the corresponding contribution to the vev of the field square behaves as @xmath298 $ ] in the region @xmath283 and like @xmath299 $ ] in the region @xmath268 . note that the corresponding factors in the vevs integrated over the internal subspace ( see eq . ( [ phi2integrated ] ) ) contain an additional factor @xmath300 coming from the volume element and are exponentially suppressed everywhere in the parameter space . hence , the strong gravitational field suppresses the vacuum fluctuations . the similar behavior for the boundary induced quantum effects in the gravitational field of the global monopole is described in ref . the branes divide the space into three distinct sections : @xmath301 , @xmath302 , and @xmath303 . in general , there is no reason for the inverse curvature radius @xmath242 to be the same in these three sections , as branes may separate different phases of theory . the vevs in the first and last regions are the same as in the corresponding geometry of a single brane described in the previous section . in this section we consider the vev for the field square in the region between two branes . note that in the braneworld scenario with two branes based on the orbifolded version of the model this region is employed only . taking the coincidence limit in the corresponding formulae for the wightman function and using the integration formula ( rel3 ) , for the vev of the field square we obtain two equivalent representations @xmath304with @xmath305 . the last term on the right of this formula is finite on the plate at @xmath306 and diverges for the points on the brane @xmath307 , @xmath308 . these divergences are the same as those for a single brane at @xmath307 . as a result if we write the vev of the field square in the form @xmath309 then the interference part @xmath310 is finite on both plates . in the case of one parameter manifold @xmath2 with the size @xmath311 and for a given @xmath242 , the vev in eq . ( [ phi2twoplates ] ) is a function on @xmath312 , @xmath221 , and @xmath220 . note that for the points on the brane @xmath313 in the last term on the right ( [ phi2twoplates ] ) one has @xmath314 and this term vanishes for dirichlet boundary conditions . by the same way , as in the single brane case , it can be seen that in the limit @xmath228 the result for the geometry of two parallel robin plates in the bulk @xmath3 is obtained : @xmath315 , \end{aligned}\ ] ] with the notation @xmath316 from eq . ( [ cj ] ) . for large kk masses along @xmath2 , @xmath317 , for the contribution of the mode with a given @xmath68 one finds @xmath318 , \ ] ] where @xmath319 is defined by the relation similar to ( [ phi2beta ] ) . additionally assuming the conditions @xmath320 and @xmath321 , for this quantity we obtain the formula @xmath322 in particular , for @xmath323 this formula takes place for all kk modes along @xmath2 with nonzero masses . now let us consider limiting cases when the general formula for the interference part @xmath310 can be simplified . first of all assume that @xmath324 and @xmath325 . this limit corresponds to large interbrane distances compared with the ads curvature radius @xmath326 and is realized in the braneworld scenarios for the solution of the hierarchy problem . in this limit the main contribution into the @xmath233-integral comes from the region near the lower limit and for the interference part to the leading order we have @xmath327 , \label{phi22pllargezbn}\end{aligned}\ ] ] for the nonzero kk modes along @xmath2 . in the limit @xmath328 the corresponding formula takes the form @xmath329 .\ ] ] for @xmath330 and @xmath331 , by using asymptotic formulae for the bessel modified function one finds @xmath332 .\end{aligned}\ ] ] and finally , in the limit @xmath333 to the leading order we can put 0 instead of @xmath250 in the lower limit of the integral over @xmath233 and by using the asymptotic formulae for the bessel modified functions for small values of the argument , it can be seen that @xmath334 . if in addition one has @xmath335 the following formula is obtained @xmath336% \int_{0}^{\infty } du\ , u^{d_1 + 2\nu -1 } \frac{\bar{k}_{\nu } ^{(b)}(u)}{\bar{i% } _ { \nu } ^{(b)}(u)},\end{aligned}\ ] ] with the exponential suppression of the interference part . as we see for these values of the parameters the interference part in the vev of the field square is mainly located near the brane at @xmath337 . in particular , from formula ( [ phi22plas2 ] ) it follows that for the zero kk mode along @xmath215 the interference part in the vev of the field square vanishes as @xmath338 when the right brane tends to the ads horizon . note that in the same limit the contribution of a given massive kk mode vanishes as @xmath339 ( see formula ( [ phi22pllargezbn ] ) ) . when the left brane tends to the ads boundary , @xmath258 , the interference part vanishes as @xmath340 . the behavior of the interference part in the vev of the field square for large values of the ads energy scale @xmath242 directly follows from formula ( [ phi22pllargelamb1 ] ) in the case of nonzero kk masses and from formula ( [ phi22plas2 ] ) in the case of the zero mode . note that the corresponding contributions to the vev of the field square integrated over the internal space contain additional factor @xmath300 coming from the volume element . in the scenario considered in ref . @xcite @xmath341 , @xmath342 , and @xmath343 . for these values of the parameters the contribution into the interference part for the vev of field square to the leading order is given by eq . ( [ phi22pllargezbn ] ) for nonzero kk modes and @xmath344 for the zero mode . if in addition the observation point is far from the brane at @xmath337 , the zero mode contribution is determined by formula ( [ phi22plas2 ] ) with an additional suppression factor @xmath345 . in the discussion above we have considered the general case of the internal space . here we take a simple example with @xmath194 . in this case the bulk corresponds to the @xmath5 spacetime with one compactified dimension @xmath346 . the length of this dimension we will denote by @xmath216 . the corresponding normalized eigenfunctions are as follows @xmath347 first of all we consider the wightman function for the bulk without boundaries given by formula ( [ wads ] ) . we apply to the sum over @xmath68 the abel - plana summation formula @xmath348 where the prime on the sum sign means that the summand @xmath349 should be taken with the weight 1/2 . now it can be easily seen that the term in the wightman function with the first integral on the right of formula ( abelplana ) corresponds to the wightman function @xmath350 for the scalar field in the bulk @xmath5 . the latter is well investigated in literature . as a result for the wightman function in the bulk @xmath351 one finds @xmath352}{\sqrt{% u^2+v^2+k^2 } } \frac{\cosh [ ( x - x^{\prime})\sqrt{u^2+v^2+k^2}]}{e^{l\sqrt{% u^2+v^2+k^2}}-1}.\end{aligned}\ ] ] to evaluate the corresponding vev for the field square we take in this formula the coincidence limit . in this limit all divergences are contained in the part @xmath353 , and the additional part on the right of formula ( wfadss1 ) induced by the compactness of a single dimension leads to the finite result . to evaluate the corresponding integrals in the coincidence limit , first we introduce polar coordinates @xmath354 in the plane @xmath355 . after the evaluation the simple @xmath356-integral , we introduce the polar coordinates @xmath357 in the plane @xmath358 . by using the formula @xcite @xmath359 the @xmath360-integral is expressed through the hypergeometric function @xmath361 , and one finds @xmath362 as we see , unlike to the case of @xmath5 bulk , the vev for the field square in the bulk @xmath351 is a function on @xmath86 . note that the second term on the right of formula ( [ phi2adss1 ] ) is always positive . for small values of the ratio @xmath363 , to the leading order the integral is equal to @xmath364 , with @xmath365 being the riemann zeta function , and this term behaves as @xmath366 . for large values of @xmath363 it can be seen that to the leading order one has @xmath367 where we have taken into account that the quantity @xmath368 does not depend on @xmath86 . hence , near the ads horizon the vev of the field square is dominated by the second term on the right of eq . ( [ phi2adss1 ] ) . note that in the limit @xmath228 the vev @xmath369 vanishes , and taking into account that @xmath370 , from ( [ phi2s1largez ] ) we obtain the standard result for the vev in the bulk @xmath371 . brane induced vevs of the field square for the geometry under consideration are obtained from general formulae ( [ phi2spl ] ) , ( [ phi2twoplates ] ) by the replacements @xmath372 as it has been mentioned in section [ sec : vevphi2 ] , for @xmath34 and conformally coupled massless scalar field ( @xmath84 ) the single brane induced vev is finite on the ads horizon . the corresponding limiting value is obtained from ( [ phi2splsmallza2 ] ) : @xmath373 for the general case , the vev in this limit behaves as @xmath374 . in figure [ fig1 ] we have plotted the dependence of the single brane induced vev @xmath375 on the ratio @xmath220 for @xmath376 minimally ( left panel ) and conformally ( right panel ) coupled massless scalar fields with @xmath377 . the full and dashed curves correspond to the values @xmath378 and @xmath379 respectively . as it has been shown in section [ sec : vevphi2 ] , in the limit @xmath380 the vev behaves as @xmath381 . for the points near the ads horizon , @xmath285 , the main contribution comes from the zero mode and the vev behaves as @xmath374 . on the brane surface the vev diverges like @xmath382 . [ cols="^,^ " , ] it is not difficult to generalize the corresponding results for the case of the internal space @xmath38 with radius @xmath383 and curvature scalar @xmath384 . now the eigenfunctions @xmath54 are expressed in terms of spherical harmonics of degree @xmath167 , @xmath385 . the vevs for the field square are obtained from the general formulae in sections [ sec : vevphi2 ] and [ sec : phi2twopl ] by the replacements @xmath386 where the factor under the sum sign on the right in eq . ( [ sd1psi ] ) is the degeneracy of each angular mode with a given @xmath167 . from the point of view of embedding the braneworld model into a more fundamental theory , such as string / m - theory , one may expect that a more complete version of this scenario must admit the presence of additional extra dimensions compactified on a manifold @xmath2 . in the present paper we have extended the previous work describing the local vacuum effects in the braneworlds with the ads bulk on a higher dimensional brane models which combine both the compact and warped geometries . this problem is also of separate interest as an example with gravitational , topological , and boundary polarizations of the vacuum , where one - loop calculations can be performed in closed form . we have investigated the wightman function and the vacuum expectation value of the field square for a scalar field with an arbitrary curvature coupling parameter satisfying robin boundary conditions on two parallel branes in @xmath131 spacetime . the kk modes corresponding to the radial direction are zeros of a combination of the cylinder functions . the application of the generalized abel - plana formula to the corresponding mode sum in the expression of the wightman function allowed us to extract the boundary - free part and to present the brane induced parts in terms of integrals exponentially convergent in the coincidence limit of the arguments . in the region between two branes the wightman function is presented in two equivalent forms given by eqs . ( w15 ) and ( [ w17 ] ) . the first term on the right of these formulae is the wightman function for the bulk without boundaries . the second one is induced by a single brane and the third term is due to the presence of the second brane . further we give an application of our results for two branes case to the higher dimensional version of the randall - sundrum braneworld with arbitrary mass terms on the branes . for the untwisted scalar the robin coefficients are expressed through these mass terms and the curvature coupling parameter by formulae ( [ atilders ] ) . for the twisted scalar dirichlet boundary conditions are obtained on both branes . the expectation values for the field square are obtained from the wightman function taking the coincidence limit . for the case of a single brane geometry this leads to formula ( [ phi2spl ] ) for the region @xmath206 . the corresponding formula in the region @xmath291 is obtained from eq . ( phi2spl ) by replacements @xmath155 . for a one parameter manifold @xmath2 , the vev of the field square induced by a single brane is a function on the proper distance of the observation point from the brane and on the ratio of the physical size of @xmath224 ( from the viewpoint of an observer residing on the brane ) to the ads curvature radius . as a partial check of our formulae , we have shown that in the limit when the ads radius goes to infinity the result for the brane in the bulk @xmath87 is obtained . on the boundary the vev for the field square diverges . in the limit when the proper distance from the brane is much smaller compared with the ads curvature radius and proper size @xmath239 of the internal manifold , the leading term of the corresponding asymptotic expansion for a contribution of a given kk mode along @xmath2 is given by expression ( [ phi2splnear ] ) . this term does not depend on the robin coefficient and has different signs for dirichlet and non - dirichlet scalars . the coefficients for the subleading asymptotic terms will depend on the bulk curvature radius , robin coefficient and on the mass of the field . in the limit when the kk masses @xmath241 are much larger compared to the ads energy scale @xmath242 , the contribution of a given mode with fixed @xmath387 to the vev of the field square is given by formula ( phi2spllargelamb1 ) . from this formula it follows that if additionally one has @xmath388 , the contribution of the corresponding kk modes is exponentially suppressed . in the region @xmath283 , for large distances from the brane , @xmath254 , and small kk masses @xmath389 compared with ads energy scale , @xmath390 , the contribution of a given kk mode along the internal space @xmath2 is approximated by formula ( [ phi2splsmallza ] ) . in both subcases @xmath391 and @xmath392 we have exponentially suppressed vevs by the factors @xmath393 $ ] and @xmath394 $ ] in the first and second cases respectively . for large distances from the brane in the region @xmath268 , the contribution of a given kk mode is presented in the form ( [ phi2splsmallz ] ) . this formula is further simplified in the subcases of large and small kk masses @xmath241 . in the case of large mass , @xmath395 , one has an exponential suppression by the factor @xmath396 $ ] . in the limit of small masses , @xmath255 , the suppression is by the factor @xmath397 $ ] . on the ads horizon ( @xmath398 ) , the brane induced vev is exponentially small for nonzero kk masses along @xmath2 and behaves as @xmath287 for the zero mode . on the ads boundary , corresponding to @xmath161 , the brane induced vev vanishes as @xmath381 . when the brane position tends to the ads boundary , @xmath258 , the brane induced vev in the region @xmath206 vanishes as @xmath340 . in the limit when the brane tends to the ads horizon , @xmath399 , the vev in the region @xmath291 behaves as @xmath400/z_a^{d_1/2}$ ] for massive kk modes along @xmath2 , and vanishes as @xmath293 for the zero mass mode . for large values of the ads energy scale @xmath242 corresponding to strong gravitational fields , the vevs integrated over the internal space ( see formula ( [ phi2integrated ] ) ) are exponentially suppressed everywhere in the parameter space . for two branes geometry , by using the corresponding expression for the wightman function , vev for the field square in the region between the branes is presented as a sum of boundary - free , single brane induced and interference parts , eq . ( [ phi2twopl1 ] ) . the latter is regular everywhere including the points on the branes . in the limit @xmath228 the result for the geometry of two branes in the bulk @xmath3 is obtained . for large kk masses along @xmath2 , @xmath401 , the contribution of the kk mode with a given @xmath402 is determined by formula ( [ phi22pllargelamb ] ) . this formula is further simplified when the branes are not too close to each other and the observation point is sufficiently far from the branes , @xmath403 and @xmath404 , @xmath305 . in this case we obtain formula ( phi22pllargelamb1 ) with exponentially small interference part . for the modes with masses in the range @xmath405 , the contribution to the vev of the field square is determined by formula ( phi22plas1a ) with two small factors , @xmath406 and @xmath407 . for small values of kk masses , @xmath408 , and large interbrane distances , to the leading order the contribution of a given kk mode does not depend on the kk mass and the corresponding vev is exponentially suppressed by the factor @xmath409 $ ] . if in addition the observation point is far from the brane at @xmath41 , additional suppression factor @xmath410 $ ] appears and the interference part is mainly located near the brane @xmath41 . when the right brane position tends to the ads horizon , @xmath411 , the interference part in the vev of the field square vanishes as @xmath412 for the zero kk mode along @xmath2 . for a given massive kk mode this quantity behaves as @xmath413/z_b^{d_1/2}$ ] . when the left brane tends to the ads boundary , the interference part vanishes as @xmath340 . as an illustration of general results , in section [ sec : example ] we consider an example of the internal space @xmath194 . using the abel - plana formula , the wightman function for the boundary - free geometry is presented in the form ( [ wfadss1 ] ) , where the first term on the right is the corresponding function for the @xmath5 bulk . the vev of the field square is obtained in the coincidence limit and has the form ( [ phi2adss1 ] ) . unlike to the case of @xmath5 bulk , this vev for the bulk @xmath414 depends on @xmath86 coordinate and behaves as @xmath415 for small values of the ratio @xmath363 and as @xmath416 for large values of this ratio . in figures [ fig1 ] and [ fig2 ] we have presented single brane induced and interference parts in the vev of the field square for minimally and conformally coupled scalars in the bulk geometry @xmath417 as functions on the distance from the brane at @xmath40 . we also discuss the generalization of the corresponding results for the internal space @xmath418 . the vevs in this case are obtained from the general formulae in sections [ sec : vevphi2 ] and [ sec : phi2twopl ] by the replacements ( [ sd1psi ] ) and ( [ sd1lambda ] ) . in particular , from the point of view of embedding the randall - sundrum model into the string theory , the case of the bulk @xmath419 is of special interest . this spacetime also plays an important role in the discussions of the holographic principle . note that in this paper we have considered boundary induced vacuum densities which are finite away from the boundaries . we expect that similar results would be obtained in the model where instead of externally imposed boundary condition the fluctuating field is coupled to a smooth background potential that implements the boundary condition in a certain limit @xcite . the work was supported by the cnr - nato senior fellowship , by ansef grant no . 05-ps - hepth-89 - 70 , and in part by the armenian ministry of education and science , grant no . 0124 . the author acknowledges the hospitality of the abdus salam international centre for theoretical physics ( trieste , italy ) and professor seif randjbar - daemi for his kind support . mostepanenko and n.n . trunov , _ the casimir effect and its applications _ ( oxford university press , oxford , 1997 ) ; g. plunien , b. muller , and w. greiner , phys . rep . * 134 * , 87 ( 1986 ) ; m. bordag , u. mohideen , and v.m . mostepanenko , phys . rep . * 353 * , 1 ( 2001 ) ; k.a . milton , _ the casimir effect : physical manifestation of zero point energy _ ( world scientific , singapore , 2002 ) ; k.a . milton , j. phys . a * 37 * r209 ( 2004 ) . saharian , izv . matematika * 22 * , 166 ( 1987 ) [ sov . j. contemp . analysis * 22 * , 70 ( 1987 ) ] ; a.a . saharian , `` the generalized abel - plana formula . applications to bessel functions and casimir effect , '' report no . ic/2000/14 , hep - th/0002239 . a.a . saharian and m.r . setare , class . . grav . * 20 * , 3765 ( 2003 ) ; a.a . saharian and m.r . setare , int . j. mod a * 19 * , 4301 ( 2004 ) ; a.a . saharian and e.r . bezerra de mello , j. phys . a * 37 * , 3543 ( 2004 ) . n. graham , r.l . jaffe , v. khemani , m. quandt , m. scandurra , and h. weigel , nucl . b645 * , 49 , ( 2002 ) ; n. graham , r.l . jaffe , and h. weigel , int . a * 17 * , 846 ( 2002 ) ; n. graham , r.l . jaffe , v. khemani , m. quandt , m. scandurra , and h. weigel , phys . b * 572 * , 196 ( 2003 ) ; k.d . olum and n. graham , phys . b * 554 * , 175 ( 2003 ) ; n. graham and k.d . olum , phys.rev . d * 67 * 085014 ( 2003 ) .

a variational framework is developed to examine the equilibrium states of a semiflexible polymer that is constrained to lie on a fixed surface . as an application the confinement of a closed polymer loop of fixed length within a spherical cavity of smaller radius , , is considered . it is shown that an infinite number of distinct periodic completely attached equilibrium states exist , labeled by two integers : and , the number of periods of the polar and azimuthal angles respectively . small loops oscillate about a geodesic circle : , is the stable ground state ; states with higher exhibit instabilities . if new states appear as oscillations about a doubly covered geodesic circle ; the state replaces the two - fold as the ground state in a finite band of values of . with increasing , loop states make a transition from oscillatory and orbital behavior on crossing the poles , returning to oscillation upon collapse to a multiple cover of a geodesic circle ( signaled , respectively , by an increase in and an increase in ) . the force transmitted to the surface does not increase monotonically with loop size , but does asymptotically . it behaves discontinuously where changes . the contribution to energy from geodesic curvature is bounded . in large loops , the energy becomes dominated by a state independent contribution proportional to the loop size ; the energy gap between the ground state and excited states disappears . _ instituto de ciencias nucleares , universidad nacional autnoma de mxico + apdo . postal 70 - 543 , 04510 mxico , df , mexico + _

1109.6555

understanding how surfaces may constrain the configuration of biopolymers on mesoscopic scales is important in a number of processes in cell biology . it is particularly relevant in the packing of dna within viral envelopes @xcite . modeling all of the relevant interactions is complicated : one needs to accommodate the competition between polymer elasticity and entropy ; in the case of dna , electric fields will be important . a nice short review in this context is provided in @xcite . recently , simulations treating various features of the confinement process have been performed @xcite ; the former focuses on entropy , the latter on the dramatic effects of friction and the finite transverse dimensions of the confined object . in this paper , we address an aspect of the problem of a fundamental nature that does not appear to have been addressed previously in any detail : how does one characterize the equilibrium states of the three - dimensional elastic energy of a polymer confined within a fixed surface ? while this description of confinement leaves out a lot of the physics that is relevant in biological systems , it presents a well - defined problem exhibiting a striking level of complexity that is worth studying in its own right . the semi - flexible polymer will be modeled as a curve in three - dimensional space parameterized by arc - length @xmath11 . the bending energy associated with a given conformation is quadratic in the frenet - serret curvature along the curve , @xmath12 , @xmath13 curves of fixed length minimizing the unconstrained energy ( [ hamk2 ] ) were first studied in depth by euler . a historical review is provided in ref . @xcite ; a more contemporary approach to the problem is presented in @xcite and reviewed in ref . an alternative framework , which lends itself to adaptation to the confinement problem , was developed in ref . @xcite . in the absence of constraints on a closed loop , there is not a lot to say : the only stable equilibrium is a circle . confinement within a surface , however , will generally oblige the loop to adopt a non - circular shape , increasing its elastic energy . the contact with the surface itself may be complete or it may be partial ; and contrary to one s initial guess , the bound state will not generally follow a surface geodesic ; nor need it be unique . the wrapping of a semi - flexible polymer around a cylinder , a closely related problem relevant in the the winding of dna around histone octamers , was first examined some time ago by nickerson and manning @xcite . their strategy was to look at the independent degrees of freedom of the surface - bound polymer . while focusing directly on these degrees of freedom makes sense , it does not exploit the symmetries of the problem . for , even though the constraint breaks the euclidean invariance of the three - dimensional bending energy , how this occurs is not arbitrary . a variational framework is developed here that involves the unconstrained degrees of freedom , imposing the constraint using a local lagrange multiplier . this multiplier will quantify the loss of euclidean invariance in the constrained system . its value at any point along the loop will be identified as the local normal force that is being transmitted to the surface . the well - known integrability of the euler - lagrange equations for the unconstrained curve is a consequence of the euclidean invariance of its energy . the constrained counterpart generally will not be integrable . in various interesting cases , however , the confining geometry will respect some subgroup of the euclidean group . in particular , the conservation of torque associated with the residual rotational invariance of a sphere permits the euler - lagrange equation to be cast as a quadrature in the geodesic curvature which can be integrated directly . it also provides a direct recipe for the reconstruction of the loop from its curvature data adapted to the conserved torque vector as an axis of symmetry . in particular , we apply this framework to examine the confinement of a closed polymer loop of fixed length @xmath0 within a sphere of radius @xmath14 . in contrast to an open polymer which will wind around a geodesic circle on the sphere when its length exceeds @xmath15 , the closure of the loop is incompatible with a geodesic unless @xmath9 is tuned to be commensurate with an integer multiple of @xmath1 . we show that there exists an infinite number of distinct completely attached states , labeled by a pair of integers , @xmath6 and @xmath10 : @xmath16 , @xmath17 , the number of periods of the polar angle and azimuthal angles in one circuit of the loop respectively . @xmath6 characterizes the dihedral symmetry with respect to the axis of symmetry ; @xmath10 characterizes the number of revolutions about this axis . small loops oscillate symmetrically about a geodesic circle with @xmath5 and an @xmath6-fold symmetry , @xmath18 . the two - fold is the stable ground state . for any finite values of @xmath9 , the higher @xmath6-folds are unstable with respect to decay toward the two - fold . states with @xmath4 are illustrated in figs . [ figure1](a ) - [ figure1](c ) for various normalized values of @xmath9 . + + in the interval @xmath19 $ ] within a unit sphere : ( a ) and ( b ) display increasingly large oscillations about the equator ; ( c ) oscillations develop overhangs ; ( d ) first self - contacts made at the two poles ; ( e ) self - intersecting triple orbit of sphere ; ( f ) orbit collapses to triple cover of the equator . the normalized local confining force @xmath20 is color coded in these figures.,title="fig : " ] beyond some critical size , the loop will exhibit self - intersections on the sphere . when @xmath4 this will occur when @xmath21 where the loop crosses the poles as illustrated in fig . [ figure1](d ) . we suppose that self - intersections are consistent with the physics and do not cost energy . as @xmath9 is increased to @xmath22 , the loop will collapse onto a geodesic circle which it will cover three times , ( see figs . [ figure1](e ) and [ figure1](f ) . the discontinuity in the number of revolutions occurs at an intermediate value of @xmath9 where the loop crosses the poles . this is accompanied by a transition from oscillatory to orbital behavior . as @xmath9 is increased above @xmath23 , the dihedral symmetry of the lowest energy state descending from the @xmath4 ground state will jump to @xmath24 with the reestablishment of oscillatory behavior . the bending energy of this state as well as the total force that gets transmitted to the surface will be determined as a function of @xmath9 . the three- dimensional bending energy decomposes on the surface into a sum of two terms : one is associated with the geodesic curvature , intrinsic to the sphere ; the other is associated with the normal curvature inherited from the surface , constant on a sphere . the latter energy thus counts the number of times the loop is wound within the sphere and it is proportional to @xmath9 . the geodesic energy periodically falls to zero whenever @xmath25 and the loop is geodesic . local maxima , associated with the incommensurability of the loop length with geodesic behavior , are displayed between these values ; we show that their values decrease monotonically with loop size . for large values of @xmath9 , the normal contribution is always dominant . it does not depend on the state in question thus , in this limit , the energy gap between the ground state and excited states disappears . the local force transmitted by the completely attached ground state loop is positive everywhere . surprisingly , the total force does not grow monotonically with loop length except asymptotically , where it grows linearly with @xmath9 and coincides with the naive expression , energy divided by @xmath1 . the change of symmetry as @xmath9 passes through odd integral multiples of @xmath1 , manifests itself in a positive jump in the transmitted force analogous to a euler instability associated with the buckling into oscillations . when @xmath7 , a new set of states appears as oscillations about a doubly covered geodesic circle , @xmath26 with an @xmath6-fold symmetry , @xmath27 ; among these new states , the lowest energy is displayed in the three - fold . also , because the geodesic energy is small when the oscillations are small , these states will have lower energy than the two - fold ground state in a finite band of values of @xmath9 beginning at @xmath28 . there will be similar behavior in an infinite set of bands of values of @xmath9 where the descendant states become geodesic . one thus needs to reassess the stability of the two - fold ground state and its descendants . we provide a heuristic argument for stability by constructing a homotopy that interpolates between the states with @xmath4 , @xmath5 and @xmath29 , @xmath26 . by examining the energy along this homotopy , we show that a steep energy barrier separates the two equilibrium states . the @xmath4 , @xmath5 state and its descendants thus appear to be stable classically . for any @xmath7 , there are two stable states that alternate as the ground state as the length is increased . the paper is organized as follows : in section [ confinedelastica ] , we describe the framework . in particular , the breaking of translational invariance will be quantified by the non - conservation of a vector along the loop . it will be shown that the euler - lagrange equation for the curve can be cast as the vanishing of a linear combination of the unconstrained euler - lagrange derivatives . the constraining force is identified as some other linear combination of these derivatives . in sect . [ spherical ] the confinement of a closed loop by spheres will be considered . in secs . [ weak ] and [ strong ] we analyze loops in the harmonic and non - linear regimes respectively . the equilibrium states of the loop will be identified and the forces they transmit to the surface determined . an assessment of the stability of these states will be provided . we conclude with a discussion and a few suggestions for future work in sec . [ discussion ] . a number of useful definitions , identities and derivations are collected in a set of appendixes . consider a space curve @xmath30 parametrized by arclength constrained to lie on a surface @xmath31 . this surface is described in parametric form by the mapping @xmath32 . the confined curve can then also be described as a surface curve @xmath33 . in order to enforce the condition that @xmath34 lie on @xmath31 , one adds to the energy @xmath35 $ ] given by eq . ( [ hamk2 ] ) a term enforcing this constraint : @xmath36 = h[\mathbf{y } ] + \int \ , ds \,\ , \bm{\lambda}(s ) \cdot \left [ \mathbf{y}(s ) - \mathbf{x } ( u^a(s ) ) \right]\,,\ ] ] where @xmath37 is a vector - valued lagrange multiplier defined along the curve . this constraint will break the manifest translational invariance of the energy @xmath38 . the variation of @xmath39 with respect to the embedding functions @xmath40 can be cast in the form @xmath41 where prime represents derivation with respect to arclength and the tension in the loop @xmath42 is given by @xmath43 here @xmath44 is the standard frenet - serret frame carried by the curve and @xmath45 is its torsion . the constant @xmath46 is associated with the constraint of fixed length . a derivation of eq . ( [ eefsforce ] ) is provided in [ stresstender ] . in equilibrium , one finds that @xmath47 . thus , in the presence of the constraint , the tension in the loop is not conserved ; the multiplier is identified as the external force associated with the constraint @xcite . the corresponding variation of @xmath39 with respect to @xmath48 is given by @xmath49 where @xmath50 , @xmath51 are the two tangent vectors to the surface adapted to the parametrization by @xmath52 . in equilibrium , @xmath53 ; in equilibrium , the force on the curve associated with the constraint always acts orthogonally to the surface . let us write @xmath54 , where @xmath55 is the unit vector normal to @xmath31 . the combined result is that @xmath56 an integrability condition for closed curves follows from eq . ( [ el ] ) : @xmath57 this identity holds whether or not contact is complete . ( [ eefsforce ] ) along with the frenet - serret equations , a straightforward calculation decomposes @xmath58 along the normals @xmath59 where the euler - lagrange derivatives of the bending energy @xmath60 and @xmath61 are given by @xmath62 the tangential euler - lagrange derivative @xmath63 vanishes identically , a consequence of the fact that the only relevant degrees of freedom are geometrical , whether the curve is constrained or not . the surface - bound curve also carries a darboux frame , @xmath64 . relevant properties of this frame are summarized in [ appdarbouxframe ] . the frenet normals are related to their darboux counterparts by a rotation about the tangent vector : @xmath65 the relationship @xmath66 permits one to express the angle of rotation in terms of a curvature ratio : @xmath67 or @xmath68 , where @xmath69 and @xmath70 are , respectively , the geodesic and normal curvatures along @xmath71 , @xmath72 using these expressions it is possible to decompose @xmath58 given by eq . ( [ eq : fprimeel ] ) in a form adapted to the surface , @xmath73 the projection of eq . ( [ el ] ) onto @xmath74 provides the euler - lagrange equation in the remarkably simple form , @xmath75 the multiplier @xmath20 does not appear . the corresponding projection onto @xmath76 determines @xmath20 , @xmath77 in this approach , one sees explicitly how both the euler - lagrange equation ( [ eulelsurfeneb ] ) and the confining force ( [ magnlambda ] ) are constructed out of the two unconstrained euler - lagrange derivatives . the normal force is completely determined when the local geometry is known . the euler - lagrange equations for an unconstrained elastic curve , given by @xmath78 and @xmath79 , are replaced by the single equation , @xmath80 . the apparent discrepancy in the number of equations reflects the fact that a space curve possesses two independent modes of deformation whereas a surface curve has only one . in general , the integrability of the former pair of equations is surrendered when the constraint is present . using the identities ( [ kaptau ] ) and ( [ dkappagn ] ) the euler - lagrange equation ( [ eulelsurfeneb ] ) can be expressed completely in terms of surface curvatures , @xmath81 this agrees with the equation derived in ref . @xcite using a very different approach . note that it involves the curvatures as well as the geodesic torsion and , in general , the curve will not follow a geodesic with @xmath82 . for a geodesic to minimize bending energy , one requires one of the following to occur : @xmath83 is constant ; the curve coincides with an asymptotic line with @xmath84 or a principal curve with @xmath85 . such conditions typically do not occur unless they do so trivially . the magnitude of the force @xmath20 transmitted to the surface , given by eq . ( [ magnlambda ] ) , assumes the form @xmath86 its magnitude will vary along the contact region , even for confinement by a sphere . this expression is missing in the framework presented in @xcite . a curious consequence of the symmetric decomposition of the frenet - serret curvature in terms of the geodesic and the normal curvatures is the fact that @xmath20 turns out to be identical to the euler - lagrange derivative @xmath87 given by ( [ eleulerelastonasurf ] ) under the interchange of @xmath69 with @xmath70 and a change of sign of the last term . in general , under confinement , one surrenders not only translational invariance but also rotational invariance . the torque about the origin per unit length of the curve , @xmath88 , is given by @xmath89 where @xmath90 . the first term on the right in eq . ( [ eq : mfs ] ) is the torque due to the force @xmath42 ; the second term is the bending moment originating in second derivatives in the bending energy . for a free curve , @xmath91 , which can be cast in the manifestly translationally invariant form @xmath92 @xcite . in general , @xmath93 is not conserved . one has instead @xmath94 thus , in a confined equilibrium with @xmath95 , @xmath93 will not generally be conserved . the source is given by the moment of the force associated with the constraint . if the confining geometry is a sphere centered on the origin , so that @xmath96 is directed along the normal vector , @xmath93 will be conserved . if it is symmetric about some axis ( say the @xmath97 axis ) , then the conserved quantity is the corresponding projection of @xmath93 , that is , @xmath98 . consider a closed curve of length @xmath99 , confined within a sphere of radius @xmath1 . we normalize lengths in terms of @xmath1 . on a sphere , the extrinsic curvature tensor is proportional to the metric @xmath100 , so that the normal curvature is constant , @xmath101 , and the geodesic torsion vanishes , @xmath102 . using eq . ( [ eq : ceef ] ) , the vector @xmath103 given by eq . [ eefsforce ] , then reduces to @xmath104 @xmath42 is everywhere tangent to the surface . it is also completely determined by the intrinsic geometry . however , it is not conserved . the torque vector defined by eq . ( [ eq : mfs ] ) is given by @xmath105 the rotational invariance of bending energy confined to a sphere implies that @xmath93 is a constant vector . in particular , its length is a constant . this provides a quadrature for the geodesic curvature ( @xmath106 ) : @xmath107 it is simple to check that the condition @xmath108 in eq . ( [ ceespherem2 ] ) reproduces the euler - lagrange eq . ( [ eleulerelastonasurf ] ) on a sphere : @xmath109 this equation is also identical to the one describing an elastic loop on a sphere with an energy density , @xmath110 . the minimization of the constrained three - dimensional bending energy coincides in this case with that of two - dimensional spherical bending energy . this is a consequence of the decomposition of the frenet - serret curvature into normal and geodesic parts , @xmath111 as well as the fact that @xmath70 is a constant . spheres are special in this respect . various mathematical properties of elastic curves on spheres were described by langer and singer in the 1980s @xcite ; see also @xcite for a numerical treatment of the problem . recently they have been reexamined in some detail in @xcite . the connection to the description of conical defects in unstretchable flat sheets was developed in @xcite and @xcite . while the mathematical literature provides a useful point of departure , the absence of any reference either to energy or to the transmitted forces limits its usefulness . note that in a geodesic , @xmath112 everywhere . equation ( [ ceespherem2 ] ) then implies that @xmath113 . unless @xmath9 is an integer , however , geodesics will be inconsistent with the boundary conditions associated with closure . if @xmath112 is accessible anywhere along the loop , there will be a non - trivial bound on @xmath114 from below : @xmath115 in equilibrium , all confined loops will satisfy this bound . it is also useful to cast eq . ( [ ceespherem2 ] ) in the alternative form @xmath116 where @xmath117 and @xmath118 is manifestly positive . ) . ] if @xmath69 is identified as the position of a particle and @xmath11 as time , then eq . ( [ ceespherefirstint ] ) described the motion of a particle of mass @xmath119 and total `` energy '' @xmath120 in the symmetric quartic potential @xmath121 . if @xmath122 , @xmath123 possesses a single minimum at @xmath112 ; if @xmath124 it possesses two symmetric wells centered at @xmath125 , separated by a local maximum at @xmath112 . while the particle analogy is useful , it does have its limitations . @xmath120 also is not the energy of the loop and the potential depends on the constant of integration @xmath126 which we are not free to tune but , like the ` energy ' , is itself determined by the boundary conditions . the qualitative behavior of the loop will depend on the turning points of the potential , and thus on the relative values of @xmath127 and @xmath120 : \(1 ) @xmath128 ( equivalently eq . ( [ eq : parabola ] ) is satisfied ) . in this parameter regime , there are only two turning points . one has @xmath129 where @xmath130 and @xmath131 . thus @xmath132 when @xmath133 . the geodesic curvature thus ranges in the symmetric interval @xmath134 $ ] . this will be independent of the sign of @xmath126 . the loop will oscillate symmetrically about the equator where @xmath112 ( see eq . ( [ eq : theta ] ) ) so that @xmath135 . \(2 ) @xmath136 . this regime is inaccessible physically if the loop is closed . so that the potential possesses two wells and the trajectory in @xmath69 is confined to oscillate in one of them . one can write @xmath137 where @xmath138 and @xmath139 . @xmath69 is then confined to the lie in one of two intervals with a definite sign . it is thus confined to inhabit a single hemisphere . it is intuitively clear that any such state will spontaneously unbind into the interior of the sphere where it may relax into a lower energy bound state . ] the reconstruction of the loop from its curvature data involves examining the conserved torque vector @xmath93 . without loss of generality , it is always possible to allign @xmath140 along the @xmath141 axis , @xmath142 ( we follow ref . the normal vector @xmath55 is parametrized in terms of spherical polar coordinates @xmath143 and @xmath144 , @xmath145 the projection of @xmath140 , given by eq . ( [ torquevctsphcnf ] ) , onto @xmath76 determines the polar angle in terms of @xmath69 : reproduces ( the derivative of ) eq . ( [ eq : theta ] ) . ] @xmath146 its projection onto @xmath74 determines the azimuthal angle @xmath144 @xmath147 thus the projections of @xmath140 onto the darboux frame determine the embedding functions of the curve on the sphere in terms of @xmath69 and the two constants , @xmath126 and @xmath114 . @xmath144 will not generally increase monotonically along the loop . it will exhibit overhangs with @xmath148 , if @xmath149 . combining eqs . ( [ eq : theta ] ) and ( [ eq : phi ] ) gives @xmath150 note that eq . ( [ eq : theta ] ) implies the bound on @xmath69 , @xmath151 , already implicit in the quadrature , eq . ( [ ceespherem2 ] ) . ) which implies , for positive @xmath126 , @xmath152 . ] as we will see , @xmath114 will always be bounded . thus @xmath69 and with it the geodesic energy will also be bounded . equation ( [ eq : theta ] ) also implies that the extremal values of @xmath69 occur where the polar angle is turning . the bound is saturated when the loop passes through the poles . using the identification , ( [ eq : theta ] ) , it is possible to recast the quadrature in terms of @xmath153 . one has @xmath154 in this form , it is clear that access to the poles is possible only when @xmath114 and @xmath126 are tuned so that the coefficient of the divergent term in the potential appearing in eq . ( [ eq : quadtheta ] ) vanishes : @xmath155 this will require @xmath156 . when eq . ( [ eq : s ] ) is satisfied , the evolution of @xmath144 simplifies . equation ( [ eq : phitheta ] ) assumes the form @xmath157 so that @xmath144 increases linearly with arclength along the loop . in general , the dependence of @xmath144 on @xmath11 is not monotonic , much less linear . this identity , in turn , implies the value @xmath158 ( and as a consequence of eq . ( [ eq : s ] ) , @xmath159 at pole crossing . let us first suppose that the loop is sufficiently small so that neither self - contact nor self - intersections occur . the periodic motion in the potential @xmath121 implies an @xmath6-fold dihedral symmetry : the loop closes upon completing @xmath6 periods of @xmath153 in one revolution of the polar axis so that @xmath160 and @xmath161 , where @xmath162 is an integer . ) implies that the former is equivalent to @xmath163 . the quadrature then implies that closure is smooth . ] closure provides a quantization of physical states . is incompatible with the four - vertex theorem for a sphere so does not occur . ] in equilibrium , the number of circuits of the polar axis , @xmath10 , will increase with loop length , so that the boundary condition on @xmath144 is replaced with @xmath164 . using eqs . ( [ eq : theta ] ) and ( [ eq : phitheta ] ) , and the quadrature ( [ ceespherefirstint ] ) it is possible to cast the boundary conditions in the form @xmath165 and @xmath166 where @xmath167 is the turning point of the potential , defined by eq . ( [ eq : turningpoints ] ) . equation ( [ eq : c1 ] ) is independent of @xmath10 , whereas eq . ( [ eq : c2 ] ) is independent of @xmath9 . together , they determine the two constants of integration @xmath126 and @xmath114 in terms of the loop radius @xmath9 and the two integers @xmath6 and @xmath10 . while eq . ( [ ceespherem2 ] ) can be integrated exactly in terms of elliptic functions @xcite ) , a perturbative approach to the problem is instructive . let us thus suppose that @xmath168 . such a loop is sufficiently small that neither self - contact nor self - intersections occur . we thus expand the function @xmath69 as well as the constants @xmath126 and @xmath114 in powers of @xmath169 , the small dimensionless parameter in the problem : @xmath170 the equilibrium states are described by small oscillations about a geodesic circle on the sphere with @xmath112 . the harmonic approximation of the quadrature ( [ ceespherem2 ] ) about @xmath112 reads @xmath171 at lowest order , the arclength coincides with the azimuthal angle , @xmath144 , and the geodesic curvature along a closed loop is given by @xmath172 where @xmath6 is an integer and @xmath173 is a constant . @xmath174 is determined by the amplitude @xmath173 and the @xmath6 ( see [ bcdetail ] ) . the quadrature implies that @xmath175 and @xmath176 . it also implies the constraint @xmath177 on the difference of their second order corrections . equation ( [ linear ] ) implements the boundary conditions at lowest order . to complete the specification of @xmath178 and @xmath179 in terms of @xmath174 , it is necessary to examine the boundary conditions ( [ eq : c1 ] ) and ( [ eq : c2 ] ) correct at next to leading order . one finds that , for @xmath162 , eqs . ( [ eq : c1 ] ) and ( [ eq : c2 ] ) together imply @xmath180 and @xmath181 the details of the derivation are provided in [ bcdetail ] . using eq . ( [ eq : k1sm ] ) in ( [ eq : delr2 ] ) one reproduces the relationship eq . ( [ eq : a1delr ] ) between @xmath173 and @xmath174 , @xmath182 . it then follows that , for @xmath183 , @xmath184 the bending energy of the loop confined by the sphere decomposes into a sum of geodesic and normal parts , reflecting the decomposition of the frenet curvature , @xmath185 , @xmath186 @xmath187 is the energy associated with an elastic rod that been wound into a circular coil of radius @xmath188 . it grows linearly with loop length @xmath0 and is state independent . for a weakly confined loop @xmath189 where @xmath190 is the bending energy of a circular loop of radius @xmath9 . the energy increases linearly with loop size . this will not be true in longer loops . for a fixed value of @xmath174 , the energy increases quadratically with @xmath6 . the ground state , as one would have predicted , is the completely attached @xmath4 state . as we show , all states with @xmath191 are unstable . the force transmitted to the sphere at any point , given in eq . ( [ ceelmagnlambda ] ) , takes the particularly simple form @xmath192 it differs from the local energy density : compare eqs . ( [ htotsphere ] ) and ( [ lambdasphere ] ) . its spatial dependence is completely determined by the geodesic curvature . it is bounded from below by @xmath126 . this lends a direct physical interpretation of @xmath126 . for small @xmath174 and each @xmath162 , the force is given by @xmath193\,;\ ] ] while it oscillates along the loop , it remains positive everywhere . @xmath194 does not vanish in the limit @xmath195 . this is interpreted as an euler instability . the buckling of the circular loop into an @xmath6-fold involves a critical compression in the loop provided by this normal force . thereafter , @xmath194 decreases linearly with @xmath174 . this may appear counterintuitive . this behavior will be put in context when we examine loops outside of perturbation theory . here we examine the stability of the weakly confined @xmath6-folds that have been described . to do this we require the second variation of the energy with respect to small deformations of the loop . this is expressed in the form @xcite @xmath196 where @xmath197 is the deformation along @xmath198 , and the self - adjoint operator @xmath199 is given by @xmath200 the fixed length constraint along the spherical surface implies a global constraint on the normal deformation , @xmath201 periodicity implies that the modes of deformation are represented by a constant ( @xmath202 ) , @xmath203 and @xmath204 , @xmath205 . for fixed @xmath6 , the eigenvalues of @xmath199 are then labeled by the integer @xmath206 , given by @xmath207 the constant mode with @xmath202 has positive @xmath208 . all eigenvalues with @xmath209 possess a two - fold degeneracy . there are four zero modes satisfying @xmath210 ; two occur at @xmath211 and two at @xmath212 . the mode @xmath213 corresponds to rotation of the loop about the axis of symmetry . the two modes with @xmath214 correspond to rotations about an orthogonal axis . these three modes are the zero modes anticipated by the rotational invariance of the bending energy . the fourth zero mode @xmath215 is inconsistent with the fixed length constraint ( [ eq : arccon1 ] ) and , so , is unphysical . it is also the only mode of deformation inconsistent with this constraint in this regime . the two - fold ground state with @xmath4 is stable . there are no modes of deformation with negative eigenvalue . all excited confined states are unstable . there will be @xmath216 unstable modes of deformation corresponding to @xmath217 lying between the zero modes at @xmath211 and @xmath218 . all modes of deformations with @xmath219 contribute a positive energy . the first excited state with @xmath29 has two modes of decay , of equal energy , into the ground state . the dominant mode of instability in higher energy states is not directly towards the ground state involving a cascade of instabilities . let us now examine the shape adopted by a loop of finite @xmath9 confined by the sphere . equation ( [ ceelsphere ] ) can be integrated in terms of elliptic functions to give @xmath69 as a function of @xmath11 @xcite @xmath220\ , , \qquad \kappa_0 = 2 \sqrt{m } q\,.\ ] ] the function @xmath221 $ ] is the jacobi elliptic cosine @xcite . the angular wavenumber @xmath222 is given in terms of the constant @xmath223 defined below eq . ( [ eq : vdef ] ) by @xmath224 ; the modulus @xmath225 is defined by @xmath226 the curvature depends on the two parameters @xmath126 and @xmath114 through the parameters @xmath222 and @xmath225 . and @xmath114 are inverted to give @xmath227 the modulus @xmath225 will lie in the interval @xmath228 $ ] ; this bounds @xmath126 in terms of @xmath222 : @xmath229 , changing sign when @xmath230 . ] these parameters will be determined explicitly in terms of @xmath9 and @xmath6 using the boundary conditions associated with the closure of the loop . using the fact that the period of @xmath231 is given by @xmath232 $ ] , where @xmath233 $ ] is the complete elliptic integral of the first kind @xcite , it is possible to cast the boundary condition on @xmath69 , given by eq . ( [ eq : c1 ] ) , in the form @xmath234/(2\pi r)\,.\ ] ] integration of eq . ( [ eq : phi ] ) gives @xmath235,\ , m\right ] - s\right)\,,\ ] ] where @xmath236,m]$ ] is the incomplete elliptic integral of the third kind and @xmath237 $ ] is the jacobi amplitude @xcite . closure of the loop after @xmath10 circuits of the polar axis , @xmath238 , then reads @xmath239 \left(q^2 -1\right ) } \ , \pi \left[-\frac{4 m q^2}{\left(q^2 - 1\right)^2},\ , m\right ] - 1\right ) = 2\pi p\,.\ ] ] modulo eq . ( [ eq : bck ] ) , eq . ( [ eq : varphism ] ) determines @xmath225 implicitly as a function of @xmath9 . this completes the formal construction of the confined loops . the two - fold state with @xmath4 is illustrated in fig . [ figure1 ] for values of @xmath9 in the interval @xmath19 $ ] . when @xmath174 is small , it consists of a small oscillation about a geodesic circle ( see fig . 1(a ) ) consistent with the perturbative description . as @xmath9 is increased the oscillations about this circle initially increase in amplitude as more of the surface of the sphere is explored ( figs . 1(b ) and 1(c ) ) . in fig . 1 ( c ) overhangs with @xmath148 occur on the increasingly crowded loop . at a critical value @xmath240 , the loop makes self - contacts at the two poles ( fig . there will be a critical value @xmath241 where this occurs , whose magnitude increases with @xmath6 . as pointed out in @xcite , this value saturates . beyond some critical loop length , all equilibrium states are self - intersecting . ] for values of @xmath9 above @xmath242 , one needs to look more carefully at the boundary conditions . if , as we assume , the microscopic physics accommodates self - intersections on the surface and they occur without costing energy , the mathematical curve described by the elliptic function continues to represent the physical loop . there is a qualitative change in the behavior as @xmath9 is increased through @xmath242 . on crossing the poles , the loop makes two additional revolutions about the polar axis so that @xmath5 must be replaced by @xmath243 in the boundary condition eq . ( [ eq : c2 ] ) . to see this , note that the loop will cross the poles four times , twice at each pole . a discontinuity of @xmath244 is introduced in @xmath144 at each crossing . these discontinuities contribute to its period ; thus , the period of @xmath245 describing a single revolution gets replaced by @xmath246 . the orientation is also reversed .- fold , the corresponding number of revolutions is @xmath247 . ] as fig . [ figure1 ] illustrates , this discontinuity marks a transition from oscillatory to orbital behavior . as @xmath9 is increased further the two regions bounded by the self - intersections grow ( fig . 1(e ) ) . when @xmath248 the loop degenerates into a triple cover of a geodesic circle ( fig . 1(f ) ) . each equilibrium state can be represented as a point on the parameter space @xmath249 . as @xmath9 increases , the state will follow a trajectory @xmath250 labeled by the two integers @xmath6 and @xmath10 . , eq . ( [ eq : c2 ] ) , does not involve @xmath9 explicitly . it thus identifies the trajectories on the @xmath251 parameter space without locating the position along these trajectories . the latter is provided by eq . ( [ eq : c1 ] ) . ] to facilitate the interpretation of trajectories in parameter space , it is useful to first locate relevant geometrical landmarks . various significant parameter curves are represented in fig . [ figure2 ] . \(1 ) the parabola @xmath252 given by @xmath253 ( saturating the inequality ( [ eq : parabola ] ) ) provides a boundary on parameter space . vs @xmath126 ; hence `` parabola '' . ] points on the boundary describe geodesic loops , with @xmath82 and @xmath132 , covering the equator an integer number of times . points below the parabola do not describe confined loops . \(2 ) points on the straight line l , given by @xmath254 , describe loops that pass through the poles . points above this line represent loops that display oscillatory behavior ; those below it display orbital behavior . \(3 ) overhangs occur to the left of the vertical line , @xmath255 . the solid curve , @xmath256 in fig . [ figure2 ] represents the trajectory of the state with @xmath4 and @xmath5 as @xmath9 is increased in the interval @xmath19 $ ] . the initial state with @xmath257a single geodesic circle is represented by the point @xmath258 lying on the boundary parabola @xmath252 . in this interval , both @xmath126 and @xmath114 decrease monotonically with @xmath9 . the trajectory terminates in the point @xmath259 on the boundary parabola , where the loop degenerates into a triple covering of a geodesic circle . is approximated accurately by the straight line , @xmath260 , with a slope @xmath261 , predicted in perturbation theory . perturbation theory thus provides an accurate description of the problem outside its expected range of validity . ] more generally , whenever @xmath262 , where @xmath10 is an integer , loops may form @xmath10-fold coverings of a geodesic circle . as @xmath9 is raised above @xmath262 , it is possible to examine the state perturbatively in a manner analogous to that of small loops . suppose that the loop oscillates with an @xmath6-fold symmetry , so that @xmath263 . consistency with @xmath264 places a lower bound on @xmath6 : @xmath265 . thus one must have @xmath266 . ) for @xmath10 revolutions implies @xmath267 ] the state with @xmath268-fold dihedral symmetry will minimize the energy among these states . it is thus identified as the ground state . in particular , when @xmath9 passes through @xmath248 , the symmetry of the ground state changes from @xmath4 to @xmath24 ( see fig . [ figure3 ] ) . it is straightforward to determine the values of @xmath126 and @xmath114 in the ground state with @xmath269 as @xmath9 passes through @xmath262 : @xmath270 , and @xmath271 , representing a point lying on the boundary parabola @xmath252 . as @xmath9 is increased the amplitude of oscillation increases . at some point the loop will ( re)cross the poles where the number of revolutions changes by two . as @xmath272 is approached , the loop morphs continuously into a @xmath273 cover of a geodesic circle . in this region , the ground state can also be treated perturbatively . now @xmath274 . the corresponding end point in parameter space as @xmath9 approaches @xmath272 is given by @xmath275 , @xmath276 , which also lies on the boundary parabola @xmath252 . the trajectory in the parameter plane representing the @xmath4 , @xmath5 ground state loop and its minimum energy descendants is illustrated in fig . [ figure2 ] . it consists of a number of disconnected curve segments , @xmath277 , @xmath278 with end points @xmath279 , and @xmath280 lying on the boundary parabola @xmath252 . the points @xmath281 and @xmath282 describe loops with @xmath262 as it is approached from below and above respectively . there are discontinuities in @xmath126 and @xmath283 at @xmath284 and the loop passes through a @xmath10-fold covering of a geodesic circle . these discontinuities are identified analytically using perturbation theory . their origin is the transition from orbital back to oscillatory behavior with a change from @xmath285-fold to @xmath268-fold symmetry . as @xmath286 , @xmath287 so that @xmath288 and @xmath289 . + , @xmath243 for @xmath9 in the interval @xmath290 $ ] : ( a ) increasingly deformed triple covering of the equator ; ( b ) pole crossing , four points at a time ) ; ( c ) state making five orbits of sphere before it becomes a quintuple covering of equator . as in fig . 1 colors represent the normalized local confining force . , title="fig : " ] in fig . [ figure4 ] @xmath126 and @xmath283 are plotted as functions of @xmath9 . the @xmath6-fold symmetry states with @xmath5 and their descendants do not exhaust all possible self - intersecting states . if @xmath291 , a set of states consisting of a double covering of a geodesic circle ( @xmath26 ) comes into existence with symmetry @xmath292 . the sequence @xmath293 with @xmath29 is illustrated in fig . [ figure5 ] for values of @xmath9 in the interval @xmath294 $ ] . as @xmath9 is increased , the state with @xmath29 has least energy . as before , one can show that it is the only stable member among these states . it also has a lower energy than its counterpart with @xmath4 , and @xmath5 illustrated in fig . [ figure1 ] . the trajectory @xmath293 representing these states in parameter space as @xmath9 is varied in the interval @xmath294 $ ] , is represented in fig . [ figure2](b ) by a solid curve . when @xmath295 a transition occurs to a state with a five - fold symmetry . the sequence of trajectories @xmath296 is generated . their endpoints converge to the same point @xmath297 as those of the sequence @xmath298 + + in the interval @xmath294 $ ] : ( a),(b),(c ) increasingly deformed double covering of the equator ; ( d ) pole crossing , three points at a time ) ; ( e ) state making four orbits of sphere ; ( f ) quadruple covering of equator . as in fig . 1 colors represent the normalized local confining force.,title="fig : " ] the bending energy eq . ( [ htotsphere ] ) can be recast explicitly as a function of @xmath9 : @xmath299 \,\left({\cal e}[m]+(m-1){\cal k}[m]\right)+\pi r\,,\ ] ] where @xmath300 $ ] is the complete elliptic integral of the second kind @xcite , and @xmath225 is determined by solving eq . ( [ eq : varphism ] ) . @xmath38 is plotted as a function of @xmath9 for the ground state with @xmath4 and @xmath5 and its descendants in fig . [ figure6 ] . expanding expression ( [ eq : hstrong ] ) to second order in @xmath301 we reproduce eq.([eq : hweak ] ) . the normal energy , represented by the last term , grows linearly with @xmath9 and is state independent . in contrast the energy associated with geodesic curvature , represented by the first two terms in eq.([eq : hstrong ] ) , depends sensitively on the state . it is bounded and its maxima fall monotonically as the loop becomes large . it is simple to place a crude upper bound on this falloff : one has @xmath302 , so that the @xmath303 . thus in a large loop , the geodesic contribution to the energy is negligible compared to its state independent normal counterpart . the geodesic energy vanishes when @xmath304 , where the loop collapses to a multiple covering of a geodesic circle . the discontinuities in the derivative of @xmath38 with respect to @xmath9 at @xmath305 , are directly associated with the change of symmetry at these values of @xmath9 . in the intervals between consecutive values , @xmath38 increases linearly from zero and rises to a maximum value before falling to zero , again linearly . behavior is not symmetrical in these intervals . the slopes are different at the two ends and the maximum is not centered . while the maxima are not exactly periodic , they do migrate to the center of their respective intervals as @xmath9 increases . the existence of these maxima can be understood as a consequence of the incommensurability of the loop with geodesic behavior . the strongly asymmetrical initial maximum is associated with the development of overhangs on the loop before it crosses the poles . as the loop becomes longer this incommensurability plays a diminishing role . this qualitative behavior is repeated for the odd ground states @xmath29 , @xmath26 and its descendants , as well as the excited counterparts of these states . strikingly , the energy of any closed equilibrium loop tends to a common value in the limit , independent of the state , that completely dominated by the normal energy . the finite gap between the ground state and excited states state disappears . ground state and its descendants as a function of @xmath9 is represented by a solid curve . it is sum of geodesic and normal parts which are plotted separately as a dashed curve and a dotted line . the normal energy grows linearly and dominates in large loops . by contrast , the geodesic energy is bounded , vanishing at odd integers where its derivative is discontinuous . its maxima between zeros decrease inversely with @xmath9 . ] if @xmath306 is not small , the stability analysis is rather more complicated than the one presented in the context of weak confinement . the results of ref . @xcite in another context implies that , although the details differ , the conclusions for weak confinement continue to hold for values of @xmath9 before the onset of self - intersection and only the state @xmath2 , @xmath5 are excited . a detailed analysis of stability has yet to be performed beyond this point . one would , however , expect that the instabilities persist . the stability of the @xmath4 , @xmath5 ground state and its descendants needs to be reassessed when @xmath307 . for now one has to accommodate the existence of a set of states described by @xmath308 @xmath26 and their descendants . in particular , within a finite interval of values of @xmath9 above @xmath291 , the state @xmath29 , @xmath26 , represented in fig . [ figure5 ] has lower energy than the state with @xmath4 and @xmath5 represented in figure [ figure1 ] ) . [ figure7 ] . the difference in energy is associated with the significant geodesic curvature of the latter when @xmath309 . its descendants with @xmath310 also will have lower energy than their even counterparts when @xmath311 . there are , of course , no smooth deformations taking one from a loop with @xmath5 to one with @xmath26 that remain attached to the sphere . a hairpin costing an infinite bending energy will necessarily always form . however , it is possible to sidestep this topological obstruction by permitting the loop to detach into the interior . while a rigorous stability analysis is beyond the scope of this paper , we will argue that a finite energy barrier will always separate the two surface - bound states . this involves the construction of a natural homotopy connecting these states . let @xmath312 where we represent by @xmath313 the initial state with @xmath314 and by @xmath315 the final state with @xmath316 . these interpolating non - equilibrium states will also be confined within the sphere due to the convexity of the latter . the two boundary states have @xmath317 . and @xmath315 appropriately at intermediate values of @xmath318 . ] a rotation @xmath319 of the final state about the axis of symmetry is introduced in order to initialize the pointwise sum so as to avoid the development of cusps . intermediates in this homotopic sequence are illustrated in fig . [ figure8 ] with the constant choice @xmath320 . the bending energy is plotted in fig . [ figure9 ] as a function of @xmath318 . it exhibits a finite potential barrier separating the two states . while this does not prove that the @xmath4 state is stable in this regime , it does suggest that it is . in the neighborhood of @xmath291 ( solid and dashed lines respectively ) . the geodesic energy , and thus the total energy , of the three - fold is lower than that of the two - fold in a finite interval of @xmath9 beginning at @xmath291 . ] + one can not rule out the existence of semiattached equilibrium states of the loop . we were , however , unable to construct any such states . if they do exist , one would expect them to be unstable . ( [ eq : kappagab ] ) , the local force per unit length transmitted to the sphere is given explicitly as the following function of @xmath11 , @xmath321 where @xmath322 = 1-m \,\text{sn}^2 \,[u , m]$ ] @xcite . whereas the transmitted force depends on the local value of @xmath69 , both @xmath69 and @xmath126 depend on the boundary conditions associated with closure . thus @xmath20 does depend on the global loop geometry . note that the expansion of the expression in ( [ eq : fstrong ] ) to second order in @xmath301 reproduces the result for weak confinement ( eq . ( [ eq : fweak ] ) ) . in fig . [ figure10 ] the maximum and minimum values of @xmath20 have been plotted as a function of @xmath9 for the two - fold `` ground '' state and its descendants . it is positive everywhere with the minimum given by @xmath126 . one observes that @xmath20 does not behave in a monotonic way , suffering a positive discontinuity whenever @xmath262 , where @xmath10 is an odd integer . this is associated with the transition from orbital to oscillatory behavior at these values and is the analog of the euler instability described in the discussion of a weakly confined loop . if the unstretchability constraint is relaxed the discontinuity will be smoothed but the jump will persist . within the individual intervals between discontinuities both the maximum and minimum values decrease monotonically with @xmath9 ; this is associated with the reduction in the force once buckling into a state of oscillation has occurred . curiously , whereas the maxima within these intervals decrease with @xmath9 , the corresponding minima increase . this occurs in such a way that the mean of their values increases monotonically . both maxima and minima approach the value @xmath323 asymptotically . as a function of @xmath174 in the two - fold ground state with @xmath4 , @xmath5 , @xmath20 in the two - fold ground state and its minimum energy descendants . ] recall that the transmitted force is bounded from below by @xmath126 ( see eq.([lambdasphere ] ) ) . thus , if @xmath324 , then @xmath20 is also ; the confined loop will then push on the sphere everywhere . if , however , @xmath325 then @xmath20 may change sign along the curve . while @xmath126 is positive in the ground state and its descendants , it may become negative in the excited states of sufficiently long loops . in particular , we find that all states with @xmath326 and @xmath5 exhibit regions in which @xmath20 turns negative at values of @xmath9 below the onset of self - contact . when @xmath327 and @xmath5 , this occurs when @xmath328 ( see fig [ figure11](a ) ) . such states are , however , unstable . + , @xmath5 with @xmath329 . above this value of @xmath9 , @xmath20 may be negative in places . ( b ) the state @xmath4 , @xmath5 with @xmath330 inhabits a single hemisphere . these orbital states always have intervals where @xmath20 is negative.,title="fig : " ] , @xmath5 with @xmath329 . above this value of @xmath9 , @xmath20 may be negative in places . ( b ) the state @xmath4 , @xmath5 with @xmath330 inhabits a single hemisphere . these orbital states always have intervals where @xmath20 is negative.,title="fig : " ] it may appear counterintuitive that @xmath20 can be negative . it has to be remembered , however , that while @xmath20 depends only on the local geometry , this geometry is determined by the global behavior of the loop through the boundary conditions . if one cuts a loop somewhere , the state will immediately relax into a geodesic state with constant positive @xmath20 everywhere . the existence of regions along the loop where @xmath331 does not necessarily signal a tendency to detach . this will depend on details of the energy landscape . the states examined here are those consistent with the bound eq . ( [ eq : parabola ] ) . as was pointed out , states that violate this bound are always unstable with respect to the unbinding of the loop into the interior . such a state is illustrated in fig . [ figure11](b ) , making two self - intersecting orbits in the northern hemisphere . the curvature is high within the two orbits so that this state is evidently an excited state of the loop . the transmitted force @xmath20 is negative everywhere below the circle with @xmath332 ( see fig . [ figure11](b ) ) . this suggests that the likely mode of instability will involve the loop unbinding along regions where @xmath20 is negative , allowing the loop to relax toward the ground state @xmath4 , @xmath5 through the unwinding of the high curvature orbits . intuitively , one might have expected the total force transmitted to the sphere , @xmath333 to increase with loop size . this is not generally the case , except in the limit of large loops , where @xmath334 increases linearly with @xmath9 , a consequence of the asymptotic behavior of @xmath126 . notice also that @xmath334 is not the same as the energy @xmath38 ( divided by the radius of the sphere ) for finite values of @xmath9 . one has @xmath335 . in the limit , however , the two do coincide . @xmath334 is plotted as a function of @xmath9 in fig . [ figure12 ] . vs. @xmath9 ( solid line ) . the dashed line indicates the contribution from geodesic curvature @xmath336 . the dotted line is the contribution @xmath337 . ] the confinement of a closed semiflexible loop by a surface has been examined . the constraint on the spatial degrees of freedom of the loop is enforced in the variational principle using the method of lagrange multipliers . the loss of euclidean invariance of the energy of the unconstrained loop under confinement is quantified by the lagrange multiplier which , in equilibrium , gets identified as the force transmitted to the surface by the confined loop . we have focused on a description of the ground state of a confined loop . if the loop is short this consists of an oscillation about a geodesic circle , exhibiting a two - fold dihedral symmetry . this is the only stable state of the loop in this regime . the description of the ground state and its excitations gets more complicated when the loop length is increased . when @xmath9 is increased above @xmath28 , a new set of states comes into existence that oscillates about a double covering of a geodesic circle . among these the state with three - fold symmetry has lowest energy ; within a finite band of values of @xmath9 , this state also replaces the two - fold as the ground state . the two states remain separated by an energy barrier so that the two - fold remains stable . as the loop size increases these states get replaced by descendants with higher @xmath6-fold symmetries which alternate as ground states . both the energy and the transmitted force suffer discontinuities associated with changes of symmetry . they do not generally increase monotonically with loop size except asymptotically . in this limit the energy gap between the ground state and excited states disappears . our examination of the confinement of a loop within a spherical cavity might lead one to expect that the equilibrium states of a confined loop will always attach . however , other confining geometries display very different behavior . consider , for example , the confinement of a loop by a cylinder of smaller radius . analogs of the @xmath6-folds exist but , in general , they do not provide the ground state as one can easily verify by playing with metal rings in a wastepaper basket . the details are surprisingly complicated . in this context , it would be interesting to understand how the spherical equilibrium states `` evolve '' under elliptical deformations of the sphere , or what effect surface irregularities or pores have on the confinement process . when contact with the confining surface is incomplete , one needs to address the issue of boundary conditions at points of contact . if the contact is due entirely to geometric hindrance , the boundary conditions at isolated points of contact are simple : the tangent vector to the curve is tangent to the surface at these points . if contact extends over a finite region , however , in addition to the tangent vectors the curvatures will also be continuous at the boundary of the region of contact . there may also be a tendency to either promote or inhibit adhesion to the confining surface . a simple way to accommodate such interactions is to introduce a local contact energy , proportional to the length of the contact interval . the boundary conditions will reflect this additional energy . while the curvature will suffer a discontinuity analogous to the discontinuity associated with the contact line bounding the region of contact between a fluid membrane and an attractive substrate @xcite , this is not the full story . the normals may rotate about the tangent vector , aligning themselves with some preferred direction on the confining geometry . the extension of our framework to accommodate contact energies will be presented elsewhere . in this paper the role played by the confining surface has been passive : we have not considered the possibility that it may deform in reaction to the presence of the confined polymer . surface deformations due to membrane bound polymers may also play a role in shaping the morphologies of biological membranes . the forces transmitted to the membrane will now provide a source for the surface stress . understanding this coupling poses technical challenges . what is clear is that there will be interesting physics associated with the competing elastic energies @xcite . we have benefited from conversations with chryssomalis chryssomalakos , markus deserno , yair gutirrez fosado , eytan katzav , martin mller and eugene starostin . partial support from dgapa papiit grant in114510 is acknowledged . there are various derivations of eq . ( [ ceefsphere ] ) . the approach adopted here involves treating the tangent vector to the curve @xmath338 as an independent variable . if the curve is also parametrized by arc - length , @xmath339 is a unit vector , @xmath340 . thus , consider the energy functional ( @xmath341 ) , @xmath342 = \int ds \ , \left(\frac{1}{2}\ , { \bf t}'{}^2 - \frac{1}{2}\lambda ( { \bf t}^2 -1 ) + { \bf f}\cdot ( { \bf t}- { \bf y}')\right)\,;\ ] ] the presence of the two constraints liberates @xmath339 to be varied independently . now the variation of @xmath343 with respect to@xmath344 is given by @xmath345 whereas its counterpart with respect to @xmath339 is @xmath346 modulo the frenet - serret equations for the curve , @xmath347 in eq . ( [ eq : ft ] ) implies @xmath348 for an isolated elastic curve , @xmath349 . if the curve is constrained to lie on a surface , it was seen in sec . [ confinedelastica ] that @xmath350 . thus the tangential equation @xmath351 continues to hold . this equation can be integrated to determine the lagrange multiplier @xmath334 associated with the parametrization by arclength @xmath352 where @xmath46 is a constant of integration . equation ( [ eefsforce ] ) follows . with respect to the darboux frame , @xmath103 is given by @xmath353 where we have used the relations ( [ dkappagn ] ) derived below . the structure equations ( analogous to the frenet - serret equations ) describing the darboux frame are given by @xmath354 the normal curvature , the geodesic curvature , and torsion are given , respectively , by @xmath355 here @xmath356 and @xmath357 are the components of the vectors @xmath339 and @xmath198 with respect to a basis of surface tangent vectors adapted to the parametrization , @xmath50 , @xmath51 : @xmath358 , @xmath359 . @xmath360 is the extrinsic curvature tensor defined on @xmath114 and @xmath361 is the covariant derivative compatible with the induced surface metric @xmath362 . whereas @xmath70 and @xmath363 depend on the extrinsic curvature , @xmath69 is defined intrinsically ; it depends only on the surface metric @xmath364 . the frenet curvatures are related to their darboux counterparts by @xmath365 so that @xmath366 . thus , the frenet curvature can be decomposed into its intrinsic and extrinsic parts . the frenet torsion @xmath45 is the sum of the derivative of the angle connecting both frames and the geodesic torsion . note that @xmath363 involves two derivatives , whereas @xmath45 involves three . the extra derivative is associated with @xmath367 , the rotation rate of one frame with respect to the other . the identities eqs . ( [ kaptau ] ) imply the following : @xmath368 these relations are used in sec . [ confinedelastica ] to express the euler - lagrange derivatives for the curve in terms of @xmath69 , @xmath70 , and @xmath363 and their derivatives . we first derive the relationship between arc - length and the angle @xmath144 correct to second order . one has @xmath369^{1/2}\,.\ ] ] using eq . ( [ eq : theta ] ) and the harmonic approximation for @xmath69 given by eq . ( [ linear ] ) , one obtains @xmath370 thus , in this approximation , @xmath371 this implies that vanishes when @xmath372 , a solution which is identified as a trivial rotation of the equatorial loop about an axis in the equatorial plane . ] @xmath373 we now derive eqs . ( [ eq : delr1 ] ) and ( [ eq : delr2 ] ) . to do this one needs to examine the boundary conditions ( [ eq : c1 ] ) and ( [ eq : c2 ] ) correct to second order in @xmath374 . at this order , the turning point @xmath167 of the potential @xmath121 , defined by eq . ( [ eq : vdef ] ) , coincides with @xmath173 given in the harmonic approximation by eq.([eq : k1sm ] ) . thus , the roots in the factorization of @xmath375 given in eq . ( [ eq : turningpoints ] ) , where @xmath223 is is the constant defined below eq . ( [ eq : vdef ] ) are @xmath376 one thus has , correct to second order , @xmath377 equation ( [ eq : c1 ] ) then implies that @xmath378 furthermore , to second order , @xmath379 equation ( [ eq : c2 ] ) then implies that for @xmath183 , @xmath380 d. a. singer , in proceedings of curvature and variational modelling in physics and biophysics , edited by o. j. garay , e. garca - ro and r. vzquez - lorenzo , ( american institute of physics , college park , md , 2008 ) .

we describe the first results from a six - month long reverberation - mapping experiment in the ultraviolet based on 171 observations of the seyfert 1 galaxy ngc 5548 with the cosmic origins spectrograph on the _ hubble space telescope_. significant correlated variability is found in the continuum and broad emission lines , with amplitudes ranging from% to a factor of two in the emission lines and a factor of three in the continuum . the variations of all the strong emission lines lag behind those of the continuum , with heii lagging behind the continuum bydays and ly , civ , and siiv lagging bydays . the relationship between the continuum and emission lines is complex . in particular , during the second half of the campaign , all emission - line lags increased by a factor of 1.32 and differences appear in the detailed structure of the continuum and emission - line light curves . velocity - resolved cross - correlation analysis shows coherent structure in lag versus line - of - sight velocity for the emission lines ; the high - velocity wings of civrespond to continuum variations more rapidly than the line core , probably indicating higher velocity blr clouds at smaller distances from the central engine . the velocity - dependent response of ly , however , is more complex and will require further analysis .

1501.05954

one of the most prominent characteristics of the ultraviolet ( uv ) , optical , and near - infrared ( nir ) spectra of active galactic nuclei ( agns ) is the presence of broad emission lines . while we know that these features arise on scales not much larger than the accretion disk , their physical nature remains one of the major unsolved mysteries in agn astrophysics . a particularly important feature of the broad emission lines is that they are , by definition , resolved in line - of - sight ( los ) velocity , and their large widths leave little doubt that the primary broadening mechanism is differential doppler shifts due to the motion of individual gas clouds , filaments , or more - or - less continuous flows around the central black hole . however , it is not possible to establish the broad - line region ( blr ) kinematics simply by inverting the line profiles because this inverse problem is degenerate , with a wide variety of simple velocity models providing satisfactory fits ( e.g. , * ? ? ? * ) . the existing evidence on the blr kinematics is ambiguous : some of this gas may flow inward , helping to feed the central black hole . extended , flattened , rotating disk - like structures seem to be important in at least some blrs , as shown statistically for radio - loud agns @xcite , by the pronounced double - peaked profiles observed in some sources ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and from spectropolarimetry @xcite . there is evidence of the importance of the black hole gravity in dominating the motion of the blr gas @xcite , although radiation pressure may also play a role @xcite . on the other hand , developments over the last two decades re - open the interesting possibility that much of the emitting blr gas is due to outflowing winds ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , perhaps connected to the outflows detected in absorption features ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? ? * ; * ? ? ? * ) , whose kinematics and energetics are also poorly understood . the unknown dynamics of the blr gas represents a serious gap in our understanding of agns and in the calibrations needed for the study of black - hole / host - galaxy co - evolution up to very high redshifts . there have been many attempts to model the physics of the blr . in general , photoionization equilibrium models can reproduce the line intensities , but self - consistent models that provide simultaneous solutions to the line intensities , profiles , and variability are lacking . the locally optimally emitting cloud model @xcite and the stratified cloud model @xcite explain most observed line intensities and some of the observed time lags between the continuum and emission lines . however , they lack the important kinematic ingredients required to explain the observed line profiles . in order to understand the structure and kinematics of the blr , we must break the degeneracy that comes from the study of the line profiles alone . we can do this by using reverberation mapping ( rm ) to determine how gas at various los velocities responds to continuum variations as a function of light travel - time delay @xcite . over the last quarter century , the rm technique has become a standard tool for investigating the blr . in its simplest form , rm is used to determine the mean time delay between continuum and emission - line variations , typically by cross - correlation of the respective light curves . it is assumed that this represents the mean light - travel time across the blr . by combining this with the emission - line width , which is assumed to reflect the velocity dispersion of gas whose motions are dominated by the mass of the central black hole , the black hole mass can be estimated . rm in this form has been used to measure the black hole masses in over 50 agns ( for a recent compilation , see * ? ? ? * ) to a typical accuracy of @xmath8dex . important findings that have arisen from these rm studies include the following : 1 . in a given agn , emission lines that are characteristic of higher - ionization gas respond more rapidly to continuum flux variations than those characteristic of lower - ionization gas , indicating ionization stratification within the blr @xcite . 2 . there is an inverse correlation between the time delay , or lag @xmath9 , for a particular emission line and the doppler width @xmath10 of that emission line . the relationship for a given agn is consistent with the virial prediction @xmath11 @xcite . without this relationship , rm masses would be highly dubious . there is an empirical relationship between the agn luminosity @xmath12 and the radius of the blr @xmath13 ( hereafter the @xmath13@xmath12 relationship ) that is well - established only for the h@xmath14 emission line @xcite . limited data on civ@xmath15 indicates a similar relationship applies to that line as well @xcite . the existence of @xmath13@xmath12 relationships for both low - ionization and high - ionization lines has been independently confirmed by gravitational microlensing observations @xcite . the @xmath13@xmath12 relationship is of particular interest as it allows estimation of the central black - hole mass based on a single spectrum from which the line width is measured and the blr radius is inferred from the agn luminosity . this neatly bypasses the need for a direct rm measurement of the emission - line time lag . rm is necessarily resource intensive : even to determine the mean time delay for an emission line typically requires some 3050 well - spaced high - quality spectrophotometric observations or a good measure of luck for fewer observations . the @xmath13@xmath12 relationship is very important as the rm - based mass determinations anchor empirical scaling relationships ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) that are used to estimate the masses of quasar black holes in large numbers ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? virtually all quasar mass estimates and their astrophysical uses are tied to rm . measurement of the mean lag and line width for a given emission line provides important , though limited , information about the blr and the central mass of the agn . we are only now beginning to realize the full power of rm through velocity - resolved investigations of the blr response . the first generation of successful rm programs provided sufficient understanding of agn variability and blr response times to design programs that could effectively extract velocity dependent information that would lead to an understanding of the structure and kinematics of the blr through recovery of `` velocity delay '' maps from rm data @xcite . the relationship between the continuum variations @xmath16 and velocity - resolved emission - line variations @xmath17 is usually described as @xmath18 where @xmath19 is the `` response function , '' or velocity delay map @xcite . as can be seen by inspection , @xmath19 is simply the observed emission - line response to a delta - function continuum outburst . the velocity delay map is simply the blr geometry and kinematics projected into the two observable quantities of los velocity and time delay relative to the continuum . this linearized echo model is justified by the fact that the continuum and emission - line variations are generally quite small ( 1020% ) on reverberation time scales ( see also * ? ? ? the technical goal of a reverberation program such as the one described here is to recover the velocity delay map @xmath19 from the data and thus infer the geometry and kinematics of the blr . time - resolved velocity delay maps have now been obtained for a handful of agns ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but only for optical lines ( the balmer lines , hei@xmath20 , and heii@xmath21 ) . in general , these suggest flattened geometries at small to modest inclinations and some combination of virialized motion and infall . an outflow signature has been observed in only one case , ngc3227 @xcite . the lack of velocity delay maps for uv lines , on the other hand , leaves us with a very incomplete understanding of the blr . it is , in fact , the high - ionization level uv resonance lines ( e.g. , civ@xmath15 , siiv@xmath6 , ly@xmath3@xmath4 ) that might be expected to dominate any outflowing component of the blr . the optical lines , in contrast , generally seem to arise in disk - like structures with infall components ( e.g. , * ? ? ? rm studies in the uv have been limited . several observing campaigns were undertaken with the _ international ultraviolet explorer ( iue ) _ or _ hubble space telescope ( hst ) _ or both on ( i ) ngc5548 @xcite , ( ii ) ngc3783 @xcite , ( iii ) fairall9 @xcite , ( iv ) 3c390.3 @xcite , ( v ) ngc 7469 @xcite , ( vi ) ngc4151 @xcite , ( vii ) akn564 @xcite , and ( viii ) ngc4395 @xcite . with the exception of akn564 , which showed essentially no emission - line variability over a comparatively short campaign , all of these programs yielded emission - line lags , but only limited information about the detailed response of the uv emission lines ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the existing velocity - delay map for ngc4151 shows some incipient structure in civ@xmath15 and heii@xmath1 and a general shape that seems to be consistent with a virialized blr @xcite . given the importance of the uv emission lines in the photoionization equilibrium of the blr gas and the probable differences between the geometry and kinematics of the high and low - ionization gas in the blr , we have undertaken a large rm program in the uv using the cosmic origins spectrograph ( cos ; @xcite ) on _ hst _ ( _ hst _ program go-13330 ) , the agn space telescope and optical reverberation mapping ( agn storm ) project , in the first half of 2014 . the program was designed with certain specific goals in mind : 1 . determine the structure and kinematics of the high - ionization blr through observations of the variations in the civ@xmath15 , ly@xmath3@xmath4 , nv@xmath22 , siiv@xmath6 , and heii@xmath1 emission lines . , siiv@xmath23 , and civ@xmath24 . moreover , the heii feature is blended with oiii]@xmath25 , 1665 and siiv is blended with the quintuplet oiv]@xmath26 , 1399.8 , 1401.2 , 1404.8 , 1407.4 , where the second , third , and fifth transitions dominate . ] carry out simultaneous ground - based observations of ( a ) the high - ionization optical line heii@xmath21 for direct comparison with heii@xmath27 and ( b ) the balmer lines , particularly h@xmath14@xmath28 , to determine the structure and kinematics of the low - ionization blr . although the optical spectrum is extremely well - studied ( * ? ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) , simultaneous observations are necessary , as the dynamical timescale for the blr in ngc5548 is only a few years . 3 . compare in detail the continuum variations in the uv ( at @xmath29 ) with those at other wavelengths ( see * ? ? ? * hereafter paper ii ) and infer the structure of the continuum - emitting region . the motivation for the uv / optical continuum comparison is multifold : 1 . delays between continuum variations at longer versus shorter wavelengths have been detected or hinted at in a number of sources ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? such delays can provide insight into the structure , geometry , and physics of the continuum - emitting region . 2 . velocity delay maps recovered using the uv continuum as the driving light curve ( equation [ eq : trans ] ) are expected to be of higher fidelity than those obtained from the optical continuum because the observable uv is closer in wavelength to the ionizing continuum ( @xmath30 ) that powers the emission lines . the optical continuum is not only a slightly time - delayed version of the uv continuum , but it seems smoothed somewhat as well @xcite , which might make it difficult to recover detailed structure in the velocity delay maps . our _ hst _ program afforded a valuable opportunity for exploring agn behavior at high time resolution for an extended period at wavelengths beyond those covered by our _ hst _ cos spectra . the _ hst_program is the anchor of a much broader agn storm project to address broader issues through observations across the electromagnetic spectrum . this paper serves as the first in a series . of special interest is the possibility of using short - timescale lags between variations in different continuum bands to map the temperature structure of the accretion disk . the _ swift _ satellite @xcite is especially suitable for such a study because of its broad wavelength coverage ( hard x - ray through @xmath31-band ) and ability to execute high - cadence observations over an extended period of time . in paper ii , we present the results of a four - month program of high - cadence ( approximately twice per day ) multiwavelength observations with _ swift_. additional papers in this series will describe high - cadence ground - based photometry from the near uv through the nir . we will also present results from a program of ground - based spectroscopy that is similar in cadence to the _ hst_cos observations , but covers a somewhat longer temporal baseline . other additional papers will present results on the variable absorption features and on our efforts to decipher the broad emission - line variations and determine the structure and geometry of the blr . in section 2 , we describe the observations and data processing , including a discussion of the program design and a complete description of how the standard data reduction pipeline was modified to meet our stringent calibration requirements . we describe our initial data analysis and results in section 3 , and in section 4 , we briefly discuss the first results from our program and place these results in the context of previous monitoring campaigns on ngc 5548 . when necessary , we assume a @xmath32cdm cosmology with @xmath33 , @xmath34 , and @xmath35 @xcite . rm is a resource - intensive activity that requires obtaining high signal - to - noise ratio ( @xmath36 ) homogeneous spectra at sufficiently high spectral resolution to resolve the gross kinematics of the blr . spectra must be obtained at a high cadence over a temporal baseline that is longer than the typical variability timescale of the agn . given the inherent risks of rm programs due to the unpredictability of agn variability , it is essential that our experimental design assures a successful outcome , yet is as economical with observing time as possible . the first consideration is that each epoch of observation should require no more than one _ hst _ orbit per `` visit '' which restricts the integration time per visit to @xmath37 minutes . this consideration limits us to relatively bright nearby seyfert 1 galaxies . cos is clearly the instrument of choice for such a project , as it is a very sensitive , high spectral resolution spectrometer . its native resolution ( @xmath38 ) is high enough to allow us to trade off resolution and @xmath36 in the data processing phase . in order to schedule the observatory efficiently , a cadence of one visit per day or longer is required . we therefore want to target an agn that has a civ - emitting region several light days in extent , and this requires a source with @xmath39 ( e.g. , * ? ? ? this led us immediately to select as a target the well - studied seyfert 1 galaxy ngc5548 ( @xmath40 ) . ngc5548 is probably the best - studied agn by rm , with historical optical spectroscopy extending as far back as the early 1970s @xcite . importantly , it has never been known to go into a `` dormant state , '' as observed recently in the case of mrk 590 @xcite , that would preclude a successful reverberation campaign and , historically , self - absorption in the uv resonance lines has been minimal @xcite , although strong absorption appeared in 2013 @xcite . the remaining adjustable parameter is the duration of the campaign . we investigated this using monte carlo simulations similar to those described by @xcite . using recent developments in statistically modeling agn light curves @xcite , we can make very robust models of the expected continuum behavior of ngc 5548 . quasar light curves are well - described by a stochastic process , the damped random walk . the process is described by an amplitude @xmath41 and a damping timescale @xmath42 , which for ngc 5548 in the optical are measured to be @xmath43 and @xmath44 days , respectively @xcite . we used these measured properties of ngc 5548 to simulate the continuum variations ; this is a conservative choice as the uv continuum can be expected to show both higher amplitude and shorter time - scale variations , both of which are an advantage . we then convolved the artificial light curves with model velocity delay maps for several lines to provide an artificial spectrum . as described by @xcite , we adopted a blr model with an extremely challenging velocity delay map for these simulations , a keplerian disk with a single two - armed spiral density wave . while this is unlikely to be the actual agn blr geometry , it provides a challenging test : if we can recover such a complex velocity delay map correctly , then we can certainly hope to recover others of comparable complexity and would have no difficulties with geometries like those that have been recovered for optical lines ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? we modeled the emissivity and response of each line realistically using a grid of photoionization equilibrium models @xcite . we sampled the artificial spectra to match our proposed observations , including noise . we then modeled the artificial spectra to recover the velocity delay maps using memecho @xcite . simulations based on characteristics of previous rm experiments yield velocity delay maps with noise levels similar to those obtained from the actual data , demonstrating the verisimilitude of our simulations ( see * ? ? ? * for examples ) . the goal of our simulations was to determine the minimum duration program that would allow us to recover a velocity delay map with a high probability of success . for cos - like observations ( in terms of @xmath36 per visit and spectral resolution ) , our initial simulations indicated that reliable velocity delay map recovery for a strong line ( e.g. , civ ) required between 130 to 200 days . a finer grid of models showed that in 10 of 10 simulations , a high - fidelity velocity delay map was recovered after 180 days , which was thus adopted as the program goal . a much longer program at this sampling rate would in any case be precluded by the accessibility of the target to _ hst_. because of the long duration of the proposed program , we also considered the possible impact of losses of data due to instrument or spacecraft safing events . short safing events occur frequently enough that we needed to assess their impact . based on the record for _ hst_and cos in cycles 1720 , there might be two spacecraft events that lose 23 days each and one cos event that loses 2 days over a stretch of 180 consecutive days . by repeating a subset of our simulations , we found that losses of such small numbers of observations would have no impact on our ability to recover the velocity delay maps . the simulations also allowed us to assess the impact of early termination of our experiment due to a major failure . if a program was terminated at @xmath45 days , the probability that the data would yield a useful ( but not a detailed ) velocity delay map would be @xmath46% . however , a program as short as 75 days would have a very low probability ( @xmath47% ) of success . the key to a successful rm campaign is that it must be long enough that favorable continuum variability characteristics become highly probable . that this is essentially guaranteed to happen during a 180-day experiment played a major role in selecting ngc 5548 as our target . observations were made in single - orbit _ hst _ cos visits approximately daily from 2014 february 1 through july 27 . of the 179 scheduled visits , 171 observations were executed successfully and 8 were lost to safing events or target acquisition failures ( very close to the expected number of losses ) . in each visit , we used the g130 m and g160 m gratings to observe the uv spectrum over the range 11531796 in four separate exposures . exposure times were selected to provide @xmath48 when measured over velocity bins of @xmath49 . during each visit , we obtained two 200-second exposures with g130 m centered at 1291 and 1327 and two 590-second exposures with g160 m centered at 1600 and 1623 . the cos far - ultraviolet detector is a windowless , crossed delay - line microchannel plate stack that is subject to long - term charge depletion . to extend the useful lifetime of the detector , we positioned the spectrum so that bright geocoronal airglow lines ( e.g. , ly@xmath50 ) and agn emission lines ( e.g. , redshifted ly@xmath3 ) would not always fall on the same area of the detector . first , we alternated the target acquisition between the g130m/1291 and the g130m/1327 configurations . the g130m/1327 configuration is then followed by a g130m/1327/fp - pos=3 exposure , and by a g130m/1291 exposure alternating among fp - pos=1 , 2 , and 4 . the g130m/1291 configuration is instead followed by a g130m/1291/fp - pos=3 exposure , and by a g130m/1327 exposure alternating among fp - pos=1 , 2 , and 4 . second , we alternated the fp - pos for the g160m/1623 exposure between fp - pos=1 and fp - pos=2 . we could not vary the settings for the g160m/1600 and g160m/1623 further because we needed to ensure the coverage of the entire wavelength range while keeping the detector gap from falling on the redshifted civ@xmath15 emission line . finally , we used four additional orbits to improve our understanding of the cos flux calibrations ( see section [ sec_data_reduction ] ) . during these additional visits , we observed two of the standard stars ( wd 0308565 and wd 1057 + 719 ) employed to obtain sensitivity functions @xcite at the same detector locations we used for the reverberation program . the observations were taken using all the instrument configurations employed in our primary observing program . we used the calcos pipeline v2.21 for the bulk of our data processing . the absolute flux calibration of the cos reduction pipeline is reported to be accurate to @xmath51 and the relative flux calibration is good to better than @xmath52 @xcite . we are primarily interested in the quality of the relative flux calibration as we are looking for very small - scale variations on short timescales ; we need the fluxes to be stable and repeatable across the spectrum . we found , however , that there were local variations in the precision of the fluxes that necessitated improvements . to produce a final dataset with a flux calibration that is everywhere precise at the 2% level , we refined the existing calibration reference files and applied a post - calcos pipeline to further process the data . the main areas of improvement include refinements to the dispersion solution , fixed - pattern noise mitigation , the sensitivity function , and the time dependent sensitivity ( tds ) functions , as outlined below . the final data product consists of one combined spectrum per grating per day . airglow emission lines ( oi@xmath53 and ni@xmath54 ) were filtered from the data by removing events detected when _ hst _ was in daylight . the spectra were further binned by 4 pixels in order to increase the @xmath36 per spectral element of the agn continuum . this binning still results in two binned pixels per cos resolution element . the cos wavelength solution has a quoted uncertainty of @xmath55kms@xmath56 @xcite . as the stis uncertainty is @xmath57kms@xmath56 @xcite , we refined the dispersion solutions for our cos dataset using previous observations of ngc 5548 taken with the stis e140m/1425 mode in 1998 ( pid 7572 , pi : kraemer ) . to accomplish this , we cross - correlated the line profiles of strong interstellar medium absorption features between each cos observation and the stis reference spectrum . we used 19 interstellar absorption features , ranging from siii@xmath58 at the short - wavelength end to alii@xmath59 at the long - wavelength end . a linear correction to the initial wavelength solution was then computed across each detector segment and applied directly to the extracted spectra . with this correction , measurements of the root mean square ( rms ) of the residual offsets decreased from @xmath55kms@xmath56 to @xmath60kms@xmath56 . the standard reference files used in the calcos pipeline correct for only the most prominent fixed - pattern noise features such as the quantum - efficiency gridwires , low - order response variations , and large geometric distortion artifacts @xcite . usually , users combine multiple fp - pos positions to smooth over the remaining features . however , this was not possible for our dataset , since only a single fp - pos setting was used for each central wavelength setting in each orbit ( see section [ sec_cos_observations ] ) . to correct these features to a higher degree we derived one - dimensional pixel - to - pixel flats ( `` p - flats '' ) . these flats were produced by combining normalized , high signal - to - noise ratio white dwarf spectra in detector space , following the method described by @xcite . the white dwarf spectra used were taken as part of the _ hst_/cal program 12806 ( pi : maasa ) and used the same detector locations as used for the ngc 5548 datasets . to test the effects of our p - flat correction , we combined the 171 spectra reduced both with and without the application of the p - flats . the _ s / n _ per pixel in 5 continuum regions increased from @xmath61 to @xmath62 for the g130 m grating , and from @xmath63 to @xmath64@xmath65 for the g160 m grating through the removal of small localized flux calibration errors by the p - flat correction . the improvement for the g130 m combined spectrum is less dramatic because ( a ) we rotated among the four fp - pos settings , and ( b ) the g130 m grating disperses more widely in the cross - dispersion direction , and thus intrinsically averages the fixed - pattern noise over a larger area of the detector . the cos flux calibration is done in two steps : ( a ) derivation of static sensitivity functions and ( b ) characterization of the time evolution of the sensitivity through the tds correction @xcite . thanks to the existing calibration program that monitors the tds variations ( pid 13520 ) , we had bi - monthly observations of the standard star wd 0308565 for 3 out of the 4 central wavelength settings we are using in our program ( g130m/1291 , g130m/1327 and g160m/1623 ) . standard star data were obtained in 2014 february , april , june , and august . by analyzing these calibration data , together with the data collected during our additional calibration orbits ( section [ sec_cos_observations ] ) , we verified that both the static and time - dependent response functions vary more with instrument configuration than currently modeled by the calcos pipeline . while calcos assumes that both the sensitivity function and the tds correction vary only as a function of wavelength , we were able to improve the relative flux calibrations and reach our required level of precision by ( a ) obtaining sensitivity functions individually for each configuration ( one function per wavelength setting per fp - pos per detector segment ) , and ( b ) computing the tds correction individually for each wavelength setting observed as part of the routine calibration program . we estimate the quality of the flux calibration by inspection of the fractional residuals @xmath66 of the calibrated standard star spectra and their respective calspec stellar model ( the same models employed by the standard pipeline reference file ) , @xmath67 where both @xmath68 and @xmath69 are binned over 1using a boxcar filter in order to increase the _ s / n _ per spectral element . while the visual inspection of the residuals as a function of wavelength allows us to identify and correct for local biases , we use the mean value of the distribution of the residuals as an indicator of a global bias in the calibration . the flux calibration uncertainty is an estimate of the limit of the stability of the flux calibration at a given time . however , since the overall instrument sensitivity evolves with time , and our final spectra are obtained from the combination of multiple settings for each grating , we conservatively define the fractional precision error @xmath70 for each grating as the maximum fractional uncertainty computed for any of the wavelength settings . the new sensitivity functions were derived from spectra of the standard star wd 0308565 for the g130 m settings , and of wd 1057 + 719 for the g160 m settings ( pid 12806 ) . while one individual sensitivity function per detector segment characterizes the full grating ( data from different settings are averaged together ) in calcos , we built one independent sensitivity function for each wavelength setting and fp - pos used in our program . by comparing the bi - monthly wd 0308565 data , we found that residuals with respect to the stellar models were greatly reduced if the tds corrections were computed individually for each of the wavelength settings ( g130m/1291 , g130m/1327 and g160m/1623 ) , instead of averaging the data over multiple modes . additional improvements were obtained by increasing the number of time intervals over which the tds trends are computed and by redefining the wavelength ranges used in the analysis . unfortunately , there are insufficient calibration data for the longest wavelengths in the g160 m spectra , so we were forced to truncate these spectra at 1750 . in spite of these improvements , the available data did not allow us to conduct any tests on the remaining setting ( g160m/1600 ) . this setting is particularly important for our scientific goals since it includes most of the broad civ emission line ( section [ section : introstorm ] ) . moreover , since all the calibration data for tds monitoring purposes are obtained only using fp - pos=3 , they did not allow us to test for any residual dependence of the tds correction on fp - pos configuration . these are the two reasons that motivated us to request further calibration data ( see section [ sec_cos_observations ] ) . these data for wd 0308565 and wd 1057 + 719 collected in 2014 september allowed us to derive an independent set of sensitivity functions . by comparing the new sensitivity functions with the originals , appropriately corrected for time evolution of the tds , we were able : 1 . to identify the best possible tds correction attainable for the g160m/1600 setting with the current tds calibration data . the current calcos tds correction for this configuration was obtained from combining both g160m/1577 and g160m/1623 data . although this correction is not ideal , it minimizes both the global bias and the flux calibration uncertainty when compared to the tds corrections obtained individually from either the g160m/1577 or the g160m/1623 settings . 2 . to verify that the tds correction does not vary strongly with fp - pos settings . while comparing the residuals with respect to the stellar models shows structure unique to each fp - pos setting , the level of the local biases is such that both the global bias and the flux calibration uncertainty can be considered stable for each wavelength setting ( the maximum deviation in the width of the residual distribution is @xmath71 ) . with the new flux calibration and tds characterization , our global biases are consistent with zero and the overall fractional precision is @xmath72% and @xmath73 , respectively , for the g130 m and g160 m settings , compared to @xmath74 for the standard pipeline . we intend to make these improvements available to other cos users . during the final week of the observing campaign , the operating high voltage ( hv ) for one of the cos detector segments was increased to combat the negative effects of `` gain - sag '' @xcite . while necessary to provide well - calibrated data , this hv change also has the effect of introducing small changes in the detector response . using wd 0308565 observations from our calibration orbits taken at the same hv as the rest of the campaign and contemporaneous tds monitoring observations of the same calibration target ( taken at the increased hv ) , we were able to estimate a hv bias correction for the g130 m grating from a direct comparison of the spectra . this bias was measured to be 1% without any detectable dependence on wavelength or cenwave setting . unfortunately , the same procedure could not be done for the g160 m grating as our additional orbits and the tds observations used different standard stars ( wd 0308565 instead of wd 1057 + 719 ) . instead , the bias estimate for this grating was obtained by analyzing the time evolution of the mean flux in overlapping regions of g130m/1327 at the lower hv setting and g160m/1600 at the higher hv setting . this analysis gave a plausible estimate of a 1% bias . however , with such a limited amount of data at the lower hv and a narrow overlapping wavelength range , the estimate lacks the accuracy of the g130 m bias estimate . for an initial look at the spectral variations , we define g130 m and g160 m mean spectra as @xmath75 where @xmath76 is the @xmath77th spectrum of the series of @xmath78 spectra . similarly , the rms residual spectrum ( hereafter referred to simply as the rms spectrum ) is defined as @xmath79 ^ 2 } } \right\}^{1/2 } \ .\ ] ] the rms spectrum is especially useful as it isolates the variable part of the spectrum ; constant components disappear , though sometimes small residuals are visible in the case of strong features . the statistical uncertainty in the mean spectra is @xmath80 where @xmath81 is the error spectrum of the @xmath77th spectrum in the series . the total uncertainty in the mean spectra consists of this statistical uncertainty and our estimate of the fractional uncertainty in precision as described above , which amounts to @xmath82 and @xmath83 . to determine the total uncertainty , the statistical uncertainty ( equation [ eq : staterror ] ) and the uncertainty in precision ( @xmath84 ) are added in quadrature . the mean and rms spectra for the g130 m and g160 m settings are shown in figures [ fig : mean_rms_g130 ] and [ fig : mean_rms_g160 ] , respectively . the rms spectrum resulting from equation ( [ eq : rmsspectrum ] ) combines both the intrinsic variability and the variance due to noise , as discussed by @xcite and @xcite . in order to isolate the rms spectrum of the intrinsic variations @xmath85 , we model the distribution of the residuals of each pixel about the mean . the combined statistical and systematic noise in each pixel of spectrum @xmath77 is thus @xmath86 . assuming that the flux measurement errors and the intrinsic variations arise from independent gaussian random processes , we find maximum likelihood estimates for the optimal average in equation [ eq : meanspectrum ] is formally distinct from the optimal average @xmath87 , though practically they are indistinguishable for these data . ] , @xmath87 , and @xmath85 by minimizing @xmath88 } \ , \ ] ] where @xmath89 and @xmath90 is the precision of the @xmath77th spectrum . this estimate of the intrinsic rms spectrum ( @xmath85 ) is also shown in figures [ fig : mean_rms_g130 ] and [ fig : mean_rms_g160 ] . the next step in our initial analysis is to produce light curves for the continuum and emission lines . at this stage , our goal is to make simple measurements from the reduced spectra , introducing as few assumptions as possible . all flux measurements are performed on spectra in the observed frame . we have not corrected the spectra for galactic extinction in order to facilitate the cleanest comparison with other measurements to be reported elsewhere in this series of papers ( e.g. , broad - band photometry ) . there are bad pixels throughout the spectrum , and their location and severity change with time , instrument settings , and airglow subtraction ( e.g. , if a spectrum is taken entirely in orbital bright time , the flux in the airglow windows is set to zero and the pixels are flagged as bad pixels ) . to prevent the introduction of artificial variations in the relative flux estimates , bad pixels are masked throughout the dataset . this means that if a pixel is bad in any of the visits , the pixel is masked out in each of the 171 spectra . we further mask galactic ly@xmath3 absorption and airglow region . integration ranges ( listed in table [ tab : int_ranges ] ) were chosen using the mean spectra in figures [ fig : mean_rms_g130 ] and [ fig : mean_rms_g160 ] as a guide . continuum ranges are chosen to be as uncontaminated as possible by absorption lines and broad emission - line wings . in the case of overlapping emission lines ( e.g. , civ and heii ) , the boundary wavelength corresponds to the wavelength at which the fluxes of the two lines are comparable . we do not mask absorption lines at this stage in our analysis . we are unable to cleanly separate nv and ly@xmath3 using this simple procedure . lccc @xmath91 & 1364.51369.5 & & + ly@xmath3@xmath92 & 1201.01255.0 & 1155.01160.0 & 1364.51369.5 + siiv@xmath6 & 1405.01455.0 & 1364.51369.5 & 1460.01463.5 + civ@xmath15 & 1520.01646.0 & 1475.01482.0 & 1743.01749.0 + heii@xmath1 & 1647.01740.0 & 1475.01482.0 & 1743.01749.0 continuum fluxes are measured as the weighted mean of the flux density in the integration region , with weights equal to the inverse of the variance , @xmath93 where @xmath94 as in equation ( [ eq : staterror ] ) . statistical uncertainties computed by calcos are corrected for low counts following @xcite . statistical uncertainties on the mean fluxes are obtained through standard error propagation , @xmath95 in all cases , bad pixels are excluded from the computation . emission - line fluxes are measured as the numerical integral of the emission flux above a locally defined continuum defined by the relatively featureless windows given in table [ tab : int_ranges ] . to estimate the local continuum underneath the line we performed a @xmath96 linear fit of the continuum flux in the selected regions . the linear local continuum is then subtracted from the emission component , again masking bad pixels . the line flux is numerically integrated over the integration limits given in table [ tab : int_ranges ] using simpson s method . we do not interpolate over bad pixels . we note , however , that the difference between integrating over the bad pixels and computing the integral excluding them is @xmath97% . statistical errors are computed numerically by creating @xmath98 realizations of the line flux and the underlying linear continuum . the flux @xmath99 is randomly generated from a gaussian distribution having mean equal to the flux of the spectral element and width @xmath41 equal to the statistical error on the flux . for the linear continuum , we generate @xmath100 fits having a mean equal to the best fit values and covariance equal to their covariance matrix . for each realization , a line - flux estimate is then obtained by subtracting the linear continuum and by performing the numerical integration of the residuals . confidence levels ( @xmath101 ) are finally obtained from the distribution of the @xmath100 line fluxes . when the error bars are asymmetric , we adopt the larger error as the statistical error associated with the integrated flux . as noted above , we adopt as the fractional error in precision @xmath102 and @xmath103 for the g130 m and g160 m settings , respectively . this is added in quadrature to statistical error of the integrated fluxes . the error in the precision dominates throughout the g160 m spectra and in the g130 m spectra as well , except at wavelengths shortward of @xmath104 longward of @xmath105 , and in the core of the ly@xmath3 complex . the final continuum and emission - line light curves are listed in table [ tab : light_curves ] and shown in figure [ fig : light_curves ] . the light curve statistics are given in table [ tab : lc_stats ] . the average interval between two consecutive observations is @xmath106 days with an rms @xmath107 days . the median interval between observations is @xmath108 days . the largest gaps between consecutive observations are three days ( on two occasions ) and two days ( on six occasions ) . cccc|ccc 6690.6120 & @xmath109 & @xmath110 & @xmath111 & @xmath112 & @xmath113 & @xmath114 + 6691.5416 & @xmath115 & @xmath116 & @xmath117 & @xmath118 & @xmath119 & @xmath120 + 6692.3940 & @xmath121 & @xmath116 & @xmath122 & @xmath123 & @xmath124 & @xmath125 + 6693.3237 & @xmath126 & @xmath127 & @xmath128 & @xmath129 & @xmath130 & @xmath131 + 6695.2701 & @xmath132 & @xmath133 & @xmath134 & @xmath135 & @xmath136 & @xmath137 + 6696.2459 & @xmath138 & @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath143 + 6697.3080 & @xmath144 & @xmath145 & @xmath146 & @xmath147 & @xmath148 & @xmath149 + 6698.3041 & @xmath150 & @xmath151 & @xmath152 & @xmath153 & @xmath154 & @xmath155 + 6699.2338 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 + 6700.2299 & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 + 6701.3588 & @xmath168 & @xmath169 & @xmath170 & @xmath171 & @xmath172 & @xmath173 + 6702.1557 & @xmath174 & @xmath175 & @xmath176 & @xmath177 & @xmath178 & @xmath179 + 6703.1518 & @xmath180 & @xmath181 & @xmath182 & @xmath183 & @xmath184 & @xmath185 + 6705.3432 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath191 lcccccc @xmath192 & @xmath193 & 0.017 & 0.201 & 64.74 & 21.87 & @xmath194 + @xmath195(ly@xmath3 ) & @xmath196 & 0.012 & 0.065 & 46.84 & 34.31 & @xmath197 + @xmath195(siiv ) & @xmath198 & 0.065 & 0.099 & 6.00 & 3.23 & @xmath199 + @xmath195(civ ) & @xmath200 & 0.015 & 0.072 & 62.97 & 47.23 & @xmath201 + @xmath195 ( heii ) & @xmath202 & 0.046 & 0.135 & 10.62 & 5.47 & @xmath203 certain simplifying assumptions underlie the rm technique . most time - series analyses start with the assumption that the emission - line light curves are simply scaled , time - delayed , and possibly smoothed versions of the continuum light curve . inspection of the light curves in figure [ fig : light_curves ] suggests that this is an entirely reasonable assumption for the first half of the campaign . however , approximately halfway through the campaign , the emission - line response becomes more complicated . between approximately hjd2456780 and 2456815 , the emission - line light curves are either flat ( ly@xmath3 , heii ) or decreasing ( civ ) while the continuum is slowly rising . moreover , the intensity ratio between the last two strong peaks in the continuum light curve at around hjd2456820 and 2456840 seems to be almost inverted in the lines , with the second peak being stronger than the first one ( especially in civ ) . there is also a small event in the continuum light curve around hjd2456785 that does not appear to have counterparts in the emission - line light curves . the line light curve that seems to best trace the continuum is the heii light curve , which is sensitive to the continuum at energies above 4 ryd . it is also the only strong line in the cos spectra that is neither a resonance line nor self - absorbed . moreover , it is the line that arises closest to the continuum source , as we will show below . because of the changing character of the emission - line response , for our initial analysis we measure emission - line lags ( a ) for the entire data set and ( b ) for subsets that divide the data into two separate halves of 85 observations each . the first subset , which we will refer to as `` t1 , '' runs from hjd2456690 to 2456780 and the second subset , `` t2 , '' runs from hjd2456781 to 2456865 . we first measured the emission - line lags relative to the continuum variations by cross - correlation of the light curves . we used the interpolation cross - correlation ( iccf ) method as implemented by @xcite . in this method , uncertainties are estimated using a model - independent monte carlo method referred to as `` flux randomization and random subset selection ( fr / rss ) . '' for each realization , @xmath204 data points are selected from a light curve with @xmath204 independent values , without regard to whether or not any particular point has been previously selected . for data points selected @xmath205 times in a given realization , the flux error associated with that data point is reduced by a factor of @xmath206 . the flux measured at each data point is then altered by adding or subtracting a random gaussian deviate scaled by the flux uncertainty ascribed to that point . each realization yields a cross - correlation function that has a maximum linear correlation coefficient @xmath207 that occurs at a lag @xmath208 . we also compute the centroid @xmath209 of the cross - correlation function using all the points near @xmath210 with @xmath211 . typically a few thousand realizations are used to construct distribution functions for the iccf centroid and peak . we adopt the median values of the cross - correlation centroid distribution and the cross - correlation peak distribution as our lag measurements . the uncertainties , which are not necessarily symmetric , correspond to a 68% confidence level . in general , @xmath209 is found to be a more reliable indicator of the blr size than @xmath208 , though we record both . the iccf measurements of @xmath208 and @xmath209 for the four strongest uv emission lines are given in the second and third columns , respectively , in table [ tab : lags ] . lccccc ly@xmath3 & @xmath212&@xmath213&@xmath214&@xmath215&@xmath216 + siiv & @xmath217&@xmath218&@xmath219&@xmath220&@xmath221 + civ & @xmath222&@xmath223&@xmath224&@xmath225&@xmath226 + heii & @xmath227&@xmath228&@xmath229&@xmath230&@xmath231 we have also estimated emission - line lags using javelin , which is an improved version of spear @xcite . javelin assumes that the emission - line light curves are shifted and smoothed versions of the continuum light curve ( as with the iccf analysis ) , where the continuum is modeled as a damped random walk @xcite with uncertainties determined using the markov chain monte carlo method . we model the full dataset , and each line light curve was run independently with the continuum . the results , given in column ( 4 ) of table [ tab : lags ] , are in good agreement with the iccf analysis , as expected . in columns ( 5 ) and ( 6 ) of table [ tab : lags ] , we also give the iccf centroid values for the t1 and t2 subsets . we also show the iccfs for the entire sample and the t1 and t2 subsamples in figure [ fig : ccf ] . in general , the lags for the t1 subsample have the smallest uncertainties and the iccfs have the largest peak correlation coefficients @xmath207 , as expected . the t2 subset , on the other hand , yields lags with larger uncertainties and iccfs with lower values of @xmath207 ( indeed , _ much _ lower in the case of siiv and civ ) , again as expected from visual inspection of the light curves . the t2 lags are also larger than those from t1 , probably only in small part because the continuum is on average brighter ( by @xmath55% on average ) during the second half of the campaign so the blr gas that is most responsive to continuum changes is farther away from the central source . as noted earlier ( section [ sec : design ] ) , this program was designed to recover kinematic information about the blr by resolving the emission - line response as a function of radial velocity . this will be explored in detail in subsequent papers in this series . here we carry out a simple preliminary analysis intended to show only whether velocity - dependent information is present in the data . we isolate the ly@xmath3 and civ profiles as described earlier ( section [ sec : intlc ] ) but then integrate the fluxes in bins of width 500 ^ -1 km s@xmath56 , except in the shortward wing of ly@xmath3 ( @xmath232 ) where we use 1000 ^ -1 km s@xmath56 bins on account of the low flux in the blue wing of this line . we show the iccf centroids for each velocity bin in the civ emission - line profile in the bottom panel of figure [ fig : civ_bin_t1t2 ] . for the two subsets as well as the entire dataset , we see that there is a clear ordered structure in the kinematics . we can not infer much from such a simple analysis , of course , because we can not accurately characterize a complex velocity field with a single number . it is reassuring , however , that the general pattern is similar to what has been seen in other objects and is qualitatively consistent with a virialized region ( i.e. , the high velocity wings respond first ) . the middle panel of figure [ fig : civ_bin_t1t2 ] shows the rms spectra for the entire dataset and the two subsets . in the bottom panel of figure [ fig : lya_bin_t1t2 ] , we show the velocity - binned iccf centroids for the ly@xmath3 emission line . again , a clear pattern emerges , as the lags in each velocity bin are highly correlated with those of adjacent bins . however , the pattern that emerges is unlike what is seen in civ ; the largest lags are at intermediate velocities and the lags decrease toward line center . however , given the severe blending and strong absorption , detailed modeling will be required before any meaningful conclusions can be drawn . the mean spectra shown in figures [ fig : mean_rms_g130 ] and [ fig : mean_rms_g160 ] can be compared with earlier uv spectra of ngc 5548 obtained with _ hst_(e.g . * ; * ? ? ? the earliest high - quality uv spectra of ngc 5548 showed only weak absorption in the resonance lines ( ly@xmath3 , nv , siiv , and civ ) , a factor that contributed to our selection of ngc 5548 as a target for this investigation . the 2013 spectra @xcite , however , revealed not only several strong narrow absorption features in the resonance lines , but also evidence for a relatively large `` obscurer '' that strongly absorbs the emission in the blue wings of the resonance lines . this feature is still present in our spectra obtained a year later , but is weaker than it was in 2013 . the presence of this absorption , combined with the blending of various emission features ( ly@xmath3 and nv , siiv and oiv ] , civ and heii ) , complicates analysis of these spectra . to characterize the emission - line structure of ngc 5548 and guide our selection of continuum windows for our emission - line flux measurements , we fit a heuristic model to the lines and continuum in the mean spectra . our model for ngc 5548 is similar to that adopted by @xcite , and it includes the broad absorption features associated with all permitted transitions in the spectrum . our adopted continuum is a reddened power law of the form @xmath233 . we correct for @xmath234mag of galactic extinction @xcite using the prescription of @xcite and @xmath235 . we do not apply any correction for possible internal extinction in ngc5548 . longward of 1550 , we also include blended feii emission as modeled by @xcite , broadened with a gaussian with full - width at half - maximum @xmath236 . we model the emission lines with multiple gaussian components . these are not an orthogonal set , and the decomposition is not rigorously unique , but they characterize each line profile well . for the brightest lines , we start with a narrow component , typically with @xmath237 . this component is essentially identical to the narrow component used by @xcite for fitting the 2004 stis spectrum of ngc 5548 . since this narrow component is difficult to deblend from the broader components of each line , and since @xcite saw little variation in narrow - line intensity over time , we fix the flux and widths of the narrow components of ly@xmath3 , nv , civ , and heii to the values we used to fit the 2004 stis spectrum . the siiv lines do not have a detectable narrow - line component . we note that while @xcite detected changes in the strength of the narrow [ oiii]@xmath238 , 5007 lines over timescales of years , the variations over the last decade have been only at the few percent level . similar variations in the narrow components of the uv lines would not be easily detected here because the narrow components are all so weak . next we add an intermediate - width component with @xmath239 without ascribing physical meaning to it , using the stis 2004 spectrum as a model . an intermediate - width component is included in ly@xmath3 , nv , siiv , civ , and heii , as well as in the fainter lines ciii*@xmath2401176 , siii@xmath2401260 , siii@xmath241oi@xmath2401304 , cii @xmath2401335 , niv]@xmath2401486 , oiii]@xmath2401663 , and niii]@xmath2401750 . for the weaker lines , this is often the only component detected , so its flux , width , and position are all allowed to vary . for the stronger lines , as for the nlr components , we again keep the fluxes and widths of the intermediate - width components fixed at the values found for the stis 2004 spectrum since @xcite found that these components vary only slightly in flux over several years . for the stronger lines , we next include broader components with @xmath242 8000 , and @xmath243 , respectively . for the doublets of nv , siiv , and civ , we assume the line - emitting gas is optically thick and fix the flux ratio of each pair to 1:1 , although for the @xmath243 component , only a single gaussian is used . finally , as can be seen in the rms spectrum in figure [ fig : mean_rms_g160 ] , there are two weak bumps that appear on the red and blue wings of the civ emission - line profile at @xmath244 and @xmath245 in the observed frame . these bumps are also present in the mean spectrum ; we include a single gaussian component to account for each of these bumps . as described by @xcite , we use an asymmetric gaussian with negative flux to model the broad absorption troughs . the asymmetry in these gaussian profiles is introduced by specifying a larger dispersion on the blue side of line center than on the red side . the asymmetry is fitted as a free parameter , and the resultant absorption line has a roughly rounded triangular shape with a blue wing extending from the deepest point in the absorption profile . during the first part of our reverberation campaign , when these absorption features were strongest , an additional depression appeared on the high - velocity blue tail of the main absorption trough . we use a single , symmetric gaussian to model the shape of this additional shallow depression . for absorption by nv , siiv , and civ , since the individual doublet profiles are unresolved , we assume the lines are optically thick , so each line in the doublet has the same strength and profile . as a final component , we include absorption by damped galactic ly@xmath3 with a column density of @xmath246 @xcite . the full spectral model for ngc 5548 , excluding the narrow absorption , is shown in the upper panels of figure [ fig : mean_rms_g130 ] and figure [ fig : mean_rms_g160 ] , superposed on the observed mean spectra . in future papers , we will apply this model to individual spectra to isolate the individual emission - line fluxes and to study absorption - line variability . to put the results reported here in context , we note that ngc5548 has been monitored in the uv for rm purposes on two previous occasions , as noted in section [ section : introrm ] . in 1989 , ngc5548 was observed once every four days for eight months with _ iue _ @xcite . in early 1993 , it was observed every other day with _ iue _ for a period of two months , and during the latter part of that campaign , it was also observed daily for 39 days with the _ hst_faint object spectrograph @xcite . the primary goals of these two experiments were quite different : the 1989 campaign was the first massive coordinated rm experiment and it was designed to measure the mean lags for the strong uv lines . the 1993 campaign was a higher time - resolution experiment that was designed to eliminate ambiguities from the 1989 campaign . specifically , its goals were : 1 . to measure the lag of the most rapidly responding line , heii@xmath1 . 2 . to determine whether or not there is a lag between the uv and optical continuum variations . 3 . to determine whether the wings and core of the civ emission line have different lags . in addition to these rm programs , several _ hst _ cos spectra of ngc5548 were obtained in 2013 with the primary goal of studying absorption features in the uv as support for an intensive x - ray monitoring program undertaken with _ xmm - newton _ @xcite . our own results on variable absorption features constitute an extension of that effort and will be the subject of a future paper . again , for broader context , during the agn storm campaign ngc5548 was at about the same mean continuum luminosity as it was during the 1989 campaign ( but with a somewhat lower amplitude of variability ) , somewhat brighter than in the 1993 campaign , and decidedly brighter than it was in 2013 , which was near the end of a lower - than - normal state that lasted several years @xcite . the resonance lines showed much more self - absorption in this campaign than in either the 1989 or 1993 observations , but less than seen in 2013 . the emission - line lags were somewhat larger during the 1989 campaign and the emission - line fluxes were higher , at least in part on account of much lower absorption in 1989 . the 1993 emission - line lags were similar to those obtained in this experiment , but again the line fluxes were larger , but less self - absorbed . as already noted , the response of the emission lines becomes complicated during the second half of the present campaign . the heii@xmath1 light curve seems to match the 1367continuum most closely ; this line responds primarily to continuum emission at @xmath247 , implying that the variations in the 1367 continuum provide a reasonable proxy for the behavior of the hydrogen - ionizing continuum ( @xmath30 ) . heii arises closer to the central source than the other emission lines , and it is also the only non - resonance line . more detailed analysis will be undertaken once the heii@xmath21 and balmer - line results become available from our contemporaneous ground - based monitoring program @xcite . in addition to determining the geometry and kinematics of the blr , we also wish to use these data to improve on previous estimates of the mass of the central black hole . however , the strong absorption in the blue wings of the resonance lines , which was very weak if even present in the 1989 and 1993 campaigns , precludes using simple measurements of the rms spectra ( e.g. , * ? ? ? * ) to make a mass estimate . more detailed modeling that we hope will lead to a more accurate black hole mass will be undertaken in future papers . to summarize briefly , we have presented the first results from a uv spectroscopic rm study of ngc5548 undertaken with _ hst _ cos in 2014 . we detect strong variations in the continuum and find clear delayed response of the strong emission lines , ly@xmath3 , nv , siiv , civ , and heii . a preliminary investigation shows that there is indeed a strong velocity - dependence of the emission - line lags , at least in the case of ly@xmath3 and civ , although blending and strong resonance absorption will make interpretation challenging . however , we have also shown that a heuristic multicomponent model can account for virtually all the spectral features . in future contributions , we will use this model as a starting point to explore the uv spectral variations in detail . we will also undertake a similar analysis of contemporaneous optical spectra in an effort to more completely understand the blr geometry and kinematics . in the accompanying paper ii , we combine the continuum light curve presented here with high - cadence observations with _ swift _ for a similar reverberation study of the accretion disk structure in ngc5548 . support for _ hst _ program number go-13330 was provided by nasa through a grant from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 . we are grateful for the dedication and hard work by the institute staff to make this program a success . gdr , bmp , cjg , mmf , and rwp are grateful for the support of the national science foundation through grant ast-1008882 to the ohio state university . ajb and lp have been supported by nsf grant ast-1412693 . mcb gratefully acknowledges support through nsf career grant ast-1253702 to georgia state university . kdd is supported by an nsf fellowship awarded under grant ast-1302093 . jmg gratefully acknowledges support from nasa under award nnh13ch61c . pbh is supported by nserc . sron is financially supported by nwo , the netherlands organization for scientific research . bck is partially supported by the uc center for galaxy evolution . csk acknowledges the support of nsf grant ast-1009756 . pl acknowledges support from fondecyt grant # 1120328 . ap acknowledges support from a nsf graduate fellowship and a ucsb dean s fellowship . jss acknowledges cnpq , national council for scientific and technological development ( brazil ) for partial support and the ohio state university for warm hospitality . tt has been supported by nsf grant ast-1412315 . tt and bck acknowledge support from the packard foundation in the form of a packard research fellowship to tt . tt thanks the american academy in rome and thee observatory of monteporzio catone for kind hospitality . the dark cosmology centre is funded by the danish national research foundation . mv gratefully acknowledges support from the danish council for independent research via grant no . dff 4002 - 00275 . this research has made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . the authors acknowledge with great sadness the loss of our long - time collaborator in the planning phases of this project , professor david j. axon , who passed away on 2012 april 5 . , k. 1994 , in astronomical society of the pacific conference series , vol . 69 , reverberation mapping of the broad - line region in active galactic nuclei , ed . m. gondhalekar , k. horne , & b. m. peterson , pp . 2351 massa , d. , ely , j. , osten , r. , penton , s. , aloisi , a. , bostroem , a. , roman - duval , j. , & proffitt , c. 2014 , updated absolute flux calibration of the cos fuv modes , instrument science report cos 2013 - 09 ( baltimore : stsci )

using non - equilibrium molecular dynamics simulations , it has been recently demonstrated that water molecules align in response to an imposed temperature gradient , resulting in an effective electric field . here , we investigate how thermally induced fields depend on the underlying treatment of long - ranged interactions . for the short - ranged wolf method and ewald summation , we find the peak strength of the field to range between and for a temperature gradient of . our value for the wolf method is therefore an order of magnitude lower than the literature value [ j. chem . phys . * 139 * , 014504 ( 2013 ) and * 143 * , 036101 ( 2015 ) ] . we show that this discrepancy can be traced back to the use of an incorrect kernel in the calculation of the electrostatic field . more seriously , we find that the wolf method fails to predict correct molecular orientations , resulting in dipole densities with opposite sign to those computed using ewald summation . by considering two different multipole expansions , we show that , for inhomogeneous polarisations , the quadrupole contribution can be significant and even outweigh the dipole contribution to the field . finally , we propose a more accurate way of calculating the electrostatic potential and the field . in particular , we show that averaging the microscopic field analytically to obtain the macroscopic maxwell field reduces the error bars by up to an order of magnitude . as a consequence , the simulation times required to reach a given statistical accuracy decrease by up to two orders of magnitude . = 1

1602.02734

a wide range of phenomena in physics , biology , chemistry and materials science are caused by strong spatial variations in thermodynamic quantities , such as pressure or temperature , on a microscopic scale . some of these effects are related to temperature gradients which may , for instance , be induced by ultrasonic insonation @xcite or heated nanoparticles @xcite . the peltier effect as well as the soret effect both fall in this category @xcite . another effect , which has received considerable attention recently , is the thermo - polarisation effect @xcite . using non - equilibrium molecular dynamics ( nemd ) simulations , bresme and co - workers demonstrated that water molecules align in response to an imposed temperature gradient , leading to electrostatic fields as high as @xmath3 v / m for gradients of 5 k / @xcite . in addition , they were able to confirm that the electric field scales linearly with the temperature gradient @xcite in accordance with the theoretical predictions of non - equilibrium thermodynamics ( net ) @xcite . in molecular simulations , coulomb interactions are regularly treated via ewald summation @xcite ( including approximations to it ) or a form of truncated interactions @xcite . in most studies on the thermo - polarisation effect @xcite , electrostatic interactions were handled with the truncated , short - ranged wolf method @xcite . it was argued that ewald summation can introduce artifacts , which can be avoided by using the short - ranged method @xcite . very recently , however , bresme and co - workers found that the wolf method overestimates the induced electric field in a spherical droplet of water by an order of magnitude as compared to ewald summation @xcite . the wolf method and other short - ranged methods @xcite are attractive because they achieve linear scaling with the number of particles as compared to the fastest approximations to ewald summation , such as particle - mesh ewald , which scale as @xmath4 @xcite however , it is well known that truncation of long - ranged coulomb interactions in simulations can lead to severe artifacts @xcite . in particular , short - ranged methods often fail for heterogeneous systems containing interfaces , even though they are known to perform well in bulk equilibrium simulations provided that the parameters are chosen carefully @xcite . in simulations of the liquid vapour interface , for example , the wolf method was found inadequate for predicting the electrostatic potential and dipole orientations , regardless of the choice of parameters @xcite . in the context of local molecular field ( lmf ) theory it has been demonstrated recently that averaged long - range effects can be taken into account self - consistently through an external potential @xcite . in this approach , short - ranged interactions are modelled through a pairwise potential which bears strong similarities to the wolf method @xcite . however , in the absence of the external potential the short - ranged method failed to reproduce the correct results as obtained with ewald summation and molecules were found to overorient @xcite . here , using a full treatment of electrostatic interactions with ewald summation we investigate the validity of the electric fields and induced orientations observed by bresme and co - workers @xcite . the field calculation requires especially careful consideration , as the large body of work published thus far relies on the formulation which is inconsistent with the dynamics of the simulation @xcite . the correct calculation of the field requires a modified kernel ( rather than @xmath5 ) that is consistent with the effective truncated coulomb interactions @xcite . we discuss this issue in detail and carry out a comparison of the thermally induced fields and multipole moments as obtained both with ewald summation and the wolf method . another important aspect that deserves consideration , is the spatial averaging of the potential and the field . in order to resolve the spatial variation of these quantities , it is advantageous to consider a quasi one - dimensional setup to enhance sampling . usually , the charge density is first spatially averaged over small slabs ( bins ) and then convoluted with an appropriate kernel to obtain , for example , the potential @xcite . as a consequence , the potential calculated in this way does , in general , not represent the exact average over the individual bin . however , as we demonstrate in this work , calculating the exact analytical average can be done straightforwardly for both summation methods and can lead to huge reductions in the error bars for low resolutions . therefore , this approach frees us from the constraint of employing an unnecessarily high , submolecular resolution . the remainder of this paper is structured as follows : in sec . [ sec : electrostatics ] , we briefly summarise the electrostatic kernels for ewald summation and the wolf method , respectively , and discuss important differences using a simple model system . then , in sec . [ sec : spatavg ] , we reduce the three - dimensional problem to one spatial dimension employing symmetry properties of the setup . the two different multipole expansions considered in this work are derived in sec . [ sec : multi ] . the simulation protocol is explained in sec . [ sec : sim ] and all simulation results are presented in sec . [ sec : res ] . in md simulations , periodic boundary conditions ( pbcs ) are usually employed to reduce finite - size or surface effects @xcite . this implies that the simulated system is infinite , but can be fully described with knowledge of the state of a reference box . the electrostatic potential , @xmath6 , is governed by poisson s equation , @xmath7 where @xmath8 is the charge density and all quantities are expressed in gaussian units . one way of determining the potential is to solve this equation directly for the fictitious infinite system . alternatively , the task can be mapped onto the problem of finding a generalised kernel or green s function , @xmath9 , compatible with a finite volume with pbcs , considering nearest images only @xcite . once @xmath9 is known , the potential and the field can then be calculated as @xmath10 and @xmath11 where @xmath12 is the simulation box of volume @xmath13 . throughout this work , we assume that pbcs are explicitly taken into account whenever expressions that depend on an argument of the form @xmath14 are evaluated ( see for example appendix [ secapp : electrostatics ] ) . although both approaches lead to the same result , there is an important conceptual difference : in the former case , we consider the infinite system of charges interacting with the potential that scales as @xmath5 ( in three dimensions ) plus surface term , whereas in the latter case , we only consider the charge distribution in our reference box with an effective interaction . the periodicity of the setup is then fully mimicked by the green s function , which no longer decays as @xmath5 and is not even spherically symmetric . let us consider a charge - neutral system consisting of @xmath15 molecules each comprising @xmath16 partial charges @xmath17 located at positions @xmath18 ( @xmath19 labels molecules and @xmath20 sites within a molecule ) . the total electrostatic energy is then given by @xcite @xmath21 \nonumber \\ & & & + \frac 12 \sum_{j}\sum_{\alpha } q_{j \alpha}^2 \lim_{r \to 0 } \left [ g(\mathbold r ) - \frac{1}{r } \right ] , \nonumber\end{aligned}\ ] ] where @xmath22 is the distance vector between the nearest pair of images , @xmath23 and @xmath24 is a @xmath25-dimensional vector . in the above equation we have omitted the summation bounds for readibility . in eq . ( [ eq : ufromg ] ) the surface term of de leeuw and co - workers @xcite has been omitted , because we employ conducting ( tin - foil ) boundary conditions . we can see that the functional form of @xmath9 directly affects the forces , which are calculated from the negative gradient of the energy , and therefore the dynamics of the simulation . in what follows , we briefly summarise the kernels for ewald summation and the wolf method . ewald summation is a numerical approximation to the exact solution of eq . ( [ eq : poisson ] ) for pbcs , whose green s function is formally given by @xmath26 here , the summation extends over reciprocal vectors @xmath27 with components @xmath28 , where @xmath29 is an integer and @xmath30 the box size in direction @xmath20 . introducing the convergence factor @xmath31 , the expression is split up into two terms , one of which is converted back to real space . this leads to the representation @xcite @xmath32 where @xmath33 is a shift vector between a molecule and its periodic image and the summation runs over all periodic images . choosing @xmath34 carefully , it is possible to achieve fast convergence of the first sum and small contributions for @xmath35 . if we ignore these terms and introduce a spherical cutoff , @xmath36 , for better performance , eq . ( [ eq : ga ] ) finally reduces to @xmath37 where @xmath38 is the heaviside function . inserting this expression back into eq . ( [ eq : ufromg ] ) yields the standard ewald summation expression @xcite as presented in textbooks , e.g. in ref . . wolf and co - workers showed that in a condensed ionic system the net coulomb potential is effectively short - ranged @xcite . based on this insight , they devised a summation method that avoids the expensive _ k_-space term in eq . ( [ eq : gesim ] ) altogether . instead , the potential is damped and shifted in a way that enforces charge neutrality within the cutoff sphere for improved convergence properties . the corresponding kernel is given by @xmath39 \end{aligned}\ ] ] and reproduces the correct madelung energy as suggested by wolf and co - workers @xcite . later the method was extended to eliminate also higher - order multipoles inside the cutoff sphere @xcite . however , it was pointed out that the entire approach embodies certain assumptions on the underlying physical system @xcite , such as the availability of charges outside the cutoff region for screening @xcite . whether these assumptions are reasonable is not always clear a priori , especially for inhomogeneuous systems such as the one considered in this work . we note that the first term in @xmath40 is identical to the one in @xmath41 , although the optimal choice of the damping parameter , @xmath42 , is not necessarily the same as for ewald summation . a good value can be found by analysing the convergence of the madelung energy per ion @xcite . furthermore , in the wolf method the force is not exactly given by the negative gradient of the potential energy . the reason for this inconsistency is that the expression @xmath43 is used for the evaluation rather than @xmath44 in order for the force to vanish at the cutoff distance @xcite . there are extensions of the wolf method which address this issue ( for example ref . ) . however , given a reasonable combination of damping parameter and cutoff value , we expect the effects of this inconsistency on the electric field to be negligible . to illustrate the difference between the electrostatic kernels , we consider a test case based on calculating the potential generated by a single spc / e water @xcite molecule . this simple example should draw attention to the fact that , for an identical arrangement of charges , the results for the wolf method sensitively depend on the choice of kernel , damping parameter and cutoff radius . the quality of the wolf approximation to the electrostatic potential , computed according to eq . ( [ eq : phi ] ) , is assessed by comparison with the results of ewald summation , which approximates the exact solution . considering only a single molecule may seem atypical for the wolf method , since it relies on the idea that long - range contributions average out in a dense system . however , this comparison serves as a guideline for the choice of new parameters which help us to reduce the dependence on this crucial assumption . this is achieved by tuning the potential to get better agreement with ewald summation already on the level of a single molecule . the comparison in sec . [ sec : res ] will then allow us to assess the performance of the wolf method for a wider range of parameters , but it is not the intention of this work to single out an optimal choice . figure [ fig : phispcemodel ] shows the potential due to a single spc / e water molecule in a fully periodic system . the molecule is located at the centre of a rectangular simulation box with dimensions @xmath45 . the three charges , @xmath46 and @xmath47 , where @xmath48 is the elementary charge , are located in the @xmath49plane at positions @xmath50 and @xmath51 , respectively . ewald summation was carried out taking @xmath52 with @xmath53 , and choosing the set of @xmath54-vectors for eq . ( [ eq : gesim ] ) such that the estimated relative error of the force was approximately @xmath55 . for the wolf method , we compare two sets of parameters : @xmath56 and @xmath57 . the latter combination was employed by armstrong and bresme @xcite and the former with considerably weaker damping and a larger cutoff is added for comparison . we note that we also investigated the effects of a large cutoff combined with strong damping , i.e. @xmath58 . however , we did not observe any substantial differences for the main results of this work compared with the @xmath59 cutoff and therefore omitted the comparison . ( dotted lines ) and @xmath60 ( dashed lines ) . with increasing distance from the origin , the isolines of the potential correspond to the values @xmath61 , @xmath62 , @xmath63 , @xmath64 and @xmath65 , respectively . ] it is obvious that for the strong damping ( dashed lines ) the potential decays too quickly as compared to the result we get with ewald summation ( solid lines ) . only the short - range behaviour in the immediate vicinity of the molecule is captured correctly . the weaker damping parameter ( dotted lines ) , on the other hand , yields a reasonable agreement with ewald summation within a distance of about 6 from the origin , but shows some deviation further away . employing even lower values for @xmath42 , for example @xmath66 , reduces the discrepancy between the outermost contour lines only minimally ( not shown ) . since the value of the potential represented by the lowest contour level in fig . [ fig : phispcemodel ] corresponds to only @xmath67 of the highest one , we conclude that the parameters @xmath56 yield a reasonable approximation to the ewald result within the cutoff sphere of @xmath59 . validation of both sets of parameters in bulk simulations also reveals good agreement with ewald summation ( see appendix [ secapp : valid ] ) . once the method to treat electrostatic interactions is chosen and optimised , one typically wishes to improve the statistics of the collected averages . for this purpose a simulation setup with high spatial symmetry is advantageous @xcite . in this work , we focus on the case where the underlying three - dimensional problem can be reduced to one spatial dimension , as illustrated in fig . [ fig : box ] . for such a system , the average charge density can only depend on @xmath68 for sufficiently long simulation times , because the system is isotropic in all other directions . therefore , this approach is justified only if one considers sufficiently long simulations . assuming @xmath69 , we can then rewrite eq . ( [ eq : phi ] ) as @xmath70 where we introduced the one - dimensional kernel @xmath71 . ] taking the negative gradient of eq . ( [ eq : phi3dto1d ] ) yields the electrostatic field @xmath72 where @xmath73 denotes the derivative of @xmath74 . the above integrals can be evaluated readily for ewald summation and the wolf method ( see appendix [ secapp : electrostatics ] ) . the results can be improved considerably by averaging the potential and the microscopic field over small spatial regions , such that we obtain the macroscopic maxwell field for the latter . the centre of each control volume then represents its exact spatial average . to this end , we consider @xmath75 bins of width @xmath76 , as depicted in fig . [ fig : box ] . the lower and upper boundaries of bin @xmath77 , where @xmath78 , are given by @xmath79 and @xmath80 , respectively . the spatial average of the potential over bin @xmath77 is then given by [ eq : phi_ex_avg_der ] @xmath81 where the overbar denotes the spatially averaged kernel @xmath82 for our effectively one - dimensional system of point charges , we can decompose the charge density according to @xmath83 where @xmath84 is the one - dimensional dirac delta function . inserting this expression back into our previous result for the potential yields @xmath85 analogously , the averaged field is given by @xmath86 the corresponding expressions for @xmath87 and @xmath88 for ewald summation are derived in appendix [ secapp : electrostatics ] . the above averages for potential and field depend on all particle positions and therefore implicitly on time . the time average of any quantity @xmath89 is defined as @xmath90 where @xmath91 is the total simulation time of the production run . it is straightforward to evaluate @xmath92 and @xmath93 for the discrete trajectory obtained from the nemd simulation . in what follows , we outline how the exact potential , as calculated from the charge density , can be decomposed into individual multipole contributions . this helps us to gain insight into how the alignment of the molecules with respect to the temperature gradient affects the field . we consider two different expansions for comparison which are illustrated in fig . [ fig : multi ] . in the slab expansion ( fig . [ fig : multi]a ) , the multipole moments due to the charges located inside a bin are calculated relative to its centre . in the molecule expansion ( fig . [ fig : multi]b ) , separate multipole expansions are carried out for each individual molecule and the multipoles are located at the respective oxygen sites . if all moments were considered in the expansion , both approaches would give rise to the same potential at a distant point @xmath94 . we note that both types of expansion have already been considered in the past for interfacial systems @xcite . however , here we use a more general formulation @xcite which is also applicable to modified kernels representing truncated coulomb interactions . ( `` slab expansion '' ) and b ) for each molecule @xmath95 individually with the oxygen site at the origin ( `` molecule expansion '' ) . both approaches give rise to the same field at a distant point @xmath94 . ] the potential generated by a charge distribution enclosed in a volume @xmath96 is given by @xmath97 from this equation we can obtain the contributions of the individual multipole moments by expanding @xmath98 into a taylor series around @xmath99 , @xmath100 where @xmath101 is the total charge in @xmath96 , @xmath102 the dipole moment and @xmath103 the quadrupole moment . the symbol @xmath104 denotes the derivative with respect to the cartesian component @xmath105 . moving the origin of the charge distribution to @xmath106 and taking into account the symmetry properties of our effectively one - dimensional system , we find @xmath107 from the simulated trajectory , we then compute time averages of the multipole densities @xmath108 , @xmath109 and @xmath110 for the monopole , dipole and quadrupole moments of every bin @xmath77 , respectively . before defining these quantities , we first introduce some additional notation to distinguish between the two types of expansion . we use superscripts @xmath111 , where @xmath112 for slabs ( fig . [ fig : multi]a ) and @xmath113 for molecules ( fig . [ fig : multi]b ) . the density of @xmath114 [ cf . ( [ eq : moments3d ] ) ] is then given by @xmath115 for the case @xmath112 and @xmath116 for the case @xmath113 , where @xmath117 is the volume of the bin . since we only consider the multipole moments @xmath101 , @xmath118 and @xmath119 , from now on we omit the subscripts for readability . in general , the multipole moments depend on the way the charge distribution is partitioned @xcite and consequently the multipole densities for slabs and molecules are not directly comparable . for example , the quadrupole moment of a reference bin will , in general , not be equal to the sum of the molecular quadrupole moments . furthermore , we make an intentional , small mistake in the evaluation of @xmath120 and @xmath121 for the sake of computational convenience , because we ignore the precise location of the molecular moments within the bin @xmath77 . however , as we will see in sec . [ sec : res ] , the error in the electrostatic potential introduced by this approximation is negligible . the electrostatic potential ( at the centre of bin @xmath77 ) is then calculated as the sum of the three contributions in eq . ( [ eq : phimulti ] ) , @xmath122 which are given by [ eq : avgphimulticontr ] @xmath123 respectively . since the molecules are charge - neutral , it follows that all values @xmath124 and consequently @xmath125 vanish identically . for production runs , we prepared the system in the same state as armstrong and bresme @xcite in order to carry out a quantitative comparison . the simulation box ( fig . [ fig : box ] ) has exactly the same dimensions as the one used for the model system . for two of the three nemd simulations we used the wolf method and the remaining one was performed with ewald summation ( the relevant parameters are summarised in sec . [ subsec : model ] ) . lennard - jones interactions were truncated at @xmath59 in all cases . the box contains @xmath126 spc / e molecules resulting in a mass density of @xmath127 . all simulations were carried out using a modified version of the software package lammps ( 9dec14 ) @xcite which we augmented with the ehex / a algorithm @xcite . the system was first equilibrated and validated . starting from an initial lattice structure with zero linear momentum , we integrated the equations of motion with the velocity verlet algorithm @xcite employing a timestep of @xmath128 . for the first @xmath129 ps we rescaled the velocities to drive the system close to the target temperature of @xmath130 . this was followed by a short 200 ps _ npt _ run using a nos hoover thermostat with a relaxation time of @xmath131 and a nos hoover barostat with a relaxation time of @xmath132 @xcite . we then rescaled the box to the target dimensions and carried out a @xmath133 ps _ nvt _ run during which we monitored the average system energy . next , we adjusted the kinetic energy of the last configuration by velocity rescaling and used it as input for another @xmath134 ns _ nve _ equilibration run . the average temperature during this run was @xmath135 , where the error bar was estimated using block average analysis @xcite . we computed the pair - correlation function , the velocity autocorrelation function , the dielectric constant and the distance - dependent kirkwood @xmath136-factor ( see appendix [ secapp : valid ] ) . the validation suggests that our implementation is correct and our choice of parameters reasonable . to investigate the effect of a thermal gradient after the equilibration , the system was driven to a non - equilibrium stationary state by imposing a constant heat flux between two reservoirs , @xmath137 and @xmath138 ( fig . [ fig : box ] ) . this was achieved by introducing an additional force , @xmath139 , to the equations of motion @xcite , such that @xmath140 where @xmath141 is the mass of atom @xmath19 and @xmath142 the force calculated from the potential . the thermostatting force is defined as @xmath143 where @xmath144 is an indicator function which maps the particle to the region @xmath145 in which it is located and @xmath146 is a constant energy flux into @xmath145 . those parts of the simulation box which are not thermostatted are labelled with @xmath147 . the non - translational kinetic energy of the region @xmath145 is given by @xmath148 where the quantities @xmath149 and @xmath150 are the centre of mass velocity and the total mass of @xmath145 , respectively , and the index set @xmath151 comprises all particles which are located inside that region @xcite . the equations were solved numerically with our recently proposed ehex / a algorithm @xcite with a timestep of @xmath152 . for the symmetric setup shown in fig . [ fig : box ] , the heat flux is trivially related to the energy flow into the reservoir by @xmath153 where the factor of 2 in the denominator accounts for the periodic setup . after switching on the thermostat , we waited for 10 ns for any transient behaviour to disappear before starting with the @xmath154 production run . the energy conservation was excellent ( @xmath155 ) and the centre of mass velocity of the simulation box remained close to machine precision throughout the simulation . the heat fluxes are input parameters of the ehex algorithm which were adjusted by trial and error . the employed values are summarised in tab . [ tab : t ] . .[tab : t]imposed heat fluxes and measured values for the tempature gradients . we note that our heat flux for the wolf ( @xmath60 ) run is about 1.7% larger than the value used by armstrong and bresme @xcite . [ cols="<,^,^",options="header " , ] we note that lower heat fluxes are required for the wolf method in order to achieve the same temperature gradient as for ewald summation . this is consistent with the observation that the truncation of electrostatic interactions results in lower thermal conductivities @xcite . in this section , we present the key results for the temperature and density profiles ( sec . [ subsec : restemp ] ) , the multipole expansions ( sec . [ subsec : resmoments ] ) , the potential ( sec . [ subsec : respot ] ) , the field ( sec . [ subsec : resfield ] ) and the polarisation ( sec . [ subsec : respol ] ) . we estimated error bars for all results in this section . to this end we divided the entire trajectory into 600 blocks ( of length @xmath156 ) and assumed the results for the individual blocks to be uncorrelated . the size of the individual error bar then corresponds to twice the standard deviation of the mean . this estimate comprises the statistical error as well as the methodological error arising , for example , from the employed quadrature . figures [ fig : trho]a - b show the spatial variations in temperature and density along the @xmath68-direction with a resolution of @xmath157 ( @xmath158 ) . the temperature of an individual bin was calculated from the non - translational kinetic energy of the atoms inside that bin @xcite . there are only small differences between the results obtained with the ewald and wolf methods . the peak temperature at the centre of the hot reservoir is about @xmath159 and the lowest temperature at the centre of the cold reservoir is about @xmath160 ( fig . [ fig : trho]a ) . the temperature profile is linear outside the reservoirs and symmetric with respect to the origin of the simulation box , which is in accordance with the setup . the measured average number densities ( fig . [ fig : trho]b ) obtained with ewald summation and the wolf method agree well apart from slight differences in the vicinity of the cold reservoir . the mass density varies by up to 15% ( cold reservoir ) with respect to @xmath161 . we note that on this scale , we did not observe any appreciable discontinuities of the temperature or density close to the reservoir boundaries , although the thermostatting force is discontinuous . in this section , we discuss the induced molecular alignment and multipole moments due to the thermal gradient for both expansions in fig . [ fig : multi ] . the left column in fig . [ fig : multicomb ] corresponds to the slab ( centre - of - bin ) expansion and the right column to the molecule expansion . the monopole in the molecule expansion vanishes identically , hence it is not shown . the spatial variations of all quantities are shown with a resolution of @xmath162 ( @xmath163 ) . let us consider the time averaged charge density for slabs first ( fig . [ fig : multicomb]a ) . for ewald summation the error of the average is so large that it swamps the signal even after 60 ns of simulation time . we also note that the curve is not symmetric in the vicinity of the cold reservoir within the statistical uncertainty shown in the plot . we believe that this may be due to the fact that we computed the error bars as if neighbouring bins were independent , which is not the case , because molecules are charge neutral . the real error bars may be larger due to long - wavelength fluctuations . we confirmed that the results become symmetric ( within the statistical error ) upon doubling the simulation time . for the wolf method there is an accumulation of positive charge in the vicinity of the hot reservoir , which is enhanced by stronger damping . this result agrees qualitatively with the findings of rodgers and weeks for a different inhomogeneous system , where the authors compared the ( gaussian - smoothed ) charge density obtained with gaussian - truncated ( gt ) water to that of ewald summation @xcite . furthermore , we note that the error bar increases by about one order of magnitude upon refining the resolution by a factor of 10 , which corresponds to @xmath164 ( @xmath165 ) used by armstrong and bresme @xcite . figures [ fig : multicomb]b and d show the dipole densities for both expansions , respectively . for the slabs ( fig . [ fig : multicomb]b ) , there is no noticable trend of the dipole density within the statistical uncertainity . however , for the molecule expansion ( fig . [ fig : multicomb]d ) we find a strong disagreement between the two electrostatic kernels . for this case , we also quantified the average molecular alignment using the order parameter @xcite @xmath166 where @xmath167 defines the orientation of a molecule and @xmath168 is the unit vector in the direction of the temperature gradient . in the case of ewald summation molecules , on average , point to the cold reservoir and the alignment is fairly constant outside the reservoirs ( see inset in fig . [ fig : multicomb]d ) . the wolf method entirely fails to capture this behaviour . for the wide range of parameters considered in this work ( including the ones previously employed in the literature ) , the method predicts opposite orientations and overestimates the magnitude of alignment by a factor of about 7 for the strong damping . employing a lower value for the damping parameter reduces the overestimation , but can not correct the wrong sign . we also note that our results for the average molecular orientation ( inset in fig . [ fig : multicomb]d ) are in agreement with the ones reported by armstrong and bresme @xcite . the quadrupole densities , shown in figs [ fig : multicomb]c and e , agree well with each other within each expansion . similarily to the dipole density , considering slabs for the expansion ( fig . [ fig : multicomb]c ) yields results which are negligible compared to the molecule expansion ( fig . [ fig : multicomb]e ) . we note that in the latter case , the profile is proportional to the oxygen number density ( fig . [ fig : trho]b ) and can lead to considerable contributions to the potential . repeating our simulation with ewald summation and vacuum boundary conditions ( see refs for more details ) , we found consistent results for the multipole densities ( not shown ) . we can therefore rule out any artifacts arising from the boundary conditions at infinity on the results shown in this section . however , we noticed that the statistical error of the molecular dipole density decays much faster for vacuum boundary conditions relative to tin - foil boundary conditions . in the previous section , we analysed the thermally induced multipole moments for two different multipole expansions , namely slabs and molecules . the aim of this section is to compare three different ways of calculating the electrostatic potential : firstly , we consider the exact analytical average given by eq . ( [ eq : phi_ex_avg ] ) . secondly , we approximate the potential using only the average charge density given by the slab expansion , eq . ( [ eq : avgphimulticontr]a ) , which is the approach regularly employed in the literature @xcite . thirdly , we approximate the potential using also the dipole and quadrupole densities , i.e. eqs ( [ eq : avgphimulticontr]b - c ) . let us consider the results for the exact calculation first , which are shown in fig . [ fig : phiex]a . all graphs are symmetric with respect to the origin of the simulation box and periodic , indicating that the field vanishes at the centres of the reservoirs . although the shape of the potential predicted by the short - ranged method is similar to the one for ewald summation , the results are sensitive to the choice of damping parameter . weak damping overestimates the potential , whereas strong damping leads to an underestimation . both our choices fail to reproduce the ewald summation result correctly , although it seems plausible that intermediate values for the damping parameter could lead to a better agreement . figure [ fig : phiex]b compares ( for ewald summation ) the exact result for the electrostatic potential to that given by the monopole density in the slab expansion . we recall that the latter approach corresponds to averaging the charge density first and integrating it with the appropriate kernel afterwards [ eq . ( [ eq : avgphimulticontr])a ] . it is clear from comparison of the two curves including error bars that the exact calculation yields a huge improvement over the approximation . for the resolution shown in the plot ( @xmath169 , @xmath170 ) , the error bars are reduced by more than one order of magnitude . the inset shows the ratio of the maximum error of the approximation to the maximum error of the exact calculation as a function of the number of bins . ( we define the maximum error to be half the length of the largest error bar throughout the entire interval . ) for a very low resolution of 10 grid points ( @xmath171 ) , the maximum error decreases by about a factor of 26 . for high resolutions of @xmath172 the error ratio approaches unity implying that both methods become comparable , which is the expected behaviour in the limit @xmath173 . at the same time the magnitude of the error naturally increases for higher resolutions because fewer molecules contribute to a particular bin ( for 400 bins the maximum error increases by about @xmath174 as compared to the resolution of 40 bins shown in the figure ) . given that molecules point , on average , in opposite directions for the two electrostatic kernels ( fig . [ fig : multicomb]d ) , it is counterintuitive that the potentials are qualitatively comparable . to understand the origin of this seeming contradiction , we singled out the individual multipole contributions , which are illustrated in figs [ fig : phimulticomb]a - d for both expansions . let us consider the slab expansion first . for both electrostatic kernels ( figs [ fig : phimulticomb]a - b ) we found the monopole contribution ( black curve ) to capture the exact potential ( red line ) reasonably well for the chosen spatial resolution ( @xmath169 , @xmath170 ) . however , if we consider a point dipole and a point quadrupole ( representative for the respective bin average ) in addition to the point monopole located at the centre of each bin , we obtain a much better approximation to the exact result ( red circles ) . in fact , for ewald summation we recover the exact potential almost perfectly , whereas we observe an overshoot inside the hot reservoir for the wolf method . we believe that a more accurate approximation for the short - ranged method might be obtained by considering octupole and hexadecapole contributions in addition , but we did not investigate this further . the situation changes entirely for the molecule expansion shown in figs [ fig : phimulticomb]c - d , where the monopole contribution is zero . for ewald summation ( fig . [ fig : phimulticomb]c ) , the dipole density leads to a linear potential outside the reservoirs ( green curve ) corresponding to a negative field in the left half of the simulation box . however , close to the hot reservoir the quadrupole contribution ( blue curve ) outweighs the dipole contribution causing the slope of the overall potential to be negative and therefore the field to be positive . in the vicinity of the cold reservoir the dipole contribution dominates over the quadrupole contribution and the field is negative . the sum of both terms ( red circles ) agrees perfectly with the exact average ( red line ) . for the wolf method we found that the quadrupole density constitutes a much smaller correction to the dipole contribution which is almost negligible outside the reservoirs this might seem surprising at first given that the results for the quadrupole densities agree well for both summation methods ( fig . [ fig : multicomb]e ) . the apparent contradiction is explained by the fact that the derivatives of the kernels in the evaluation of the potential are very different for both methods . we will get back to this point in sec . [ subsec : respol ] when we discuss the macroscopic polarisation . with regard to the accuracy of the full multipole approximations ( up to the quadrupole term ) , we observed different trends for the maximum error of the potential within each expansion . for the slab expansion we found the maximum error to be about 6 times larger than the error of the exact potential for the lowest resolution ( @xmath175 , @xmath176 ) . upon increasing the resolution , the error ratio approaches unity , which is the expected behaviour . however , this is not the case for the molecule expansion , where the error is only about @xmath177 larger than the error of the exact potential initially , but the difference increases to about @xmath178 for the highest resolution ( @xmath179 , @xmath180 ) . we believe that this behaviour is reasonable , because we never intersect molecules and can not resolve the potential inside a molecule correctly . the higher the resolution the worse we expect the approximation to become in the vicinity of the point multipoles . averaging the potential exactly is preferable on all scales , rendering it clearly the method of choice . the exact results for the field in the sense of eq . ( [ eq : ez_ex_avg ] ) are shown in fig . [ fig : eex]a . focusing on the left half of the simulation box , we notice that the field is positive and strongest in the vicinity of the hot reservoir . for the peak field strength we measured values of about @xmath181 , @xmath182 and @xmath183 for ewald summation and the wolf method with weak and strong damping , respectively . close to the hot reservoir , the short - ranged method overshoots the ewald summation result for low damping and vice versa for high damping . we also infer from the figure that the field changes its sign in the vicinity of the cold reservoir . from the discussion of the potentials in the previous section ( fig . [ fig : phimulticomb]c ) we know that the inversion happens exactly when the dipole contribution to the field dominates over the quadrupole contribution . comparing our results to the ones reported by bresme and co - workers , we find a major discrepancy : in the original work @xcite the reported fields are about one order of magnitude higher than what we found . recently , however , it was suggested that the thermally induced field in a spherical droplet of spc / e water is of the order of @xmath184 after comparison with ewald summation ( pppm ) @xcite . nevertheless , the discrepancy still persists as the authors @xcite suggest that the wolf method itself is responsible for the overestimated field , whereas , in fact , the opposite is true for the set of parameters employed in ref . . the wolf method slightly underestimates the field if it is calculated consistently , namely using the correct kernel ( see fig . [ fig : eex]a ) . we can reproduce the results of armstrong and bresme closely if we calculate the field as @xcite @xmath185 considering gaussian units and taking the lower integration bound to be @xmath186 rather than @xmath187 . ( a comparison is omitted for brevity . ) for ewald summation this expression is correct and equivalent to eq . ( [ eq : ez3dto1d ] ) with @xmath188 as long as the net dipole density of the box , @xmath189 vanishes . considering sufficiently long simulations , this is necessarily the case for our system because of the symmetric setup ( see figs [ fig : box ] , [ fig : multicomb]b and d ) . if this was not the case , an additional term @xmath190 would have to be added to the right - hand side of eq . ( [ eq : eiso ] ) . the equivalence is trivially shown by rewriting the integral in eq . ( [ eq : ez3dto1d ] ) taking into account periodicity and charge neutrality . alternatively , one can integrate poisson s equation directly and impose periodicity by choosing the integration constants accordingly @xcite . however , applying eq . ( [ eq : eiso ] ) for the wolf method is wrong and the discrepancy between our result and the one of armstrong and bresme @xcite can therefore be traced back to using the incorrect expression in the calculation . similarly to what we observed for the potential , considering exact averages rather than estimating the field from the average charge density yields a huge improvement for low resolutions . the comparison in fig . [ fig : eex]b is carried out for a resolution of @xmath175 ( @xmath176 ) and , as shown in the inset , the error of the approximative field , i.e. using the negative derivative of @xmath191 in eq . ( [ eq : avgphimulticontr]a ) , is about 10 times larger than the exact one . for resolutions higher than @xmath192 ( @xmath193 , both approaches yield similar errors . comparing the insets of figs [ fig : phiex]b and [ fig : eex]b , we notice that the enhancement of the exact method over the approximative one is much higher for the potential . this can be partly explained by looking at the functional form of @xmath188 ( eq . ( [ eqapp : geders]a ) in appendix [ secapp : electrostatics ] ) . the function is piecewise linear and the midpoint rule , which corresponds to multiplying the function value at the centre of the bin by @xmath76 , is exact in the absence of any discontinuity . therefore , the advantage of using @xmath194 over @xmath195 for the evaluation of the field is less significant than for the potential . figure [ fig : e_error ] compares the spatial maximum errors for varying resolutions . interestingly , for sufficiently high resolutions of @xmath196 we found the maximum error of the approximative method to be up to almost 30% lower than the one for the exact average . we attribute this to cancellation of errors , since convergence tests support a correct implementation . far more important is the magnitude of the error for high resolutions . for simulation time scales of @xmath197 the error is comparable to the signal itself requiring even longer runs for the statistics to be satisfactory . suppose we wanted to get a rough idea of what the field looked like . with the conventional method , i.e. averaging the charge density first and then integrating it , the best we can do is to calculate the results on a sufficiently high resolution and then perform some sort of averaging . on the one hand , this approach is problematic because the coarse - grained values do not represent the correct bin averages . on the other hand , it is not straightforward to propagate the statistical errors from the fine resolution to the coarse one since the values are highly correlated . our proposed method of averaging the potential and the field analytically eliminates both issues and yields a huge improvement for low resolutions reducing the required simulation time scales by up to two orders of magnitude for the same quality of statistics . our final goal in this section is to relate the molecular multipole densities to the macroscopic polarisation . we show that the macroscopic maxwell equation @xmath198 holds locally for the bin averages calculated with ewald summation , where @xmath199 and @xmath200 are the @xmath68-components of polarisation and displacement field , respectively . we do not make any a priori assumptions about the locality @xcite and use the multipole expansions developed in sec . [ sec : multi ] as a general starting point for the discussion . we then identify the quantities on the right - hand side of eq . ( [ eq : maxwelledp ] ) after simplifying the expressions . we note that our analysis only holds in the context of sufficiently long simulations ( like in sec . [ sec : multi ] ) , because we use @xmath201 in place of the full @xmath202 . this simplifies the discussion in that we only need to consider the @xmath68-component of the spatially averaged dipole density , @xmath203 , and the density of @xmath119 given by @xmath204 , respectively . the water molecules comprise the polarisable background medium and there are no free charges . from our discusion in sec . [ subsec : respot ] , we know that the dipole contribution alone yields a poor approximation to the potential ( figs [ fig : phimulticomb]c and d ) . as a natural extension we considered the quadrupole contribution @xcite , which was also found to be important in simulation studies of interfacial electric fields @xcite . with the inclusion of this contribution , the potentials from the molecular multipole expansions match the exact potentials very well for both methods , respectively . the corresponding expression for the field extends to [ eq : efrompquad ] @xmath205 \\ & = & & \int\limits_{-\frac{l_z}{2}}^{\frac{l_z}{2 } } \mathrm dz'\ g_\text{1d}^{''}(z - z ' ) \left [ { \bar \rho}_{\mu}(z ' ) - { \bar \rho}^{'}_{q}(z ' ) \right ] , \end{aligned}\ ] ] where the derivatives of the kernels are given in appendix [ secapp : electrostatics ] . to get to eq . ( [ eq : efrompquad]b ) we integrated the second term in eq . ( [ eq : efrompquad]a ) by parts taking into account the periodicity . we can solve the above integral analytically for ewald summation and find that @xmath206,\ ] ] where @xmath207 is the box average of @xmath208 . in general , we can identify this contribution with @xmath200 as it corresponds to the ( constant ) field arising from an induced surface charge density at infinity ( tin - foil boundary conditions ) . we refer to refs and for a more general discussion . although the instantaneous value of @xmath200 may fluctuate , we know that its time average vanishes , because our system does not exhibit a net dipole moment ( figs [ fig : multicomb]b and d ) . for ewald summation the definition of polarisation as @xmath209 therefore naturally leads to the correct proportionality of @xmath210 . for the wolf method the relation between electric field and polarisation ( as defined above ) is more complicated , because we can not solve the integral in eq . ( [ eq : efrompquad]b ) analytically . more importantly , we can not expect the short - ranged method to predict fields accurately in general , because its kernel is not a solution of poisson s equation . the estimates for the thermally induced fields might be reasonable , but it is trivial to come up with an example , such as a plate capacitor , for which the method would fail . finally , we would like to discuss the macroscopic maxwell equation ( [ eq : maxwelledp ] ) in the context of the slab expansion . as shown in figs [ fig : phimulticomb]a - b , we can identify all relevant multipole contributions to the potential and recover a good approximation to the exact solution implying overall consistency . due to the nature of the spatial averaging , we obtain a non - vanishing charge density ( fig . [ fig : multicomb]a ) for our inhomogeneous system . this is inconsistent , however , with the derivation of eq . ( [ eq : maxwelledp ] ) , where charges within a molecule are summed first in order to get from the microscopic to the macroscopic description @xcite and the charge density vanishes identically . identification of displacement field and polarisation is therefore not meaningful for the slab expansion . this problem is avoided altogether in the molecule expansion , which is consistent with eq . ( [ eq : maxwelledp ] ) , and we can unambiguously identify all terms in the macroscopic maxwell equation . in this paper we have analysed the electric fields and multipole moments induced by a strong thermal gradient in nemd simulations of water in a setup which was previously studied by armstrong and bresme @xcite . our comparison comprises results for two different treatments of coulomb interactions , namely ewald summation and the short - ranged wolf method . the latter was employed in most of the previous studies on the thermo - polarisation effect @xcite . we identified two key differences to the literature data : firstly , the wolf method fails to reproduce the dipole density correctly for parameters that work well in equilibrium simulations . the molecules point , on average , in opposite directions as compared to ewald summation and the alignment is strongly enhanced . secondly , for both methods the peak field strength is of the order of @xmath184 . however , for the wolf method the result depends sensitively on the employed parameters . for low damping the wolf method slightly overestimates the field obtained with ewald summation and vice versa for high damping . the results are therefore in direct constrast to very recent findings of bresme and co - workers @xcite who reported that the wolf method overestimates the field by an order of magnitude . in fact , we argue that the employed formula for the calculation of the field is incorrect . taking such truncation into account correctly results in comparable results for the electric field . another key result of this paper are the highly improved spatial averages of the potential and the field for low resolutions . we propose to integrate these quantities analytically over the bins rather than calculating them from the time - averaged charge density , as is usually done in the literature . potentials and fields then truly represent the exact spatial averages over the microscopic or macroscopic control volumes . we showed that this procedure is straightforward for both summation methods and requires no computational overhead . comparing the ratio of maximum errors , we found a more than 20-fold reduction of the error for the potential and a 10-fold reduction for the field at the coarsest resolution of @xmath211 . consequently , employing the new method can reduce the simulation time scales by up to two orders of magnitude for the same quality of statistics . the advantage of calculating analytical averages becomes less significant with increasing spatial resolution and both methods are comparable for resolutions of @xmath196 . however , in this case the magnitude of the statistical error is comparable to the signal itself rendering the results meaningless . in addition , we found that accurate estimates of the potential and the field can be obtained by approximating the water molecules as ideal point dipoles and quadrupoles . for low spatial resolutions we found this approach to yield considerably better results than the calculation from the averaged charge density . our detailed comparison of the results for the slab and molecule expansions illustrates that the ratio of the individual contributions depends on the control volume we choose for the expansion . for slabs almost all the information can be recovered by considering the monopole , as is usually done in the literature . however , in the molecule expansion the dipole and the quadrupole contributions are significant and both have to be considered in order to recover results from the exact calculation accurately . finally , taking into account the quadrupole contribution leads to the expected proportionality between the polarisation and the macroscopic maxwell field in accordance with the macroscopic maxwell equations . the wolf method fails to satisfy this relation entirely . based on its shortcomings , we therefore conclude that the method is not suitable for reproducing the electrostatic key quantities in inhomogeneous systems reliably . this is in agreement with the findings of takahashi and co - workers @xcite , who reported poor predictions for the electrostatic potential and dipolar orientations in simulations of the liquid vapour interface , even for cutoff radii almost six times larger than the maximum value considered in this work . the authors should like to dedicate this paper to the memory of simon de leeuw , who was a pioneer in the calculation of coulomb effects in simulations . w. would like to thank the austrian academy of sciences for financial support through a doc fellowship , and for covering the travel expenses for the cecam workshop in zaragoza in may 2015 , where these results were first presented . p. w. would also like to thank chao zhang for pointing out the equivalence of the two expressions for the electric field discussed in sec . [ subsec : resfield ] , michiel sprik for emphasising the importance of the quadrupole contribution in experimental studies of interfacial systems , as well as aleks reinhardt and other members of the frenkel and dellago groups for their advice . we further acknowledge support from the federation of austrian industry ( iv ) carinthia ( p. w. ) , the university of zagreb and erasmus smp ( d. fijan ) , the human frontier science program and emmanuel college ( a. . ) , the austrian science fund fwf within the sfb vicom project f41 ( c. d. ) , and the engineering and physical sciences research council programme grant ep / i001352/1 ( d. f. ) . additional data related to this publication are available at the university of cambridge data repository ( http://dx.doi.org/10.17863/cam.118 ) . our goal is to integrate @xmath40 over the entire cutoff sphere in order to obtain @xmath212 . to this end we have to evaluate the integral @xmath213 \label{eq : gw1db},\end{aligned}\ ] ] where @xmath214 . we first consider the integral @xmath215 and use the substitution @xmath216 to rewrite the expression as @xmath217 using integration by parts it is easy to show that the result is @xmath218 for @xmath219 and zero otherwise . the integration of the second term in eq . ( [ eq : gw1db ] ) is trivial and the averaged kernel is given by @xmath220 for @xmath219 and it vanishes otherwise . the first three derivatives of this function are [ eqapp : gwders ] @xmath221 \nonumber , \end{aligned}\ ] ] respectively . instead of integrating the kernel @xmath41 ( eq . ( [ eq : gesim ] ) ) directly , we replace it by @xmath222 ( eq . ( [ eq : gpbc3d ] ) ) in order to simplify the problem . the sum in eq . ( [ eq : gpbc3d ] ) is only conditionally convergent , which is why we are formally not allowed to change the order of integration and summation . however , if we ignore this fact we arrive at the same result that we would have obtained by considering @xmath41 directly . this yields [ eqapp : ge1d ] @xmath223 in the last step , we make use of the fact that the integration eliminates all terms in the summation for which @xmath224 or @xmath225 . the inverse fourier transform in eq . ( [ eqapp : ge1d]c ) is @xmath226 and the first three derivatives of this expression are given by [ eqapp : geders ] @xmath227 respectively . the aim is to average the one - dimensional kernel analytically for any bin @xmath77 of width @xmath228 to obtain @xmath229 taking into account the periodicity . as mentioned in sec . [ sec : electrostatics ] , in our notation we understand the argument @xmath230 to be mapped back to the interval @xmath231 $ ] implicitly . the interesting case , where the separation of the charge at @xmath232 and the bin covering the interval @xmath233 $ ] is such that periodicity has to be taken into account in the integration , is illustrated in fig . [ fig : g1de ] . the first step is to map the distances from @xmath232 to the bin boundaries back into the reference interval using the function @xmath234 where @xmath235 gives the nearest integral number to @xmath68 . applying this function yields the shortest distances to the nearest images which we label with [ eqapp : pbczjl ] @xmath236 respectively . for the case shown in fig . [ fig : g1de ] , where @xmath237 , we can split the original expression into the two integrals @xmath238.\ ] ] in order to simplify the integration further , we focus on the case of ewald summation . application of the procedure to the wolf method is omitted for brevity , because the integration is tedious . we know that the average of @xmath191 over the reference interval vanishes because the term corresponding to @xmath239 in eq . ( [ eqapp : ge1d]c ) is absent . therefore , the special case shown in fig . [ fig : g1de ] reduces to the ordinary case @xmath240 in which the entire bin is located inside the reference box . all possible scenarios are therefore taken into account by straightforward integration of eq . ( [ eqapp : ge1dreal ] ) , which yields @xmath241 likewise , we find @xmath242 for the average of the derivative . equations ( [ eqapp : g1danaavgres ] ) and ( [ eqapp : g1deranaavgres ] ) along with eqs ( [ eqapp : pbczjl]a b ) can be substituted into eqs ( [ eq : phi_ex_avg ] ) and ( [ eq : ez_ex_avg ] ) to calculate the exact averages of the potential and the field , respectively . and the bin @xmath77 covering the interval @xmath233 $ ] is such that periodicity has to be taken into account . @xmath243 and @xmath244 are the nearest images of the bin boundaries . ] in this section , we compare the oxygen - oxygen pair correlation function , @xmath245 , the oxygen - oxygen velocity autocorrelation function , @xmath246 , a cumulative estimate of the dielectric constant , @xmath247 , and the distance - dependent kirkwood @xmath136-factor , @xmath248 . we refer to refs and for a detailed discussion and the relevant formulae . all quantities were sampled during 2 _ nve _ simulations before imposing the temperature gradients . the results are shown in figs [ fig : validcomb]a d . as we can see , all sets of parameters lead to excellent agreement for @xmath245 and @xmath246 ( figs [ fig : validcomb]a b ) . the dielectric constant ( fig . [ fig : validcomb]c ) is well reproduced by the wolf method with strong damping , whereas weak damping leads to an overestimation . more insight about the structural properties can be gained by looking at @xmath248 in fig . [ fig : validcomb]d . for very short distances both sets of parameters for the wolf method yield a good agreement with ewald summation . we note that for the weak damping the agreement extends a bit further than for strong damping , which is consistent with our observations for the model system . we also note that the shape of @xmath248 looks different for our elongated box as compared to a cubic box . 60ifxundefined [ 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 10.1126/science.2309118 [ * * , ( ) ] link:\doibase 10.1016/s1748 - 0132(07)70017 - 8 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.020602 [ * * , ( ) ] link:\doibase 10.1039/c1cp21895f [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.105901 [ * * , ( ) ] link:\doibase 10.1063/1.4811291 [ * * , ( ) ] link:\doibase 10.1021/jp408485d [ * * , ( ) ] @noop _ _ , ed . ( , , ) @noop * * , ( ) link:\doibase 10.1063/1.2206581 [ * * , ( ) ] link:\doibase 10.1063/1.478738 [ * * , ( ) ] link:\doibase 10.1063/1.4927229 [ * * , ( ) ] link:\doibase 10.1002/jcc.540150702 [ * * , ( ) ] link:\doibase 10.1021/jp025949h [ * * , ( ) ] link:\doibase 10.1063/1.1836733 [ * * , ( ) ] link:\doibase 10.1063/1.4898147 [ * * , ( ) ] link:\doibase 10.1063/1.3582791 [ * * , ( ) ] link:\doibase 10.1063/1.4827055 [ * * , ( ) ] link:\doibase 10.1063/1.4896627 [ * * , ( ) ] link:\doibase 10.1073/pnas.0600282103 [ * * , ( ) ] link:\doibase 10.1021/jp510612w [ * * , ( ) ] link:\doibase 10.1063/1.464397 [ * * , ( ) ] link:\doibase 10.1063/1.477414 [ * * , ( ) ] link:\doibase 10.1080/00268978000100361 [ * * , ( ) ] link:\doibase 10.1021/bi00140a022 [ * * , ( ) ] link:\doibase 10.1021/jp9614658 [ * * , ( ) ] link:\doibase 10.1073/pnas.0807623105 [ * * , ( ) ] link:\doibase 10.1063/1.474295 [ * * , ( ) ] link:\doibase 10.1021/ct0502256 [ * * , ( ) ] link:\doibase 10.1021/cr300461d [ * * , ( ) ] link:\doibase 10.1063/1.3670965 [ * * , ( ) ] link:\doibase 10.1063/1.2948951 [ * * , ( ) ] link:\doibase 10.1063/1.3578473 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/20/49/494206 [ * * , ( ) ] link:\doibase 10.1098/rspa.1980.0135 [ * * , ( ) ] link:\doibase 10.1080/00268978400101081 [ * * , ( ) ] link:\doibase 10.1063/1.2218840 [ * * , ( ) ] link:\doibase 10.1063/1.453923 [ * * , ( ) ] link:\doibase 10.1063/1.479595 [ * * , ( ) ] link:\doibase 10.1016/0013 - 4686(96)00046 - 1 [ * * , ( ) ] @noop _ _ , ed . ( , , ) link:\doibase 10.1063/1.1301522 [ * * , ( ) ] link:\doibase 10.1063/1.476834 [ * * , ( ) ] link:\doibase 10.1021/j100308a038 [ * * , ( ) ] link:\doibase 10.1063/1.453919 [ * * , ( ) ] @noop * * , ( ) @noop _ _ , ed . ( , ) link:\doibase 10.1016/j.jssc.2012.05.010 [ * * , ( ) ] http://www.sciencedirect.com/science/article/pii/s002199918571039x [ * * , ( ) ] link:\doibase 10.1063/1.4931597 [ * * , ( ) ] link:\doibase 10.1063/1.442716 [ * * , ( ) ] link:\doibase 10.1063/1.447334 [ * * , ( ) ] link:\doibase 10.1103/physreva.31.1695 [ * * , ( ) ] link:\doibase 10.1063/1.3548836 [ * * , ( ) ] link:\doibase 10.1080/00268978600100081 [ * * , ( ) ] @noop _ _ , ed . ( , ) link:\doibase 10.1080/002689797169916 [ * * , ( ) ] link:\doibase 10.1038/nphys1185 [ * * , ( ) ] link:\doibase 10.1103/physrevb.93.144201 [ * * , ( ) ] link:\doibase 10.1063/1.451198 [ * * , ( ) ]

collapse of massive stars may result in formation of accreting black holes in their interior . the accreting stellar matter may advect substantial magnetic flux onto the black hole and promote release of its rotational energy via magnetic stresses ( the blandford - znajek mechanism ) . in this paper we explore whether this process can explain the stellar explosions and relativistic jets associated with long gamma - ray - bursts . in particularly , we show that the blandford - znajek mechanism is activated when the rest mass - energy density of matter drops below the energy density of magnetic field in the very vicinity of the black hole ( within its ergosphere ) . we also discuss whether such a strong magnetic field is in conflict with the rapid rotation of stellar core required in the collapsar model and suggest that the conflict can be avoided if the progenitor star is a component of close binary . in this case the stellar rotation can be sustained via spin - orbital interaction . in an alternative scenario the magnetic field is generated in the accretion disk but in this case the magnetic flux through the black hole ergosphere is not expected to be sufficiently high to explain the energetics of hypernovae by the bz mechanism alone . however , this energy deficit can be recovered via additional power provided by the disk . black hole physics supernovae : general gamma - rays : bursts methods : numerical mhd relativity

0902.2881

the most popular model for the central engines of long gamma - ray - burst ( grb ) jets is based on the `` failed supernova '' scenario of stellar collapse , or `` collapsar '' , where the iron core of progenitor star forms a bh @xcite . if the specific angular momentum in the equatorial part of progenitor exceeds that of the marginally bound orbit of the bh then the collapse becomes highly anisotropic . while in the polar region it may proceed more or less uninhibited , at least for a while , the equatorial layers form dense and massive accretion disk . the gravitational energy released in this disk can be very large , more then sufficient to overturn the collapse of outer layers and drive grb outflows @xcite . a similar configuration can be produced via inspiraling of a bh or a neutron star into the companion star during the common envelope phase of a close binary @xcite . in any case , the observed duration of long grbs , between 2 and 1000 seconds , imposes a strong constraint on the size of progenitors - their light crossing time should be significantly shorter then the burst duration . this implies that the progenitors must be compact stars stripped of their hydrogen envelopes . this conclusion agrees with the absence of hydrogen lines in the spectra of supernovae identified with grbs ( moreover , since some of these supernovae are of type ic their progenitors may have lost their helium envelopes as well . ) . massive solitary stars can lose their envelopes via strong stellar winds . however , this may lead to unacceptably low rotation rates by the time of stellar collapse @xcite . in the case of binary system the loss of envelope can be caused by the gravitational interaction with companion . the currently most popular mechanism of powering grb jets is the heating via annihilation of neutrinos and anti - neutrinos produced in the disk @xcite . the energy deposited in this way is a strong function of the mass accretion rate which must be above @xmath0 in order to agree with the observational constraints on the energetics of grbs @xcite . such high accretion rates can be provided in the collapsar scenario and the numerical simulations by @xcite and @xcite have demonstrated that sufficiently large energy deposition in the polar region above the disk may indeed result in fast collimated jets . however , we have to wait for simulations with proper implementation of neutrino transport before making final conclusion on the suitability of this model the long and complicated history of numerical studies of neutrino - driven supernova explosions teaches us to be cautious . the first attempt to include neutrino transport in collapsar simulations was made by @xcite and they did not see neutrino - driven polar jets - the heating due to neutrino annihilation could not overcome the cooling due to neutrino emission.s , whereas the jet simulation of @xcite started at @xmath1s when the density in the polar regions above the black hole has dropped down to @xmath2 leading to much lower cooling rates . ] in any case , it is unlikely that this model can explain the bursts that are longer than @xmath3s as by this time the mass accretion rate is expected to drop significantly . for example the accretion rate in the simulations by @xcite has already drifted below the critical @xmath4 by @xmath5s . the two alternatives to the neutrino mechanism which are also frequently mentioned in connection with grbs are the magnetic braking of the accretion disk @xcite and the magnetic braking of the central black hole @xcite . a number of groups have studied the potential of magnetic mechanism in the collapsar scenario using newtonian mhd codes and implementing the paczynski - witta potential in order to approximate the gravitational field of central bh @xcite . in this approach it is impossible to capture the blandford - znajek effect and only the magnetic braking of accretion disk can be investigated . the general conclusion of these studies is that the accretion disk can launch magnetically - driven jets provided the magnetic field in the progenitor core is sufficiently strong . this field is further amplified in the disk , partly due to simple winding of poloidal component and partly due to the magneto - rotational instability ( mri ) , until the magnetic pressure becomes very large and pushes out the surface layers of the disk . in particular , @xcite studied the collapse of a star with a similar structure to that considered @xcite and with purely monopole initial magnetic field of strength @xmath6 g at @xmath7 . the corresponding total magnetic flux , @xmath8 , is in fact within the observational constraints on the field of magnetic stars , including white dwarfs @xcite , ap - stars @xcite and massive o - stars @xcite . they used realistic equation of state and included the neutrino cooling but not the heating . the simulations lasted up to @xmath9s in physical time during which a poynting - dominated polar jet has developed in the solution . the jet power seemed to show strong systematic decline , from @xmath10 down to @xmath11 or even less , which might signal transient phenomenon . @xcite studied the collapse of rotating stellar cores and formation of bhs and their disks in the collapsar scenario using the full grmhd approximation . their simulations show powerful explosions soon after the accretion disk is formed and the free falling plasma of collapsing star begins to collide with this disk . however , they did not account for the neutrino cooling and the energy losses due to photo - dissociation of atomic nuclei so these explosions could be similar in nature to the `` successful '' prompt explosions of early supernova simulations @xcite . @xcite carried out grmhd simulations in the time - independent space - time of a central bh . their computational domain did not capture the bh ergosphere and thus they could not study the role of the blandford - znajek effect @xcite . in addition , the energy gains / losses due to the neutrino heating / cooling were not included and the equation of state ( eos ) was a simple polytrope . the simulations run for a rather short time , @xmath12 where @xmath13 , and jets were formed almost immediately , presumably due to the extremely strong initial magnetic field . the original theory of the blandford - znajek mechanism was developed within the framework of magnetodynamics ( the singular limit of rmhd approximation where the inertia of plasma particles is fully ignored ) . the pioneering numerical studies of this mechanism have shown that it can also operate within the full rmhd approximation , at least in the regime where the energy density of matter is much smaller compared to that of the electromagnetic field , and that it can be captured by modern numerical techniques @xcite . however , no systematic numerical study has been carried out in order to establish the exact conditions for activation of the bz mechanism . the first simulations demonstrating the potential of the bz mechanism in the collapsar scenario were carried out by @xcite who also used simple polytropic eos and considered purely adiabatic case , but captured the effects of black hole ergosphere . the initial configuration included a torus with poloidal magnetic and a free - falling stellar envelope . the results show ultrarelativistic jets with the lorentz factor up to 10 emerging from the black hole indicating that the blandford - znajek effect may play a key role in the production of grb jets . @xcite added a realistic eos and included the energy losses due to neutrino emission ( assuming optically thin regime ) and photo - dissociation of nuclei . they also reported the development of relativistic outflows powered by the central black hole via the blandford - znajek mechanism ( the magnetic braking of accretion disk did not seem to contribute much to the jet power though ) . the black hole - driven magnetic stellar explosions and relativistic jets observed in these exploratory computer simulations invite a deeper analysis of this model . obviously , such an outcome can not be a generic feature of core - collapse supernovae . indeed , only a very small fraction of sne ib / ic seem to produce grbs @xcite . so what are the relevant conditions and can they be reproduced as the results of stellar evolution ? these are the main issues we address in this paper . the rotational energy of kerr black hole is @xmath14 where @xmath15^{1/2},\ ] ] @xmath16 is the bh mass and @xmath17 is its dimensionless rotation parameter . for @xmath18 and @xmath19 this gives the enormous value of @xmath20erg which is about fifty times higher than the rotational energy of a millisecond pulsar and well above of what is needed to explain the energy of grbs and associated hypernovae . even for a relatively slowly rotating black hole with @xmath21 this is still a respectable number , @xmath22erg . moreover , because in all versions of the collapsar model the black hole spin is aligned with the spin of accretion disk this energy reserver is continuously replenished via accretion . thus , as far as the availability of energy is concerned the black hole model looks very promising indeed . the energy release rate is usually estimated using the blandford - znajek power for the case of force - free monopole magnetosphere @xmath23 where @xmath24 is the angular velocity of the bh and @xmath25 is the magnetic flux threading one hemisphere of the bh horizon ( here it is assumed that the angular velocity of the bh magnetosphere @xmath26 . ) . this formula is quite accurate not only for slowly rotating black holes considered in @xcite but also for rapidly rotating bhs @xcite . in application to the collapsar problem it gives us the following estimate @xmath27 where @xmath28 and @xmath29 . one can see that the power of bz - mechanism is rather sensitive to black hole s mass and magnetic flux . since , the black hole mass is likely to be @xmath30 the observed energetics of hypernovae and long grbs requires @xmath31 . this value is comparable with the maximal surface flux observed in magnetic stars , ap stars , magnetic white dwarfs , and magnetars ( e.g. * ? ? ? * ) . thus , the magnetic field of grb central engine may well be the original field of the progenitor star . the estimates ( [ e - rot],[e - bz ] ) show that braking of black holes alone can explain the energetics of grbs and this is why this mechanism is often mentioned in the literature on grbs . however , there is another issue to take into consideration . the equation ( [ e - bz0 ] ) is obtained in the limit where the inertia of magnetospheric plasma and , to large degree , its gravitational attraction towards the black hole are ignored . in contrast , the mass density of plasma in the collapsing star may be rather high and has to be taken into account . for example , the magnetohydrodynamic waves may become trapped in the accretion flow and unable to propagate outwards . in such a case , it would not be possible to extract magnetically any of the black hole rotational energy and drive stellar explosion irrespectively of how high this energy is . since the bz mechanism can only operate within the black hole ergosphere ( e.g. * ? ? ? * ; * ? ? ? * ) , the magnetohydrodynamic waves must at least be able to cross the ergosphere in the outward direction for its activation . this agrees with the conclusion reached in @xcite that the alfven surface of a steady - state ingoing wind must be located inside the ergosphere to have outgoing direction of the total energy flux . thus we should consider the following condition for activation of the bz - mechanism in accretion flows : the alfven speed has to exceed the local free fall speed at the ergosphere , @xmath32 in order to simplify calculations let us apply the newtonian expressions for both speeds . the alfven speed is @xmath33 where @xmath34 is the magnetic field strength and @xmath35 is the plasma mass density , whereas the free - fall speed is @xmath36 at @xmath37 the local criticality condition reads @xmath38 that is the energy density of magnetic field has to exceed that of plasma in the vicinity the black hole . for spherically symmetric flows the condition @xmath39 can be written as a constraint on the mass accretion rate , @xmath40 , and the flux of radial magnetic field , @xmath41 , @xmath42 since @xmath43 this condition is bound not to be satisfied at large @xmath44 but we only need to apply it at the ergosphere . using @xmath45 we then obtain @xmath46 where @xmath47 and @xmath48 . in fact , the critical value of @xmath49 must depend on the black hole spin and tend to infinity for small @xmath50 . the newtonian analysis can not capture this effect . the relativistic one , on the other hand , appears rather complicated @xcite . fortunately , nowadays this issue can be investigated via numerical simulations . another , interesting and important issue for time - dependent simulations is whether there is only one bifurcation separating solutions with switched - on and switched - off bz mechanism or the transition is reacher and allows more complicated time - dependent solutions , e.g. quasi - periodic or chaotic ones . the first type of simulations presented in this paper are designed to test the validity of our arguments behind the criterion for activation of bz - mechanism which we derived in sec.[bz - theory ] using a combination of newtonian and relativistic physics . for this purpose we consider a more or less spherical accretion of cold magnetized plasma with vanishing angular momentum onto a rotating black hole . in order to focus on the mhd aspects of the problem we ignore the microphysics important in the collapsar problem and consider plasma with simplified polytropic eos . the main details of our numerical method and various test simulations are described in @xcite . the only really new feature here is the introduction of hll - solver which is activated when our linear riemann solver fails . this usually occurs in regions of relativistically high magnetization where the magnetic energy density exceeds that of matter by more than one order of magnitude or in strong rarefactions . this makes the scheme a little bit more robust though we are still forced to use a `` density floor '' , that is a lower limit on the value of mass - energy density of matter . both in the numerical scheme and throughout this section of the paper we utilize geometric units where @xmath51 . the kerr spacetime is described using the kerr - schild coordinates so the metric form reads @xmath52 where @xmath53 @xmath54 @xmath55 and @xmath56 the initial solution has to describe both the flow and the magnetic field . in this case the flow is given by the steady - state solution for unmagnetised cold plasma ( dust ) . at infinity it is spherically symmetric and has vanishing specific angular momentum and radial velocity . the components of 4-velocity of this flow in the coordinate basis of kerr - schild coordinates , @xmath57 , are @xmath58 where @xmath59 and its rest mass density is @xmath60 where @xmath61 is the rest mass density at the outer event horizon radius ( see appendix [ a1 ] ) . the initial magnetic field is radial with monopole topology @xmath62 the azimuthal component is introduced in order to ensure that the electromagnetic fluxes of energy and angular momentum vanish ( see appendix [ a2 ] ) and , thus , the initial magnetic configuration is more or less dynamically passive . the two - dimensional axisymmetric computational domain is @xmath63 , where @xmath64 and @xmath65 . notice that the inner boundary is inside of the outer event horizon , @xmath66 . this can be done because of the non - singular nature of the horizon in the kerr - schild coordinates . the computational grid is uniform in @xmath67 , where it has 100 cells , and almost uniform in @xmath68 , where it has 170 cells . the metric cell sizes are the same in both directions . the grid is divided into rings such that for any ring the metric size of its outer cells is twice that of its inner cells . the solution on each ring is advanced with its own time step , the outer ring time step being twice that of the inner ring . at the interface separating cells of different rings the conservation of mass , energy , and momentum and magnetic flux is ensured via simple averaging procedure . at the inner boundary , @xmath69 , we impose the `` non - reflective '' free - flow boundary conditions . the use of these conditions is justified by the fact that this boundary lies inside the outer event horizon in the region where all physical waves move inwards . at the outer boundary , @xmath70 we fix the flow parameters corresponding to the initial solution . in all runs the simulations are terminated well before any strong perturbation generated near the centre reaches this boundary . at the polar axis , @xmath71 , the boundary conditions are dictated by the axial symmetry . namely , quantities that do not vanish on the axis are reflected without change of sign , whereas quantities that vanish on the axis are reflected with change of sign . we consider models which differ only by the rotation rate of black hole , @xmath50 , and the mass accretion rate , @xmath40 . for each value of @xmath50 , we first deal with the model with @xmath72 . depending on whether the bz - mechanism is activated in this model or not we either increase or reduce @xmath61 ( and hence @xmath40 ) by the factor of three and compute another model . we proceed in this fashion until the solution becomes qualitatively different . then we compute one more model based on the mean value of @xmath61 for the previous two models . after that we select the closest two models with qualitatively different solutions and show them on the bifurcation diagram summarising the results of our study . the simulations do confirm the anticipated bifurcation and show that for a wide of parameter @xmath50 the critical value of @xmath49 is indeed close to unity . for example , the model with @xmath19 and @xmath73 is qualitatively different from the model the model with @xmath19 and @xmath74 . the first model quickly settles to a steady state accretion solution ( see the top panels of fig.[sp - outer ] and fig.[sp - inner ] ) . in fact , on the large - scales , this solution is almost indistinguishable from the initial solution . on the small scales , near the event horizon , the differences become more pronounced . in particular , a stationary accretion shock appears just outside of the ergosphere . it is responsible for the jump in the rest mass density and the kinks of magnetic field lines seen in the top left panel of fig.[sp - inner ] . inside the ergosphere the magnetization parameter @xmath75 is larger than unity but only just . the magnetic field lines show no rotation and thus the bz - mechanism remains completely switched - off . in contrast , the model with @xmath19 and @xmath74 exhibits powerful bipolar outflow which drives a blast wave into the accreting flow ( see the middle panel of fig.[sp - outer ] ) . this blast waves overturns the accretion both in the polar and in the equatorial direction . the black hole develops highly rarefied magnetosphere ( see the middle panels of fig.[sp - inner ] ) which rotates with about half of black hole s angular velocity . the shocked plasma of accretion flow is prevented from accretion onto the black hole instead it participates in the large scale circulations above and below the equatorial plane . the unphysical monopole configuration of magnetic field set for this model prevents any escape of magnetic flux from the black hole . this ensures operation of the bz - mechanism even if the confining accreting envelope is fully dispersed into the surrounding space . for a more realistic dipolar configuration one would expect at least partial escape of magnetic flux once the outflow is developed and thus less power in the outflow . however , this is unlikely to effect the value of @xmath76 . indeed , consider the case with monopole field where the bz - mechanism remains switched off . then , change the polarity of magnetic field lines in the northern hemi - sphere . because the speeds of mhd waves and the magnetic stresses remain invariant under change of magnetic polarity this will have no effect on the flow . for the very same reason , we expect activation of the bz mechanism for exactly the same value of @xmath49 in both magnetic configurations . in order to verify this conclusion we computed models with the initial magnetic field of split - monopole topology and the results fully agree with the expectations the critical value of @xmath49 remains unaffected but the power of bz - mechanism is reduced due partial escape of magnetic flux from the black hole . the bottom panel of fig.[sp - outer ] shows the large scale flow developed in the model with @xmath19 , @xmath74 , and split - monopole initial magnetic field . the flow pattern is very similar to that of the model with monopole field but the expansion rate is noticeable lower , approximately by the factor of 1.8 in the polar direction and even more in the equatorial direction . the bottom panels of fig.[sp - inner ] show the inner region of this model . one can see that the accretion is no longer fully halted the accreting plasma finds its way into the black hole in the equatorial zone and the black hole magnetosphere is confined only to the polar region of the funnel shape . the results for @xmath19 show that not only there is a bifurcation between the regimes with switched - on and switched - off bz process but also that the critical value of the control parameter @xmath49 , and hence @xmath77 , is indeed close to unity at least for rapidly rotating black holes . the results of our parametric study , summarised in fig.[kappa ] , show that in fact @xmath78 is a relatively weak function of @xmath50 . it is confined within the range @xmath79 for @xmath80 . which still show activation of the bz mechanism . the circles show models with the highest value of @xmath49 where the bz mechanism remains turned off . the pair of points with @xmath19 in the right bottom corner of the diagram correspond to proper collapsar simulations . the other points show the results for test simulations . , width=340 ] the test simulations described in sec.[test - sim ] allow us to determine the basic criteria for activation of the bz mechanism in principle . however , their set - up is rather artificial as in many astrophysical applications the accreting plasma has sufficient angular momentum to ensure formation of accretion disk and thus to render the approximation of spherical accretion unsuitable . this anisotropy of mass inflow is likely to make the integral activation condition ( [ cond-2 ] ) less helpful . on the other hand , one would expect the local condition condition ( [ cond-1 ] ) to be more or less robust . in order to check this we analysed the results obtained in our current numerical study of magnetic collapsar model . like in the test simulations we use the kerr metric in kerr - schild coordinates , @xmath81 . we selected @xmath82 and the rather optimistic value of @xmath19 for the central black hole . the two - dimensional axisymmetric computational domain is @xmath63 , with @xmath83 km and @xmath84 km . the computational grid is of the same type as in the test simulations ( sec.[setup-1 ] ) but with more cells , @xmath85 . the boundary conditions are also the same as in the test simulations . for these simulations we use realistic equation of state ( eos ) that takes into account the contributions from radiation , lepton gas including pair plasma , and non - degenerate nuclei ( hydrogen , helium , and oxygen ) . this is achieved via incorporation of the eos code helm @xcite , which can be downloaded from the web - site http://www.cococubed.com/code_pages/eos.shtml . the neutrino cooling is computed assuming optically thin regime and takes into account urca - processes @xcite , pair annihilation , photo - production , and plasma emission @xcite , as well as synchrotron neutrino emission @xcite . in fact , urca - processes strongly dominate over other mechanisms in this problem . photo - disintegration of nuclei is included via modification of eos following the prescription given in @xcite . the equation for mass fraction of free nucleons is adopted from @xcite . we have not included the radiative heating due to annihilation of neutrinos and antineutrinos produced in the accretion disk mainly because this requires elaborate and time consuming calculations of neutrino transport . the collapsing star is introduced using the simple free - fall model by @xcite . in this model it is assumed that immediately before the collapse the mass distribution in the star satisfies the law @xmath86const and that the infall proceeds with the free - fall speed in the gravitational field of the central black hole . this gives us the following radial velocity @xmath87 and mass density @xmath88 for the collapsing star , where @xmath89 is a constant that depends on the star mass @xcite and @xmath90 is the time since the start of collapse . the corresponding accretion rate and ram pressure are @xmath91 @xmath92 respectively . for a core of radius @xmath93 cm and mass 3.0@xmath94 the collapse duration can be estimated as @xmath95s . since , in this study we explore the possibility of early explosion , that is soon after the core collapse , we set @xmath96s . because grbs are currently associated with more massive progenitors we consider @xmath97 . this gives us the accretion rates of @xmath98 and @xmath99 respectively . on top of this we endow the free - falling plasma with angular momentum and poloidal magnetic field . the angular momentum distribution is @xmath100 where @xmath101 km and @xmath102 . , however , by mistake an additional factor , @xmath103 , was introduced in the start - up models ( the same mistake was made in the simulations described in @xcite . ) although somewhat embarrassing , this does not affect the main results of this study . ] the origin of magnetic field is probably the most important and difficult issue of the magnetic model of grbs . one possibility is that this field already exists in the progenitor and during the collapse it is simply advected onto the black hole . the recent numerical studies indicate that the poloidal and the azimuthal components of the progenitor field should have similar strengths @xcite . the magnetic flux freezing argument then suggests that during the collapse the poloidal field becomes dominant and thus in our simulations we may ignore the initial azimuthal field . as to the initial poloidal component we adopt the solution for a uniformly magnetized sphere in vacuum . to be more specific , we assume that inside the sphere of radius @xmath104 km the magnetic field is uniform and has strength of either @xmath105 or @xmath106 g , whereas outside it is described by the solution for magnetic dipole . the selected values for the field strength allow to capture the bz bifurcation . given the limitations of axisymmetry we can only consider the case of aligned magnetic and rotational axes . table [ model - tab ] describes the key parameters of six models investigated in this study and also summarises the outcome of simulations . one of the parameters , @xmath107 , is the total magnetic flux enclosed by the equator of the uniformly magnetized sphere in the initial solution . obviously , this gives us the highest magnetic flux that can be accumulated by the black hole in the simulations . whether this maximum value is actually reached at some point depends on the progenitor rotation . for slow rotation the whole of the uniform magnetized core can be swallowed by the hole before the development of accretion disk . at this point @xmath25 equals to @xmath107 and then it begins to decline as the result of advection of the oppositely directed parts of dipolar loops ( these loops are clearly seen in fig.[after ] ) . for fast rotation the accretion disk forms before @xmath25 reaches @xmath107 and after this point the growth of @xmath25 slows down noticeably as the disk accretion is much slower compared to the free - fall accretion . once the oppositely directed parts of magnetic loops begin to cross the accretion shock and enter the turbulent domain that surrounds the disk the annihilation of poloidal magnetic field in this domain becomes an important factor . as the result the maximum value of magnetic flux available in the solution begins to decrease before @xmath25 reaches @xmath107 . .summary of numerical models in collapsar simulations . @xmath108 is the initial magnetic field strength in units of @xmath109 g , @xmath89 is the density parameter in eq.[b2 ] , @xmath110 is the maximum value of magnetic flux that can be accumulated by the black hole in units of @xmath111 , @xmath112 is the value of @xmath49 corresponding to @xmath110 , `` exp '' is the explosion indicator , @xmath113 is the actual magnetic flux accumulated by the black hole by the time of explosion , @xmath114 is the mean power of the bz mechanism in units of @xmath115erg / s . [ cols="^,^,^,^,^,^,^,^",options="header " , ] @xmath112 is the value of parameter @xmath49 computed using @xmath107 instead of the actual value of black hole flux in eq.[cond-2 ] . quick inspection shows that the critical value of @xmath49 based on the values of @xmath112 is about @xmath116 ( see models c and d in table [ model - tab ] . ) , almost five time lower than that in the test simulations of spherical collapse . in fact , the actual value of magnetic flux accumulated by the black hole , @xmath117 , by the time of explosion is always noticeably lower than @xmath112 ( see table [ model - tab ] ) . in model c it is 2.4 times lower thus driving the actual critical value of @xmath49 down to the value around 0.1 . this result is undoubtedly connected to the highly anisotropic nature of the flow in the close vicinity of the black hole that develops in the simulations . most importantly , there is a clear separation of the mass and magnetic fluxes : while a large fraction of mass is accreted through the accretion disk , the bulk of magnetic field lines connected to the black hole occupy the space above ( and below ) the disc ( see fig.[before ] ) . all models considered in this study first pass through the passive phase of almost spherical accretion . during this phase the black hole accumulates magnetic flux but as @xmath107 is not sufficiently high the blandford - znajek mechanism remains switched - off . then , as the angular momentum of plasma approaching the black hole exceeds that of the marginally bound orbit , the accretion disk forms in the equatorial plane and an accretion shock separates from its surface . this shock is best seen in the density plot of fig.[before ] where it appears as a sharp circular boundary between the light blue and dark blue areas of radius @xmath118 km . in fact , the accretion shock oscillates in a manner reminding the oscillations found earlier in supernova simulations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , as well as in simulations of accretion flows onto black holes @xcite , and attributed to the so - called stationary accretion shock instability ( sasi ) . most likely , the physical origin of shock oscillations in our simulations is the same but we have not explored this issue in details . during this second phase the differential rotation results in winding up of the magnetic field in the disk and its corona where the azimuthal component of magnetic field begins to dominate ( top - right and bottom - left panels of fig.[before ] ) . the corona acts as a `` magnetic shield '' deflecting the plasma coming at intermediate polar angles towards either the equatorial plane or the polar axis ( see fig.[before ] ) . in models d , e and f the second phase continues till the end of run and the blandford - znajek process remains switch - off permanently . there are no clear indications suggesting that this phase may end soon although the accretion disk grows monotonically in radius and this may eventually bring about a bifurcation . in models a , b , c after several oscillations the solution enters the third phase which is characterized by a non - stop expansion of the accretion shock which then turns into a blast wave . during this transition the black hole develops a low mass density magnetosphere which occupies the axial funnel through which the rotational energy of the black hole is transported away away in the form of poynting flux and powers the blast wave . the typical large - scale structure of the the solution during the third phase is shown in fig.[after ] ( see also @xcite ) . it is rather similar to the structure observed in the simulations of spherical accretion in the case of split - monopole magnetic field ( see sec.[results-1 ] ) . figure [ before ] shows the solution for model b at time @xmath119ms , which is close to the end of the second phase , and helps to understand how exactly the bz mechanism is activated in our simulations . at this stage , the accretion disk is well developed and accounts for @xmath120 of mass flux through the event horizon whereas most of the magnetic field lines threading the horizon avoid both the disk and corona that surrounds the disk and contains predominantly azimuthal magnetic field . the strong turbulent motion in the disk and its corona must be the reason for this expulsion of magnetic flux @xcite . in any case , the observed separation of magnetic and mass fluxes results in a rather inhomogeneous magnetization of plasma near the black hole . most importantly , there are regions that have much higher local magnetization compared to the case of spherical accretion with the same value of integral control parameter @xmath49 . moreover , the magnetic shield action of the disk corona can promote further local increase of magnetization . indeed , the corona can deflect plasma passing through the accretion shock at intermediate latitudes in the direction which does not necessarily coincide with the local direction of the magnetic field . for example , in figure [ before ] the stagnation point of post - shock flow in the upper hemisphere is located at @xmath121 whereas the separation between the magnetic field lines connected to the hole and to the disk occurs at much lower latitude , around @xmath122 . thus , the direct mass loading of the field lines entering the shock between these points with is currently suspended . on the other hand , at the other end of these field lines , where they penetrate the event horizon , their unloading continues as usual . as the result , the magnetization along these lines increases allowing to reach the local criticality condition , @xmath123 ( see the intense blue regions in the left panel of fig.[before ] ) . once this condition is satisfied the bz mechanism turns on , at this point only very locally and the initial rotation rate of the field lines is significantly lower then @xmath124 . however , an additional magnetic energy is now being pumped into the shield via the bz mechanism promoting its expansion and increasing its efficiency as a flow deflector . this causes further increase of the magnetization at its base , further increase of magnetosphere s rotation rate and thus higher bz power and so on . the process develops in a runaway fashion . the colour image gives @xmath125 and the contours show the magnetic field lines . , width=275 ] figure [ loc - cond ] shows the magnetization parameter @xmath77 at @xmath126 as a function of the polar angle for model b ( at the same time as in fig.[before ] ) and for model d at a similar time . one can see that for model d the magnetization is lower . in fact , the local criticality condition , @xmath127 , is not satisfied anywhere for this model and this explains why the explosion is not produced in this model . ] . at @xmath126 for models b at @xmath119ms ( solid line ) and d at @xmath128ms ( dashed line ) . , width=340 ] there seems to be only two possible origins for the magnetic field in the collapsar problem . either this is a fossil field of the progenitor star of the field generated in the accretion disk and its corona . here we consider both possibilities . the blunt application of the integral condition ( [ cond-2 ] ) for activation of the blandford - znajek mechanism gives us the critical value of magnetic flux accumulated by the black hole @xmath129 although the strength of magnetic field in the interior of massive stars which end up as failed supernovae is not known this is uncomfortably high compared to fluxes deduced from the observations of `` magnetic stars '' and rises doubts about the fossil hypothesis . however , the actual values of the critical magnetic flux found in our simulations are significantly lower . as we have already commented in sec.[coll - sim ] this is due to the anisotropy of accretion in the collapsar model . this anisotropy is expected to increase with time as more distant parts of the progenitor with higher angular momentum enter the accretion shock ( we could not study this late phase of the collapse , @xmath130s , because of various technical issues . ) . due to the higher angular momentum the predominant motion downstream of the shock will be towards the equator and this should result in ( i ) further reduction of the mass flux and the ram pressure of accreting plasma in the polar region and ( ii ) more effective mass - unloading of magnetic field lines entering the black hole ( see sec.[results-2 ] ) . both effects promote higher plasma magnetization in the polar region around the black hole which may eventually exceed the local criticality condition ( [ cond-1 ] ) even in models with @xmath131 , making the fossil hypothesis more plausible . our results seem at odds with the newtonian simulations by @xcite who obtain magnetically driven explosions powered by the disk wind already within @xmath132ms in physical time since the start of simulations for a model with similar mass accretion rate , @xmath133s , but much lower magnetic field , @xmath134 . since their initial model is similar to that of @xcite the difference can not be attributed to the development of highly rarefied polar channel seen in the simulations of @xcite as this occurs only several seconds later . other possible reasons include differences in the framework , details of the initial solutions , resolution , implementation of microphysics etc . hopefully , simulations by other groups will soon clarify this issue . another factor that is likely to reduce the constraint on the strength of the fossil field , and the one that is completely ignored in our simulations , is the plasma heating due to neutrino - antineutrino annihilation . in the most extreme scenario this heating completely overpowers accretion in the polar direction and drives a relativistic grb jet @xcite . although this is an attractive model of grb explosions by itself we also note that the creation of low density funnel in this model helps to activate the bz mechanism along the magnetic field lines that are advected inside this funnel . thus , the neutrino heating and bz mechanism may work hand in hand in driving grbs and hypernovae . even in failed supernovae the stellar core collapse does not lead directly to formation of a black hole but still proceeds through the proto - neutron star ( pns ) phase first . if the period of pns is very short , around several millisecons , its rotational energy is sufficient to strongly modify the direction of collapse and the initial conditions for the black hole phase . in particular , the magnetic stresses due to either fossil or mri - generated magnetic field can drive powerful bipolar outflows @xcite . as reported by @xcite these outflows can coexist with equatorial accretion and may not be able to prevent eventual collapse of pns . however , they certainly overcome accretion in the polar direction and create favorable conditions for activation of the bz mechanism . the very fast rotation of progenitor s fe core required in this scenario , with the period around few seconds , is unlikely to be achieved in stars with strong fossil magnetic field ( see sec.[rotation ] ) . a related issue is whether a significant fraction of fossil magnetic flux available in the progenitor can actually be accumulated by the black hole . the traditional viewpoint is that the inward advection of magnetic field in accretion disk is effective and can result in build - up of poloidal magnetic field up to the equipartition with the disk gas pressure @xcite . on the other hand , it has also been argued that in standard viscosity driven accretion disk the turbulent outward diffusion of poloidal field may easily neutralise the inward advection well before it reaches the equipartition strength@xcite . however , the disk angular momentum can also be removed by magnetic stresses via interaction with the disk corona and wind and this may change the balance in favour of the inward accretion of poloidal magnetic flux @xcite . this has been demonstrated in recent 3d newtonian simulations of radiatively inefficient accretion disks where the poloidal magnetic field reached the equipartition strength near the central `` black hole''@xcite . in our simulations , a significant fraction of black hole s magnetic flux is accumulated before the development of accretion disk . however , the flux does not escape after the disk is formed . on the contrary , it keeps increasing until the oppositely directed magnetic field lines begin to enter the accretion shock . thus , our results are in agreement with those by @xcite and suggest strong magnetic interaction between the disk and its corona / wind . the alternative origin of magnetic field is via dynamo process in the turbulent accretion disk . in this case the saturated equilibrium magnetic field strength can be estimated from the energy equipartition argument @xmath135 where @xmath136 and @xmath137 are the turbulent magnetic field and velocity respectively . however , the typical scale of this field is rather small , of the order of the disk hight , @xmath138 . when magnetic loops of such small scale are advected onto the black hole one would expect quick , on the time scale of @xmath139 , reconnection and annihilation of magnetic field , resulting in low residual magnetic flux penetrating the black hole . @xcite argued that similar reconnection processes in the disk corona may lead to inverse cascade of magnetic energy on scales exceeding @xmath140 . the expected spectrum in this range is @xmath141 and thus on the scale of @xmath142 one could find the regular magnetic field of strength @xmath143 the standard theory of @xmath144-disks gives @xmath145 where @xmath146 is the keplerian frequency , @xmath140 is the half - thickness of the disk and @xmath147 is a free parameter of this theory . according to the theory of thin neutrino cooled accretion disks by @xcite @xmath148 and @xmath149 putting all these equations together we find the typical black hole magnetic flux @xmath150 the corresponding power of bz mechanism is @xmath151 almost independently on the black hole mass . one can see that even for maximally rotating black holes one should not expect much more than @xmath115erg from a collapse of a @xmath152 progenitor . on the other hand , in this model one should expect a much more powerful magnetically driven wind from the accretion disk @xcite . mass loading of this wind is likely to be much higher due to the processes similar to solar coronal mass injections are the terminal lorentz factors are not expected to be as high as required from the grb observations . however , the disk wind could be responsible for the @xmath153ergs required to drive hypernova . having said that , we need to warn that the above equations are purely newtonian and thus their application to the very vicinity of black hole may introduce large errors . moreover , the current theory of magnetic dynamo in accretion disks can hardly be called well developed and so one could expect few surprises as our understanding of the dynamo improves . finally , the above picture is in conflict with our numerical models , particularly those which do not show bz driven explosions . it is difficult to pin down the exact reasons for this but the key issues are likely to be the condition of axisymmetry and numerical resolution . with a typical rotation rate of @xmath154km / s on the zero - age main - sequence ( e.g. * ? ? ? * ; * ? ? ? * ) the standard evolutionary models of solitary stars predict very rapid rotation of stellar cores by the time of collapse @xcite . according to these models the typical periods of newly born pulsars in successful supernovae are around one millisecond and `` failed supernovae '' conceive rapidly rotating black holes with accretion disks , as required for grbs . these results , however , contradict to observations which clearly indicate that young millisecond pulsars and grbs are much more rare . when the magnetic torque due to the relatively weak magnetic field generated via the tayler - spruit dynamo @xcite is included in the evolutionary models they result in significant spin down of stellar cores during the red giant phase and during the intensive mass loss period characteristic for massive stars at the wolf - rayet phase @xcite . as the result the predicted rotation rates for newly born pulsars come to agreement with the observations but on the other hand , the core rotation becomes too low compared to what is required for the collapsar model of grbs @xcite . recently , it was proposed that low metalicity could be the factor that allows to reduce the efficiency of magnetic braking in grb progenitors @xcite . on one hand , the mass - loss rate of wr - stars decreases significantly with metalicity . on the other hand , the evolutionary models of @xcite show that low metalicity stars rotating close to break - up speed may become fully convective , chemically homogeneous , and by - pass the red giant phase . this result has been confirmed by @xcite but their models also show that even stars with metalicity two thousand times below that of sun inflate above 50 solar radii by the stage of carbon burning . for a potential progenitor of even long duration grbs the corresponding light - crossing time is uncomfortably high , @xmath155s . indeed , this makes very difficult to explain the large observed fraction of long grbs with the gamma ray burst duration of only 3 - 30 seconds . it is somewhat curious but the high mass loss rates of massive stars with solar metalicity has long been considered as an important factor in the evolution of grb progenitors as it can result in the formation of a compact helium star ( wr - star ) via total loss of extended hydrogen envelop @xcite . obviously , these complications are not specific to the standard fireball model of grbs . in fact , the magnetic field of magnetic stars is much stronger than that predicted by the taylor - spruit theory of stellar dynamo and thus the magnetic braking of grb progenitors is more severe in our magnetic model of grb central engine which relies on strong fossil magnetic field . to a large extent the above complications are specific to solitary stars and can be avoided in close binaries . firstly , the extended envelope of a giant star is only weekly bounded by gravity and is easily dispersed in close binary , predominantly in the orbital plane @xcite . secondly , even in the case of high metalicity and/or strong magnetic field the stellar rotation rate may remain high due to the synchronisation of spin and orbital motions . indeed , the minimum separation between the components of binary star , @xmath156 , is determined by the size of the critical roche surfaces as @xmath157 where @xmath158 is the radius of more massive component and @xmath159 is the ratio of component masses @xmath160 . the rotational period is given by @xmath161 where @xmath162 is the actual separation of component and @xmath163 is the mass of more massive component . using @xmath164 , @xmath165 , @xmath166 and @xmath167 we find that the specific angular momentum on the stellar equator is about @xmath168 . on the other hand the specific angular momentum of marginally bound orbit around schwarzschild black hole is @xmath169 for @xmath170 this gives us only @xmath171 . thus , when the star collapses one would expect the black hole to develop substantial rotation and to get surrounded by an accretion disk towards the final phase of the collapse @xcite . the lower accretion rates expected in this phase make the neutrino mechanism much less effective but do not have much of effect on the bz mechanism . indeed , assuming the equipartition of gas and magnetic pressure we find that @xmath172 applying this at the inner edge ( @xmath173 ) of neutrino - cooled accretion disk @xcite we can estimate the maximum magnetic flux that can be kept on the black hole by the disk advection as @xmath174 for @xmath170 and @xmath175 this gives us the formidable @xmath176 . according to eq.[e - bz ] much lower fluxes are needed to explain the observed energetics of grbs . even faster rotation is expected in the case of merger of helium star with its compact companion , neutron star or a black hole , during the common envelope phase of a close binary @xcite . since the orbital angular momentum of binary star is @xmath177 where @xmath178 is the mass of the compact star , we can roughly estimate the specific orbital angular momentum deposited at the distance @xmath44 from the centre of the helium star as @xmath179 where @xmath180 is the helium star mass within radius @xmath44 . when @xmath180 is a slow function , as it is in bethe s model where @xmath181 , we may use in eq.[b-2 ] the total mass of the helium star , @xmath163 , in place of @xmath180 . for @xmath182 and @xmath183 this gives us @xmath184 where @xmath185 cm . in spite of all the approximations , such estimates agree quite well with the results of numerical simulations @xcite . for @xmath186 this is higher than the upper value of @xmath187 imposed in the collapsar simulations by @xcite . the higher angular momentum is one of the factors that reduces the disk accretion rates and hence the efficiency of the neutrino annihilation mechanism @xcite but , as we have already commented , for the blandford - znajek mechanism this is much less of a problem . the compact companion may initially be a neutron star but during the in - spiral it is expected to accumulate substantial mass and become a black hole @xcite . in fact , during the inspiraling the compact companion accumulates not only mass but spin as well . the final value of the black hole rotation parameter computed in @xcite varies from model to model between @xmath188 to @xmath189 . thus , provided that the normal star of the common envelope binary is a magnetic star we have all the key ingredients for a successful magnetically driven grb explosion : ( i ) a rapidly rotating black hole , ( ii ) a rapidly rotating and ( iii ) compact collapsing star , and ( iv ) a `` fossil '' magnetic field which is strong enough to tap the black hole energy at rates required by observations ( see eq.[e - bz ] ) . it has been noted many times that the magnetic fluxes of magnetic ap stars , magnetic white dwarfs , and pulsars are remarkably similar ( e.g. * ? ? ? * ) , suggesting that they could be related . now one may consider adding another type of objects to this list - grb collapsars . in this paper we continued our study into the potential of the blandford - znajek mechanism as main driver in the central engines of gamma ray bursts . in particularly , we analysed the conditions for activation of the mechanism in the collapsar model where accumulation of magnetic flux by the central black hole is likely to be a byproduct of stellar collapse . it appears that the rotating black hole begins to pump energy into the surrounding and overpowers accretion only when the rest mass energy density of matter drops below the energy density of the electromagnetic field near the ergosphere ( see eq.[cond-1 ] ) . only under this condition the mhd waves generated in the ergosphere can transport energy and angular momentum away from the black hole . in the case of spherical accretion the above criterion can be written as a condition on the total magnetic flux accumulated by the black hole and the total mass accretion rate ( see eq.[cond-2 ] ) . however , in the case of nonspherical accretion , characteristic for the collapsar model , this integral condition becomes less applicable due to the spacial separation of magnetic field and accretion flow inside the accretion shock . in fact , our numerical simulations show explosions for significantly lower magnetic fluxes compared to those expected from the integral criterion . we have not included the neutrino heating in our numerical models . this is another factor that can make activation of the bz mechanism in the collapsar setup easier . both the general energetic considerations and the condition for activation of the blandford - znajek mechanism require the central black hole to accumulate magnetic flux comparable to the highest observed values for magnetic stars , @xmath190 . current evolutionary models of massive solitary stars indicate that fossil field of such strength should slow down the rotation of helium cores below the rate required for formation of accretion disks around black holes of failed supernovae . the conflict between strong fossil magnetic field and rapid stellar rotation required by the bz mechanism may be resolved in the models where the grb progenitor is a component of close binary . in this case the high spin of the progenitor can be sustained at the expense of the orbital rotation . moreover , the strong gravitational interaction between components of close binary helps to explain the compactness of grb progenitors deduced from the observed durations of bursts . the magnetic field can also be generated in the accretion disk and its corona but in this case the magnetic flux through the black hole ergosphere is not sufficiently high to explain the power of hypernovae by the bz mechanism alone . on the other hand , in this case the disk is expected to drive mhd wind more powerful than the bz jet . although such a wind is unlikely to reach the high lorentz factors deduced from the observations of grb jets ( due to high mass loading ) it could be behind the observed energetics of hypernovae . , willmarth d.w.,2000 , in eds . et al . , _ stellar astrophysics _ , kluwer , p.175 aloy m.a . , muller e. , abner j.m . , marti j.m . , macfadyen a.i . , 2000 , apj , 531 , l119 ardeljan n.v . , bisnovatyi - kogan g.s . , moiseenko s.g . , 2005 , mnras , 359 , 333 barkov m.v . , komissarov s.s , 2008a , mnras , 385 , l28 barkov m.v . , komissarov s.s , 2008b , in _ high energy gamma - ray astronomy _ , aip conference proceedings , v.1085 , p.608 barkov m.v , komissarov s.s , in prerparation . begelman m.c . , blandford r.d . , rees m.j . , 1984 , rev.modern phys . , 56 , 255 bethe h.a . , 1990 , rev.mod.phys.,62,801 bezchastnov v.g . , haensel p. , kaminker a.d . , yakovlev d.g . , 1997 , a&ap , 328,409 bisnovatyi - kogan g.s . , ruzmaikin a.a . , 1974 , ap&ss , 28 , 45 bisnovatyi - kogan g.s . , ruzmaikin a.a . , 1976 , ap&ss , 42 , 401 blandford r.d . and znajek r.l . , 1977 , mnras , 179 , 433 blandford r.d . , payne d.g . , 1982 , mnras , 199 , 883 blondin j.m , mezzacappa a. , demarino c. , 2003 , apj , 584,971 braithwaite j. , spruit h.c . , 2004 , nature , 431 , 819 buras r. , janka h .- th . , rampp m. , kifonidis k. , 2006 , a&a 457 , 281 burrows a. , dessart l. , livne e. , ott c.d . , murphy j. , 2007 , apj,664,416 camenzind m. , in _ accretion disks and magnetic fields in astrophysics _ , g.belvedere , kluwer , dordrecht , p.129 , 1989 donati j .- f . , babel j. , harries t.j . , howarth i.d . , petit p. and semel m. , 2002 , mnras , 333 , 55 ferrario l. , wickramasinghe d.t . , 2005 , mnras , 356 , 615 frayer c.l , woosley s.e . , 1998 , apj lett . , 502 , l9 fujimoto s. , kotake k. , yamada s. , hashimoto m. , sato k. , 2006 , apj , 644 , 1040 ghosh p. , abramowicz m.a . , 1997 , mnras , 292 , 887 heger a. , langer n. , woosley s.e . , 2000 , apj , 528 , 368 heger a. , woosley s.e . , spruit h.c . , 2005 , apj , 626 , 350 mszros p. , rees m.j . , 1997 , apj lett . , 482 , l29 hirschi r. , meynet g. , maeder a , 2004 , a&a , 425 , 649 hirschi r. , meynet g. , maeder a , 2005 , a&a , 443 , 581 heyvaerts j. , priest e.r . , bardou a. , 1996 , apj , 473 , 403 howarth i.d . , siebert k.w . , hussain g.a . , prinja r.k , 1997 , mnras , 284 , 265 igumenshchev i.v . , 2008 , apj , 677 , 317 ivanova l.n . , imshennik v.s . , nadezhin d.k . , 1969 , sci.inf.astr.council.acad.sci , 13,3 koide s.,2004,apj lett . , 606,l45 komissarov s.s . , 1999 , mnras , 303 , 343 komissarov s.s . , 2001,mnras,326,l41 komissarov s.s . , 2004a , mnras,350,427 komissarov s.s . , 2004b , mnras , 350 , 1431 komissarov s.s . , 2006 , mnras , 368 , 993 komissarov s.s . , 2008 , arxiv0804.1912 lee h.k . , brown g.e . , wijers r.a.m.j , 2000 , apj , 536 , 416 livio m. , ogilvie g.i . , pringle j.e . , 1999 , apj , 512 , 100 lubow s.h . , papaloizou j.c.b . , pringle j.e . , 1994 , mnras , 267 , 235 macdonald d.a . and thorne k.s . , 1982 , mnras , * 198 * , 345 . macfadyen a.i . & woosley s.e , 1999 , apj , 524 , 262 maeder a. , meynet g. , 2005 , a&a , 440 , 1041 mckinney j.c . , gammie c.f . , 2004 , apj , 611 , 977 mckinney j.c . , 2006 , mnras,368,1561 mignone a. , bodo g. , 2006 , mnras , 368 , 1040 misner c.w . , thorne k.s . , wheeler j.a . , _ gravitation _ , san francisco : w.h . freeman and co. , 1973 mizuno y. , yamada s. , koide s. , shibata k. , 2004a , apj , 606 , 395 mizuno y. , yamada s. , koide s. , shibata k. , 2004b , apj , 615 , 389 moiseenko s.g . , bisnovatyi - kogan g.s . , ardeljan n.v . , 2006 , mnras , 370 , 501 moss d.,1987 , mnras , 226 , 297 nagakura h. , yamada s. , 2009 , arxiv:0901.4053 nagataki s. , takahashi r. , mizuta a. , takiwaki t. , 2007 , apj , 659 , 512 nagataki s. , 2009 , arxiv:0902.1908 narayan r. , paczyski b. , piran t. , 1992 , apj lett . , 395 , l8 obergaulinger m. , aloy m.a . , dimmelmeier h. , muller e. , 2006,a&a,457,209 penny l.r . , 1996 , apj . , 463 , 737 piran t. , 2005 , rev.mod.phys . , 76 , 1143 popham r. , woosley s.e . and fryer c.l . , 1999 , apj , 518 , 356 proga d. , macfadyen a.i . , armitage p.j . , begelman m.c . , 2003 , apj , 629 , 397 rdler k .- h . , 1968 , z.naturforsch . , a , 23 , 1851 robstein d.m . , lovelace r.v.e . , 2008 , apj , 677 , 1221 schinder p.j . , schramm d.n . , wiita p.j . , margolis s.h . , tubbs d.l . , 1987 , apj , 313 , 531 . sekiguchi y. , shibata m.,2007,prog.theor.phys,117,1029 scheck l. , janka h .- th . , foglizzio t. , kifonidis , k. , 2008 , a&a , 477 , 931 schmidt et al . , 2003 , apj , 595 , 1101 spruit h.c . , 2002 , a&a , 381 , 923 spruit h.c . , uzdensky d.a . , 2002 , apj , 629 , 960 takahashi m. , niita s. , tatematsu y. , tomimatsu a. , 1990 , apj , 363 , 206 . tao l. , proctor m.r.e . , weiss n.o . , 1998 , mnras , 300 , 907 taam r.e . , sandquist e.l . , 2000 , ann.rev.a&a , 38 , 113 tayler r. , 1973 , mnras , 161 , 365 timmes f.x . , swesty f.d . , 2000 , apjss , 126 , 501 tout c.a . , pringle j.e . , 1996 , mnras , 281 , 219 tutukov a. , iungelson l. , in _ mass loss and evolution of o - type stars _ , redel , dordrecht , p.401 , 1979 uzdensky d.a . & macfadyen a.i . , 2006 , apj , 647 , 1192 woosley s.e . , 1993 , apj , 405 , 273 woosley s.e . , baron e. , 1992 , apj , 391 , 228 woosley s.e . , bloom j.s . , 2006 , ann.rew.a&a , 44 , 507 woosley s.e . , heger a. , 2006 , apj , 637 , 914 yoon s .- c . , langer n. , norman c. , 2006 , a&a , 460 , 199 . yoon s .- c . , cantiello m. , langer n.,2008 , arxiv:0801.4373 zeldovich ya.b . , 1957 , sov.phys.jetp , 4 , 460 zhang w. , fryer c.l . , 2001 , apj , 550 , 357 the problem of dust accretion has been considered in @xcite . in the coordinate basis of the boyer - lindquist coordinates , @xmath191 the components of 4-velocity of a dust particle accreting from rest at infinity and having zero angular momentum are @xmath192 where @xmath59 . in order to find the 4-velocity components in the coordinate basis of kerr - schild coordinate one can simply use the corresponding transformation law @xcite : @xmath193 this gives us @xmath194 notice that in both coordinate systems the particles move over a conical surface , @xmath195const . their rotation about the symmetry axis is due to the inertial frame dragging effect . in order to find the density distribution consider the steady - state version of the continuity equation : @xmath196 where @xmath197 is the determinant of the metric tensor in the kerr - schild ( and boyer - lindquist ) coordinates . this shows that @xmath198 where @xmath199 is some arbitrary function and then that @xmath200 at infinity this function does not depend on the polar angle only if @xmath201 which implies that @xmath35 does not depend on @xmath67 for any @xmath44 . denoting as @xmath61 the rest mass density at the outer event horizon we may now write @xmath202 in this section we employ the 3 + 1 splitting of electrodynamics described in @xcite where the electromagnetic field is represented by four 3-vectors , @xmath203 , and @xmath204 , defined via where @xmath209 is the faraday tensor , @xmath210 is the maxwell tensor , @xmath211 is the levi - civita tensor of space and @xmath144 is the lapse function . in stationary metric , @xmath212 , the evolution of these vector fields is described by the maxwell equations for electromagnetic field in matter . the effects of curved space time are incorporated via the non - euclidian spatial metric and the constitutive equations @xmath213 where @xmath144 is the lapse function and @xmath215 is the shift vector . in terms of these vectors the condition of perfect conductivity , @xmath216 , can be written in the familiar form @xmath217 where @xmath218 is the usual 3-velocity vector . both in the boyer - lindquist and the kerr - schild coordinates the poloidal component of poynting flux , @xmath219 , is given by @xmath220 whereas the the poloidal component of the angular momentum flux , @xmath221 , is @xmath222 where @xmath223 . suppose that in the boyer - lindquist coordinates the magnetic field is purely radial , @xmath224 . since @xmath225 eq.[h3 ] then implies that @xmath226 provided this magnetic field is frozen into the flow described by eq.[u_bl ] we also find that @xmath227 given these results the equations [ l ] and [ s ] ensure that @xmath228 in order to find the corresponding electromagnetic field in the kerr - schild coordinates we simply apply the transformation law [ transf ] ) . this way we find that @xmath229 and @xmath230 the divergence - free condition requires @xmath231 and hence @xmath232 where @xmath233 is the determinant of the spatial metric . provided that at infinity ( where @xmath234 ) @xmath235 does not depend on the polar angle we then have @xmath236

a general analysis is performed on the dimension - six operators mixing an almost hidden to the standard model ( sm ) , when the communicates with the sm via heavy mediators . these are fermions charged under both and the sm , while all sm fermions are neutral under . we classify the operators as a function of the gauge anomalies behaviour of mediators and explicitly compute the dimension - six operators coupling to gluons , generated at one - loop by chiral but anomaly - free , sets of fermion mediators . we prove that only one operator contribute to the couplings between charged matter and on - shell gluons . we then make a complete phenomenological analysis of the scenario where the lightest fermion charged under is the dark matter candidate . combining results from wmap / planck data , mono - jet searches at lhc , and direct / indirect dark matter detections restrict considerably the allowed parameter space . lpt orsay-13 - 49 , cpht - rr054.0613 , ift - uam / csic-13 - 76 = 20pt * emilian dudas , lucien heurtier , yann mambrini and bryan zaldivar * + : _ cpht , ecole polytechnique , 91128 palaiseau cedex , france _ + : _ laboratoire de physique thorique , universit paris - sud , f-91405 orsay , france _ + : _ instituto de fisica teorica , ift - uam / csic , 28049 madrid , spain _

1307.0005

new abelian gauge symmetries are arguably the simplest extensions of the standard model ( sm ) ( for a recent review , see @xcite ) . if sm fermions are charged under a new abelian @xmath7 , its couplings are strongly constrained by direct searches and especially by fcnc processes . the simplest and widely studied possibility in the literature is when sm fermions have flavor - independent charges . most popular examples in this class are @xmath8 or linear combinations @xmath9 . they are actually the only family - independent , anomaly - free gauged symmetries commuting with the sm gauge group in case where there are no new fermions beyond the ones of the sm . family - dependent anomaly - free models with no extra fermions were also extensively studied . in all such cases , the @xmath0 should be heavy enough to escape detection , at least in the multi - tev range . there is also a large literature on light @xmath10 s of string or field theory origin with anomaly cancellation a la green - schwarz @xcite , with low - energy anomalies canceled by axionic couplings and generalized chern - simons terms , or in other models with stueckelberg realization of @xmath0 @xcite . a radically different option is to have no sm fermions charged under @xmath0 . this is a relatively natural framework in string theory with d - branes . but it is also natural from a field theory viewpoint , with additional heavy fermions @xmath11 , called mediators " in what follows , which mediate effective interactions , described by the dimension - four kinetic mixing and higher - dimensional operators between the @xmath0 and the sm sector @xcite . if one wants mediators parametrically heavier than the electroweak scale ( say in the tev range ) , we need , in addition to possible sm higgs contributions , an additional source to their mass . a purely dirac mass is of course a simple viable option . however as argued in @xcite , because of the furry theorem , the only low - dimensional induced effective operator is the kinetic mixing , whereas the next higher - dimensional ones are of dimension eight . throughout our paper , we consider the kinetic mixing to be small enough . if we are interested in @xmath0 couplings to gluons , this can be achieved for example by having colored mediators with no hypercharge . in this case , the main couplings between the hidden " @xmath0 and the sm are generated by higher - dimensional effective operators ( hdo s ) , the lowest relevant ones being of dimension six . however , we will show that in the parameter space allowed by the planck / wmap data , the phenomenological consequences induced by the presence of a kinetic mixing allowed by various constraints are negligible . the simplest and natural option to obtain dimension - six effective operators is to generate the mediator masses by the vev of the scalar field @xmath12 breaking spontaneously the @xmath0 gauge symmetry . the corresponding induced mediator masses , called generically @xmath13 in what follows , determine the mass scale of the hdo s and also the uv cutoff of the effective theory . there could also be contributions to their mass from the sm higgs field @xmath14 , which are considered to be smaller , such that we can expand in powers of @xmath15 and obtain operators invariant under the sm gauge group . such a framework was already investigated in @xcite from the viewpoint of the effective couplings of @xmath0 to electroweak gauge bosons . the potential implications to dark matter , considered to be the lightest fermion in the dark sector was also investigated , with the outcome that a monochromatic gamma ray line from the dark matter annihilation is potentially observable . the potential existence of a signal in the fermi data was largely discussed in the recent literature @xcite and will not be discussed further here . in this paper we extend the previous works by allowing the mediators to be colored and therefore the @xmath0 to couple to gluons . we restrict ourselves throughout the paper to cp even couplings for simplicity . these couplings are more restricted by symmetries than the ones to the electroweak gauge bosons and their presence change significantly the phenomenology of such models . whereas at dimension - six order four such operators are possible , only two of them are induced by heavy fermion mediators loops . moreover , only one operator contributes to amplitudes in which at least one of the gluons is on - shell , as will be the case throughout our paper . we analyze in detail the corresponding phenomenology from the viewpoint of the dark matter relic abundance , direct and indirect dark matter detection and lhc constraints . allowing couplings to gluons and at the same time to electroweak gauge bosons does not change significantly the phenomenology of the @xmath0 compared to the case where only couplings to gluons are allowed . one interesting conceptual difference is that , whereas the @xmath0 couplings to gluons and photons _ vanish _ for an on - shell @xmath0 due to the landau - yang theorem @xcite , the couplings to the electroweak gauge bosons @xmath16,@xmath17 do not vanish ; they lead on the contrary to an _ enhancement _ close to the @xmath0 pole . another interesting result is that , unlike the case of kinetic mixing , the dark matter annihilation into gluons induced by virtual @xmath0 exchange can give correct relic density for heavy dark matter and @xmath0 masses , well above the electroweak scale . since our interest here is to have complementary constraints from dark matter searches and lhc , we nonetheless confine our analysis to masses below than or of the order tev in what follows . the paper is organized as follows . section 2 introduces the basic formalism we will use , which is stueckelberg realization of @xmath0 symmetry . it contains the list of the lowest dimensional effective operators generated by integrating - out heavy fermionic mediators , their classification depending on the nature of messenger masses and charges and the explicit loop computation of the @xmath0 couplings to gluons . section 3 deals with the consequences of the model for dark matter generation in the early universe , focusing on the annihilation to a gluon pair . section 4 contains the various phenomenological constraints coming from the unique @xmath0 coupling to gluons generated at one - loop by heavy colored mediators . section 5 contains the re - analysis of the various constraints when @xmath0 couplings to electroweak gauge bosons are also added . appendices contain more details about the gauge independence of the @xmath0 mediated hidden - sector - sm couplings , the effective operator couplings @xmath0 to gluons induced by heavy mediator loops and the complete cross - sections of the s- and t - channel annihilation of the dark matter . the effective lagrangian generated by loops of heavy mediators is generically invariant under sm and has a non - linear ( stueckelberg ) realization for @xmath0 , for the following reason . if the mediator masses are invariant under both the sm and the @xmath0 gauge symmetry , the induced operators would be gauge invariant in the usual sense . if the mediator masses are however generated by the breaking of @xmath7 , in the broken phase below the mass of the heavy higgs @xmath12 breaking @xmath7 , the symmetry is still present but realized a la stueckelberg . indeed , in the limit where @xmath12 is much heavier than the @xmath0 , in the effective theory we keep only the axionic component of the original @xmath0 higgs field @xmath18 . we define the dimensionless axion @xmath19 in what follows . the axion transforms non - linearly under @xmath10 transformations @xmath20 the exact lagrangian , describing all the microscopic physics , including the mediator fields @xmath11 , is then of the form @xmath21 where @xmath22 is the standard model lagrangian and where @xmath23 . this lagrangian is indeed invariant under ( [ inv1 ] ) , with non - linear shifts of the axion @xmath24 crucial for restauring gauge invariance . if the original high - energy lagrangian is anomaly - free and the sm fermions are neutral under @xmath0 , then the mediators have to form an anomaly - free set . we are considering this class of models in most of this paper . in this case , the induced effective operators are gauge invariant a la stueckelberg . throughout the paper we restrict ourselves to cp even operators for simplicity . in the case where the mediators are not an anomaly - free set , then either low - energy fermions have to be charged under @xmath0 , or there are axionic couplings and gcs terms in order to cancel anomalies . for notational convenience we define : @xmath25 \,\ , , & & ~~{\cal t}r ( e f g ) \equiv { \rm tr } [ e_{\mu}^{~\lambda } f_{\lambda \nu } g^{\nu \mu } ] \ , \label{inv5}\end{aligned}\ ] ] where @xmath26 takes into account a possible trace over non - abelian indices . in summary , there are three distinct possibilities : * i ) * the mediators are completely non - chiral , i.e. vector - like both respect to the sm and @xmath7 . in this case , there are no dimension - six induced operators , since the only one that can be potentially written , @xmath27 vanishes exactly as shown in the appendix . + * ii ) * the mediators form an anomaly - free set , but are chiral with respect to @xmath7 and vector - like with respect to the sm . the induced dimension - six operators in this case are @xmath28\right.\ok & + & \left.\partial^md_m \theta_x ( d_1\mc{t}r(f^y\tilde{f}^y ) + 2d_2\mc{t}r(f^w\tilde{f}^w ) ) + d'_{ew } \partial^{\mu } d^{\nu } \theta_x { \rm tr}(f_{\mu \rho}\tilde{f}^{\rho}_{\nu } ) \nonumber \right . \\ & + & \left . e_{ew } d^{\mu } \theta_x { \rm tr}(f_{\nu \rho } { \cal d}_{\mu } \tilde{f}^{\rho \nu } ) + e'_{ew } d_{\mu } \theta_x { \rm tr}(f_{\alpha \nu } { \cal d}^{\nu } \tilde{f}^{\mu \alpha})\right\}\ , \ , \label{gluon1 } \end{aligned}\ ] ] where @xmath29 denotes the gluon covariant derivative , in components @xmath30 the last three terms in ( [ gluon1 ] ) refer to all electroweak gauge bosons . * iii ) * the mediators do not form an anomaly - free set . it means that some low - energy fermions have to be charged in order to compensate the resulting anomaly . the induced dimension - six operators in this case are not gauge invariant , but include axionic couplings and eventually gcs terms , schematically of the form @xmath31 this case was studied from various perspectives in the past @xcite and will not be considered anymore here . in all cases , there is potentially a kinetic mixing term @xcite @xmath32 mediators generate at one - loop @xmath33 , where @xmath34 are the mediators charges to @xmath7 and @xmath35 , respectively . if @xmath36 has its natural one - loop value , then its effects are more important than most of the ones we will discuss in what follows . this is the most plausible case and was investigated in many details within the last years . in what follows , we will place ourselves in the mostly ` orthogonal ' case in which @xmath36 is small enough such that its effects are subleading compared to the dimension - six operators . this is the case , for example , if messengers are in complete representations of a non - abelian gauge group ( gut groups are of course the best such candidates ) , or if the mediators have no hypercharge . then , at low energy , the mediators being integrated out give rise to a new effective lagrangian @xmath37 where @xmath38 and @xmath39 represent the new effective operators generated separately in the sm gauge sector and @xmath0 one , whereas in @xmath40 we collect all the induced terms mixing @xmath0 with the standard model . notice that @xmath41 also contains the dm particle ( i.e. the lightest mediator ) which is not integrated out . the mediators mass matrix has the symbolic form @xmath42 where @xmath43 is the vev breaking the @xmath0 gauge group @xmath7 and @xmath44 is the electroweak vev . if the heavy higgs @xmath12 has a charge @xmath45 , then the renormalizable yukawas ( [ inv3 ] ) exist provided @xmath46 since none of our results in what follows depend on the assumption that the heavy fermions masses arise through renormalizable interactions , in the rest of the paper we include the more general case where these masses arise from arbitrary yukawas of type @xmath47 where @xmath48 is an uv cut off , such that @xmath49 corresponds to non - renormalizable interactions . for phenomenological applications , we consider here a model in which the dark matter is represented by the lightest stable fermion @xmath50 charged under @xmath0 and uncharged under sm ( the mass of dark matter will be denoted by @xmath51 in what follows ) . the _ mediators _ @xmath11 are considered to be heavy enough so that they have not been discovered yet in colliders . they can be integrated out so that we have to deal with effective operators , including new parameters . at the one - loop perturbative level , mediators generate only @xmath0 couplings to the sm gauge fields and the sm higgs as represented in fig . [ fig : gg1 ] in the case of @xmath0 coupling to gluons . indeed , in the absence of kinetic mixing , one - loop couplings to sm fermions can be generated only if there are yukawa couplings mixing mediators with sm fermions . we forbid such couplings in what follows . one ( clearly not unique ) way of achieving this is by defining a @xmath52 parity , under which all mediator fields are odd and all sm fields are even . in what follows we work in the unitary gauge where the axion is set to zero @xmath53 . as usual , gauge invariance allows to work in any gauge . in the appendix we discuss the issue of gauge independence in more details . _ when heavy fermions are integrated out , they generate dimension - six effective operators of strength @xmath54._,width=566 ] in the case of exact cp invariance that we restrict for simplicity , the three - point gauge boson amplitude can be generally be written as @xcite @xmath55 \ k_{1 \alpha } k_{2 \beta } \ , \label{g1 } \end{aligned}\ ] ] where @xmath56 are lorentz - invariant functions of the external momenta @xmath57 . the functions @xmath58 which encode the generalized chern - simon terms ( gcs ) @xcite are superficially logarithmically divergent , whereas the functions @xmath59 are uv finite . however , @xmath58 are determined in terms of @xmath59 by using the ward identities , which in the case where the heavy fermions form an anomaly - free set , are given by @xmath60 the violation of the @xmath0 current conservation may seem surprising . it encodes actually the fact that one generates dimension - six operators , for which gauge invariance is realized la stueckelberg and indeed in the appendix b it will be shown explicitly that @xmath61 . there are several contributions to @xmath62 . the first is the triangle loop diagram with no chirality flip / mass insertions , given by @xmath63 \ . \label{g4}\ ] ] where @xmath64 . as shown in the appendix b by using ward identities , computing this diagram is enough in order to find the full amplitude . the final result for the @xmath0 couplings and the details of the computation are described in the appendix b. after symmetrization among the two gluon legs , one finds @xmath65 k_1^{\alpha } k_2^{\beta } + \epsilon_{\mu \nu \rho \alpha } k_1 k_2 ( k_2-k_1)^{\alpha } \ } \ , \label{g6}\ ] ] where @xmath66 . the corresponding dimension - six operator for the triangle diagram represented in fig . [ fig : triangle ] is then @xmath67 \ , \label{g5}\ ] ] where @xmath68 is the qcd strong coupling . , width=302 ] on the other hand , by using the identities ( [ identity ] ) in appendix c , it can be shown that the antisymmetric part of the amplitude in the gluonic legs is zero , which is consistent with the fact that there is no possible dimension - six operator mixing @xmath0 to gluons , that is antisymmetric in the gluon fields . as a byproduct , we also find that the heavy mediators we are considering do not induce operators of the type @xmath69 ) \ , \label{g7}\ ] ] that are completely antisymmetric in the three gluon fields [ g5 ] , and similar operators for electroweak gauge fields . this means that there are no constraints from purely sm dimension - six operators induced in this setup and all the phenomenological constraints come from the mixing of @xmath0 with sm fields . until now we have made the important assumption that no sm fermion is charged under @xmath0 and the only couplings arise through gauge - invariant higher - dimensional operators generated by integrating out heavy fermions forming an anomaly - free set . a more subtle option , in the spirit of @xcite is to integrate - out a set of heavy fermions which do contribute to gauge anomalies . in this case there are non - decoupling effects leading to axionic couplings and eventually generalized chern - simons terms . let us consider two simple examples in order to exemplify the main points . * i ) example with no colour anomalies * @xmath70 + in this case , after defining the anomaly coefficients @xmath71 and @xmath72 , the low - energy effective theory has the following mixed anomalies : @xmath73 * i ) example with colour anomalies * @xmath74 + in this case , the low - energy effective theory has the following anomalies : @xmath75 in such examples , the heavy - fermion spectrum has to exactly cancel the low - energy gauge anomalies . in the decoupling limit there is an axionic coupling with a coefficient exactly determined by the low - energy induced anomalies @xmath76 \ . \label{an3}\ ] ] as shown in the appendix b , we can also capture the effect of these axionic couplings in the unitary gauge , where the axionic effect is encoded in the particular high - energy behaviour of the anomalous three gauge boson amplitude with light fermions in the loop . this is strictly speaking true in the large ( infinite ) mass limit of heavy fermions . for finite mass , there are corrections and the low - energy description in the unitary gauge with three - gauge anomalous couplings is corrected by finite mass effects . we start by first discussing the @xmath0 couplings to gluons . the cp and gauge invariant dimension - six operators coupling @xmath0 and the gluons are given by @xmath77 due to the fact that at one - loop only the operators with coeff . @xmath78 and @xmath79 are generated and only the first one contributes to the amplitude with on - shell gluons , we consider only @xmath78 in what follows and disregard the effects of the other operators in ( [ gluon01 ] ) . the _ dark matter _ couples minimally to the @xmath0 boson as : @xmath80 which provides us two ways of annihilating dark matter into gluons . the first one is an _ s - channel _ production of a @xmath0 boson decaying into a pair of gluons . the second one is a _ t - channel _ process , leading to two @xmath0 bosons , which will mostly decay into gluons . the associated feynman diagrams are presented in fig.[fig : gg ] . _ gluon pair production via two different processes , that are the s - channel @xmath81 and the t - channel @xmath0 pair production @xmath82 , that decay subsequently into two gluons each._,width=453 ] in the unitary gauge , the @xmath0-gluon - gluon vertex coming from the operator @xmath78 is @xmath83 where the coeff . @xmath78 was redefined compared to ( [ gluon01 ] ) in a convenient way for our purposes . the propagator of the vector boson @xmath0 in the unitary gauge is @xmath84 for dark matter fermions of mass smaller than @xmath85 , the main contribution to the @xmath0 width @xmath86 is @xmath87 , which is computed to be @xmath88 \sqrt{m_{z'}^2 - 4m_{\psi}^2}\ , \ .\ ] ] for heavier masses of dark matter , one has to consider the @xmath0 decay width into gluons and @xmath89 gauge bosons . however , it can be readily checked that the detailed values of these widths do not influence much the results in what follows approaching @xmath90 , as a consequence of the landau - yang theorem @xcite . in the non - relativistic approximation , this happens in the energy region closed to @xmath91 . the suppression therefore occurs for a mass @xmath51 significantly lower than @xmath85 , where the decay width is essentially that of decay into two dark matter particles . ] . in the case of a vector - like coupling of dm fermion to @xmath0 boson , one obtains the interaction lagrangian @xmath92 now we can perform the tree - level diagram cross section . we find that the amplitude vanishes @xmath93 and therefore the @xmath78-term does not contribute to the final cross section at all . the reason is that , due to the effective coupling of the form @xmath94 , the vertex @xmath95 gets multiplied by the virtual momentum and is of the form @xmath96 after using dirac equation for the spinors describing the wavefunctions of the dark matter fermions . in the general case we get also an axial - vector coupling in addition to the vector one @xmath97 one then gets , as far as the annihilation of dark matter into a gluon pair is concerned , the total cross section @xmath98 the cross section is suppressed for energies of order @xmath85 due to the landau - yang theorem . there is also a helicity suppression for light dark matter case , that can be easily understood by writing the vertex @xmath95 in this case @xmath99 after using dirac equation . this finally gives in the non - relativistic approximation @xmath100 , with @xmath101 being the relative velocity between the two colliding dark matter fermions , the total cross section @xmath102 as mentioned earlier , we also have to consider a t - channel process , producing pairs of @xmath0 bosons in dark matter annihilation for @xmath0 mass below the dark matter mass . considering that the only non vanishing coupling is the one in @xmath78 , each @xmath0 will decay into gluons ; this process will then produce gluons in the final state . after expanding in powers of @xmath103 , the cross - section in this case can be expressed as : @xmath104 a @xmath105 coupling can be tested in several laboratories , from direct detection experiments to indirect detection , relic abundance or lhc searches . we present in the following the constraints obtained from these different searches , before summarizing all of them at the end of the section . the reader can also find a nice recent complementary analysis of gluonic effective couplings to dark matter in @xcite . recently , planck collaboration released its latest results concerning the composition of the universe @xcite . it confirms the results of wmap experiment @xcite obtaining for the relic abundance of non baryonic matter @xmath106 at 68% of cl . with such a level of precision , it is interesting to know what is the effective scale @xmath13 which is able to produce sufficient dark matter from the thermal bath to respect the previous planck / wmap results . depending on the spectrum , two annihilation processes allow the dark matter candidate to keep thermal equlibrium with the standard model particles of the plasma : the @xmath107channel exchange of a @xmath0 ( eq.[eq : schannel ] ) , and the @xmath108-channel production of the @xmath0 ( eq.[eq : tchannel ] ) , as long as this channel is kinematically open . concerning the numerical analysis , we solved the boltzmann equations by developing a code and adapting the public software micromegas @xcite to our model . we then extracted the relic abundance and checked that our analytical solutions ( [ eq : schannel]-[eq : tchannel ] ) gives similar results to the numerical procedure gives reasonable good approximations to the full boltzmann system of equations . ] at a level of 20 to 30% . we noticed in section [ sec : vector ] that the coupling of the dark matter should be axial , as the vectorial part of the current coupling to @xmath109 does not gives any contribution to the process @xmath110 . for simplicity , we will set charges @xmath111 . our results for a different set of charges are modified in a straightforward way . to keep our results as conservative as possible , we plotted the wmap limits @xmath112 at 5@xmath113 . [ cols="^,^ " , ] as explained earlier , parameters are allowed to lie below the red / full lines ( overdensity of the universe ) , above the orange / full line ( lhc bounds on monojets production ) . since the whole study has been released using effective dimension six operators generated by integrating out heavy fermions loops , one has to check that the parameter range is still in the window where @xmath114 . this is indicated on fig.([fig : synt ] ) where we considered natural values of @xmath78 varying between @xmath115 and @xmath45 ( purple and green / dashed line , respectively ) . thus one can easily distinguish between the two regions @xmath116 ( upper region ) and @xmath117 ( lower region ) . in the case where @xmath118 , it is important to notice that low values of the coupling constant @xmath119 provide almost no validity region in the parameter space since parameters have to lie above the purple / dashed line . on the other hand , for @xmath120 one can also notice that the allowed region is much larger in the case of a heavy @xmath0 . the case @xmath121 considerably relax the constraints since the validity zones are almost in the region where @xmath116 ( below the green / dashed line ) . in the same way the @xmath0 boson couples to gluons via operators of dimension six , mediators with electroweak quantum numbers can generate operators coupling the @xmath0 boson to gauge bosons of the @xmath122 electroweak sector . they can be parametrized as @xmath123\right.\ok & + & \left.\partial^md_m \theta_x ( d_1\mc{t}r(f^y\tilde{f}^y ) + 2d_2\mc{t}r(f^w\tilde{f}^w ) ) + d'_{ew } \partial^{\mu } d^{\nu } \theta_x { \rm tr}(f_{\mu \rho}\tilde{f}^{\rho}_{\nu } ) \nonumber \right . \\ & + & \left . e_{ew } d^{\mu } \theta_x { \rm tr}(f_{\nu \rho } { \cal d}_{\mu } \tilde{f}^{\rho \nu } ) + e'_{ew } d_{\mu } \theta_x { \rm tr}(f_{\alpha \nu } { \cal d}^{\nu } \tilde{f}^{\mu \alpha } ) \right\}\ , . \label{ew1}\end{aligned}\ ] ] these effective operators give contributions to @xmath124 , @xmath125 and @xmath126 processes . we neglected such operators until now , since they induce new free parameters in the model . they can contribute to sm matter production in the universe , which in turn can slightly relax our previous constraints on the parameter @xmath127 . let us now consider the @xmath0 couplings to electroweak gauge bosons coming from the dimension - six operators @xmath128 and @xmath129 in ( [ ew1 ] ) , by ignoring the others . the reason for ignoring the last ones @xmath130 and @xmath131 is the same as for the gluonic couplings . on the other hand , although beyond the goals of the present paper , we believe that the operators @xmath128 are induced and do contribute in a computation with heavy loop of mediators , provided that part of mediator masses come from couplings to the sm higgs . the interaction lagrangian of the couplings @xmath132 to the electroweak sector are then given by * @xmath133 process : @xmath134 * @xmath135 process : @xmath136 * @xmath137 process : @xmath138 * @xmath139 process : @xmath140 these interaction terms give rise to the cross sections for the s - channel displayed in appendix d. they have to be added to the t - channel cross section . we can now add the resulting cross sections to the one of gluons production to consider a more precise constraint about universe overdensity , which is @xmath141 then , assuming for simplicity that all the couplings appearing in the different six - dimensional operators are equal to @xmath127 , which is a very strong hypothesis of course , we can plot a new constraint on this parameter , in a similar way we did before . this provides a new validity zone in the parameter space , as represented in fig.[fig : figure ] ( in the case where @xmath142 and @xmath120 ) , in which we added the electroweak processes to the gluon couplings of section 3 . , width=529 ] the resulting constraints are slightly relaxed , but the validity zones are not greatly enlarged , as anticipated earlier . one notice that the behaviour of the cross sections around @xmath143 is modified here , compared to the gluon production process . this happens because the electroweak gauge bosons @xmath144 and @xmath145 are massive , unlike the gluons . thus the landau - yang theorem does not apply and a real @xmath0 can be created , relaxing the constraints on @xmath146 parameter . implications of landau - yang theorem can yet be extended to express some constraints on what kind of cp even operators can be written down to produce electro - weak gauge bosons ; this has been done previously for @xmath147 process in @xcite . our results are in agreement with theirs in the form of operators and resulting cross sections . * acknowledgements . * we would like to thank massimo bianchi , j.m . moreno and v. martin - lozano for very useful discussions . e.d . , l.h and y.m thank the galileo galilei institute for theoretical physics for the hospitality and the infn for partial support during the completion of this work . this work was supported in part by the european erc advanced grant 226371 masstev , the french anr tapdms anr-09-jcjc-0146 the contract pitn - ga-2009 - 237920 unilhc and the spanish micinns consolider - ingenio 2010 programme under grant multi- dark * csd2009 - 00064*. y.m . acknowledges partial support from the european union fp7 itn invisibles ( marie curie actions , pitn- ga-2011- 289442 ) , the erc advanced grant higgs@lhc . b.z . acknowledges the support of micinn , spain , under the contract fpa2010 - 17747 , as well as the hospitality of lpt , orsay , during the completion of this project . in this appendix we discuss the gauge independence of @xmath0 induced effective couplings . in the stueckelberg phase and after integrating out the heavy mediators , the effective action in @xmath148 gauges is @xmath149 in ( [ a1 ] ) , @xmath150 describes the local ( non - local ) coupling between @xmath0 and sm gauge fields generated in the case where some heavy ( light ) fermions are charged under @xmath0 . @xmath151 is the axionic coupling generated in this case by the heavy set of mediator fermions cancelling an eventual gauge anomaly , which captures the low - energy remnant of the heavy mediator fermions in the infinite mass limit . gauge invariance implies @xmath152 at the abelian ( three - point function ) level , we can write @xmath153 where @xmath154 denotes symbolically the sm gauge fields . as concrete examples , the operator @xmath155 coupling gluons to the axion is of the form @xmath156 for the operators induced by chiral but anomaly - free set of heavy mediators in section 2.1 , whereas is of the form @xmath157 for the anomalous sets of fermion mediators considered in section 2.2 . in momentum space , the gauge invariance conditions for the three point function @xmath158 are @xmath159 the @xmath0 and axion propagators are @xmath160 and the unitary gauge corresponds to the limit @xmath161 . whereas the issue of gauge - fixing independence can be discussed in more general terms , we prefer to analyse it in the relevant context for our work , fermions- 2 sm gauge fields interactions mediated by the @xmath0 exchange . in an arbitrary @xmath148 gauge , there are two contributions : the @xmath0 and the axionic exchange : @xmath162 u ( p_1 ) \left(- i \frac{\eta_{\mu \nu } + ( \xi-1 ) \frac{q_{\mu } q_{\nu}}{q^2 - \xi m_{z'}^2}}{q^2 - m_{z'}^2 } \right ) \gamma^{\nu } \nonumber \\ & & + { \bar v } \gamma_5 ( x_l - x_r ) \frac{m_{\psi}}{v } \frac{i}{q^2 - \xi m_{z'}^2 } \gamma_a u ( p_1)\ , \label{a4}\end{aligned}\ ] ] where @xmath163 are the three - point functions coming from the operators present in ( [ a1 ] ) , @xmath164 is the @xmath0 virtual momentum and @xmath165 the dirac spinors associated to the fermion ( antifermion ) @xmath166 coupling to @xmath0 , to be identified with the dark matter candidate in our paper . by using dirac equation for the fermion @xmath166 and the gauge invariance condition ( [ a2 ] ) in momentum space @xmath167 , with @xmath168 the momenta of the two gauge bosons in the final space , we find @xmath169 u ( p_1 ) \left(\frac { -i \gamma^{\mu}}{q^2 - m_{z'}^2 } \right ) \nonumber \\ & & + \ { \bar v}(p_2 ) \gamma_5 ( x_l - x_r ) \frac{m_{\psi}}{v } \frac{i}{q^2 - m_{z'}^2 } \gamma_a u ( p_1 ) \ . \label{a5}\end{aligned}\ ] ] as expected , due to gauge invariance , the @xmath170-dependence cancelled in the final result . moreover , the result can also be directly found in the unitary gauge with no axion field present . in this case , the result is fully encoded in the unitary gauge computation @xmath171 u ( p_1 ) \left(- i \frac{\eta_{\mu \nu } - \frac{q_{\mu } q_{\nu}}{m_{z'}^2}}{q^2-m_{z'}^2 } \right ) \gamma^{\nu } \ . \label{a6}\ ] ] notice that in the unitary gauge the lagrangian can be expressed entirely in terms of @xmath172 unitary gauge captures correctly the result in the infinite mass limit of the heavy fermions . for finite masses , there are corrections which are not captured by the naive unitary gauge computation . in the case of cp invariance , the three - point gauge boson amplitude can be generally be written as @xcite @xmath55 \ k_{1 \alpha } k_{2 \beta } \ , \label{a8 } \end{aligned}\ ] ] where @xmath56 are lorentz - invariant functions of the external momenta @xmath57 . the functions @xmath58 which encode the generalized chern - simon terms ( gcs ) @xcite are superficially logarithmically divergent , whereas the functions @xmath59 are uv finite . however , @xmath58 are determined in terms of @xmath59 by using the ward identities , which in case the heavy fermions form an anomaly - free set , are given by @xmath173 the last current conservation is nontrivial in our case , since gauge invariance is realized through an additional axionic coupling to gauge fields generated by heavy fermions , such that we find ( [ a03 ] ) . after comparison with ( [ a08 ] ) , this implies @xmath174 the situation here is different compared to the usual discussion of anomalies . the usual axionic couplings compensating triangle gauge anomalies are generated by chiral and non - anomaly free set of fermions . if the heavy fermions form an anomaly - free set , they do not generate such couplings , but dimension six operators for gauge fields and dimension seven axionic couplings , which cancel between themselves their gauge variation . there are two contributions to @xmath62 . the first is the triangle loop diagram with no chirality flip / mass insertions , given by @xmath63 \ . \label{a9}\ ] ] where @xmath64 . there are also three other contributions with two mass insertions , of the type @xmath175 + \cdots \ , \label{a10}\ ] ] where @xmath176 denote two similar contributions with the mass insertions permuted among the three propagators . by using a feynman parametrization and after performing a shift of the momentum integral @xmath177 , we find @xmath178 ^ 3 } \ , \label{a11}\ ] ] where @xmath179_i$ ] and where @xmath180 \gamma_{\rho } [ { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$}}+ \beta { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 -\alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\nu } [ { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$}}- ( 1- \beta ) { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 -\alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\mu } \gamma_5 \ } = \nonumber \\ & & - { \rm tr } \ \ { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$}}\gamma_{\rho } { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$ } } [ ( 1- \beta ) { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 + \alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\mu } \gamma_5 \ } + { \rm tr } \ \ { [ \beta { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 + ( 1-\alpha){{\raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\rho } { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$}}\gamma_{\nu } { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$}}\gamma_{\mu } \gamma_5 \ } \label{a12 } \\ & & + { \rm tr } \ \ { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$}}\gamma_{\rho } [ \beta { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 -\alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\nu } { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$p$}}\gamma_{\mu } \gamma_5 \ } - { \rm tr } \ \ { [ \beta { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 + ( 1-\alpha){{\raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\rho } [ \beta { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 -\alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\nu } [ ( 1- \beta ) { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 + \alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\mu } \gamma_5 \ } \nonumber \end{aligned}\ ] ] the first three terms in ( [ a12 ] ) contribute to the ambiguous @xmath58 functions which will be however uniquely determined by the ward identities ( [ a08 ] ) . the last one , on the other hand , is contributing to @xmath59 and equals @xmath181 \gamma_{\rho } [ \beta { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 -\alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\nu } [ ( 1- \beta ) { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_1 + \alpha { { \raise.15ex\hbox{$/$}\kern-.63em\hbox{$k$}}}_2 ] \gamma_{\mu } \gamma_5 \ } = \nonumber \\ & & - 4 i \ \ { [ \beta ( 2 \alpha + \beta -1 ) k_{1 \rho } + \alpha ( 2- 2 \alpha - \beta ) k_{2 \rho } ] \epsilon_{\mu \nu \alpha \beta } k_1^{\alpha } k_2^{\beta } \nonumber \\ & & - \beta [ ( 1-\beta)k_{1 \mu } + \alpha k_{2 \mu } ] \epsilon_{\nu \rho \alpha \beta } k_1^{\alpha } k_2^{\beta } + \beta [ ( 1-\beta)k_{1 \mu } + \alpha k_{2 \mu } ] \epsilon_{\rho \mu \alpha \beta } k_1^{\alpha } k_2^{\beta } \nonumber \\ & & - \epsilon_{\mu \nu \rho \alpha } [ \beta^2 k_1 ^ 2 - \alpha ( 1-\alpha ) k_2 ^ 2 + ( 1- 2 \alpha ) \beta k_1 k_2 ] [ ( 1-\beta ) k_1^{\alpha } + \alpha k_2^{\alpha } ] \ } \label{a13}\end{aligned}\ ] ] integrating over the internal momentum @xmath182 and over the feynman parameters @xmath183 one finally finds @xmath184 where the a terms in ( [ a8 ] ) are determined at the end by the ward identity ( [ a08 ] ) . the last step is the symmetrization in the two gluonic legs , which leads to the final result @xmath185 where in order to find the last line we used the identities @xmath186 the contribution with two mass insertions @xmath187 are easily seen to give terms correcting the coefficients @xmath58 in ( [ a8 ] ) . as such , they are fixed by the ward identities ( [ a08 ] ) . the complete three - point function , including the @xmath58 coefficients defined in ( [ a8 ] ) , is then given by @xmath65 k_1^{\alpha } k_2^{\beta } + \epsilon_{\mu \nu \rho \alpha } k_1 k_2 ( k_2-k_1)^{\alpha } \ } \ . \label{a18}\ ] ] notice that ( [ a18 ] ) can be cast in the general form ( [ a8 ] ) . indeed , by using identities of the type ( [ a16 ] ) , one can also write @xmath188 k_1^{\alpha } k_2^{\beta } + \epsilon_{\mu \nu \rho \alpha } [ ( 2 k_1 ^ 2 + 3k_1 k_2 ) k_2^{\alpha } - ( 2 k_2 ^ 2 + 3k_1 k_2 ) k_1^{\alpha } ] \ } \ , \label{a018}\ ] ] from which the coeff . @xmath56 in ( [ a8 ] ) can be readily identified . the final result for the @xmath0 couplings is then described by the operator @xmath189 \ . \label{a17}\ ] ] the antisymmetric part of ( [ a14 ] ) , which is relevant if one replaces @xmath0 by another gluon , can be shown to vanish , by using the identities ( [ a16 ] ) . therefore , one - loops of heavy mediators do not generate triple sm gauge boson vectors operators of the type ( [ g7 ] ) and there are no new phenomenological constraints coming from purely sm contact operators . here we show that the operator @xmath27 is identically zero . the proof is the same for any sm gauge field , so we consider the gluons for definiteness . in the unitary gauge , the @xmath0-gluon - gluon vertex coming from this operator is proportional to @xmath190 in momentum space , denoting by @xmath168 the momenta of the two gluons , the linearized ( abelian ) @xmath191 vertex , after symmetrization of the two gluons , is given by @xmath192 its vanishing can be seen by starting from the identity @xmath193 the identity is actually valid for any vector @xmath194 , that can be chosen , as in ( [ a16 ] ) , to be one of the gluon momenta @xmath195 , or the @xmath0 momentum @xmath196 . if the linearized abelian part of the operator vanishes , it has to completely vanish because of gauge invariance . * @xmath133 process : @xmath197 * @xmath198 process : @xmath199 * @xmath200 process : @xmath201 notice the vanishing of the cross - section for the on - shell @xmath0 case @xmath202 , in agreement with the landau - yang theorem @xcite . * @xmath203 process : @xmath204 @xmath206\ok & -&\frac{4 m_{z'}^4}{m_{\psi}^2 \left(s-4 m_{z'}^2\right)+m_{z'}^4 } \left[m_{\psi}^4 \left(x_l^2 - 6 x_l x_r+x_r^2\right)^2\right.\ok & + & \left.\left . 2 m_{\psi}^2 \left(m_{z'}^2 \left(-3 x_l^4 + 6 x_l^3 x_r+2 x_l^2 x_r^2 + 6 x_l x_r^3 - 3 x_r^4\right)+s \left(x_l^2-x_r^2\right)^2\right)+2 m_{z'}^4 \left(x_l^4+x_r^4\right)\right]\right\}\ok\end{aligned}\ ] ] see e.g. p. langacker , rev . * 81 * ( 2009 ) 1199 [ arxiv:0801.1345 [ hep - ph ] ] ; t. g. rizzo , arxiv : hep - ph/0610104 . m. s. carena , a. daleo , b. a. dobrescu and t. m. p. tait , phys . d * 70 * ( 2004 ) 093009 [ hep - ph/0408098 ] ; e. salvioni , g. villadoro and f. zwirner , jhep * 0911 * ( 2009 ) 068 [ arxiv:0909.1320 [ hep - ph ] ] ; e. salvioni , a. strumia , g. villadoro and f. zwirner , jhep * 1003 * ( 2010 ) 010 [ arxiv:0911.1450 [ hep - ph ] ] . for a very incomplete list of early papers , see e.g. m. cvetic and p. langacker , phys . d * 54 * ( 1996 ) 3570 [ hep - ph/9511378 ] and mod . a * 11 * ( 1996 ) 1247 [ hep - ph/9602424 ] ; m. cvetic , d. a. demir , j. r. espinosa , l. l. everett and p. langacker , phys . d * 56 * ( 1997 ) 2861 [ erratum - ibid . d * 58 * ( 1998 ) 119905 ] [ hep - ph/9703317 ] ; d. m. ghilencea , l. e. ibanez , n. irges and f. quevedo , jhep * 0208 * ( 2002 ) 016 [ hep - ph/0205083 ] . p. anastasopoulos , m. bianchi , e. dudas and e. kiritsis , jhep * 0611 * ( 2006 ) 057 [ arxiv : hep - th/0605225 ] ; c. coriano , n. irges and e. kiritsis , nucl . b * 746 * ( 2006 ) 77 [ arxiv : hep - ph/0510332 ] ; p. anastasopoulos , f. fucito , a. lionetto , g. pradisi , a. racioppi and y. s. stanev , phys . d * 78 * ( 2008 ) 085014 [ arxiv:0804.1156 [ hep - th ] ] . c. coriano , n. irges and s. morelli , jhep * 0707 * ( 2007 ) 008 [ hep - ph/0701010 ] and nucl . b * 789 * ( 2008 ) 133 ; [ hep - ph/0703127 [ hep - ph ] ] c. coriano , m. guzzi and s. morelli , eur . j. c * 55 * ( 2008 ) 629 [ arxiv:0801.2949 [ hep - ph ] ] . j. kumar , a. rajaraman and j. d. wells , phys . d * 77 * ( 2008 ) 066011 [ arxiv:0707.3488 [ hep - ph ] ] ; i. antoniadis , a. boyarsky , s. espahbodi , o. ruchayskiy and j. d. wells , arxiv:0901.0639 [ hep - ph ] ; y. mambrini , jcap * 0912 * , 005 ( 2009 ) [ arxiv:0907.2918 [ hep - ph ] ] ; m. goodsell , j. jaeckel , j. redondo and a. ringwald , jhep * 0911 * ( 2009 ) 027 [ arxiv:0909.0515 [ hep - ph ] ] ; g. shiu , p. soler and f. ye , arxiv:1302.5471 [ hep - th ] ; m. cvetic , j. halverson and h. piragua , jhep * 1302 * ( 2013 ) 005 [ arxiv:1210.5245 [ hep - ph ] ] . b. kors and p. nath , phys . b * 586 * ( 2004 ) 366 [ hep - ph/0402047 ] and jhep * 0412 * ( 2004 ) 005 [ hep - ph/0406167 ] ; d. feldman , z. liu and p. nath , phys . d * 75 * ( 2007 ) 115001 [ hep - ph/0702123 [ hep - ph ] ] ; d. feldman , b. kors and p. nath , phys . rev . d * 75 * , 023503 ( 2007 ) [ arxiv : hep - ph/0610133 ] . e. dudas , y. mambrini , s. pokorski and a. romagnoni , jhep * 0908 * ( 2009 ) 014 [ arxiv:0904.1745 [ hep - ph ] ] ; e. dudas , y. mambrini , s. pokorski , a. romagnoni and , jhep * 1210 * ( 2012 ) 123 [ arxiv:1205.1520 [ hep - ph ] ] . c. b. jackson , g. servant , g. shaughnessy , t. m. p. tait , m. taoso and , jcap * 1004 * ( 2010 ) 004 [ arxiv:0912.0004 [ hep - ph ] ] ; c. b. jackson , g. servant , g. shaughnessy , t. m. p. tait , m. taoso and , arxiv:1302.1802 [ hep - ph ] . t. bringmann , x. huang , a. ibarra , s. vogl and c. weniger , jcap * 1207 * ( 2012 ) 054 [ arxiv:1203.1312 [ hep - ph ] ] ; c. weniger , jcap * 1208 * ( 2012 ) 007 [ arxiv:1204.2797 [ hep - ph ] ] ; for a recent update on the prospects to confirm or to infirm this signature , see e.g. c. weniger , m. su , d. p. finkbeiner , t. bringmann and n. mirabal , arxiv:1305.4710 [ astro-ph.he ] . b. de wit , p. g. lauwers and a. van proeyen , nucl . b * 255 * ( 1985 ) 569 ; i. antoniadis , e. kiritsis and t. n. tomaras , phys . b * 486 * ( 2000 ) 186 [ arxiv : hep - ph/0004214 ] ; e. dudas , a. falkowski and s. pokorski , phys . b * 568 * ( 2003 ) 281 [ arxiv : hep - th/0303155 ] ; l. andrianopoli , s. ferrara and m. a. lledo , jhep * 0404 * ( 2004 ) 005 [ arxiv : hep - th/0402142 ] ; e. dudas , t. gherghetta and s. groot nibbelink , phys . d * 70 * ( 2004 ) 086012 [ arxiv : hep - th/0404094 ] ; j. de rydt , j. rosseel , t. t. schmidt , a. van proeyen and m. zagermann , class . * 24 * ( 2007 ) 5201 [ arxiv:0705.4216 [ hep - th ] ] ; j. de rydt , t. t. schmidt , m. trigiante , a. van proeyen and m. zagermann , jhep * 0812 * ( 2008 ) 105 [ arxiv:0808.2130 [ hep - th ] ] . b. holdom , phys . lett . b * 166 * ( 1986 ) 196 ; k. r. dienes , c. f. kolda and j. march - russell , nucl . b * 492 * ( 1997 ) 104 [ hep - ph/9610479 ] ; s. a. abel , m. d. goodsell , j. jaeckel , v. v. khoze and a. ringwald , jhep * 0807 * ( 2008 ) 124 [ arxiv:0803.1449 [ hep - ph ] ] ; m. pospelov , phys . d * 80 * ( 2009 ) 095002 [ arxiv:0811.1030 [ hep - ph ] ] ; e. j. chun , j. c. park and s. scopel , jhep * 1102 * ( 2011 ) 100 [ arxiv:1011.3300 [ hep - ph ] ] . g. belanger , f. boudjema , p. brun , a. pukhov , s. rosier - lees , p. salati and a. semenov , comput . commun . * 182 * ( 2011 ) 842 [ arxiv:1004.1092 [ hep - ph ] ] ; g. belanger , f. boudjema , a. pukhov and a. semenov , comput . commun . * 180 * ( 2009 ) 747 [ arxiv:0803.2360 [ hep - ph ] ] ; g. belanger , f. boudjema , a. pukhov and a. semenov , arxiv:1305.0237 [ hep - ph ] . a. a. abdo _ et al . _ [ fermi - lat collaboration ] , astrophys . j. * 712 * ( 2010 ) 147 [ arxiv:1001.4531 [ astro-ph.co ] ] ; a. drlica - wagner [ fermi lat collaboration ] , arxiv:1210.5558 [ astro-ph.he ] ; y. -l . s. tsai , q. yuan and x. huang , jcap * 1303 * ( 2013 ) 018 [ arxiv:1212.3990 [ astro-ph.he ] ] . a. e. egorov and e. pierpaoli , arxiv:1304.0517 [ astro-ph.co ] ; r. laha , k. c. y. ng , b. dasgupta and s. horiuchi , phys . d * 87 * ( 2013 ) 043516 [ arxiv:1208.5488 [ astro-ph.co ] ] ; y. mambrini , m. h. g. tytgat , g. zaharijas and b. zaldivar , jcap * 1211 * ( 2012 ) 038 [ arxiv:1206.2352 [ hep - ph ] ] . p. j. fox , r. harnik , j. kopp and y. tsai , phys . d * 84 * ( 2011 ) 014028 [ arxiv:1103.0240 [ hep - ph ] ] . h. m. lee , m. park and v. sanz , jhep * 1303 * ( 2013 ) 052 [ arxiv:1212.5647 [ hep - ph ] ] . y. mambrini and b. zaldivar , jcap * 1110 * ( 2011 ) 023 [ arxiv:1106.4819 [ hep - ph ] ] . j. goodman , m. ibe , a. rajaraman , w. shepherd , t. m. p. tait and h. -b . yu , phys . b * 695 * ( 2011 ) 185 [ arxiv:1005.1286 [ hep - ph ] ] ; j. goodman , m. ibe , a. rajaraman , w. shepherd , t. m. p. tait and h. -b . yu , phys . d * 82 * ( 2010 ) 116010 [ arxiv:1008.1783 [ hep - ph ] ] ; a. rajaraman , w. shepherd , t. m. p. tait and a. m. wijangco , phys . d * 84 * ( 2011 ) 095013 [ arxiv:1108.1196 [ hep - ph ] ] . s. andreas , c. niebuhr and a. ringwald , phys . d * 86 * ( 2012 ) 095019 [ arxiv:1209.6083 [ hep - ph ] ] ; s. andreas , m. d. goodsell and a. ringwald , phys . d * 87 * ( 2013 ) 025007 [ arxiv:1109.2869 [ hep - ph ] ] ; n. fornengo , p. panci and m. regis , phys . rev . d * 84 * ( 2011 ) 115002 [ arxiv:1108.4661 [ hep - ph ] ] . y. mambrini , jcap * 1107 * ( 2011 ) 009 [ arxiv:1104.4799 [ hep - ph ] ] ; y. mambrini , jcap * 1009 * ( 2010 ) 022 [ arxiv:1006.3318 [ hep - ph ] ] . m. t. frandsen , f. kahlhoefer , a. preston , s. sarkar and k. schmidt - hoberg , jhep * 1207 * ( 2012 ) 123 [ arxiv:1204.3839 [ hep - ph ] ] . m. goodsell , j. jaeckel , j. redondo and a. ringwald , jhep * 0911 * ( 2009 ) 027 [ arxiv:0909.0515 [ hep - ph ] ] ; s. a. abel , m. d. goodsell , j. jaeckel , v. v. khoze and a. ringwald , jhep * 0807 * ( 2008 ) 124 [ arxiv:0803.1449 [ hep - ph ] ] ;

the equivalence ( or weak equivalence ) classes of orientation - preserving free actions of a finite group on an orientable-dimensional handlebody of genus can be enumerated in terms of sets of generators of . they correspond to the equivalence classes of generating-vectors of elements of , where , under nielsen equivalence ( or weak nielsen equivalence ) . for abelian and dihedral , this allows a complete determination of the equivalence and weak equivalence classes of actions for all genera . additional information is obtained for solvable groups and for the groups with prime . for all , there is only one equivalence class of actions on the genus handlebody if is at least , where is the maximal length of a chain of subgroups of . there is a stabilization process that sends an equivalence class of actions to an equivalence class of actions on a higher genus , and some results about its effects are obtained .

math0109166

the orientation - preserving free actions of a finite group @xmath0 on 3-dimensional orientable handlebodies have a close connection with a long - studied concept from group theory , namely _ nielsen equivalence _ of generating sets . indeed , as we observe in section [ sec : theory ] below , the free actions of @xmath0 on the handlebody of genus @xmath7 , up to equivalence , correspond to the nielsen equivalence classes of @xmath3-element generating sets of @xmath0 , where @xmath4 . we will utilize this to prove a number of results about equivalence and weak equivalence of free actions . these results are summarized in concise form in section [ sec : results ] , which also contains precise definitions of equivalence , weak equivalence , and other concepts that we shall use . a special feature of free actions on handlebodies is that there is a stabilization process relating actions on different genera . when a handlebody @xmath10 with a free @xmath0-action contains a @xmath0-invariant handlebody @xmath11 such that @xmath12 consists of disjoint @xmath13-handles , the action on @xmath10 is called a stabilization of the action on @xmath11 . inequivalent actions can become equivalent after stabilization , indeed we do not know an example of actions that remain inequivalent after even an elementary stabilization ( i. e. a stabilization for which @xmath12 consists of @xmath14 @xmath13-handles ) . such an example could not involve a solvable group , since a result of m. dunwoody implies that for solvable @xmath0 , any two actions on a handlebody of more than the minimum possible genus for a @xmath0-action are equivalent ( see corollary [ coro : solvable corollary 2 ] below ) . for an arbitrary @xmath0 , proposition [ prop : stabilization ] shows that any two actions become equivalent after @xmath15 elementary stabilizations , where @xmath15 is the minimum number of generators of @xmath0 . we remark that there is an interesting theory of stabilization of actions on @xmath16-manifolds @xcite . the connection between free actions on handlebodies and nielsen equivalence is well - known in some circles , although we can not find an explicit statement in the literature . it was known to j. kalliongis and a. miller and is is a direct consequence of theorem 1.3 in their paper @xcite ( for free actions , the graph of groups will have trivial vertex and edge groups , and the equivalence of graphs of groups defined there is readily seen to be the same as nielsen equivalence on generating sets of @xmath0 ) . indeed , more delicate classifications of nonfree actions on handlebodies have been examined in considerable depth . a general theory of actions was given in @xcite and @xcite , and the actions on very low genera were extensively studied in @xcite . actions with the genus small relative to the order of the group are investigated in @xcite , and the special case of orientation - reversing involutions is treated in @xcite . the first focus on free actions seems to be @xcite , where it was proven that for a cyclic group , any free action on a handlebody of genus above the minimal one is the stabilization of an action on minimal genus , and that any two free actions on a handlebody are weakly equivalent . these results were generalized to dihedral and abelian groups in @xcite , whose results are reconfirmed and extended in section [ sec : abelian and dihedral ] below . some of the arguments in this paper can be shortened by invoking results from @xcite . since the general theory given there is much more elaborate than the elementary methods needed for the present work , we have chosen to make our arguments self - contained . after giving some more precise definitions and stating our main results in section [ sec : results ] , we develop the general theory relating free actions to nielsen equivalence in section [ sec : theory ] . we apply this in section [ sec : solvable groups ] to treat the case when @xmath0 is solvable , and in section [ sec : abelian and dihedral ] we examine the specific cases of abelian and dihedral groups . in section [ sec : free actions of psl(2,3p ) ] , we show that for @xmath6 prime , two free actions of @xmath5 on a handlebody of genus above the minimal one are equivalent . by work of evans @xcite and gilman @xcite , it is known that the same is true for @xmath17 ( for all @xmath18 ) and @xmath19 ( for @xmath6 prime ) . section [ sec : stabilization of actions ] gives some general results on stabilization , in particular , we prove that if the genus of @xmath10 is at least @xmath20 , where @xmath9 is the maximum length of a decreasing chain of nonzero subgroups of @xmath0 , then any two free @xmath0-actions are equivalent . in section [ sec : questions ] , we state some open problems . in particular , do there even exist inequivalent actions that are not minimal genus actions ? two ( effective ) actions @xmath21 are said to be _ equivalent _ if they are conjugate as representations , that is , if there is a homeomorphism @xmath22 such that @xmath23 for each @xmath24 . they are _ weakly equivalent _ if their images are conjugate , that is , if there is a homeomorphism @xmath22 so that @xmath25 . said differently , there is some automorphism @xmath26 of @xmath0 so that @xmath27 for all @xmath7 . in words , equivalent actions are the same after a change of coordinates on the space , while weakly equivalent actions are the same after a change of coordinates on the space and possibly a change of the group by automorphism . if @xmath28 is homeomorphic to @xmath29 , then the sets of equivalence ( or weak equivalence ) classes of actions on @xmath28 and on @xmath29 can be put into correspondence using any homeomorphism from @xmath28 to @xmath29 . henceforth the term _ action _ will mean an orientation - preserving _ free _ action of a finite group on a @xmath1-dimensional orientable handlebody of genus @xmath2 ( only the trivial group can act freely on the handlebody of genus @xmath30 , the @xmath1-ball ) . one may work with either piecewise - linear or smooth actions ; we assume that one of these categories has been chosen , and that all maps , isotopies , etc . lie in the category . for a finite group @xmath0 we denote by @xmath15 the minimum number of generators in any generating set for @xmath0 , and by @xmath9 the maximum @xmath31 such that @xmath32 is a properly descending chain of subgroups of @xmath0 . when there is only one group @xmath0 under consideration , we often write @xmath33 for @xmath15 and @xmath31 for @xmath9 . fix a finite group @xmath0 , and consider an action of @xmath0 on a handlebody @xmath10 . the quotient map @xmath34 is a covering map , so the action corresponds to an extension @xmath35 a torsionfree finite extension of a finitely generated free group is free ( by @xcite any finitely generated virtually free group is the fundamental group of a graph of groups with finite vertex groups , and if the group is torsionfree , the vertex groups must be trivial ) . so @xmath36 is free . since @xmath10 is irreducible , so is @xmath37 , and theorem 5.2 of @xcite shows that @xmath37 is a handlebody . from equation ( @xmath38 ) , the genus of @xmath37 must be at least @xmath33 , so its euler characteristic is at most @xmath39 . therefore the euler characteristic of @xmath10 is at most @xmath40 , and the genus of @xmath10 is at least @xmath41 . on the other hand , if @xmath42 is a handlebody of genus @xmath33 , we may fix any surjective homomorphism from @xmath43 to @xmath0 , and the covering space @xmath10 corresponding to its kernel has genus @xmath41 and admits an action of @xmath0 . so @xmath41 is the minimal genus among the handlebodies that admit a @xmath0-action . we denote this minimal genus by @xmath44 , and we call an action of @xmath0 on a handlebody of genus @xmath44 a _ minimal genus action . _ if @xmath10 is any handlebody that admits a @xmath0-action , then the genus of @xmath37 is @xmath45 for some integer @xmath46 , so the genus of @xmath10 is @xmath47 . denote by @xmath48 the set of equivalence classes of actions of @xmath0 on a handlebody of genus @xmath47 , and by @xmath49 the set of weak equivalence classes . in particular , @xmath50 and @xmath51 are the equivalence classes of minimal genus actions . there is a natural surjection from @xmath48 to @xmath49 . in the introduction we explained that an action of @xmath0 on @xmath10 is a stabilization of an action on @xmath11 if and only if there is a @xmath0-equivariant imbedding of @xmath11 into @xmath10 such that @xmath12 consists of @xmath13-handles . we now give a more convenient description . for @xmath0 acting on @xmath10 , choose a disc @xmath52 which is disjoint from all of its @xmath0-translates . attach a @xmath13-handle @xmath53 to @xmath10 using an orientation - reversing imbedding @xmath54 . for each @xmath24 attach a @xmath13-handle to @xmath55 using the imbedding @xmath56 . the @xmath0-action on @xmath10 extends to an action of @xmath0 on the union of @xmath10 with these @xmath13-handles , and this action is called an _ elementary stabilization _ of the original action . alternatively , we may think of this as attaching a @xmath13-handle to a disc in the boundary of @xmath37 and extending the homomorphism @xmath57 that determines the action in the exact sequence @xmath58 to a homomorphism from @xmath59 to @xmath0 by sending the generator of @xmath60 to @xmath13 . since any two discs in @xmath61 are isotopic , and any orientation - reversing imbeddings of @xmath62 into a disc are isotopic , the equivalence class of the resulting action is well - defined . the result of applying some number of elementary stabilizations is called a _ stabilization _ of the original action . for each @xmath46 and @xmath63 , stabilization gives well - defined functions from @xmath48 to @xmath64 and from @xmath49 to @xmath65 . let @xmath66 denote the cardinality of @xmath48 , and @xmath67 the cardinality of @xmath49 . clearly @xmath68 , and @xmath69 for @xmath46 . recall that the euler @xmath70-function is defined on positive integers by @xmath71 , and for @xmath72 , @xmath73 is the number of integers @xmath74 with @xmath75 such that @xmath76 . notice that for @xmath77 , @xmath73 is even , since if @xmath76 then @xmath78 . using these notations , we can give concise statements of our main results : 1 . if @xmath0 is solvable , then @xmath79 for all @xmath80 ( corollary [ coro : solvable corollary 2 ] ) , while @xmath81 can be arbitrarily large ( theorem [ thm : n minimal classes ] ) . 2 . if @xmath0 is abelian , with @xmath82 , where @xmath83 for @xmath84 , then @xmath85 , and @xmath86 if @xmath87 , otherwise @xmath88 ( theorem [ thm : abelian ] ) . 3 . if @xmath0 is dihedral of order @xmath89 , then @xmath85 , and @xmath86 if @xmath90 , otherwise @xmath91 ( theorem [ thm : dihedral ] ) . 4 . if @xmath92 with @xmath6 prime , then @xmath79 for all @xmath80 ( corollary [ coro : actions of psl(2,3p ) ] ) . 5 . the smallest genus of handlebody admitting inequivalent actions of a group is @xmath93 , which has two equivalence classes of actions of the dihedral group of order @xmath94 ( corollary [ coro : smallest genus actions ] ) . the smallest genus of handlebody admitting inequivalent actions of an abelian group is @xmath95 , which has two equivalence classes of @xmath96-actions ( corollary [ coro : smallest genus actions ] ) . 7 . for all @xmath0 and all @xmath46 , @xmath97 and @xmath98 are constant ( proposition [ prop : stabilization ] ) . 8 . for all @xmath0 and all @xmath99 , @xmath79 ( corollary [ coro : limit stabilization ] ) . in addition to these general results , we will see calculations for several specific groups . the results given above indicate that there is a strong tendency for actions to become equivalent after stabilization , and as mentioned in the introduction , we do not know even a single example of inequivalent actions which do not become equivalent after one elementary stabilization . even in the simplest of cases , however , the underlying topology of an equivalence between the stablizations of two inequivalent actions can be surprisingly complicated . figure [ fig : actions ] gives the steps of a visualization of an explicit equivalence between two actions of the cyclic group @xmath100 of order @xmath101 on the handlebody of genus @xmath102 which are the stabilizations of inequivalent actions of @xmath100 on the solid torus ( the classification of the actions of cyclic groups is detailed in corollary [ coro : cyclic actions ] ) . the main result of this section , theorem [ thm : equivalence classes ] , gives the correspondence between actions of @xmath0 and nielsen equivalence classes of generating vectors of @xmath0 . the proof is elementary , requiring only the basic theory of covering spaces and some well - known facts about free groups . we deduce , in corollary [ coro : stabilization criterion ] , an algebraic criterion for an action to be a stabilization of another action . we close with a few examples that illustrate the theory . consider two actions of @xmath0 on genus @xmath7 handlebodies @xmath103 and @xmath104 . since the euler characteristics of the quotients must be equal ( to @xmath105 ) , we may choose diffeomorphisms from them to a single handlebody @xmath42 . the original actions are then equivalent to the actions of @xmath0 by covering transformations on the covering spaces of @xmath42 determined by two surjective homomorphisms @xmath106 . that is , in classifying actions of a fixed group @xmath0 on handlebodies of a fixed genus @xmath7 , up to equivalence or up to weak equivalence , we may assume that their quotients are the same handlebody @xmath42 . a _ generating vector _ for a group is a tuple @xmath107 such that @xmath108 is a generating set . a generating vector @xmath109 of @xmath110 is said to be obtained from @xmath111 by a _ nielsen move _ if there is a @xmath112 so that @xmath113 for all @xmath114 , and for some @xmath115 , @xmath116 equals @xmath117 or @xmath118 . also , an interchange of two of the entries , or the replacement of an entry by its inverse , is a nielsen move . generating vectors are called _ nielsen equivalent _ if there is a sequence of nielsen moves that changes one to the other . generating vectors @xmath107 and @xmath109 are called _ weakly nielsen equivalent _ if there is an automorphism @xmath26 of @xmath0 such that @xmath119 is equivalent to @xmath120 , where @xmath121 . note that equivalent or weakly equivalent generating vectors must have the same number of elements . a minimal generating vector for a free group is called a _ basis . _ if @xmath110 is a free group with basis @xmath111 , then any nielsen move on @xmath111 induces an automorphism of @xmath110 . nielsen proved @xcite that any two bases for @xmath110 are nielsen equivalent , consequently the nielsen moves generate the automorphism group of @xmath110 . associated to a given handlebody structure on a handlebody @xmath42 is a standard basis of the free group @xmath43 , where the @xmath122 generator corresponds to a loop that goes once around the @xmath122 @xmath13-handle and not around any other handle . any nielsen move on this basis is induced by an orientation - preserving diffeomorphism of @xmath42 ( see for example @xcite ) , so any automorphism of @xmath43 can be induced by an orientation - preserving diffeomorphism of @xmath42 . this also shows that associated to any basis of @xmath43 is a handlebody structure with respect to which the basis is standard . let @xmath42 be a handlebody , and let @xmath123 be surjective homomorphisms to a finite group @xmath0 . let @xmath124 be a basis for @xmath43 , so that @xmath125 and @xmath126 are generating vectors for @xmath0 . then @xmath111 and @xmath120 are weakly nielsen equivalent if and only there are an isomorphism @xmath127 and an isomorphism @xmath26 of @xmath0 such that @xmath128 . they are nielsen equivalent if and only if @xmath26 can be taken to be the identity automorphism of @xmath0 . [ lem : nielsen equivalence ] suppose @xmath129 and @xmath26 exist . since @xmath130 is a basis for @xmath43 , there is a sequence of nielsen moves that carries @xmath130 to @xmath124 . applying @xmath131 shows that the corresponding nielsen moves in @xmath0 carry @xmath132 to @xmath133 . the latter is @xmath120 , and since @xmath134 , the former is @xmath119 . conversely , suppose the generating vectors are weakly nielsen equivalent . for a sequence of nielsen moves carrying @xmath120 to @xmath119 , lifting the nielsen moves to corresponding nielsen moves starting from @xmath124 yields an isomorphism @xmath129 of @xmath43 carrying @xmath124 to a basis @xmath130 such that @xmath135 . this proves the lemma for weak equivalence . the same argument , taking @xmath26 to be the identity automorphism , gives it for equivalence . to translate this algebraic information into statements about equivalence of actions , we use the following consequence of the theory of covering spaces . let @xmath136 determine actions on handlebodies . the actions are weakly equivalent if and only if there are an isomorphism @xmath137 and an isomorphism @xmath138 such that @xmath128 . they are equivalent if and only if @xmath26 may be taken to be the identity . [ lem : equivalence ] suppose that @xmath129 and @xmath26 exist . for @xmath139 , let @xmath140 be the covering space of @xmath42 corresponding to the kernel of @xmath141 . choose a diffeomorphism @xmath142 inducing @xmath129 . since @xmath129 must take the kernel of @xmath143 to the kernel of @xmath131 , @xmath144 lifts to a diffeomorphism from @xmath103 to @xmath104 . also , @xmath129 induces @xmath26 from @xmath145 to @xmath146 . using covering space theory , one can check that this determines a weak equivalence of the actions , and an equivalence when @xmath26 is the identity automorphism . conversely , if the weak equivalence ( or equivalence , when @xmath147 ) @xmath148 exists , then since @xmath149 , @xmath110 induces a diffeomorphism @xmath150 . again by covering space theory , the induced automorphism @xmath129 of @xmath144 on @xmath43 satisfies @xmath151 . putting these two lemmas together gives our main classification theorem . let @xmath0 be finite , let @xmath152 , and let @xmath153 . then the weak equivalence classes of actions of @xmath0 on genus @xmath7 handlebodies correspond bijectively to the weak equivalence classes of @xmath3-element generating sets for @xmath0 . the equivalence classes of actions correspond to the equivalence classes of @xmath3-element generating sets . [ thm : equivalence classes ] for @xmath154 , both sets are empty , so we assume that @xmath155 . let @xmath42 be a handlebody of genus @xmath3 and fix a basis @xmath156 for @xmath43 . any action is equivalent to one determined by a surjective homomorphism @xmath157 . by lemmas [ lem : equivalence ] and [ lem : nielsen equivalence ] , sending the action determined by @xmath158 to the basis @xmath159 determines a bijection from the set of ( equivalence or ) weak equivalence classes to the set of ( equivalence or ) weak equivalence classes of @xmath3-element generating vectors of @xmath0 . let @xmath160 determine a @xmath0-action on a handlebody @xmath10 , and let @xmath161 be obtained from @xmath42 by attaching a single @xmath13-handle . then @xmath162 , where the generator of @xmath60 corresponds to a loop that goes once around the additional @xmath13-handle . we have noted that the action resulting from an elementary stabilization of the action on @xmath10 is determined by the homomorphism @xmath163 which equals @xmath158 on @xmath43 and sends the generator of @xmath60 to @xmath13 . suppose @xmath0 acts on handlebodies @xmath11 and @xmath10 , where the genus of @xmath11 is less than the genus of @xmath10 . let @xmath164 be a basis for @xmath165 and let @xmath166 be a basis for @xmath36 , and let @xmath167 and @xmath168 determine the actions . then the action on @xmath10 is a stabilization of the action on @xmath11 if and only if the generating @xmath3-vectors @xmath169 and @xmath170 of @xmath0 are equivalent . [ coro : stabilization criterion ] put @xmath171 and @xmath172 . since for any basis there is a handlebody structure for which the basis is standard , there is an inclusion @xmath173 so that @xmath174 carries @xmath175 to @xmath176 for @xmath177 , and so that @xmath178 is a disjoint union of @xmath13-handles . then , the stabilized action is determined by the homomorphism @xmath179 for which @xmath180 for @xmath177 , and @xmath181 for @xmath182 . by theorem [ thm : equivalence classes ] , the stabilized action and the action on @xmath10 are equivalent if and only if @xmath169 and @xmath170 of @xmath0 are nielsen equivalent . this provides a simple criterion for an action to be a stabilization . a vector of elements of @xmath0 is called _ redundant _ if it is nielsen equivalent to a vector with an entry equal to @xmath13 . notice that a vector is redundant if it is even weakly nielsen equivalent to a vector with an entry equal to @xmath13 . an action of @xmath0 on a handlebody @xmath10 is a stabilization if and only if a generating vector @xmath183 of @xmath0 corresponding to the action as in theorem [ thm : equivalence classes ] is redundant . [ coro : redundant vector ] the literature contains a number of results on nielsen equivalence of generating vectors . they were used to count @xmath0-defining subgroups of free groups in @xcite . the action of the automorphism group of the free group on generating vectors appears to have been introduced in @xcite . for infinite groups , there are quite a few instances of inequivalent generating vectors of cardinality greater than @xmath33 . the paper of m. evans @xcite gives general constructions of these , as well as a summary of earlier results . but for finite groups , no such example is known ( see section [ sec : questions ] below ) . in the remainder of this section , we will collect some of the known calculations for specific finite groups , and give their consequences for group actions . some important general results of m. dunwoody for solvable groups will be stated and used in the next section . for @xmath184 , b. neumann and h. neumann @xcite ( see also @xcite ) showed that there are two weak equivalence classes of generating @xmath16-vectors for @xmath185 . thus there are two weak equivalence classes of actions on the handlebody of genus @xmath186 . d. stork @xcite carried out similar calculations for @xmath187 and @xmath188 . these show , for example , that there are @xmath189 weak equivalence classes of @xmath188-actions on the handlebody of genus @xmath190 , the minimal genus . techniques developed by m. lustig @xcite using the fox calculus yield additional examples of inequivalent generating systems for certain groups . gilman @xcite proved that for @xmath6 prime , all @xmath1-element generating sets of @xmath19 are equivalent . that is , @xmath79 for all @xmath80 for these groups . in particular , this holds for @xmath191 , so the two equivalence classes of actions of @xmath185 on the handlebody of genus @xmath192 become equivalent after a single elementary stabilization . in @xcite , m. evans proved that when @xmath0 is @xmath193 or the suzuki group @xmath194 for @xmath18 , then for any @xmath195 , all @xmath3-element generating vectors are equivalent . since @xmath196 for any of these groups , this says that all actions of one of these groups on any genus above the minimal genus are equivalent . in section [ sec : free actions of psl(2,3p ) ] , we will prove a similar result for @xmath5 with @xmath6 prime . this includes the case of @xmath197 , so the four inequivalent @xmath188-actions on the handlebody of genus @xmath190 all become equivalent after a single elementary stabilization . in this section we use results of dunwoody to examine the actions of solvable groups . they show that although there can be an arbitrarily large number of weak equivalence classes of minimal genus actions , all actions on a handlebody whose genus is above the minimal one are equivalent . from @xcite , we have the following fact . let @xmath0 be a solvable group , and let @xmath198 . then any two @xmath3-element generating vectors are nielsen equivalent . [ thm : dunwoody theorem ] applying theorem [ thm : equivalence classes ] , we have immediately : let @xmath0 be solvable and let @xmath7 be greater than @xmath44 . then any two actions of @xmath0 on handlebodies of genus @xmath7 are equivalent . consequently , any action of a solvable group on a handlebody not of minimal genus can be destabilized to any given action of minimal genus . [ coro : solvable corollary 2 ] some examples due to dunwoody also have implications for free actions . the following examples are from @xcite : for every pair of positive integers @xmath3 and @xmath199 , there exists a @xmath6-group @xmath200 , nilpotent of length @xmath16 and with @xmath201 , which has at least @xmath199 weak nielsen equivalence classes of @xmath3-element generating sets . [ thm : n minimal classes ] this yields immediately : for every pair of positive integers @xmath3 and @xmath199 , there exists a @xmath6-group @xmath200 , nilpotent of length @xmath16 , with @xmath202 and @xmath203 . [ coro : n minimal actions ] of course , by corollary [ coro : solvable corollary 2 ] , all of the actions in @xmath50 become equivalent after a single stabilization . we include here an elementary and transparent example of two weakly inequivalent @xmath0-actions of a nilpotent group of order @xmath204 on the handlebody of genus @xmath205 . b. neumann @xcite gave a nilpotent group of order @xmath206 admitting weakly inequivalent actions on the handlebody of this same genus ; his example is slightly more complicated , but requires only two generators , whereas ours requires @xmath1 . dunwoody s examples in @xcite are considerably more sophisticated renderings of the one we give here . let @xmath0 be the group with presentation @xmath207=[y , z]=1 , [ x , y]=z^8\rangle\ .\ ] ] there is an extension@xmath208 where @xmath209 is the subgroup generated by @xmath210 , and the images of @xmath211 and @xmath212 generate @xmath213 . now @xmath214 , from which it follows that @xmath215 for all integers @xmath216 and @xmath217 . we will show that the generating vectors @xmath218 and @xmath219 are not weakly nielsen equivalent . by theorem [ thm : equivalence classes ] , these correspond to two actions of @xmath0 on the handlebody of genus @xmath220 which are not weakly equivalent . every element of @xmath0 can be written in the form @xmath221 , where @xmath216 and @xmath217 are integers mod 8 , and @xmath222 is mod 64 , and the inverse of @xmath221 is @xmath223 . using this , we can calculate that @xmath224=z^{8(ps - rq)}$ ] . sending @xmath211 to @xmath225 , @xmath212 to @xmath226 , and @xmath210 to @xmath227 defines a homomorphism @xmath228 . regarding these as vectors of integers mod 8 , any three - element generating set determines a @xmath229 matrix with entries mod 8 . nielsen moves on the generating set only change the determinant of the associated matrix by multiplication by @xmath230 . the determinant of the matrix associated to @xmath231 is @xmath1 . the subgroup @xmath209 is central . if @xmath221 has @xmath232 , then @xmath233 , so @xmath221 is not central . similarly , if @xmath234 , the element is not central . therefore the center of @xmath0 is exactly @xmath209 , and any automorphism of @xmath0 must carry @xmath210 to @xmath235 for some @xmath236 . now consider any automorphism @xmath26 of @xmath0 , with @xmath237 , @xmath238 , and @xmath239 . from above , @xmath240)=z^{8(ps - rq)}$ ] . since this must equal @xmath241 , it follows that @xmath242 is congruent to @xmath236 modulo @xmath243 . the matrix associated to the generating vector @xmath244 is @xmath245 which has determinant @xmath246 . this is congruent to @xmath247 . since the only squares modulo @xmath243 are @xmath13 and @xmath189 , it follows that @xmath244 can not be nielsen equivalent to the generating set @xmath219 . therefore the generating vectors @xmath218 and @xmath219 are not weakly nielsen equivalent . in this section we examine the actions of abelian and dihedral groups . we will see that for either of these two kinds of groups , all actions on the minimal genus are weakly equivalent , but that there can be arbitrarily large numbers of equivalence classes . let @xmath248 be a finite abelian group , @xmath249 where @xmath83 for @xmath84 . then @xmath250 . if @xmath251 , then any two @xmath248-actions on a handlebody of genus @xmath7 are equivalent . any two @xmath248-actions on a handlebody of genus @xmath252 are weakly equivalent . if @xmath87 , then all @xmath248-actions on the handlebody of genus @xmath252 are equivalent , while if @xmath253 , then there are exactly @xmath254 equivalence classes , where @xmath70 denotes the euler @xmath70-function . [ thm : abelian ] in preparation for the proof , we will prove a general lemma about vectors of elements in solvable groups . by a _ cyclic tower _ for a group @xmath0 we mean a descending sequence of subgroups @xmath255 such that @xmath256 is normal in @xmath257 and @xmath258 is cyclic . we allow @xmath257 to equal @xmath256 for some @xmath259 , also we define @xmath260 for all @xmath261 . let @xmath0 be solvable and let @xmath262 be a cyclic tower for @xmath0 . let @xmath109 be a vector of elements of @xmath0 ( not necessarily a generating vector ) . then @xmath120 is nielsen equivalent to a vector with @xmath263 for all @xmath259 ( in particular , @xmath264 for any @xmath182 ) . [ lem : cyclic towers ] let @xmath110 be the subgroup of @xmath0 generated by @xmath120 . replacing each @xmath257 by @xmath265 , we may assume that @xmath120 is a generating set for @xmath0 . suppose first that @xmath0 is a cyclic group @xmath266 , generated by @xmath267 , and that at least two of the @xmath268 , say @xmath269 and @xmath270 , are not equal to @xmath13 . write @xmath271 and @xmath272 , where @xmath273 . if @xmath274 , replace @xmath269 by @xmath275 , then @xmath276 and @xmath272 . if @xmath277 , replace @xmath270 by @xmath278 . repeat this process until either @xmath279 or @xmath280 . by an interchange , we may assume that @xmath281 and @xmath280 . repeating the process with the other elements , we eventually achieve that @xmath264 for all @xmath282 . in the general case , we may regard @xmath120 as a vector of elements in the cyclic group @xmath283 . by the cyclic case , we may assume that @xmath284 for all @xmath285 . the subgroup of @xmath0 generated by @xmath286 has a cyclic tower of length @xmath287 . by induction on @xmath288 , @xmath289 is nielsen equivalent to a vector with @xmath263 . we have @xmath290 , since the minimal number of generators of @xmath291 is @xmath288 , so @xmath250 . if @xmath251 , then corollary [ coro : solvable corollary 2 ] shows that any two @xmath248-actions on a handlebody of genus @xmath7 are equivalent . regard @xmath0 as solvable with cyclic tower given by @xmath292 . let @xmath107 be any generating vector . by lemma [ lem : cyclic towers ] , we may assume that @xmath293 , and consequently @xmath294 generates @xmath283 . now @xmath295 generate @xmath296 , since otherwise the quotient of @xmath248 by the subgroup that it generates is of the form @xmath297 with @xmath298 and the order of @xmath299 not relatively prime , but this quotient could not be generated by @xmath294 . so we may apply nielsen moves changing @xmath294 by multiples of the other @xmath300 until @xmath301 . inductively , we may assume that @xmath302 and generates @xmath303 for @xmath177 , and @xmath304 for @xmath182 . let @xmath305 be any another @xmath3-element generating vector for @xmath248 . again , we may assume that @xmath306 and @xmath268 generates @xmath303 , so @xmath307 and @xmath308 for some @xmath309 and @xmath310 relatively prime to @xmath311 . by nielsen moves , replace @xmath270 by @xmath312 , then @xmath269 by @xmath313 . since @xmath314 is relatively prime to @xmath298 and @xmath315 divides @xmath298 , @xmath314 is also relatively prime to @xmath315 . so there exists @xmath316 with @xmath317 , and by nielsen moves we may replace @xmath312 by @xmath318 . interchanging @xmath269 and @xmath270 , we have that @xmath319 and @xmath270 still generates @xmath320 . continuing , we may assume that @xmath321 for all @xmath322 . if @xmath323 , then @xmath321 for all @xmath259 since both equal @xmath13 for @xmath182 . this proves that all actions on genera greater than @xmath252 are equivalent . if @xmath324 , then we have only that @xmath325 , with @xmath6 relatively prime to @xmath326 . if @xmath87 , then @xmath120 must be nielsen equivalent to @xmath111 . from now on , assume that @xmath253 . the automorphism @xmath26 of @xmath248 defined by @xmath327 for @xmath328 and @xmath329 shows that all @xmath288-element generating vectors are weakly nielsen equivalent . since @xmath330 is nielsen equivalent to @xmath331 , there are at most @xmath332 equivalence classes . to show that this is a lower bound for the number of equivalence classes , we define a function from ( ordered ) generating sets to @xmath333 matrices as follows . regard @xmath300 as an @xmath288-tuple with all entries @xmath30 except for a @xmath13 in the @xmath122 place . working @xmath334 , any generating @xmath288-vector then determines an @xmath333 matrix with @xmath122 row the vector corresponding to @xmath268 . a nielsen move corresponds to multiplying by an elementary matrix ( that adds one row to another , or interchanges two rows , or multiplies one row by @xmath335 ) . these elementary matrices have determinant @xmath336 , so nielsen equivalent generating sets have determinants that are either equal or are negatives , as elements of @xmath337 . the determinant of the matrix corresponding to @xmath330 is @xmath6 . this gives the lower bound of @xmath332 on the number of equivalence classes . specializing to the case of a cyclic group , we have the following : for the cyclic group @xmath338 , @xmath339 . if @xmath340 , then any two @xmath338-actions on a handlebody of genus @xmath7 are equivalent . on the solid torus , any two @xmath338-actions are weakly equivalent , any two @xmath341-actions are equivalent , and if @xmath342 , then there are exactly @xmath343 equivalence classes of @xmath338-actions . explicitly , if @xmath344 is the action on @xmath345 defined by @xmath346 where @xmath267 is a fixed generator of @xmath338 and @xmath74 is relatively prime to @xmath347 , then @xmath348 and @xmath349 are equivalent if and only if @xmath350 . [ coro : cyclic actions ] the corollary is immediate from the statement of theorem [ thm : abelian ] , except for the explicit description of the equivalence classes . the matrix corresponding to @xmath344 in the proof of theorem [ thm : abelian ] is @xmath351 $ ] , and the last paragraph of the proof shows the condition for equivalence of @xmath348 and @xmath349 . let @xmath352 be the dihedral group of order @xmath89 . then @xmath353 . any two @xmath352-actions on the handlebody of genus @xmath354 are weakly equivalent . if @xmath90 , all actions on the handlebody of genus @xmath354 are equivalent , and if @xmath77 then there are exactly @xmath355 equivalence classes . [ thm : dihedral ] regard @xmath352 as @xmath356 , and let @xmath266 be the cyclic subgroup of @xmath352 generated by @xmath217 . by lemma [ lem : cyclic towers ] , applied to the tower @xmath357 , any two - element generating vector @xmath358 is nielsen equivalent to one of the form @xmath359 . since @xmath360 has order 2 , @xmath361 must generate @xmath266 , so @xmath362 and the vector is equivalent to @xmath363 . for @xmath90 , the proof is complete . for @xmath77 , there are at most @xmath355 equivalence classes , since @xmath363 is equivalent to @xmath364 . on the other hand , it is essentially an observation of d. higman ( see @xcite ) that the pair ( possibly equal ) of conjugacy classes of @xmath365 $ ] and @xmath366 $ ] is an invariant of the nielsen equivalence class of @xmath367 . in our case , @xmath368^{\pm1}=b^{\mp 2j}$ ] since @xmath369 is conjugate to @xmath370 only when @xmath371 , there are exactly @xmath355 equivalence classes . the smallest genus of handlebody admitting two inequivalent actions is genus @xmath372 , which has two equivalence classes of @xmath373-actions . the smallest genus of handlebody admitting two inequivalent actions of an abelian group is genus @xmath374 , which has two equivalence classes of @xmath375-actions . [ coro : smallest genus actions ] theorem [ thm : abelian ] verifies the assertion about abelian groups . by theorem [ thm : dihedral ] , the smallest - genus inequivalent actions of dihedral groups are the two @xmath373-actions in the corollary . only cyclic groups have @xmath376 , so @xmath377 for a noncyclic group . the only nonabelian and nondihedral group of order smaller than @xmath372 is the quaternion group , which is easily checked to have only one equivalence class of generating pair . we have mentioned that gilman @xcite proved that for @xmath6 prime , all @xmath1-element generating sets of @xmath19 are equivalent , and evans @xcite proved the same for @xmath17 and for the suzuki groups @xmath194 for @xmath18 . in this section , we prove the same for all @xmath5 with @xmath6 prime . the main result is the following . let @xmath6 be prime . if @xmath378 , then any @xmath3-element generating vector for @xmath5 is redundant . [ thm : psl(2,3p ) ] in particular , these groups include the case of @xmath197 ( @xcite , p. 412 ) . before proving theorem [ thm : psl(2,3p ) ] , we deduce a corollary . let @xmath6 be prime . if @xmath378 , then any two @xmath3-element generating vectors for @xmath5 are nielsen equivalent . consequently , for any handlebody of genus above the minimal one , @xmath379 , all @xmath5-actions are equivalent . [ coro : actions of psl(2,3p ) ] we recall from @xcite that a @xmath16-generator group @xmath0 is _ of spread @xmath16 _ when for any pair @xmath380 of nontrivial elements of @xmath0 , there exists @xmath381 such that @xmath382 . when @xmath383 , theorem 4.02 of @xcite shows that @xmath5 has spread @xmath16 . for @xmath384 , @xmath385 is isomorphic to @xmath188 ( @xcite , p. 412 ) , so has spread @xmath16 by proposition 3.02 of @xcite . let @xmath183 and @xmath386 be any two @xmath3-element generating sets . by repeated use of theorem [ thm : psl(2,3p ) ] , we may assume they are of the form @xmath387 and @xmath388 , where @xmath389 indicates a possibly empty string of @xmath13 s . choose @xmath212 such that @xmath390 . as in lemma 2.8 of @xcite , we have equivalences @xmath391 . to prepare for the proof of theorem [ thm : psl(2,3p ) ] , we will list several group - theoretic results . the first is lemma 4.10 of @xcite . let @xmath0 be a simple group generated by involutions @xmath392 , @xmath393 . then @xmath394 is redundant . [ lem : involutions ] by direct calculation , we have the following information . let @xmath211 and @xmath212 be elements of @xmath395 such that @xmath396 and @xmath397 . 1 . if @xmath398 , then @xmath399 and @xmath400 , where @xmath267 is either @xmath30 or @xmath13 . 2 . if @xmath401 and @xmath402 , then @xmath399 and @xmath403 where @xmath267 is either @xmath30 or @xmath16 . moreover , @xmath404 and @xmath405 . 3 . if @xmath406 , then @xmath407 and @xmath408 . [ lem : equations ] since @xmath409 , theorem @xmath13 of @xcite implies let @xmath410 be a generating vector for @xmath185 . then @xmath120 is redundant . [ lem : a5 redundant ] another useful property of @xmath185 follows from @xcite . let @xmath411 be a generating vector for @xmath185 , and let @xmath412 . then @xmath411 is nielsen equivalent to a generating vector @xmath413 with @xmath414 . [ lem : nn ] let @xmath415 denote the free group on two letters @xmath416 and @xmath417 . according to @xcite , there is an automorphism @xmath418 such that as elements of @xmath185 , @xmath419 . since @xmath420 , it can be written as a product of nielsen moves on the basis @xmath411 , so as vectors of elements of @xmath185 , @xmath411 is nielsen equivalent to @xmath421 . we will use a nice observation from @xcite . let @xmath422 be a generating vector for a nonabelian simple group @xmath0 , with @xmath423 . then @xmath120 is equivalent to a generating vector @xmath424 with @xmath425 conjugate to @xmath426 and @xmath427\neq 1 $ ] . [ lem : nonabelian ] the subgroup generated by the conjugates of @xmath426 by elements of @xmath428 is normal , hence equals @xmath0 . since @xmath429 can not be central , there must be some @xmath430 such that @xmath429 does not commute with @xmath431 , and @xmath120 is equivalent to @xmath432 . we will use some well - known information about subgroups of @xmath433 . specializing theorems 3(6.25 ) and 3(6.26 ) of suzuki @xcite to the case at hand , we obtain the following description . let @xmath6 be a prime . then every subgroup of @xmath5 is contained in one of the following kinds of subgroups : 1 . the dihedral groups of orders @xmath434 and @xmath435 . 2 . a group @xmath436 of order @xmath437 . a sylow @xmath1-subgroup @xmath438 of @xmath436 is elementary abelian , normal , and the factor group @xmath439 is a cyclic group of order @xmath440 . 3 . @xmath441 . @xmath395 and @xmath185 , which are subgroups only when @xmath384 . [ thm : subgroups ] notice that a sylow @xmath1-subgroup @xmath438 in theorem [ thm : subgroups ] is also a sylow @xmath1-subgroup of @xmath5 , since @xmath442 , and the group @xmath436 is its normalizer , since @xmath436 is maximal . we will refer to a subgroup @xmath436 as in theorem [ thm : subgroups](b ) as a _ sylow @xmath1-normalizer . _ we will also need the following facts about centralizers , which follow from 3(6.5 ) and 3(6.8 ) of @xcite . the centralizer of any element in @xmath443 is a maximal abelian subgroup , and is either cyclic of order prime to @xmath1 , isomorphic to @xmath444 , or is a sylow @xmath1-subgroup of @xmath443 . distinct maximal abelian subgroups have trivial intersection . [ lem : centralizers ] in particular , any two distinct sylow @xmath1-subgroups have trivial intersection . fix an @xmath3-element generating vector @xmath109 of @xmath5 . suppose first that @xmath445 . we will show that @xmath120 is redundant . put @xmath446 . if @xmath447 , then @xmath120 is redundant . if @xmath110 is cyclic , dihedral , @xmath441 or @xmath395 , then @xmath120 is redundant by theorem [ thm : dunwoody theorem ] , and if @xmath110 is @xmath185 , then lemma [ lem : a5 redundant ] applies . if @xmath110 is contained in a sylow @xmath1-normalizer @xmath436 of a sylow @xmath1-subgroup q , then @xmath448 is ( nielsen ) equivalent to @xmath449 such that @xmath450 ( and @xmath451 , otherwise @xmath120 is redundant ) . as @xmath452 , there is a @xmath453 for which @xmath454 is not a subgroup of @xmath436 . since the sylow @xmath1-subgroups intersect trivially , @xmath454 can not be contained in any other sylow @xmath1-normalizer . so @xmath455 is redundant or @xmath456 . in either case , @xmath120 is redundant . in the rest of the proof , we may assume that @xmath410 . we assume at every stage of the argument that no two elements of @xmath120 generate a cyclic group or all of @xmath5 , since otherwise @xmath120 is clearly redundant . we will argue that @xmath120 either is redundant or is equivalent to @xmath218 with @xmath211 , @xmath212 , and @xmath210 all of order @xmath16 . this will prove the theorem , since lemma [ lem : involutions ] shows that the latter is also redundant . we first show that @xmath120 is ( redundant or ) equivalent to @xmath457 with @xmath211 of order @xmath16 . if none of @xmath416 , @xmath417 , or @xmath458 already has order @xmath16 , put @xmath459 . we may assume that @xmath110 is not contained in any sylow @xmath1-normalizer @xmath436 . for if it is , then since @xmath439 is cyclic , we may assume that @xmath460 ( and @xmath461 ) . interchanging @xmath417 and @xmath458 if necessary , we may assume that @xmath110 is not contained in @xmath436 . since the sylow @xmath1-subgroups intersect trivially , @xmath110 can not be contained in any other sylow @xmath1-normalizer . since neither of its generators has order @xmath16 , @xmath110 can not be dihedral . if @xmath110 is @xmath441 or @xmath395 , then lemma [ lem : equations ] can be used to find @xmath211 . if it is @xmath185 , then lemma [ lem : nn ] can be used . so we may write @xmath120 as @xmath457 with @xmath211 of order @xmath16 . next , we will show that @xmath457 is equivalent to @xmath462 with @xmath212 also of order @xmath16 . put @xmath463 . by lemma [ lem : nonabelian ] , we may assume that @xmath110 is nonabelian . we may assume that @xmath110 is not contained in any sylow @xmath1-normalizer . for suppose it lies in @xmath436 . since @xmath439 is cyclic , we may asssume that @xmath464 . since @xmath211 can not be in @xmath436 , @xmath465 can not be in @xmath436 . since the sylow @xmath1-subgroups are disjoint , @xmath466 can not be in any sylow @xmath1-normalizer . if @xmath110 is dihedral , then one of @xmath417 or @xmath458 already has order @xmath16 and will be @xmath212 . if @xmath110 is one of @xmath441 , @xmath395 , or @xmath185 , then as before , either lemma [ lem : equations ] or [ lem : nn ] produces @xmath212 . we now have @xmath467 where @xmath211 and @xmath212 are elements of order @xmath16 . we may assume that @xmath468 and , using lemma [ lem : nonabelian ] , that @xmath365\neq 1 $ ] , so @xmath469 . put @xmath470 . it can not be dihedral , since neither generator is of order @xmath16 . suppose that @xmath110 is contained in a sylow @xmath1-normalizer @xmath436 . if @xmath471 is of order @xmath1 , then since @xmath211 inverts @xmath471 , @xmath472 and hence @xmath473 , so @xmath474 , a contradiction . so @xmath475 . assume first that @xmath383 , so that @xmath476 is odd . then @xmath477 is not contained in a sylow @xmath1-normalizer . if it is dihedral , then we can take @xmath478 . otherwise , it is isomorphic to @xmath441 , so @xmath458 has order @xmath1 . similarly , considering @xmath479 , we may assume that @xmath480 has order @xmath1 . since this element lies in @xmath436 , it must lie in @xmath438 , forcing the contradiction that @xmath481 . when @xmath384 , we have from p. 398 of @xcite that @xmath436 is a split extension of @xmath482 by @xmath483 . if @xmath347 generates @xmath483 , then @xmath484 must act by inverting each element of @xmath482 ( otherwise it would act trivially , but then the centralizer of an element of @xmath438 would be larger than @xmath438 , contradicting lemma [ lem : centralizers ] ) so @xmath485 has order @xmath16 for any @xmath486 . since @xmath475 and @xmath469 , we have @xmath487 . if @xmath458 has order @xmath1 , then @xmath488 is of order @xmath16 . if @xmath458 is of order @xmath189 , then @xmath489 is of order @xmath16 . so we may assume that @xmath110 is not contained in a sylow @xmath1-normalizer . if @xmath110 is isomorphic to @xmath441 or @xmath395 , then lemma [ lem : equations ] applies . the case of @xmath383 is complete , so we may assume that @xmath384 and @xmath490 . since @xmath120 is equivalent to @xmath491 , we may apply lemma [ lem : nn ] to @xmath492 to obtain a new generating vector of the form @xmath462 where @xmath211 and @xmath212 have order @xmath16 , @xmath458 has order @xmath101 , and @xmath493 . by the previous arguments , we may also assume that @xmath477 and @xmath492 are isomorphic to @xmath185 . in the remainder of the proof , it is convenient to regard @xmath494 as @xmath188 . also , we may apply any automorphism to all elements of the generating vector , since any vector weakly equivalent to a redundant vector is redundant . according to 3(2.19 ) of @xcite , there are two conjugacy classes of @xmath185-subgroups in @xmath188 . one class consists of the six @xmath185-subgroups that stabilize one of the six letters , and the other class consists of six @xmath185-subgroups that act transitively on the six letters . moreover , there exists an automorphism @xmath26 of @xmath188 that interchanges the two classes . applying @xmath26 to each of the generating elements , if necessary , we may assume that @xmath492 fixes a letter . since @xmath458 is a @xmath101-cycle , this must be the unique letter fixed by @xmath458 . neither of @xmath477 and @xmath495 can fix a letter . for if so , it would fix the same letter as @xmath492 , and one of @xmath211 or @xmath212 would be in @xmath492 , forcing the latter to be all of @xmath188 . applying @xmath26 to all generators , @xmath477 and @xmath495 become @xmath185-subgroups fixing the letter fixed by @xmath458 , so are equal , achieving the same contradiction . the principal difficulties in extending the proof of theorem [ thm : psl(2,3p ) ] to more general cases @xmath496 seem to be the more complicated sylow @xmath74-normalizers , when @xmath497 , the presence of subgroups of the form @xmath498 when @xmath316 is not prime , and analyzing the case when any two generators generate an @xmath185 subgroup . throughout this section , @xmath0 is an arbitrary finite group , and as usual @xmath15 or just @xmath33 will denote the minimal number of generators for @xmath0 , and @xmath9 or just @xmath31 the maximum length of a chain of strictly decreasing nonzero subgroups of @xmath0 . clearly @xmath499 , and if @xmath500 , then @xmath501 . any two actions of a group @xmath0 on handlebodies of the same genus become equivalent after at most @xmath15 stabilizations . [ prop : stabilization ] let @xmath502 be a generating vector of minimal length , and let @xmath305 be any generating vector . put @xmath503 , where @xmath33 @xmath13 s have been added . then @xmath504 is equivalent to @xmath505 , and hence to @xmath506 . so any two generating vectors of the same length become equivalent after @xmath33 stabilizations . if @xmath507 , then any two generating @xmath288-vectors for @xmath0 are equivalent . [ prop : supraminimal generating sets ] fix a generating vector @xmath502 of minimal length , and let @xmath305 be any generating vector with @xmath288 elements . we will show that @xmath120 is equivalent to @xmath506 . put @xmath508 , so @xmath509 . since @xmath510 , @xmath511 for some @xmath512 , so @xmath268 can be written as a word in @xmath513 , @xmath514 . therefore @xmath120 is equivalent to @xmath515 . so we may assume that @xmath279 . since @xmath279 , @xmath120 is equivalent to @xmath516 . let @xmath517 and @xmath518 for @xmath285 . since @xmath519 , we must have @xmath511 for some @xmath285 . so @xmath520 , and therefore @xmath120 is equivalent to @xmath521 . we may reselect notation so that @xmath120 is equivalent to @xmath522 . inductively , suppose that @xmath120 is of the form @xmath523 , for some @xmath524 . put @xmath525 for @xmath526 and to @xmath527 for @xmath528 . for @xmath526 , since @xmath111 is a minimal generating set , we have @xmath529 , so we must have @xmath530 for some @xmath531 . so @xmath532 . this implies that @xmath120 is equivalent to @xmath533 , and , perhaps after reselecting notation , to @xmath534 . so @xmath120 is equivalent to @xmath535 , and hence to @xmath506 . since @xmath536 is at most @xmath537 , proposition [ prop : supraminimal generating sets ] improves theorem 3 of @xcite , which shows that any two generating vectors of length at least @xmath538 are equivalent . if @xmath539 , then any two actions of @xmath0 on a handlebody of genus @xmath7 are equivalent . [ coro : limit stabilization ] if @xmath540 , for @xmath6 prime , then @xmath541 . thus by theorem [ thm : abelian ] , if @xmath542 , there are inequivalent actions on the handlebody of genus @xmath543 , showing that the estimate in corollary [ coro : limit stabilization ] is the best possible , in general . on the other hand , it appears to be far from the best possible for many cases . frequently there is a large gap between @xmath15 and @xmath9 ( for example , all symmetric groups can be generated by two elements , but have values of @xmath9 that are arbitrarily large ) . our results on free actions are far from complete . the most obvious question is : that is , is @xmath79 for all @xmath544 ? algebraically , if @xmath198 are any two @xmath3-element generating sets of @xmath0 nielsen equivalent ? according to p. 92 of @xcite , this algebraic version was first asked by f. waldhausen . it has been resolved negatively for infinite groups . the first example appears to be due to g. a. noskov @xcite , and general constructions are given in @xcite . an affirmative answer to question 1 for finite groups would imply affirmative answers to the next two questions :

we analyse the diagonal quotient for the product of certain artin schreier curves . the smooth models are almost always surfaces of general type , with chern slopes tending asymptotically to 1 . the calculation of numerical invariants relies on a close examination of the relevant wild quotient singularity in characteristic . it turns out that the canonical model has rational double points of type , and embeds as a divisor of degree in , which is in some sense reminiscent of the classical kummer quartic .

1103.0088

it is a classical fact in complex geometry that the singular kummer surface @xmath5 attached to an abelian surface @xmath6 has sixteen rational double points , and an irreducible principal polarization embeds it as a quartic surface in @xmath4 , compare hudson s classical monography @xcite , or for a modern account @xcite . our starting point was an analogous computation in characteristic @xmath7 for the diagonal action of the additive group @xmath8 on the selfproduct @xmath9 , where @xmath10 is the supersingular elliptic curve , viewed as an artin schreier covering . it turns out that this is a special case of a rather general construction , which works for all primes @xmath0 , in fact for all prime powers @xmath11 . it starts with certain _ artin schreier curves _ and leads , with a few exceptions for small prime powers , to _ surfaces of general type_. the goal of this paper is to describe the geometry of these surfaces , and we obtain a fairly complete description . the construction goes as follows : fix an algebraically closed ground field @xmath12 of characteristic @xmath13 and consider artin schreier curves of the form @xmath14 where @xmath15 is a monic polynomial of degree @xmath16 . these curves carry a translation action of the additive group @xmath17 , and we consider the diagonal action on the product @xmath18 of two such artin schreier curves . the quotient @xmath19 is a normal surface containing a unique singularity . such singularities are examples of _ wild quotient singularities _ , i.e. , the characteristic of the ground field divides the order of the group @xmath20 . few examples of wild quotient singularities occur in the literature , and little is known in general . artin @xcite gave a complete classification for wild @xmath21-quotient singularities in dimension two , and general @xmath22-quotient singularities were studied further by peskin @xcite . peculiar properties of wild @xmath23-quotient singularities in relation to punctual hilbert schemes appear in @xcite . in light of the scarcity of examples , it is useful to have more classes of wild quotient singularities in which computations are feasible . lorenzini initiated a general investigation of wild quotient singularities on surfaces @xcite , which play an important role in understanding the reduction behaviour of curves over discrete valuation fields . he compiled a list of open questions @xcite . in a recent paper , lorenzini studied wild @xmath22-quotient singularities resulting from diagonal actions on products of @xmath18 , where one or both factors are ordinary curves @xcite . in some sense , we treat the opposite situation , as artin - schreier curves have vanishing @xmath0-rank , and our main concern is the interplay between the local structure coming from the the wild quotient singuarity and the global geometry of the algebraic surface . we have choosen artin schreier coverings that are concrete enough so that explicit computations are possible . note that our set - up includes wild quotient singularities with respect to elementary abelian groups , and not ony cyclic groups . consider the minimal resolution of singularities @xmath24 of our normal surface . according to lorenzini s general observation , the exceptional divisor of a wild quotient singularity in dimension two consists of projective lines , and its dual graph is a tree @xcite . using an explicit formal equation for the singularitiy , we show that the dual graph is even star - shaped , with @xmath25 terminal chains attached to the central node , each of length @xmath3 , as depicted in figure 1 . the two basic numerical invariants of surface singularities are the genus @xmath26 of the fundamental cycle @xmath27 and the _ geometric genus _ @xmath28 , @xmath29 . the latter is usually very difficult to compute . we obtain @xmath30 the latter under the assumption that our prime power @xmath3 is prime . this relies on a computation of the global @xmath31-adic euler characteristic of the surface , combined with a determination of the @xmath20-invariant part in @xmath32 , which in turn depends on a problem in modular representation theory related to tensor products of jordan matrices . in contrast , the singularities occuring in @xcite , where at least one factor in @xmath18 is an ordinary curve , are all rational . having a good hold on the structure of the resolution of singularities , we determine the global invariants of the smooth surface @xmath33 . it turns out that @xmath34 , and that @xmath35 is a free abelian group of finite rank . moreover , the algebraic fundamental group @xmath36 vanishes . passing to the minimal model @xmath37 we show that the minimal surface @xmath38 is of general type for @xmath39 , a k3 surface for @xmath40 , and a weak del pezzo surface for @xmath41 . their chern invariants are given by the formula @xmath42 the resulting _ chern slopes _ asymptotically tend to @xmath43 , and one may say that our surfaces show no pathological behaviour with respect to surface geography . the determination of the euler characteristic @xmath44 depends on dolgachev s formula @xmath45 for @xmath31-adic euler characteristic for schemes fibered over curves @xcite , where @xmath46 is _ serre s measure of wild ramification_. its computation is quite easy in our situation , given the explicit nature of the artin schreier curves . the surface @xmath38 has a surprisingly simple _ projective description _ , which is reminiscent to kummer s quartic surfaces @xmath47 . passing to the normal surface @xmath48 with @xmath1 rational double points of type @xmath2 obtained by contracting all terminal chains in the fundamental cycle , the image @xmath49 of the fundamental cycle remains cartier , and defines the embedding : the invertible sheaf @xmath50 is very ample , has @xmath51 and embeds the normal surface @xmath48 as a divisor of degree @xmath3 in @xmath4 , sending the rational double points into a line . in fact , a canonical divisor is @xmath52 . so for @xmath39 , the closed embedding @xmath53 can be viewed as an _ @xmath54-canonical map _ , for the fractional value @xmath55 . in the simplest case @xmath56 , this is in line with classification results of horikawa on minimal surfaces of general type with @xmath57 and vanishing irregularity @xcite . the projective description also allows us to deduce that our surfaces @xmath38 admit a lifting into characteristic zero , at least in the category of algebraic spaces . it would be interesting to determine the homogeneous polynomial describing the image @xmath58 , but we have made no attempt to do so . the paper is organized as follows : in section [ artin schreier curves ] we review some relevant facts on artin schreier curve , all of them well - known . in section [ products of artin schreier curves ] we study the normal surface @xmath19 , obtained as the quotient of the product of two artin schreier curves with respect to the diagonal action . we find an explicit equation for the singularitiy , and a dimension formula for the global sections of the dualizing sheaf . section [ invariants of the singularity ] contains an analysis of the minimal resolution of singularities @xmath59 . notable results are formulas for the fundamental cycle and its genus , as well as some bounds on the arithmetic genus . in section [ vanishing of irregularity ] we prove that @xmath34 , such that the picard scheme is reduced and 0-dimensional . this relies on a general fact about group actions with fixed points , which seems to be of independent interest , and is verified with grothendieck s theory of @xmath20-equivariant cohomology . in section [ place in the enriques classification ] we determine the place of the smooth surface @xmath33 in the enriques classification . among other things , this depends on the geometry of the fibration @xmath60 induced from the projections on @xmath18 . section [ canonical model and canonical map ] contains our analysis of projective models for the surfaces . finally , in section [ numerical invariants and geography ] we take up questions from surface geography and compute chern invariants . this mainly relies on dolgachev s formula for @xmath31-adic euler characteristics for fibered schemes . the first author would like to thank the mathematisches institut of the heinrich heine universitt dsseldorf , where this work has begun , for its warm hospitality . research of the first author was partially supported by grand - in - aid for scientific research ( c ) 20540044 , the ministry of education , culture , sports , science and technology . we thank the referee for bringing to our attention lorenzini s preprints @xcite , @xcite , @xcite . [ artin schreier curves ] let @xmath13 be a prime number and @xmath12 be an algebraically closed ground field of characteristic @xmath0 . consider _ artin schreier curves _ of the form @xmath61 where the left side of the defining equation is a monic polynomial @xmath62 of degree @xmath1 with coefficients from the ground field @xmath12 . in other words , @xmath63 is defined by the homogeneous equation @xmath64 of degree @xmath65 inside the projective plane @xmath66)$ ] . homogeneous and inhomogeneous coordinates are related by @xmath67 and @xmath68 . the curve @xmath69 is smooth , with numerical invariants @xmath70 now let @xmath71 be the additive group of all scalars @xmath72 satisfying @xmath73 , viewed as an elementary abelian @xmath0-group . we may also regard it as a subgroup @xmath74 thus the elements @xmath75 act on @xmath76 via @xmath77 . this action leaves the homogeneous equation ( [ homogeneous equation ] ) invariant , whence induces an action on @xmath69 . this action is free , except for a single fixed point @xmath78 . dehomogenizing in another way by setting @xmath79 , @xmath80 , we see that an open neighborhood of the fixed point is the spectrum of the coordinate ring @xmath81/(u^q - uw^{q-1}-p(w)),\ ] ] where @xmath82 , and the group elements @xmath75 act via @xmath83 , @xmath84 . since @xmath85 becomes a unit in the local ring @xmath86 , there is a unique way to write the indeterminate @xmath87 as a formal power series @xmath88 in the variable @xmath89 so that @xmath90 ( @xcite , 4 , no . 7 , corollary to proposition 10 ) . using the latter condition , one easily infers that the initial coefficients are @xmath91 the upshot is that the inclusion @xmath92\subset k[[u , w]]/(u^q - uw^{q-1}-p(w))={\mathcal{o}}_{c , a}^\wedge\ ] ] is bijective , and the group elements @xmath75 act on the formal completion @xmath93={\mathcal{o}}_{c , a}^\wedge$ ] via @xmath94 from this we infer that the filtration given by the _ higher ramification subgroups _ @xmath95 takes the simple form @xmath96 recall that @xmath97 is defined as the _ decomposition group _ of the @xmath98-th infinitesimal neighborhood of the closed point , that is , the subgroup of those @xmath99 with the property @xmath100 . the corresponding function @xmath101 on @xmath20 is given by @xmath102 we refer to serre s monograph @xcite for the theory of higher ramification groups . these groups will play a crucial role in section [ numerical invariants and geography ] in the determination of euler characteristics . the hurwitz formula for the quotient map @xmath103 of degree @xmath3 , in the form of @xcite , chapter vi , proposition 7 , gives @xmath104 where @xmath105 , @xmath106 denotes genus , and @xmath107 is the character of the artin representation attached to the fixed point , which by definition has @xmath108 . it follows that @xmath109 , whence @xmath110 . in light of the defining equations , this was of course clear from the very beginning : the quotient map @xmath111 is a classical artin schreier covering of the projective line . using that the quotient map is tale away from and totally ramified at the fixed point , one deduces : [ canonical curve ] a canonical divisor for the curve @xmath69 is given by @xmath112 , where @xmath113 is the fixed point for the @xmath20-action . it is not difficult to give an explicit basis for the vector space of global @xmath114-forms on @xmath69 : [ rational differentials ] the rational differentials @xmath115 , @xmath116 are everywhere defined and constitute a basis for @xmath117 . consider the two coordinate rings @xmath118/(x^q - x - y^qp(1/y)){\quad\text{and}\quad}r'=k[u , w]/(u^q - uw^{q-1}-p(w))\ ] ] for our curve @xmath69 . the relation @xmath119 reveals that @xmath120 is freely generated by @xmath121 . similarly , the relation @xmath122 and @xmath123 shows that @xmath124 is freely generated by @xmath125 , locally at the point @xmath126 . given a polynomial @xmath127 , we thus express the differential @xmath128 in terms of @xmath129 , using @xmath130 and @xmath131 : @xmath132 hence the rational differential @xmath128 is everywhere defined if @xmath133 , which lies in the field of fractions for @xmath134 , actually lies in @xmath134 . this indeed holds for the monomials @xmath135 , provided @xmath116 , by ( [ formal substitution ] ) . the resulting @xmath136 elements @xmath137 are clearly linearly independent over @xmath12 . they must constitute a basis , because the vector space in question is of dimension @xmath136 . the induced action of the @xmath75 on @xmath117 is @xmath138 . whence @xmath117 is the direct sum of the @xmath20-invariant subspaces @xmath139 that are generated by @xmath115 , @xmath140 . obviously , @xmath141 . [ indecomposable subspaces ] the @xmath20-invariant subspaces @xmath142 are indecomposable as @xmath20-modules , and the fixed spaces @xmath143 are @xmath114-dimensional . we first check that the fixed space @xmath143 are @xmath114-dimensional . clearly , @xmath144 is invariant . seeking a contradiction , we suppose there is a monic polynomial @xmath145 with @xmath146 so that the differential @xmath147 is invariant . factoring @xmath148 , we see that the set of roots @xmath149 is invariant under the substitution @xmath150 , @xmath75 . hence @xmath151 , contradiction . suppose @xmath152 is decomposable , such that we have decomposition @xmath153 into nonzero @xmath20-invariant subspaces . since @xmath20 is commutative , there is a basis of @xmath154 in which all @xmath75 act via lower triangular matrices ( see @xcite , chapter vii , 5 , no . 9 proposition 19 ) . the last member @xmath155 of such a basis is then a common eigenvector for all @xmath75 . since each @xmath75 has order @xmath0 , all eigenvalues are @xmath156 , whence @xmath155 is @xmath20-fixed . the same applies to @xmath157 , giving a contradiction to @xmath158 . note this implies that the fixed space @xmath159 is of dimension @xmath160 . now choose a second artin schreier curve @xmath161 of the form discussed in the previous section , and consider the product @xmath18 , endowed with the diagonal @xmath20-action . in this section we start to study the quotient @xmath19 , which is a normal surface whose singular locus consists of one point @xmath162 , the image of the fixed point @xmath163 . the projections @xmath164 and @xmath165 induce fibrations @xmath166 respectively . choose coordinates on the copies of projective lines so that the fixed points @xmath113 and @xmath167 map to the origin @xmath168 , and consider the fibers @xmath169 . recall that the _ multiplicity of a fiber _ is the greatest common divisor for the multiplicities of its integral components . [ canonical quotient ] the fibers @xmath170 have multiplicity @xmath3 , and a canonical divisor is given by @xmath171 . its selfintersection number is @xmath172 . the fiber @xmath173 is clearly irreducible . according to ( [ i invariant ] ) , the @xmath20-action on the preimage of @xmath174 in @xmath69 is trivial . whence outside the singularity , the fiber @xmath173 is the quotient of @xmath175 , where the action on the right factor is trivial , which is an artin ring of length @xmath3 . it follows that the fiber has multiplicity @xmath3 . as to the canonical divisor , we have @xmath176 by proposition [ canonical curve ] . this is the preimage of @xmath177 . using that the projection @xmath178 is tale in codimension one , together with @xcite , theorem 2.7 , we infer that @xmath177 is indeed a canonical divisor . its selfintersection number is @xmath179 by the projection formula . the canonical divisor @xmath171 is clearly cartier , such that the normal surface @xmath19 is gorenstein . let us now examine its singularity . throughout the paper , it will be crucial to understand the local invariants of this singularity , in order to determine global invariants for smooth models of @xmath19 . let @xmath180 be four variables and set @xmath181=k[[u , w , u',w']]/(u^q - uw^{q-1}-p(w),u'^q - u'w'^{q-1}-q(w')),\ ] ] which is the complete local ring at the fixed point @xmath182 . here @xmath183 are polynomials stemming from the left hand side of the artin - schreier equations , as discussed in section [ artin schreier curves ] . the group elements @xmath75 act via @xmath184 clearly , the elements @xmath185 are invariant and satisfy the relation @xmath186 therefore , we obtain a homomorphism of @xmath12-algebras @xmath187/(c^q-(ab)^{q-1}c - a^qq(b)+b^qp(a ) ) { \longrightarrow}a^g,\ ] ] where @xmath188 are indeterminates and @xmath189 , @xmath190 , @xmath191 . [ invariant ring ] the preceding homomorphism ( [ quotient singularity ] ) is bijective . clearly , the local ring @xmath192/(c^q-(ab)^{q-1}c - a^qq(b)+b^qp(a))$ ] is 2-dimensional and cohen macaulay . computing the jacobian ideal , one sees that the singular locus consists of the closed point , whence @xmath193 is normal . according to galois theory , the finite extension @xmath194 has generically rank @xmath195 . by the main theorem of zariski ( @xcite , corollary 4.4.9 ) , it therefore suffices to check that @xmath196 has generically rank @xmath3 . obviously , the extensions @xmath197\subset r$ ] and @xmath197\subset a$ ] have rank @xmath3 and @xmath198 , respectively , and the statement follows by transitivity of ranks . this local description of the singularity enables us to determine the scheme structure for the canonical divisor @xmath199 , viewed as a reduced weil divisor : [ fiber union ] the reduced weil divisors @xmath200 are isomorphic to the projective line , and the schematic intersection @xmath201 has length one . we saw in the proof of proposition [ canonical quotient ] that @xmath200 is isomorphic to @xmath202 , at least outside the singular point on the ambient surface . however , at this singularity , @xmath173 is formally isomorphic to the spectrum of @xmath203/(c^q , a)$ ] , which follows from proposition [ invariant ring ] . hence its reduction is regular . similarly , the union @xmath204 is formally isomorphic to the spectrum of @xmath203/(c , ab)$ ] , and the result follows . we would like to compute the dimension of the vector space of global sections for the dualizing sheaf @xmath205 , but are only able to do so directly in the special case @xmath206 : [ invariant differentials ] for @xmath206 , we have @xmath207 . we have a commutative diagram @xmath208 where @xmath209 is the locus where @xmath20 acts freely , @xmath210 is the smooth locus , and @xmath211 is the induced finite tale galois covering with galois group @xmath20 . the latter ensures that the vertical map on the left is bijective . since our dualizing sheaves are invertible , the horizontal maps are bijective ( compare , for example , @xcite , proposition 1.11 ) . it follows that that the canonical map @xmath212 is bijective . recall that @xmath213 , where @xmath152 are indecomposable @xmath20-submodule of dimension @xmath214 . since @xmath215 , the action of the generator @xmath216 is given , in a suitable basis , by the jordan matrix @xmath217 which determines the @xmath20-module up to isomorphism . computing the dimension of the @xmath20-fixed part for @xmath218 thus reduces to extracting the number of blocks in the jordan normal form of tensor products of certain jordan matrices . write @xmath219 for certain multiplicities @xmath220 . note that it is a notorious unsolved problem in linear algebra to find the jordan decomposition of tensor products of nilpotent jordan matrices in positive characteristics . however , it is well - known that @xmath221 ( see , for example , @xcite , proposition 3.2 ) . hence the @xmath20-fixed part of such a tensor product representation is of dimension @xmath222 , and we get @xmath223 summing over @xmath224 rather than @xmath225 , we rewrite the latter sum as @xmath226 the statement follows by applying the formulas @xmath227 and @xmath228 . in order to make similar computations in the general case @xmath11 , one would need to understand the modular representations of the elementary abelian group @xmath229 . such representations can be expressed in terms of @xmath230 commuting nilpotent matrices , or equivalently via modules of finite length over the polynomial ring in @xmath230 indeterminates over @xmath231 . for this situation , little seems to be known about multiplicities in tensor products . in this section we study in more detail the 2-dimensional ring @xmath232/(c^q-(ab)^{q-1}c - a^qq(b)+b^qp(a)),\ ] ] whose formal completion gives the singularity of the surface @xmath19 . as in the proof of proposition [ invariant ring ] , one sees that @xmath193 is normal , and the maximal ideal @xmath233 corresponds the unique singularity @xmath234 . according to @xcite , theorem 2.5 , the exceptional divisor on the minimal resolution of singularities consists of projective lines , and has as dual graph a tree . to understand its structure , we first consider a partial resolution of singularities : let @xmath235 be the blowing - up of the maximal ideal @xmath236 , and denote by @xmath237 the exceptional divisor . this is the cartier divisor @xmath238 with ideal @xmath239 . [ first blowing - up ] the surface @xmath240 is normal , and @xmath241 consists of @xmath25 closed points , whose local rings are rational double points of type @xmath2 . we have @xmath242 and @xmath243 . furthermore , the inclusions @xmath244 are infinitesimal extensions by the sheaf @xmath245 , @xmath246 . the blowing - up @xmath240 is covered by three affine charts , the @xmath247-chart , the @xmath248-chart , and the @xmath249-chart . the coordinate ring of the @xmath247-chart is generated by @xmath250 , subject to the relation @xmath251 computing partial derivatives , one sees that the singular locus is given by @xmath252 and @xmath253 , and the latter means @xmath254 , @xmath255 . writing the left hand side of the preceding equation in the form @xmath256 one easily computes that @xmath257 form a regular system of parameters in the formal completion @xmath258 $ ] . the upshot is that singularity is a rational double point of type @xmath2 . the situation on the @xmath248-chart is symmetric , whereas the @xmath249-chart turns out to be disjoint from the exceptional divisor . on the @xmath247-chart , the exceptional divisor is given by @xmath259 , whence its coordinate ring is @xmath260/(c / a)^q$ ] . its reduction is defined by @xmath261 . we infer that @xmath262 is isomorphic to @xmath263 $ ] . the ideal of @xmath244 is generated by @xmath264 and @xmath265 on the @xmath247- and @xmath248-chart , respectively . the statement about the infinitesimal extensions follows . now let @xmath266 be the minimal resolution of singularities of @xmath240 , the latter being a partial resolution of @xmath267 . for each singular point @xmath268 , @xmath269 , write @xmath270 as a chain of projective lines @xmath271 , such that @xmath272 . let @xmath273 be the strict transform of @xmath274 , which is another projective line . according to lorenzini @xcite , theorem 2.5 , the dual graph for the exceptional divisor on the resolution of a wild quotient singularity is necessarily a tree . to our knowledge , no examples are known where the tree is not star - shaped . in @xcite , question 1.1 , lorenzini askes whether trees with more that one node are possible . in our situation , equation ( [ rdp equation ] ) immediately shows that @xmath275 intersects each chain of rational curves @xmath276 in a terminal component of the chain , say @xmath277 . thus the reduced exceptional divisor @xmath278 is a strictly normal crossing made out of projective lines , and its dual graph looks like this : figure : dual graph for exceptional divisor on @xmath33 . [ selfintersection numbers ] we have @xmath279 , and @xmath280 , and @xmath281 . in particular , the composite map @xmath282 is the minimal resolution of singularities . as in the proof of proposition [ first blowing - up ] , we have @xmath283 \cup { \operatorname{spec}}k[b / a],\ ] ] and the invertible sheaf @xmath284 is given by @xmath285 on the two charts . viewing @xmath286^\times$ ] as a cocycle for @xmath284 , one immediately verifies that this invertible sheaf has degree @xmath114 . thus @xmath287 since @xmath288 , the selfintersection number @xmath279 also follows . now consider the first infinitesimal neighborhood @xmath289 . in light of proposition [ first blowing - up ] , this is an infinitesimal extension of @xmath242 by the invertible sheaf @xmath290 . let @xmath291 be the blowing - up of the singular points @xmath292 . then the exceptional divisors are disjoint unions of pairs of rational curves , each pair intersecting transversely at one point whose local ring on @xmath293 is a rational double point of type @xmath294 . moreover , the strict transform of @xmath262 lies in the smooth locus of @xmath293 . according to the theory of _ ribbons _ developed by bayer and eisenbud , the strict transform of @xmath295 on @xmath293 , that is , the blowing - up of the scheme @xmath295 with respect to the centers @xmath296 , is an infinitesimal extension of @xmath297 by the invertible sheaf @xmath298 , compare @xcite , theorem 1.9 . it follows that @xmath299 . in particular , the exceptional divisor for the resolution of singularities @xmath282 contains no @xmath300-curve , hence is minimal . we next compute the _ fundamental cycle _ @xmath27 for the resolution of singularities @xmath301 , a notion introduced by m. artin @xcite . by definition , this is the smallest effective cycle @xmath302 whose support equals the exceptional divisor and has intersection number @xmath303 on each irreducible component of the exceptional divisor . one way to compute the fundamental cycle is with a _ computation sequence _ @xmath304 , where in each step @xmath305 is an integral component of the exceptional divisor with @xmath306 . using the exact sequence @xmath307 one inductively infers that @xmath308 , in particular @xmath309 . consequently the schematic image @xmath310 is nothing but the reduced singular point @xmath234 . indeed , one should view the fundamental cycle as an approximation to the schematic fiber @xmath311 , the latter usually containing embedded components . [ fundamental cycle ] the fundamental cycle is given by the formula @xmath312 and its selfintersection number is @xmath313 . let @xmath314 be the cycle on the right hand side . one easily computes the intersection numbers @xmath315 whence the fundamental cycle @xmath302 is contained in @xmath314 , by the minimality property of fundamental cycles . seeking a contradiction , we suppose @xmath316 . the effective cycle @xmath317 has intersection numbers @xmath318 since nonzero effective exceptional cycles are not nef on all exceptional curves ( see @xcite , proof of proposition 2 ) , there is no possibility but @xmath319 . now write @xmath320 with coefficients @xmath321 and @xmath322 . note that the coefficients @xmath323 do not depend on @xmath98 , due to the obvious symmetry of the dual graph in figure 1 . we have @xmath324 whence @xmath325 , contradicting @xmath326 . recall that the _ canonical cycle _ @xmath327 is the divisor supported by the exceptional divisor that satisfies the equations @xmath328 , where @xmath69 runs through the integral exceptional divisors . [ canonical cycle ] the canonical cycle is given by @xmath329 , with selfintersection number @xmath330 . using that @xmath275 and @xmath271 are copies of the projective lines with selfintersection numbers @xmath331 , @xmath332 , one deduces the result from the intersection numbers ( [ intersection fundamental ] ) . the selfintersection then follows from proposition [ fundamental cycle ] . we are now in position to compute the arithmetic genus of the fundamental cycle @xmath333 which is also called the _ fundamental genus _ of the singularity , confer @xcite . [ fundamental genus ] the fundamental genus of the singularity @xmath193 is given by the formula @xmath334 . riemann roch yields @xmath335 . we have @xmath336 , and finally obtain @xmath337 as claimed . from this we deduce : [ rdp elliptic ] the singularity @xmath193 is a rational double point if and only if @xmath338 , and minimally elliptic if and only if @xmath339 . according to @xcite , theorem 3 , the singularity is rational if and only if @xmath340 . in light of corollary [ fundamental genus ] , this happens precisely when @xmath338 . by laufer @xcite , theorem 3.4 , the condition @xmath341 is one of several equivalent defining property of _ minimally elliptic singularities_. by corollary [ canonical cycle ] , this happens if and only if @xmath339 . for @xmath338 , this singularity is actually a rational double point of type @xmath342 , according to artin s list @xcite . indeed , the equation @xmath343 is a special case of artin s normal form for wild @xmath21-quotient singularities in dimension two @xcite , compare also @xcite . for @xmath339 , the minimally elliptic singularity appears in laufer s classification ( @xcite , table 3 on page 1294 ) under the designation @xmath344 . shioda @xcite and katsura @xcite obtained rather similar results for the action of the sign involution on abelian surfaces in characteristic @xmath345 . we next want to compute the _ geometric genus _ @xmath346 of the singularity . except for rational double points and minimally elliptic singularities , this invariant is difficult to compute . we have at least some bounds . it will turn out later that in the special case @xmath206 , these bounds are actually equalities ( corollary [ local geometric genus ] ) . [ geometric genus ] the geometric genus of the singularity @xmath193 satisfies the inequalities @xmath347 . for @xmath39 , we moreover have @xmath348 , in particular @xmath349 . it is convenient to work on the partial resolution @xmath240 rather than on the full resolution @xmath33 . the leray serre spectral sequence gives an exact sequence @xmath350 the term on the right vanishes , since @xmath240 has only rational singularities , whence @xmath351 . let @xmath352 be the anticanonical cycle , where @xmath237 . it follows with the grauert riemenschneider vanishing theorem ( see @xcite , theorem 1.5 ) that @xmath353 , whence the canonical surjection @xmath354 is bijective . moreover , riemann roch gives @xmath355 , such that @xmath356 . according to proposition [ first blowing - up ] and proposition [ selfintersection numbers ] , we have @xmath357 consider the integral weil divisors @xmath358 . we deduce from proposition [ first blowing - up ] that the kernel @xmath359 in the exact sequence @xmath360 is an invertible sheaf on @xmath242 of degree @xmath361 where @xmath362 and @xmath363 denotes integral and fractional parts , respectively . let us tabulate the kernels @xmath359 for @xmath364 in a matrix of size @xmath365 : @xmath366 only kernels of degree @xmath367 may contribute to @xmath368 , and the total possible contribution is @xmath369 but some coboundary maps in the exact sequence @xmath370 might be nonzero , so we only get an upper bound @xmath347 , rather than an equality . to obtain a lower bound , we consider the cycle @xmath371 . the second row in ( [ kernel array ] ) reveals that some cycle @xmath372 has @xmath373 now suppose that we would have @xmath374 . then in each step leading from @xmath375 to @xmath376 the coboundary map in ( [ coboundary maps ] ) must surjects onto the respective cohomology groups @xmath377 it follows that @xmath378 whence @xmath379 . this already implies the inequality @xmath380 for @xmath381 . it remains to rule out the case @xmath382 and @xmath56 . the preceding paragraph reveals that than @xmath383 , and the short exact sequence @xmath384 yields @xmath385 . on the other hand , riemann roch gives @xmath386 contradiction . we continue to study the normal surface @xmath19 . let @xmath387 be the blowing - up of the unique singularity . we saw in the preceding section that the singular locus @xmath388 consists of @xmath25 rational double points of type @xmath2 . let @xmath266 be the minimal resolution of these double points . we now dispose off the _ irregularity _ @xmath389 . [ irregularity ] the irregularity @xmath389 vanishes . since @xmath240 contains only rational singularities , it suffices to check @xmath390 . let @xmath391 be the blowing - up of the reduced fixed point @xmath163 . we claim that the schematic preimage on @xmath392 of the singular point @xmath393 is a cartier divisor , such that the universal property of blowing - ups gives a commutative diagram @xmath394 this is a local problem . using the notation from section [ products of artin schreier curves ] , we have to understand what happens with the ideal @xmath395\ ] ] on @xmath392 . but @xmath396 according to ( [ formal substitution ] ) , and it obvious that the latter ideal becomes invertible upon blowing - up of @xmath397 . the @xmath20-action on @xmath18 induces a @xmath20-action on @xmath392 , and the canonical morphism @xmath398 is an isomorphism by the main theorem of zariski . it follows from ( [ formal substitution ] ) that the @xmath20-action on the cotangent space @xmath399 for the fixed point @xmath182 is trivial , whence the @xmath20-action on the exceptional curve @xmath400 is trivial as well . now lemma [ injective pullback ] below ensures that the induced map @xmath401 is injective . to finish the argument , consider the reduced fiber union @xmath402 for the two projections @xmath403 . its preimage on @xmath392 contains the strict transform @xmath404 of @xmath405 , which is isomorphic to a disjoint union @xmath406 , and we have a commutative diagram @xmath407 the term @xmath408 vanishes by corollary [ fiber union ] . since the map on the left is injective , it suffices to check that the the restriction map @xmath409 is injective . indeed , the maps @xmath410 are all bijective . in the course of the preceding proof we have used a fact that appears to be of independent interest . let us formulate it in a rather general way : suppose @xmath33 is a scheme over a field @xmath12 , and @xmath20 be a finite group acting on @xmath33 so that the quotient @xmath411 exists as a scheme . [ injective pullback ] assumptions as above . suppose additionally that @xmath412 and that there is a rational fixed point @xmath413 . then the canonical map @xmath414 is injective . the idea is to use @xmath20-equivariant cohomology @xmath415 , which was introduced in @xcite . consider the two spectral sequences with @xmath416-terms @xmath417 abutting to @xmath418 , where @xmath419 denotes the sheaf of cohomology groups . they give rise to a commutative diagram @xmath420\\ & & h^1(y,{\mathcal{h}}^0(g,{\mathcal{o}}_x))\ar[d]\ar[dr]\\ 0\ar[r ] & h^1(g , h^0(x,{\mathcal{o}}_x))\ar[r]\ar[dr ] & h^1(x , g,{\mathcal{o}}_x)\ar[r]\ar[d ] & h^0(g , h^1(x,{\mathcal{o}}_x))\\ & & h^0(g,{\mathcal{h}}^1(g,{\mathcal{o}}_x))\\ } \end{xy}\ ] ] with exact row and column , and the composition given by the upper diagonal arrow @xmath421 is our map in question . by a diagram chase , it therefore suffices to check that the other composition @xmath422 is injective . now comes in our rational fixed point @xmath413 : composing further with the restriction map induced by @xmath423 , we obtain @xmath424 by assumption , the map @xmath425 is bijective , whence the assertion . let @xmath426 be the quotient map . if @xmath33 is normal , and the order of @xmath20 is prime to the characteristic of @xmath12 , then the existence of a trace map shows that @xmath427 is a direct summand , such that @xmath428 is injective for all @xmath429 . on the other hand , if @xmath240 is an enriques surface in characteristic @xmath345 with @xmath430 , and @xmath431 is the k3-covering , such that @xmath432 is cyclic of order two and @xmath411 , then @xmath433 is @xmath114-dimensional , whereas @xmath434 vanishes . we note in passing that this situation is somewhat typical : [ almost injective ] suppose @xmath20 is cyclic , acts freely on @xmath33 , and @xmath412 . then the kernel of the canonical map @xmath414 is at most @xmath114-dimensional . making a diagram chase in ( [ spectral sequence ] ) , the dimension of the kernel is bounded by the dimension of @xmath435 . clearly , @xmath436 is the trivial @xmath20-module . let @xmath437 . then @xmath438 is isomorphic to the kernel of the multiplication map @xmath439 , whence a @xmath12-vector space of dimension at most one . we now study the global geometry of the normal surface @xmath19 in more detail . recall that @xmath387 is the blowing - up of the singularity . we saw in section [ invariants of the singularity ] that @xmath240 is normal and contains @xmath440 rational double points of type @xmath2 . let @xmath266 be the minimal resolution of these singularities . then the composite map @xmath441 is the minimal resolution of singularities . let @xmath442 be the contraction to a minimal model @xmath38 . we display our surfaces and maps in a commutative diagram : @xmath443 & & & x\ar[dl]^{g}\ar[rr]\ar[dll]_{h}\ar[dd]^{\psi_1,\psi_2 } & & s \\ & ( c\times c')/g\ar[drr]_{\varphi_1,\varphi_2 } & y\ar[l]^ { \quad\qquad f}\\ & & & { \mathbb{p}}^1 } \end{xy}\ ] ] the top row contains the smooth surfaces , the middle row the normal surfaces , and the arrows @xmath444 , @xmath445 are the maps induced from the two projections @xmath164 and @xmath165 . the goal of this section is to determine the place of @xmath33 , or rather its minimal model @xmath38 , in the enriques classification of surfaces , in dependence on the prime power @xmath11 . an elementary argument involving only intersection numbers already gives : [ x general type ] if @xmath446 , then the surface @xmath33 is of general type . we first compute the number @xmath447 on the smooth surface @xmath33 . obviously @xmath448 , such that @xmath449 . but @xmath330 by proposition [ canonical cycle ] , whereas @xmath450 according to proposition [ canonical quotient ] . the upshot is @xmath451 now suppose that @xmath446 . then @xmath452 , and the theorem of riemann roch gives @xmath453 so for @xmath454 either @xmath455 or @xmath456 is effective . the latter is impossible , because the canonical divisor on @xmath33 maps to the canonical divisor on @xmath19 , which is effective . thus @xmath455 , and in turn @xmath457 is effective . since @xmath458 , it follows from the enriques classification of surfaces that the minimal surface @xmath38 is of general type . to understand the remaining cases , and the geometry of the contraction @xmath442 as well , we have to analyze the two fibrations @xmath459 , @xmath460 induced by the projections @xmath164 and @xmath165 . let us first record : [ geometrically connected ] we have @xmath461 . by the main theorem of zariski , it suffices to check this at the generic point . let @xmath462 be the function field of @xmath69 , such that @xmath463 is the function field of the projective line . by construction , the generic fiber @xmath464 of @xmath465 is isomorphic to @xmath466 . in other words , the generic fiber is a twisted form of @xmath467 with respect to the tale topology . this description ensures that @xmath468 , and the result follows . now let @xmath469 and @xmath470 be the respective schematic fibers for the projections @xmath471 and @xmath472 containing the exceptional divisor @xmath473 for the resolution of singularities @xmath59 . up to multiplicities , the fiber @xmath474 , @xmath460 is the union of this exceptional divisor @xmath473 with another integral curve @xmath475 , which is birational to @xmath476 . one can say more about the curves @xmath477 : [ minus - one curves ] the curves @xmath477 are @xmath300-curves , and the reduction of the fiber union @xmath478 has only simple normal crossings . by symmetry , it suffices to treat @xmath479 . being an integral component of a reducible fiber , it has selfintersection @xmath480 . thus it suffices to check that @xmath481 . in light of corollary [ canonical cycle ] and proposition [ fundamental cycle ] , we have @xmath482 . the image of @xmath483 on the normal surface @xmath19 is a schematic fiber . using the projection formula and computing intersections on @xmath18 , we deduce the value @xmath484 . the upshot is that @xmath485 whence @xmath479 is a @xmath300-curve . consequently , @xmath486 . in light of corollary [ canonical cycle ] , it follows that @xmath487 intersects the exceptional divisor @xmath473 in precisely one component where the fundamental cycle attains its minimal multiplicity , that is , @xmath488 . we conclude that the reduced fiber @xmath489 is simple normal crossing . let us choose the indices @xmath98 for the @xmath271 so that @xmath490 . thus the dual graph of the fiber union @xmath491 looks like this : figure : dual graph for fiber union @xmath491 on @xmath33 . we are now in position to compute the multiplicities occurring in the schematic fibers @xmath492 : [ schematic fibers ] the schematic fibers are given by @xmath493 and @xmath494 . it suffices to treat @xmath495 . by definition of the fundamental cycle @xmath302 , we have @xmath496 and it is a straightforward computation that this cycle is numerically trivial on all irreducible components of @xmath495 . it is therefore a rational multiple of @xmath495 . our cycle contains @xmath479 with multiplicity @xmath3 . but @xmath483 is the strict transform of the fiber for @xmath497 , by proposition [ canonical quotient ] . thus our cycle coincides with @xmath495 . note that the component @xmath498 has multiplicity one in the fiber @xmath470 , and similarly for @xmath499 . in particular , the fibers are neither multiple nor wild . moreover : [ direct images ] the sheaf @xmath500 is locally free of rank @xmath136 , and the formation of the direct image @xmath501 commutes with base change . this follows from @xcite , theorem 7.2.1 , because each geometric fiber of @xmath445 contains a reduced irreducible component . this leads to a very useful consequence concerning fundamental group : [ fundamental group ] the fundamental group @xmath36 vanishes . let @xmath502 be a finite tale covering with @xmath503 nonempty . we have to check that it has a section . consider the projection @xmath504 , and let @xmath505 be the stein factorization for the composition @xmath506 , which is given by @xmath507 . as explained in @xcite , remark 7.8.10 , the finite morphism @xmath508 is tale , since the equality @xmath509 commutes with base change . moreover , the fiber @xmath510 is simply connected , by the description in proposition [ schematic fibers ] . using this as in the proof for @xcite , expos x , theorem 1.3 , we infer that the canonical map @xmath511 , which is finite tale , is actually an isomorphism . but the projective line is simply connected , so @xmath508 , whence also @xmath502 has a section . for later use , we record : [ picard group ] the picard group @xmath35 is a finitely generated free abelian group . it already follows from proposition [ irregularity ] that @xmath512 , whence @xmath513 is finitely generated . if @xmath514 is a prime different from the characteristic , then the elements @xmath515 of order @xmath31 yield nontrivial @xmath516-torsors @xmath502 , which are finite tale coverings of degree @xmath31 . since the fundamental group vanishes , there are no such elements . finally , suppose there is an element @xmath517 of order @xmath0 . to handle this case , first note that the artin schreier curve @xmath69 has @xmath0-rank @xmath518 . this follows from @xcite , theorem 3.5 , applied to @xmath111 . in other words @xmath519 , or equivalently , @xmath520 contains no element of order @xmath0 . then the generic fiber @xmath464 for the second projection @xmath472 , which is a twisted form of @xmath69 over the function field of @xmath297 , admits no invertible sheaf of order @xmath0 . thus @xmath521 is trivial . it follows that @xmath522 , where @xmath523 is a divisor supported by fibers of @xmath472 . now recall that the intersection form on the integral components of a given fiber is negative semidefinite . moreover , the radical is generated by the fiber , since the fibers have multiplicity one by proposition [ schematic fibers ] . using that @xmath517 is numerically trivial , we first deduce that each @xmath524 , @xmath525 is a multiple of the fiber , and then that @xmath523 is linearly equivalent to the zero divisor , contradiction . the geometry of our surface @xmath33 simplifies further if we contract in a different way . indeed , the curves @xmath526 and @xmath527 are two disjoint _ exceptional curves of the first kind _ , comprising altogether @xmath528 irreducible components . let @xmath529 be their contraction , such that @xmath530 is a smooth surface with @xmath531 by choosing another minimal model @xmath38 if necessary , we may tacitly assume that the contraction @xmath442 factors over @xmath530 . let @xmath532 be the images of the curves @xmath533 for @xmath534 . then @xmath535 is the image of the fundamental cycle @xmath27 for the resolution of singularities @xmath59 . [ canonical divisor ] we have @xmath536 , and @xmath537 . recall that @xmath538 by ( [ selfintersection resolution ] ) . the map @xmath539 contracts successively @xmath528 @xmath300-curves , such that @xmath540 the canonical class @xmath541 is the image of the canonical class @xmath542 . recall that we have @xmath543 the latter coincides with the cycle @xmath544 , up to components that are contracted by @xmath539 . consequently @xmath545 . we now have an explicit description of the minimal model : [ minimal surface ] for @xmath546 , the surface @xmath530 is minimal , such that @xmath547 . we have @xmath548 . consequently , there is no @xmath300-curve supported by @xmath549 . since @xmath541 is effective , there is no other @xmath300-curve on @xmath530 , and the result follows . from this we easily determine the place in the enriques classification of surfaces . recall that a _ weak del pezzo surface _ is a surface whose anticanonical divisor is nef and big . [ enriques classification ] the mimimal surface @xmath38 is of general type if @xmath39 , a k3-surface for @xmath40 , and a weak del pezzo surface for @xmath41 . first suppose @xmath39 . then @xmath547 is minimal , and @xmath550 is effective with @xmath551 . by the enriques classification , @xmath38 is of general type . for @xmath40 , we have @xmath552 . whence @xmath38 is either abelian , bielliptic , quasibielliptic , k3 or enriques . but @xmath553 by proposition [ irregularity ] , whence @xmath38 is either k3 or a classical enriques surface . in the latter case , @xmath554 is cyclic of order two . in light of proposition [ fundamental group ] , our @xmath38 must be a k3 surface . finally , suppose @xmath555 . then @xmath556 is effective , and one easily checks that @xmath557 is not an exceptional curve of the first kind . thus @xmath558 is nef . it is also big , according to proposition [ canonical divisor ] . the same necessarily holds for @xmath559 , hence @xmath38 is a weak del pezzo surface . in this section we introduce projective models for our surfaces . if @xmath39 then @xmath547 is a minimal surface of general type , and the homogeneous spectrum @xmath560 is called the _ canonical model _ of @xmath38 . the canonical morphism @xmath561 is the contraction of all @xmath562-curves , and the singularities on the normal surface @xmath48 are at most rational double points . clearly , the integral curves @xmath563 are @xmath562-curves , which get contracted . it turns out that there are no more : [ canonical model ] for @xmath564 , the canonical model @xmath48 is obtained by contracting the @xmath562-curves @xmath565 , @xmath566 , such that the singular locus of @xmath48 comprises exactly @xmath1 rational double points of type @xmath2 . the exceptional curve for the contraction @xmath567 is the union of all @xmath562-curves , so all @xmath565 , @xmath566 get contracted . suppose there would be another @xmath562-curve @xmath568 , such that @xmath569 is disjoint from the canonical divisor @xmath570 . this implies that its strict transform @xmath571 is contained in a smooth fiber of one of the projections @xmath572 , contradiction . note that the contraction @xmath567 of the @xmath562-curves @xmath565 makes sense for all prime powers @xmath11 , and it turns out that the normal surface @xmath48 has a very satisfactory projective description . to see this , let @xmath49 be the weil divisor defined as the image of @xmath27 . recall that the latter is nothing but the fundamental cycle for the resolution of singularities @xmath573 . then clearly @xmath574 since the local picard group of a rational double point of type @xmath2 is cyclic of order @xmath3 , the weil divisor @xmath49 is actually cartier . the invertible sheaf @xmath50 defines a rational map @xmath575 , with @xmath576 . the invertible sheaf @xmath50 is very ample , has @xmath51 , and the image of the closed embedding @xmath53 is a normal surface of degree @xmath3 whose singular locus consists of @xmath1 rational double points of type @xmath2 , all lying on a line in @xmath4 . we first check that @xmath577 is globally generated . let @xmath578 , @xmath460 be the image of the schematic fibers @xmath579 for the two projections @xmath459 . using the multiplicities computed in proposition [ schematic fibers ] , we deduce @xmath580 . now let @xmath581 be the image of the schematic fibers @xmath582 . then @xmath583 are linearly equivalent . using that @xmath584 intersects @xmath585 only in the component @xmath586 , we deduce that and @xmath587 is the image of the contracted curve @xmath586 under the canonical map @xmath588 . by symmetry , @xmath589 is the image of @xmath479 . the upshot is that @xmath590 is empty , hence @xmath577 is globally generated . we also showed that @xmath591 . by the same ideas we verify that the resulting morphism @xmath592 is generically injective , where @xmath576 . let @xmath593 be two points not lying on the cycle @xmath594 with @xmath595 . the image on @xmath48 of the schematic fiber @xmath596 is a divisor linearly equivalent to @xmath597 , which contains the image of @xmath155 , but not the image of @xmath598 . consequently , our map @xmath599 is generically injective . next , we verify @xmath51 . to this end , consider the short exact sequence @xmath600 since @xmath601 , we have @xmath602 . the weil divisors @xmath603 yield short exact sequences @xmath604 for @xmath605 , the kernels are @xmath606 , which follows from the intersection number @xmath607 and proposition [ first blowing - up ] , together with an application of @xcite , theorem 1.9 . this gives the estimate @xmath608 . if @xmath609 , then @xmath610 would be a finite surjective map of degree @xmath3 , contradicting generic injectivity . thus @xmath51 . summing up , we have a morphism @xmath53 whose image @xmath611 is a divisor of degree @xmath3 , and the induced map @xmath612 is the normalization map . clearly , @xmath613 is cohen macaulay and gorenstein . using @xmath614 and @xmath615 , we deduce @xmath616 , such that relative dualizing sheaf @xmath617 is trivial . from this it follows that the conductor locus for the finite birational morphism @xmath612 is empty , such that @xmath618 is an isomorphism . consequently @xmath577 is very ample . finally observe that the image of @xmath619 , which contains the singular locus of @xmath48 , is a line , because @xmath607 . recall that a proper @xmath12-scheme @xmath620 is called _ liftable in the category of schemes _ , if there exists a local ring @xmath621 of characteristic zero with @xmath622 , together with a proper flat @xmath193-scheme @xmath623 with @xmath624 . if such @xmath623 exists at least as an algebraic space , we say that @xmath620 is _ liftable in the category of algebraic spaces_. the surface @xmath38 is liftable in the category of algebraic spaces . we first consider the projective normal surface @xmath625 . let @xmath626 $ ] be a homogeneous polynomial of degree @xmath627 so that @xmath628 as subscheme of @xmath629 $ ] . choose a homogeneous polynomial @xmath15 of degree @xmath630 with coefficients in the ring of witt vectors @xmath631 reducing to @xmath632 modulo @xmath0 , and consider the proper @xmath193-scheme @xmath633 . then @xmath634 is flat and projective , such that @xmath48 is liftable in the category of schemes . according to a general result of artin and brieskorn @xcite , there exists a finite extension @xmath635 and a simultaneous minimal resolution of singularities @xmath636 . here , however , one knows only that the total space @xmath637 is an algebraic space , although the individual fibers are projective . thus @xmath638 is liftable in the category of algebraic spaces . the _ chern invariants _ of a smooth proper surface are the numbers @xmath639 they are paramount for minimal surfaces of general type , and the study of occurrence and distribution of chern invariants for minimal surfaces of general type is referred to as _ surface geography_. we already computed the value of @xmath642 in proposition [ canonical divisor ] . recall that @xmath539 is a sequence of @xmath528 blowing - ups of @xmath530 , such that @xmath643 . to determine the euler characteristic , we first examine the surface @xmath33 and its fibration @xmath471 . according to dolgachev s formula @xcite we have @xmath644 where @xmath645 is the geometric generic fiber , the sum runs over all closed point @xmath525 , and @xmath46 is _ serre s measure of wild ramification _ attached to the galois module @xmath646 at the point @xmath525 . here @xmath31 is any prime number different from @xmath0 . this invariant is given by @xmath647 } \dim_{{\mathbb{f}}_l}(m_a / m_a^{g_i}),\ ] ] here @xmath20 is the galois group of a finite galois extension of the function field @xmath648 trivializing the galois module @xmath649 , and @xmath97 are the ramification subgroups for the induced extension of discrete valuation rings . by the very construction , we may choose this extension induced from @xmath178 , such that @xmath650 . in light of ( [ ramification groups ] ) , we have @xmath651 . according to proposition [ picard group ] , the picard group @xmath35 is finitely generated . since the map @xmath652 is surjective , the group @xmath653 is finitely generated as well . thus we may choose our prime @xmath31 so that @xmath653 contains no nontrivial @xmath31-torsion . since the brauer group of the function field of @xmath297 vanishes by tsen s theorem , it then follows that @xmath654 , whence @xmath655 . in turn , we obtain the value for @xmath656 . finally , riemann roch for surfaces @xmath657 yields the formula for @xmath658 . there are two unfortunate misprints in @xcite : the correction terms in theorem 1.1 should appear with the sign lost in the proof while passing from equation ( 3.2 ) to ( 3.3 ) . in the definition of serre s measure of wild ramification @xmath659 on top of page 305 , the sum should run only for @xmath660 , which comes from the fact that the swan character is the difference of the artin character and the augmentation character . compare also @xcite . using our global invariants , we now can show that our bound on the geometric genus @xmath665 of the singularity on @xmath19 in proposition [ geometric genus ] is actually an equality , at least in a special case : the leray serre spectral sequence for @xmath441 gives an exact sequences @xmath667 the term on the left vanishes by proposition [ irregularity ] . we thus have @xmath668 the first summand was computed in proposition [ invariant differentials ] , the second in corollary [ global geometric genus ] , and the result follows . the _ chern quotients _ of our surfaces @xmath38 asymptotically tend to @xmath673 note also that we always have @xmath674 , such that the _ bogomolov miyaoka yau inequality _ @xmath675 holds . therefore our surfaces @xmath38 show no exotic behavior with respect to geography .

in this article we present a non - linear dynamic programming algorithm for the computation of forward rates within the maximum smoothness framework . the algorithm implements the forward rate positivity constraint for a one - parametric family of smoothness measures and it handles price spreads in the constraining dataset . we investigate the outcome of the algorithm using the swedish bond market showing examples where the absence of the positive constraint leads to negative interest rates . furthermore we investigate the predictive accuracy of the algorithm as we move along the family of smoothness measures . among other things we observe that the inclusion of spreads not only improves the smoothness of forward curves but also significantly reduces the predictive error .

math0305307

within the financial industry forward rate curves play a central role in fixed - income derivative pricing and risk management . notwithstanding , such curves are not empirical or directly measurable objects but rather useful abstract concepts from where observed prices can be derived . moreover , given a finite set of market prices , we can construct in general an infinite number of compatible forward rate curves . to avoid such ambiguity several approaches have been proposed in the literature trying to capture a reasonable or natural functional form within the set of compatible possibilities . in historical order the first kind of methods proposed to solve this problem make use of the so - called parametric approach . in this approach a particular functional form for the forward rate curve is assumed leaving a certain number of free parameters to be fixed from the calculation of a given set of quoted prices . an extensive literature exists advocating for this approach . we can cite as examples the works of mcculloc @xcite , vasicek and fong @xcite , chambers , carleton and waldman @xcite , shea @xcite , nelson and siegel @xcite and more recently the works of svensson @xcite , fisher , nychka and zervos @xcite and waggoner @xcite . in most of these works we notice the privileged role played by polynomial and exponential splines as the preferred functional forms for the forward rate curves . the second kind of methods has been termed in the literature as non - parametric or maximum - smoothness approach . here instead of advocating for an a priori functional form for the forward rate a given measure of smoothness is chosen and then the forward rate curve is obtained as the one maximizing this measure subject to the constraints imposed by market prices . examples where these methods have been investigated include the works of adams and van deventer @xcite , delbaen and lorimier @xcite , kian guan lim , qin xiao et al @xcite , frishling and yamamura @xcite and yekutieli @xcite . in these works three different smoothing measures have been proposed . we also have the works of forsgren @xcite and kwon @xcite that generalize these methodologies and clarify the connection between splines and certain smoothness measures . finally we point out the work of wets , bianchi and yang @xcite that can be located somewhere in - between both approaches since here the number of functional parameters is finite ( albeit arbitrarily large ) and the functional behavior is restricted to a subfamily of @xmath0 curves . the purpose of this article is two - fold : firstly we want to present an efficient maximum - smoothing algorithm that handles the presence of spreads and implements the positivity constraint . secondly we want to investigate the predictive power of a linear combination of two quadratic measures , namely the one proposed by delbaen et al . and frishling et al . @xcite and the one by adams and van deventer @xcite . here it is worth remarking that once the compatibility with market prices is fulfilled the only guiding principle that should be taken as definition of reasonable or natural is the predictive power and not other ad - hoc criteria . in this article we will only use as constraining data coupon bearing bonds . the inclusion of treasury bills , zero coupon bonds or bill futures is straightforward and amounts to adding the corresponding linear constraints . since our objective in this article is focusing on an algorithm dealing with non - linear constraints and inequality constraints ( spreads and positivity constraint ) we have not included such data . with these objectives in mind we organize the article as follows : in section [ sec algorithm ] we present the objective function that we will use throughout the article and we establish the basic notation . in this section we present a sketch of the complete algorithm leaving the details for the appendices . in section [ sec results ] we present the results of the article including examples where the absence of the positivity constraint or the adequate spreads leads to negative rates . here we present also a study of the predictive power of the one - parametric family of smoothness measures that include as extreme cases the measures used by delbaen et al . , frishling et al . and adams and van deventer . finally in section [ sec conclusions ] we present the conclusions a bond , @xmath1 , is an instrument that gives future coupons , @xmath2 , at time stages @xmath3 and a final payment , @xmath4 we mean the complete last cash flow of bond @xmath1 , typically that includes a principal plus a last coupon . ] . the bond price , @xmath5 , can be determined from the discrete forward rate curve , @xmath6 , as follows @xmath7 where @xmath8 is the length of the time period between time stage @xmath9 and @xmath10 ( in our implementation we have used @xmath11 day ) . the objective function , or smoothing measure , is defined as a linear combination of the one used by delbaen et al . and frishling et al . ( df ) @xcite and the one used by adams and van deventer ( ad ) @xcite @xmath12 the first term in eq.([objective ] ) ( df ) is a discrete approximation of the integral of the square of the first derivative of @xmath13 and the second term ( ad ) is a discrete approximation of the integral of the square of the second derivative of @xmath13 . this objective function is to be minimized subject to the consistency constraints@xmath14 where @xmath15 is to be determined along with @xmath13 and where we have used the definitions @xmath16 with @xmath17 @xmath18 the respective bid and ask prices of bond @xmath1 ( @xmath19 ) . note that the equality constraint given by eq.([eq_const ] ) is just eq.([bondprice ] ) rewritten taking logarithms and using the definitions ( [ definitions ] ) . eq.([ineq_constr ] ) introduces two inequality constraints . the first one is the positivity constraint over the forward rate curve and the second one is the requirement that the single price given by eq.([bondprice ] ) must lie in - between the bid and ask prices . we define the spread of bond @xmath1 as the quantity @xmath20 . we take the largest time to maturity in eq.([objective ] ) equal to the largest time to maturity in the constraining dataset , namely@xmath21 the constraints reflecting bond prices ( [ bondprice ] ) have been rewritten in a way such that they become linear when no coupons are present ( @xmath22 for a zero coupon bond @xmath1 ) . constraints given by eqs.([eq_const ] ) and ( [ ineq_constr ] ) are moved to the objective function defining@xmath23 -\mu\sum _ { r=1}^{n}\ln\left ( f_{r}\right ) -\tilde{\mu}\sum_{j=1}^{m}\left ( \ln\left ( \rho_{j}-\rho_{j}^{b}\right ) + \ln\left ( \rho_{j}^{a}-\rho _ { j}\right ) \right ) , \label{total_obj}\ ] ] with the lagrange multipliers @xmath24 @xmath25 and the logarithmic barriers with parameters @xmath26 and @xmath27 ( in the solution procedure we take @xmath28 , @xmath29 ) . the use of log barriers to deal with inequality constraints is a standard methodology in interior point methods for optimization problems @xcite . an explanation of this methodology adapted to our problem is given in appendix a. briefly the minimization algorithm is structured as follows:@xmath30 } \text { , } \tilde{\mu}^{\left [ 0\right ] } \text { and set ( } f,\rho\text { ) } = \text{(}f^{\left [ 0\right ] } , \rho^{\left [ 0\right ] } \text { ) } \nonumber\\ & \text{with the seed ( } f^{\left [ 0\right ] } , \rho^{\left [ 0\right ] } \text { ) satisfying the inequality constraints ( \ref{ineq_constr } ) . let } k=0.\nonumber\\ \text{step 1}\text { : } & \text{make a second order approximation of } z\text { at ( } f^{\left [ k\right ] } , \rho^{\left [ k\right ] } \text { ) ( see appendix a).}\nonumber\\[0.03 in ] \text{step 2}\text { : } & \text{determine the newton step ( } \hat{f}^{\left [ k+1\right ] } , \hat{\rho}^{\left [ k+1\right ] } \text { ) using dynamic programming ( see appendix b).}\nonumber\\[0.03 in ] \text{step 3}\text { : } & \text{modify ( } \hat{f}^{\left [ k+1\right ] } , \hat{\rho}^{\left [ k+1\right ] } \text { ) to get a solution ( } f^{\left [ k+1\right ] } , \rho^{\left [ k+1\right ] } \text { ) that satisfies the inequality}\nonumber\\ & \text{constraints ( \ref{ineq_constr } ) . update log barriers ( } \mu^{\left [ k\right ] } \rightarrow0\text { and } \tilde{\mu}^{\left [ k\right ] } \rightarrow0\text { as } k\rightarrow\infty\text { ) and}\nonumber\\ & \text{calculate the values of } w\text { and of constraints ( \ref{eq_const } ) . check if a termination criterion}\nonumber\\ & \text{is satisfied , otherwise let } k = k+1\text { and go to step 1 ( see appendix c ) . } \label{algorithm}\ ] ] computing times involved in step 2 are summarized in subsection b.1 . the solution typically stabilizes in approximately 6 iterations as can be seen in fig.([graph2 ] ) . on a pentium 4 , 2.4 ghz computer the algorithm coded in c++ takes around 1/4 sec . to compute a forward rate curve like any of the ones seen in fig.([graph1 ] ) . ( [ objective ] ) ) vs. newton iterations ( starting from the seeds : @xmath31}=0.04/year$ ] , @xmath32}=(\rho_{j}^{a}+\rho_{j}^{b})/2 $ ] ) . the plots correspond to the forward rate curves of fig.([graph1 ] ) . the converge behavior is affected by the termination criterion dictated by eq.([criterion ] ) and by the initial values of the log barriers dictated by eq.([values ] ) . note the exponential convergence of the algorithm in the first steps . note also that on friday 06 the minimal solution for ad measure is , for all practical purposes , a straight line . , width=566 ] year@xmath33 , @xmath34 ) while the ones in plot ( b ) correspond to the ad measure ( @xmath35 , @xmath36 year@xmath37 ) . in both plots we have used a tolerance spread of 1% ( @xmath38 ) . , width=566 ] in this section we present some examples of the behavior of algorithm ( [ algorithm ] ) and we investigate the predictive power of the resulting forward rate curves . in fig.([graph1 ] ) we present a series of forward rate curves calculated using df and ad smoothing measures . there we can see that for both measures the resulting curves share some similar traits like the positions of most peaks and dips . clear differences between both sets of curves are found at their end - points and in their behavior in the presence of high spreads . in the set of curves obtained from the df measure we have vanishing first derivative at the end - points and curves that tend to constants for high spreads . for the ad measure we have vanishing second and third derivatives at end - points @xcite and curves for high enough spreads given by straight lines . from the financial point of view these features are , in principle , just different aesthetic possibilities . in order to choose a particular measure the guiding principles should be , in the first place , the fulfillment of the consistency constraints given by eqs.([eq_const]-[ineq_constr ] ) and after this is guaranteed the predictive performance . let us start now with the analysis of consistency . as can be found in @xcite if we do not consider the positivity constraint the local minima of objective ( [ objective ] ) are given by exponential splines with exponents @xmath39 or polynomial splines of order 2 or 4 when @xmath40 or @xmath41 are respectively zero . the main problem with these exponential or polynomial splines is that there is no warranty that they fulfill the positivity constraint . negative rates are not admissible in the absence of arbitrage opportunities and the risk of obtaining this unwanted feature is illustrated in fig.([graph3 ] ) . in this figure we have concentrated on the swedish bond data on monday , july 09 , 2001 . there we have tested three spread patterns for both df and ad measures with and without the positivity constraint . from these plots it is evident that the inclusion of spreads in the calculation of the forward rates is a necessary ingredient that can have a major impact in the resulting functional behavior . once we have an algorithm that insures the consistency of forward rates we can concentrate on the predictive accuracy of different measures . however , before starting the analysis of this issue let us make a brief digression to comment a point regarding measure ( [ objective ] ) . if we want to have both @xmath41 and @xmath40 different from zero and we want to compare the effects of each term it is important to realize that df and ad measures scale differently under changes of time units . in other words , @xmath41 and @xmath40 have different units . a practical way to define their units is to consider the objective @xmath42 as an adimensional quantity . by doing so and remembering that @xmath13 has units of inverse time , it is immediate to obtain that @xmath41 has units of time@xmath33 and @xmath40 units of time@xmath37 . the importance of keeping this in mind becomes apparent in results like the ones presented in figs.([graph4 ] ) and ( [ graph5 ] ) . figs.([graph4 ] ) and ( [ graph5 ] ) summarize our results regarding the predictive accuracy of the algorithm as a function @xmath43 there it is clear that the characteristic time span where df and ad compete is not the day or the century , but clearly the year . to construct these figures we have calculated the forward rate curves for different values of @xmath44 when one bond is removed from the constraining dataset . the price of this missing bond is used afterwards as a benchmark to test the accuracy of the resulting curves . since we are interested in the statistical performance we have done such comparison for 335 consecutive trading days starting on wednesday , november 08 , 2000 and ending on thursday , march 07 , 2002 . we are also interested in studying the impact of spreads in the constraining dataset over the predictive accuracy . therefore we present our results for three spread patterns , namely constant spreads of 0% , 0.5% and 1% in the constraining dataset . fig.([graph4 ] ) concentrates on the predictive accuracy for the first 9 bonds of table ( [ table2 ] ) and fig.([graph5 ] ) presents the same analysis for the remaining 2 bonds of table ( [ table2 ] ) . these last 2 bonds are the ones with the larger maturities in the complete dataset . in particular for the last one with the largest maturity we have to decide upon the methodology to extrapolate the forward curve outside the range of the constraining dataset . therefore in fig.([graph5 ] ) we present the prediction accuracy using constant extrapolation from the last maturity in the constraining dataset and @xmath42-generated extrapolation that consists in utilizing the @xmath42-optimal forward rate curve even outside the range of the constraining dataset . and [ table2 ] ) and allowing for a large 5 % in the rest ( almost unconstraining this subset ) . doing so we observe in ( b1 ) and ( b2 ) curves that are still negative in the region where the nominal cash flows of the 0-spread bonds take place . , width=604 ] . these prediction errors are obtained removing the indicated bonds from the constraining dataset . the statistical sample comprise the forward curves corresponding to 335 consecutive trading days starting on wednesday , november 08 , 2000 and ending on thursday , march 07 , 2002 . within this set of days we calculate the relative errors of the predicted prices given by @xmath45 where @xmath46 is the actual price of the bond given by the market and @xmath47 is the price predicted using the prices of all other bonds in the dataset . hence in plots ( a1 ) , ( a2 ) and ( a3 ) we show the average of the absolute value of these errors and the plots ( b1 ) , ( b2 ) and ( b3 ) just their average . consecutive columns show averages obtained using spreads of @xmath48 , @xmath49 and @xmath50 in the constraining dataset . the set of predicted bonds shown in this figure is given by the first 9 bonds of table [ table2 ] . the remaining bonds of this table ( the last two bonds with the larger maturities ) are analyzed in fig.([graph5 ] ) . note how prediction errors tend to decrease when spreads are considered albeit not significantly ( compare with fig.([graph5 ] ) ) . note also that in the absence of spreads the df measure ( @xmath51 ) systematically exhibits a better performance than the ad one ( @xmath52)).,width=566 ] ) but now the plots correspond to the last two bonds of table [ table2 ] that are the ones with larger maturities within our dataset ( we have also included the bond with the shortest maturity to facilitate the comparison with fig.([graph4 ] ) ) . for the bond with the largest maturity the forward curve has to be extrapolated to reach its last cash flows ( see fig.([graph3 ] ) ) . we consider two extrapolation possibilities , one where we continue the forward rate curve from the last cash flow as a constant and the other where the curve is dictated by the minimization of functional @xmath42 even beyond the last cash flow in the constraining dataset . note that constant extrapolation gives better results than w - generated extrapolation in the region close to the ad measure ( @xmath52 ) . again like in fig.([graph4 ] ) we observe that in the absence of spreads the df measure ( @xmath51 ) systematically exhibits a better performance than the ad one . the most striking difference with fig.([graph4 ] ) is that here we observe that for these two long maturing bonds the introduction of spreads drastically reduce the prediction error all along the family of measures.,width=566 ] in this article we have presented a non - linear dynamic programming algorithm designed for the calculation of positive definite forward rate curves using data with or without spreads . we have included multiple details of the algorithm aiming at practitioners not familiar with the techniques of dynamic programming . we have illustrated the results of this algorithms using the swedish bond data for a one - parametric family of smoothness measures . the results and conclusions are the following : * the proposed algorithm calculates forward rate curves in real time and admits , without time or complexity penalizations , the use of any non - linear _ local _ objective function . since it also handles non - linear constraints it is possible to include within the constraining dataset any derivative products with prices bearing some dependence on forward rates . * this is the first algorithm proposed in the literature that implements the positivity constraint in the maximum smoothness framework . to the knowledge of the authors the only other work that implements such constraint outside this framework is the one of wets et al . @xcite . the proposal in @xcite has the advantage of using simple linear programming but do not consider the presence of spreads minimizing instead the sum over the _ modulus _ of the difference between calculated prices and market prices . * for the objective functions and constraining datasets like the ones we have used or more generally for the ones studied in @xcite the algorithm proposed in that reference offers better computing times at the expense of ignoring the positivity constraint . essentially the complexity in @xcite is @xmath53 and in ours is @xmath54 @xmath55 where @xmath54 is the number of time steps and @xmath56 the number of constraints . for that reason , when this class of objective functions and constraining datasets are used , a well coded algorithm might try first the proposal in @xcite ( improving its treatment of the spreads using e.g. log - barriers ) and later , only if the result is not positive definite , use our approach . * since the optimization problem we are solving is non - convex we can not discard the presence of several local minima ( this is independent of the presence of the positivity constraint @xcite ) . however , we have tested our algorithm starting from different seeds and in all cases we have arrived at the same minima . these tests included hundreds of searches with initial log - prices given by @xmath57 } = \rho_{j}^{b}+x\left ( \rho_{j}^{a}-\rho_{j}^{b}\right ) $ ] ( with @xmath58 $ ] a flat random variable ) and initial forward rates given by several constant and oscillating functions . with this comment we only want to convey our practical experience and by no means we intend to say that we have exhaustively explored the presence or absence of local minima . * it is clear that one way to avoid negative rates is just by increasing spreads by hand . the advantage of using real spreads at a given moment is that one can be sure of being consistent with market prices . if one observes in fig.([graph3 ] ) the large forward rate variations taking place for different spread patterns no doubts should remain about the relevance of a careful treatment of this issue . * from fig.([graph5 ] ) we conclude that the inclusion of spreads can remarkably improve the accuracy of resulting forward curves in the prediction of market prices of long maturing bonds . in @xcite it was pointed out that the inclusion of spreads notably improved the smoothness of the forward curve . to the knowledge of the authors this is the first time it is shown that their presence also improves the prediction accuracy . for that reason we believe that spreads should be considered even when the market data does not provide such information . in that case the approach should consist in using a cross validation technique to asses the optimal spread minimizing an error criterion based in the prediction of market prices ( for cross validation methods see for example @xcite ) . one possibility is using the well know leave - one - out cross - validation to select the optimal spread much in the spirit suggested by figs.([graph4 ] ) and ( [ graph5 ] ) . given a constraining set of @xmath59 products , the method actually consists in obtaining @xmath59 forward rate curves for a given spread , each curve leaving out one of the constraining products ( bonds in our case ) but using only the @xmath59 omitted products to compute an error criterion like @xmath60 thus we obtain a quantitative criterion to select an optimal level of spread . for example for our dataset and our family of models ( measures ) it can be conjectured from figs.([graph4 ] ) and ( [ graph5 ] ) that this optimal level of spread is typically around 0.5% . * figs.([graph4 ] ) and ( [ graph5 ] ) strongly suggest that for low spread patterns df measure is more accurate than ad one . for bigger spreads results do not clearly favor any particular measure . j. m. acknowledges the financial support from _ tekniska hgskolan , inst . fr fysik och mtteknik _ , linkping university , thanks j. shaw ( barclays ) for calling his attention to ref . @xcite , f. delbaen for ref . @xcite and j. m. eroles ( hsbc ) for proof reading the manuscript . j. m. would also like to thank dr . per olov lindberg for the hospitality during his stay at the division of optimization ; linkping university . the authors would like to thank the referees for their helpful comments on the manuscript . since the objective function @xmath61 in eq.([total_obj ] ) is a non - quadratic function of @xmath62 , @xmath63 we will use an iterative quadratic approximation ( newton steps ) to find its minima . we start with feasible seeds @xmath64 } $ ] and @xmath57 } $ ] fulfilling inequality constraints ( [ ineq_constr ] ) and we set up initial log - barriers coefficients @xmath65 } > 0 $ ] and @xmath66 } > 0.$ ] for newton iteration number @xmath67 we define @xmath68 } & : = f_{r}^{\left [ s-1\right ] } + \delta _ { r}^{\left [ s\right ] } , \nonumber\\ \hat{\rho}_{j}^{\left [ s\right ] } & : = \rho_{j}^{\left [ s-1\right ] } + \sigma_{j}^{\left [ s\right ] } , \label{dynamic_sol}\ ] ] with @xmath69 the hat over @xmath13 and @xmath70 indicate that at each step @xmath67 , @xmath71 } $ ] and @xmath72 } $ ] are the minima of the quadratic approximation and may not fulfill the inequality constraints ( [ ineq_constr ] ) . to assure constraints ( [ ineq_constr ] ) are fulfilled a final redefinition @xmath71 } \rightarrow f^{\left [ s\right ] } , $ ] @xmath72 } \rightarrow \rho^{\left [ s\right ] } $ ] is necessary after each newton step . this redefinition is explained in appendix c. expanding @xmath61 up to second order in @xmath73 } $ ] and @xmath74 } $ ] we write@xmath75 } & : = \left . z\left ( \hat{f}^{\left [ s\right ] } , \hat{\rho}^{\left [ s\right ] } , \lambda^{\left [ s\right ] } \right ) \right\vert _ { o\left ( 2\right ) } \nonumber\\ & = \frac{1}{2}\delta^{\left [ s\right ] t}q^{\left [ s\right ] } \delta^{\left [ s\right ] } + \delta^{\left [ s\right ] t}b^{\left [ s\right ] } \lambda^{\left [ s\right ] } + \delta^{\left [ s\right ] t}c^{\left [ s\right ] } + \lambda^{\left [ s\right ] t}a^{\left [ s\right ] } \nonumber\\ & + \sigma^{\left [ s\right ] t}\lambda^{\left [ s\right ] } + \frac{1}{2}\sigma^{\left [ s\right ] t}m^{\left [ s\right ] } \sigma^{\left [ s\right ] } + \sigma^{\left [ s\right ] t}d^{\left [ s\right ] } + b^{\left [ s\right ] } , \label{quadratic}\ ] ] where @xmath76 } $ ] collects all terms not depending on @xmath77 } $ ] , @xmath78 } $ ] or @xmath79 } $ ] . we will use square brackets around newton step indices and parenthesis around dynamic programming ones . let us now work - out the matrices involved in the quadratic approximation . from eqs.([objective ] ) and ( [ quadratic ] ) we immediately obtain@xmath80 t}q^{\left [ s\right ] } \delta^{\left [ s\right ] } + \delta^{\left [ s\right ] t}c^{\left [ s\right ] } \\ & = \gamma\sum_{r=1}^{n-1}\left ( \frac{f_{r+1}^{\left [ s-1\right ] } -f_{r}^{\left [ s-1\right ] } } { \xi_{r}}\right ) \left ( \frac{\delta _ { r+1}^{\left [ s\right ] } -\delta_{r}^{\left [ s\right ] } } { \xi_{r}}\right ) \xi_{r}+\frac{\gamma}{2}\sum_{r=1}^{n-1}\left ( \frac{\delta_{r+1}^{\left [ s\right ] } -\delta_{r}^{\left [ s\right ] } } { \xi_{r}}\right ) ^{2}\xi_{r}\\ & + 4\varphi\sum_{r=2}^{n-1}\left ( \frac{1}{\left ( \xi_{r-1}+\xi_{r}\right ) } \left ( \frac{1}{\xi_{r}}f_{r+1}^{\left [ s-1\right ] } + \frac{1}{\xi_{r-1}}f_{r-1}^{\left [ s-1\right ] } \right ) -\frac{1}{\xi_{r}\xi_{r-1}}f_{r}^{\left [ s-1\right ] } \right ) \\ & \times\left ( \frac{1}{\left ( \xi_{r-1}+\xi_{r}\right ) } \left ( \frac { 1}{\xi_{r}}\delta_{r+1}^{\left [ s\right ] } + \frac{1}{\xi_{r-1}}\delta _ { r-1}^{\left [ s\right ] } \right ) -\frac{1}{\xi_{r}\xi_{r-1}}\delta _ { r}^{\left [ s\right ] } \right ) \xi_{r}\\ & + 2\varphi\sum_{r=2}^{n-1}\left ( \frac{1}{\left ( \xi_{r-1}+\xi_{r}\right ) } \left ( \frac{1}{\xi_{r}}\delta_{r+1}^{\left [ s\right ] } + \frac{1}{\xi _ { r-1}}\delta_{r-1}^{\left [ s\right ] } \right ) -\frac{1}{\xi_{r}\xi_{r-1}}\delta_{r}^{\left [ s\right ] } \right ) ^{2}\xi_{r},\end{aligned}\ ] ] and defining @xmath81 } & : = \sum_{j=1}^{m}\lambda_{j}^{\left [ s\right ] } \left [ \hat{\rho}_{j}^{\left [ s\right ] } -\ln\left ( v_{j}^{\left [ s\right ] } \right ) + \sum_{r=1}^{r_{n_{j}}^{\left ( j\right ) } -1}\hat{f}_{r}^{\left [ s\right ] } \xi_{r}\right ] , \label{cosntr}\\ y_{\mu}^{\left [ s\right ] } & : = -\mu^{\lbrack s-1]}\sum_{r=1}^{n}\ln\left ( \hat{f}_{r}^{\left [ s\right ] } \right ) , \label{barrier1}\\ y_{\tilde{\mu}}^{\left [ s\right ] } & : = -\tilde{\mu}^{\left [ s-1\right ] } \sum_{j=1}^{m}\left ( \ln\left ( \rho_{j}^{\left [ s\right ] } -\rho_{j}^{b}\right ) + \ln\left ( \rho_{j}^{a}-\rho_{j}^{\left [ s\right ] } \right ) \right ) , \label{barrier2}\ ] ] we have @xmath81 } & = \sum_{j=1}^{m}\lambda_{j}^{\left [ s\right ] } \left [ \rho_{j}^{\left [ s-1\right ] } + \sigma_{j}^{\left [ s\right ] } -\ln\left ( v_{j}^{\left [ s-1\right ] } \right ) + \sum _ { r=1}^{r_{n_{j}}^{\left ( j\right ) } -1}\left ( f_{r}^{\left [ s-1\right ] } + \delta_{r}^{\left [ s\right ] } \right ) \xi_{r}\right . \\ & \left . -\frac{1}{v_{j}^{\left [ s-1\right ] } } \sum\limits_{i=1}^{n_{j}-1}\sum_{r = r_{i}^{\left ( j\right ) } } ^{r_{n_{j}}^{\left ( j\right ) } -1}\alpha_{ij}\exp\left ( \sum_{z = r_{i}^{\left ( j\right ) } } ^{r_{n_{j}}^{\left ( j\right ) } -1}f_{z}^{\left [ s-1\right ] } \xi_{z}\right ) \delta_{r}^{\left [ s\right ] } \xi_{r}+o\left ( \delta_{r}^{\left [ s\right ] } { } ^{2}\right ) \right ] , \\ y_{\mu}^{\left [ s\right ] } & = -\mu^{\left [ s-1\right ] } \sum_{r=1}^{n}\left ( \frac{\delta_{r}^{\left [ s\right ] } } { f_{r}^{\left [ s-1\right ] } } -\frac{1}{2}\left ( \frac{\delta_{r}^{\left [ s\right ] } } { f_{r}^{\left [ s-1\right ] } } \right ) ^{2}+o\left ( \delta_{r}^{\left [ s\right ] } { } ^{3}\right ) \right ) , \\ y_{\tilde{\mu}}^{\left [ s\right ] } & = -\tilde{\mu}^{\left [ s-1\right ] } \sum_{j=1}^{m}\left ( \ln\left ( \rho_{j}^{\left [ s-1\right ] } -\rho_{j}^{b}\right ) + \ln\left ( \rho_{j}^{a}-\rho_{j}^{\left [ s-1\right ] } \right ) \right ) \\ & + \tilde{\mu}^{\left [ s-1\right ] } \sum_{j=1}^{m}\left ( \frac{1}{\rho _ { j}^{a}-\rho_{j}^{\left [ s-1\right ] } } -\frac{1}{\rho_{j}^{\left [ s-1\right ] } -\rho_{j}^{b}}\right ) \sigma_{j}^{\left [ s\right ] } \\ & + \frac{\tilde{\mu}^{\left [ s-1\right ] } } { 2}\sum_{j=1}^{m}\left [ \frac { 1}{\left ( \rho_{j}^{\left [ s-1\right ] } -\rho_{j}^{a}\right ) ^{2}}+\frac{1}{\left ( \rho_{j}^{\left [ s-1\right ] } -\rho_{j}^{b}\right ) ^{2}}\right ] \sigma_{j}^{\left [ s\right ] 2}+o\left ( \sigma_{j}^{\left [ s\right ] 3}\right ) .\end{aligned}\ ] ] thus defining @xmath82{cc}1 & x\leq r\leq y\\ 0 & \mathrm{otherwise}\end{array } \right . , \qquad\delta_{i , j}:=\left\ { \begin{array } [ c]{cc}1 & i = j\\ 0 & \mathrm{otherwise}\end{array } \right . , \ ] ] from above expansions and eq.([quadratic ] ) we immediately obtain@xmath83 } & = \xi_{r}\chi\left ( r,1,r_{n_{j}}^{\left ( j\right ) } -1\right ) -\frac{1}{v_{j}^{\left [ s-1\right ] } } \sum \limits_{i=1}^{n_{j}-1}\alpha_{ij}\exp\left ( \sum_{z = r_{i}^{\left ( j\right ) } } ^{r_{n_{j}}^{\left ( j\right ) } -1}f_{z}^{\left [ s-1\right ] } \xi _ { z}\right ) \xi_{r}\chi\left ( r , r_{i}^{\left ( j\right ) } , r_{n_{j}}^{\left ( j\right ) } -1\right ) , \\ a_{j}^{\left [ s\right ] } & = \rho_{j}^{\left [ s-1\right ] } -\ln\left ( v_{j}^{\left [ s-1\right ] } \right ) + \sum_{r=1}^{r_{n_{j}}^{\left ( j\right ) } -1}f_{r}^{\left [ s-1\right ] } \xi_{r},\\ q_{r , x}^{\left [ s\right ] } & = 4\varphi\left [ \left ( \frac{\left ( 1-\delta_{r,1}\right ) \left ( 1-\delta_{r,2}\right ) \xi_{r-1}}{\left ( \xi_{r-2}+\xi_{r-1}\right ) ^{2}\xi_{r-1}^{2}}+\frac{\left ( 1-\delta _ { r , n}\right ) \left ( 1-\delta_{r,1}\right ) \xi_{r}}{\xi_{r}^{2}\xi _ { r-1}^{2}}+\frac{\left ( 1-\delta_{r , n-1}\right ) \left ( 1-\delta _ { r , n}\right ) \xi_{r+1}}{\left ( \xi_{r}+\xi_{r+1}\right ) ^{2}\xi_{r}^{2}}\right ) \delta_{r , x}\right . \\ & + \frac{\xi_{r+1}\delta_{r+2,x}}{\left ( \xi_{r}+\xi_{r+1}\right ) ^{2}\xi_{r+1}\xi_{r}}+\frac{\xi_{x+1}\delta_{r-2,x}}{\left ( \xi_{x}+\xi _ { x+1}\right ) ^{2}\xi_{x+1}\xi_{x}}-\frac{\left ( 1-\delta_{r,1}\right ) \xi_{r}\delta_{r+1,x}}{\left ( \xi_{r-1}+\xi_{r}\right ) \xi_{r-1}\xi_{r}^{2}}\\ & \left . -\frac{\left ( 1-\delta_{x,1}\right ) \xi_{x}\delta_{r-1,x}}{\left ( \xi_{x-1}+\xi_{x}\right ) \xi_{x-1}\xi_{x}^{2}}-\frac{\left ( 1-\delta_{r , n}\right ) \xi_{r}\delta_{r-1,x}}{\left ( \xi_{r-1}+\xi _ { r}\right ) \xi_{r}\xi_{r-1}^{2}}-\frac{\left ( 1-\delta_{x , n}\right ) \xi_{x}\delta_{r+1,x}}{\left ( \xi_{x-1}+\xi_{x}\right ) \xi_{x}\xi_{x-1}^{2}}\right ] \\ & + \gamma\left [ \left ( 1-\delta_{r , n}\right ) \frac{\xi_{r}}{\xi_{r}^{2}}\delta_{r , x}+\left ( 1-\delta_{r,1}\right ) \frac{\xi_{r-1}}{\xi_{r-1}^{2}}\delta_{r , x}-\frac{\xi_{r}}{\xi_{r}^{2}}\delta_{r+1,x}-\frac{\xi_{x}}{\xi _ { x}^{2}}\delta_{r-1,x}\right ] + \frac{\mu^{\left [ s-1\right ] } } { f_{r}^{\left [ s-1\right ] 2}}\delta_{r , x},\\ c_{r}^{\left [ s\right ] } & = -\frac{\mu^{\left [ s-1\right ] } } { f_{r}^{\left [ s-1\right ] } } + \gamma\frac{f_{r}^{\left [ s-1\right ] } -f_{r-1}^{\left [ s-1\right ] } } { \xi_{r-1}}\frac{\left ( 1-\delta _ { r,1}\right ) } { \xi_{r-1}}\xi_{r-1}-\gamma\frac{f_{r+1}^{\left [ s-1\right ] } -f_{r}^{\left [ s-1\right ] } } { \xi_{r}}\frac{\left ( 1-\delta_{r , n}\right ) } { \xi_{r}}\xi_{r}\\ & + 4\varphi\left ( \frac{f_{r}^{\left [ s-1\right ] } -f_{r-1}^{\left [ s-1\right ] } } { \xi_{r-1}}+\frac{f_{r-2}^{\left [ s-1\right ] } -f_{r-1}^{\left [ s-1\right ] } } { \xi_{r-2}}\right ) \frac{\left ( 1-\delta _ { r,1}\right ) \left ( 1-\delta_{r,2}\right ) } { \left ( \xi_{r-2}+\xi _ { r-1}\right ) ^{2}\xi_{r-1}}\xi_{r-1}\\ & + 4\varphi\left ( \frac{f_{r+2}^{\left [ s-1\right ] } -f_{r+1}^{\left [ s-1\right ] } } { \xi_{r+1}}+\frac{f_{r}^{\left [ s-1\right ] } -f_{r+1}^{\left [ s-1\right ] } } { \xi_{r}}\right ) \frac{\left ( 1-\delta_{r , n}\right ) \left ( 1-\delta_{r , n-1}\right ) } { \left ( \xi_{r}+\xi_{r+1}\right ) ^{2}\xi_{r}}\xi_{r+1}\\ & -4\varphi\left ( \frac{f_{r+1}^{\left [ s-1\right ] } -f_{r}^{\left [ s-1\right ] } } { \xi_{r}}+\frac{f_{r-1}^{\left [ s-1\right ] } -f_{r}^{\left [ s-1\right ] } } { \xi_{r-1}}\right ) \frac{\left ( 1-\delta_{r,1}\right ) \left ( 1-\delta_{r , n}\right ) } { \left ( \xi_{r-1}+\xi_{r}\right ) \xi_{r}\xi_{r-1}}\xi_{r},\\ m_{j , k}^{\left [ s\right ] } & = \tilde{\mu}^{\left [ s-1\right ] } \left [ \frac{1}{\left ( \rho_{j}^{a}-\rho_{j}^{\left [ s-1\right ] } \right ) ^{2}}+\frac{1}{\left ( \rho_{j}^{b}-\rho_{j}^{\left [ s-1\right ] } \right ) ^{2}}\right ] \delta_{j , k},\\ d_{j}^{\left [ s\right ] } & = \tilde{\mu}^{\left [ s-1\right ] } \left ( \frac{1}{\rho_{j}^{a}-\rho_{j}^{\left [ s-1\right ] } } + \frac{1}{\rho_{j}^{b}-\rho_{j}^{\left [ s-1\right ] } } \right ) .\end{aligned}\ ] ] given positive definite symmetric matrices @xmath84 and @xmath85 and the @xmath54-vector @xmath86 and @xmath56-vector @xmath87 , we want to minimize the objective function @xmath88 subject to the set of constraints@xmath89 thus defining a new objective function @xmath90 for any @xmath91 we define@xmath92{cc}1 & x\neq0\\ 0 & x=0 \end{array } \right . , \ ] ] where e.g. @xmath93 , @xmath94 for diagonal , tridiagonal matrices respectively . when @xmath95 and @xmath96 then @xmath61 can be solved efficiently with dynamic programming . to set up the notation and for those readers not familiar with dynamic programming let us here briefly explain the basics of this well known method @xcite . we start defining @xmath97 where @xmath98 satisfies@xmath99 for @xmath100 we use the inductive hypothesis@xmath101 where @xmath102 @xmath103 and @xmath104 . the final step of this backwards process consists in obtaining @xmath105 and @xmath106 satisfying@xmath107 from eqs.([extrema ] ) and ( [ hypothesis ] ) we obtain @xmath108 and plugging eq.([first ] ) into eq.([fundamental ] ) we obtain@xmath109 hence obtaining@xmath110 where@xmath111{cc}1 & x\geq0,\\ 0 & x<0 . \end{array } \right.\ ] ] now from eq.([lagrange ] ) we obtain @xmath112 or@xmath113 and using eq.([g ] ) we obtain@xmath114 finally from @xmath106 and eq.([sig ] ) we obtain @xmath115 and then using eqs.([first ] ) and ( [ q]-[c ] ) we obtain @xmath116 moving forward in the @xmath117 index . the time necessary to compute all @xmath118 is proportional to@xmath119 to calculate the inverse of @xmath120 , that is a @xmath121 symmetric matrix , we require a computing time proportional to@xmath122 all @xmath123 and @xmath124 require , respectively , computing times proportional to@xmath125 finally the calculation of @xmath126 requires a computing time proportional to @xmath127 in the forward rate calculation typically we have @xmath128 ( in eq.([objective ] ) we have @xmath129 ) and therefore the maximum delay would be given by @xmath130 the newton step obtained from eq.([dynamic_sol ] ) may not fulfill the inequality constraints ( [ ineq_constr ] ) . to satisfy such constraints we search for a @xmath131}$ ] in the interval @xmath132 $ ] such that@xmath133 } + \alpha^{\lbrack s]}\delta_{r}^{\left [ s\right ] } \geq0,\qquad\rho_{j}^{b}\leq\rho_{j}^{\left [ s-1\right ] } + \alpha^{\lbrack s]}\sigma_{j}^{\left [ s\right ] } \leq\rho_{j}^{a}.\label{constr_alpha}\ ] ] in order to do this we first determine the maximum @xmath134}$ ] in the interval @xmath132 $ ] satisfying eq.([constr_alpha ] ) . that is , given the sets of points@xmath135 } \leq0\right\ } , \\ \tilde{a}_{\leq } & : = \left\ { j,\text { } \hat{\rho}_{j}^{\left [ s\right ] } \leq\rho_{j}^{b}\right\ } , \\ \tilde{a}_{\geq } & : = \left\ { j,\text { } \hat{\rho}_{j}^{\left [ s\right ] } \geq\rho_{j}^{a}\right\ } , \end{aligned}\ ] ] we have@xmath136}=\min\left ( \min\limits_{r\in a}\left ( \frac { -f_{r}^{\left [ s-1\right ] } } { \delta_{r}^{\left [ s\right ] } } \right ) , \min\limits_{j\in\tilde{a}_{\leq}}\left ( \frac{\rho_{j}^{b}-\rho _ { j}^{\left [ s-1\right ] } } { \sigma_{j}^{\left [ s\right ] } } \right ) , \min\limits_{j\in\tilde{a}_{\geq}}\left ( \frac{\rho_{j}^{a}-\rho _ { j}^{\left [ s-1\right ] } } { \sigma_{j}^{\left [ s\right ] } } \right ) \right ) , \ ] ] and then we take@xmath137}:=\beta\alpha_{\max}^{[s]},\ ] ] with @xmath138 ( in our implementation we have taken @xmath139 that is a standard election in the optimization literature ) . once we have @xmath140}$ ] we define@xmath141 } & = f_{r}^{\left [ s-1\right ] } + \alpha^{\lbrack s]}\delta_{r}^{\left [ s\right ] } , \nonumber\\ \rho_{j}^{\left [ s\right ] } & = \rho_{j}^{\left [ s-1\right ] } + \alpha^{\lbrack s]}\sigma_{j}^{\left [ s\right ] } , \nonumber\\ \mu^{\left [ s\right ] } & = \max(\psi\left ( \alpha^{\lbrack s]}\right ) \mu^{\left [ s-1\right ] } , \mu_{\min}),\nonumber\\ \tilde{\mu}^{\left [ s\right ] } & = \max(\psi\left ( \alpha^{\lbrack s]}\right ) \tilde{\mu}^{\left [ s-1\right ] } , \tilde{\mu}_{\min } ) , \label{update}\ ] ] where @xmath142 and @xmath143 are positive small values guaranteeing that matrices @xmath91 and @xmath144 are positive definite and @xmath145\rightarrow r^{+}$ ] is a monotonically decreasing function satisfying @xmath146 . in our implementation we have taken @xmath147 of the form@xmath148 with @xmath149 , @xmath150 . the value of @xmath151 controls how fast barriers are reduced . in our implementation we have found that adequate values for @xmath151 range @xmath152 . the value of @xmath153 controls the non - linearity of @xmath147 . we have found that the simple linear response provides good performance albeit convergence time is not significantly affected for @xmath153 in the range @xmath154 in this way we iterate the algorithm @xmath155 times until a given termination criterion is met . defining@xmath156 } & : = \ln\left ( w\left ( f^{[s]}\right ) \right ) -\ln\left ( w\left ( f^{[s-1]}\right ) \right ) , \\ \epsilon^{\left [ s\right ] } & : = \max\limits_{j}\left ( \left\vert \rho _ { j}^{\left [ s\right ] } -\ln\left ( v_{j}^{\left [ s\right ] } \right ) + \sum_{r=1}^{r_{n_{j}}^{\left ( j\right ) } -1}f_{r}^{\left [ s\right ] } \xi_{r}\right\vert \right ) , \end{aligned}\ ] ] we have chosen the following termination criterion@xmath157}<w_{zero}~~\mathrm{or}~~\left ( \left\vert \delta_{\ln w}^{\left [ s\right ] } \right\vert < \delta_{\ln w}^{\max}~~\mathrm{and}~~\left\vert \delta_{\ln w}^{\left [ s-1\right ] } \right\vert < \delta_{\ln w}^{\max}\right ) \right ] \right . \nonumber\\ & \left . \mathrm{and}~~n_{it}>n_{\min}~~\mathrm{and}~~\mu^{\left [ s\right ] } < \mu_{\max}~~\mathrm{and}~~\tilde{\mu}^{\left [ s\right ] } < \tilde{\mu}_{\max}~~\mathrm{and}~~\epsilon^{\left [ s\right ] } < \epsilon_{\max}\right\ } . \label{criterion}\ ] ] in our implementation we have taken@xmath158 } & = 10^{-1},\quad\mu_{\min}=10^{-10},\quad\mu_{\max}=10^{-6},\nonumber\\ \tilde{\mu}^{[0 ] } & = 10^{+1},\quad\tilde{\mu}_{\min}=10^{-10},\quad \tilde{\mu}_{\max}=10^{-6},\nonumber\\ \epsilon_{\max } & = 10^{-8},\quad\delta_{\ln w}^{\max}=10^{-2},\quad w_{zero}=10^{-9},\nonumber\\ n_{\min } & = 5,\quad n_{\max}=60 . \label{values}\ ] ] obviously there is considerable latitude to change the heuristic values assigned to the above parameters . let us finish this appendix making some comments regarding their robustness . @xmath159 is there to guarantee a minimum number of iterations so as to have @xmath160 } $ ] and @xmath161 } $ ] well defined and to avoid premature termination in the improbable case where the other criteria incorrectly suggest convergence . for this purpose is enough to take @xmath162 . @xmath163 serves as a maximum limit to secure termination even if convergence is not achieved and therefore an alarm should be provided whenever @xmath164 . from our experience we observe that is more than enough to take @xmath165 @xmath166 sets our precision to consider a given forward rate curve as a straight line . the order of magnitude of @xmath166 should be taken much lower than the typical order of magnitude of the observed optimal @xmath167 the value of the optimal @xmath42 depends not only on the constraining data but also on @xmath41 and @xmath168 we have adopted the practise of spanning the range @xmath169 year@xmath170 taking @xmath171 year@xmath33 , @xmath36 year@xmath37 and the range @xmath94 year@xmath172 taking @xmath173 year@xmath33 , @xmath174 year@xmath37 . with this convention we have found reasonable to take @xmath175 for the full range @xmath176 @xmath177 sets the maximum variation of @xmath178 between newton steps that is accepted before termination . note that in ( [ criterion ] ) to have ( @xmath179 } \right\vert < \delta_{\ln w}^{\max}~~\mathrm{and}~~\left\vert \delta_{\ln w}^{\left [ s-1\right ] } \right\vert < \delta_{\ln w}^{\max}$ ] ) is a necessary but not sufficient condition for termination . therefore sending @xmath180 has no major impact and is equivalent to rely only on the error @xmath181 } $ ] and the barrier coefficients as indicators of convergence . on the contrary excessively reducing @xmath182 can generate unnecessary iterations . we have tested that for values satisfying @xmath183 we do not have any drastic increase in convergence time . @xmath184 controls the error in the constraints and is the most important parameter in ( [ criterion ] ) . a too large value of @xmath185 reduces the accuracy of the result and a too small value can give rise to unnecessary iterations . we have observed acceptable results for @xmath184 in the range @xmath186 @xmath187 and @xmath188 control the maximum allowed values for the logarithmic barriers implementing the positivity and spread constraints respectively . large values for these parameters can make log barriers to have a residual influence in the feasible region . acceptable values of these maximum weights range in @xmath189 and @xmath190 as explained above keeping parameters @xmath142 and @xmath143 positive guarantees that matrices @xmath91 and @xmath144 are positive definite which in turn is sufficient to guarantee the existence of an optimal solution in each iteration . hence @xmath142 and @xmath143 should be chosen as small as possible without interfering with the numerical stability of the algorithm . using double precision in our program we have found the values given in ( [ criterion ] ) as a good compromise . finally @xmath191}$ ] and @xmath192}$ ] are the initial values for the log barriers coefficients . taking very large values for @xmath191}$ ] and @xmath192}$ ] increases the convergence time because we need more time to reduce the barriers . taking too small values for @xmath191}$ ] and @xmath192}$ ] also increases the convergence time because time is wasted exploring unfeasible solutions . moreover , we have observed that convergence and stability is improved if the contributions to the objective of the two barrier terms are kept balanced . this is achieved setting @xmath192}\simeq\frac{n}{2m}\mu^{\lbrack0]}$ ] ( see eqs.([barrier1 ] ) and ( [ barrier2 ] ) ) and using the same updating factor @xmath193}\right ) $ ] for @xmath194}$ ] and @xmath195 } $ ] ( see eq.([update ] ) ) . keeping @xmath192}\simeq\frac { n}{2m}\mu^{\lbrack0]}$ ] we have found that the time of convergence is stable for @xmath191}$ ] in the range @xmath196}\lesssim10^{+1}.$ ] in this work we have used the following data tables and conventions . [ c]|l||l|l|l|l|l|l|l|l|l|l|l|@xmath197bond ( so ) & 1033 & 1042 & 1035 & 1044 & 1038 & 1037 & 1040 & 1043 & 1034 & 1045 & 1041 + & 4.86 & 4.92 & 5.06 & 5.15 & 5.26 & 5.27 & 5.355 & 5.4 & 5.395 & 5.46 & 5.655 + & 4.905 & 4.965 & 5.11 & 5.2 & 5.05 & 5.325 & 5.4 & 5.455 & 5.23 & 5.51 & 5.7 + & 4.885 & 4.945 & 5.095 & 5.185 & 4.93 & 5.295 & 5.395 & 5.435 & 5.24 & 5.495 & 5.68 + & 4.865 & 4.92 & 5.06 & 5.15 & 4.93 & 5.275 & 5.355 & 5.41 & 5.415 & 5.465 & 5.65 + & 4.835 & 4.885 & 5.025 & 5.125 & 4.93 & 5.25 & 5.355 & 5.405 & 5.405 & 5.465 & 5.66 + & 4.84 & 4.89 & 5.045 & 5.145 & 4.93 & 5.27 & 5.38 & 5.43 & 5.43 & 5.495 & 5.69 + & 4.825 & 4.87 & 5.035 & 5.13 & 4.93 & 5.25 & 5.36 & 5.415 & 5.41 & 5.475 & 5.66 + & 4.805 & 4.85 & 5.015 & 5.11 & 4.93 & 5.23 & 5.34 & 5.395 & 5.395 & 5.455 & 5.64 + & 4.8 & 4.86 & 5.005 & 5.1 & 4.93 & 5.22 & 5.335 & 5.38 & 5.395 & 5.445 & 5.64 + & 4.77 & 4.83 & 4.97 & 5.06 & 4.93 & 5.165 & 5.27 & 5.34 & 5.34 & 5.39 & 5.585 + the price of the bond @xmath1 is calculated using the formula@xmath198 where @xmath199 is the nominal amount ( sek 40 millions for all bonds in table [ table1 ] ) , @xmath200 is the coupon rate ( given in table [ table2 ] ) , @xmath201 is the total number of remaining coupons ( each paid at time @xmath202 ) , @xmath203 is the quoted rate given in table [ table1 ] and @xmath204 is a time difference between @xmath202 and the settlement day . this time difference is calculated according to the isma 30e/360 convention defined as follows : given two dates @xmath205 and @xmath206 , their isma 30e/360 time difference @xmath207 is given by@xmath208 [ c]|c||c|c| & @xmath209 & @xmath210 + & 05/05/2003 & 10.25 + & 15/01/2004 & 5 + & 09/02/2005 & 6 + & 20/04/2006 & 3.5 + & 25/10/2006 & 6.5 + & 15/08/2007 & 8 + & 05/05/2008 & 6.5 + & 28/01/2009 & 5 + & 20/04/2009 & 9 + & 15/03/2011 & 5.25 + & 05/05/2014 & 6.75 + kian guan lim , qin xiao , and jimmy ang , _ estimating forward rate curve in pricing interest rate derivatives _ , derivatives use , trading & regulation , an international journal of the futures and options association uk , vol . 6 no.4 , pp . 299 - 305 , 2001 . kwon , oh kang , _ a general framework for the construction and the smoothing of forward rate curves_. qfrg , university of technology , sydney . http://www.business.uts.edu.au/finance/qfr/cfrg_papers.html , march 2002 .

we propose a phenomenological model for the polygonal hydraulic jumps discovered by ellegaard _ et al . _ , based on the known flow structure for the type ii hydraulic jumps with a `` roller '' ( separation eddy ) near the free surface in the jump region . the model consists of mass conservation and radial force balance between hydrostatic pressure and viscous stresses on the roller surface . in addition , we consider the azimuthal force balance , primarily between pressure and viscosity , but also including non - hydrostatic pressure contributions from surface tension in light of recent observations by bush _ et al_. the model can be analyzed by linearization around the circular state , resulting in a parameter relationship for nearly circular polygonal states . a truncated , but fully nonlinear version of the model can be solved analytically . this simpler model gives rise to polygonal shapes that are very similar to those observed in experiments , even though surface tension is neglected , and the condition for the existence of a polygon with n corners depends only on a single dimensionless number . finally , we include time - dependent terms in the model and study linear stability of the circular state . instability occurs for sufficiently small bond number and the most unstable wave length is expected to be roughly proportional to the width of the roller as in the rayleigh - plateau instability . _ _

1111.4599

the term hydraulic jump " refers to the sudden jump in fluid height , as for example observed in the outward spreading water layer in a kitchen sink , resulting from the impact of the water from the tap with the horizontal bottom of the sink @xcite . similar phenomena are seen in rivers with large tidal variation at the outlet and are known as river bores " . river bores move up the rivers , whereas hydraulic jumps are stationary , either due to spatial inhomogeneities or to the geometric configuration of the flow . in the kitchen sink , the jump occurs on a more or less circular locus . close to the point of impact the water level is very thin , but at a certain radius , @xmath1 , the level increases abruptly forming a circular jump . the water flow and thus the jump shape in a kitchen sink fluctuate , but by building a more symmetric experimental setup and/or using a more viscous fluid , one may obtain a completely stationary and axisymmetric flow , where the jump occurs on a surprisingly well - defined circle ( fig . [ fig : circularjump ] ) . it has been shown @xcite that circular hydraulic jumps can undergo a sequence of structural changes seen by varying the fluid height @xmath2 downstream of the jump . this can be achieved by inserting an adjustable circular weir at the rim of the circular impact plate . when increasing @xmath2 , the jump becomes steeper until , at a critical value of @xmath2 , the jump becomes unstable and loses its balance . like a breaking wave , it creates a new stationary state with a wider jump region and a new flow structure in which the surface flow is reversed due to a separation vortex ( referred to in the following as the `` roller '' ) . following @xcite , his new flow structure is called a _ type ii jump _ to distinguish it from the `` ordinary '' _ type i jump_. the most remarkable observation about the type ii state is that the jump typically loses its azimuthal symmetry and attains the shape of a regular , though not necessarily straight - edged , polygon ( fig . [ fig : polyjump ] ) . in this paper , we present a phenomenological model for the type ii jump , which we believe to include the basic mechanisms of the flow rendering the circular state unstable and giving the polygonal jump its peculiar shape . this allows us to obtain the qualitative dependence of the jump shape on the control parameters available , primarily the outer fluid height @xmath2 , and the properties of the fluid , i.e. , viscosity and surface tension . for experimental data , we refer to earlier work . our own experiments are performed with the same setup as in @xcite and is only used qualitatively as a visual guide ( see also supplementary movies @xcite ) . for the type i state , averaging theory with a variable profile gives at least qualitatively the right flow structure including the separation region on the bottom , outside the jump @xcite . for the type ii state with separation at the surface , no such theory exists , and we therefore from the outset assume a flow structure derived from experiments . our model extends earlier models proposed in @xcite , based on the competition between gravitational and viscous effects . in @xcite a radial force balance for the roller was used together with a line tension " to derive polygonal shapes . in @xcite it was shown that such a line tension is not necessary and that a more satisfactory model is obtained by taking into account the azimuthal flow . this is the approach used in the present paper . it was recently pointed out @xcite , however , that the instability forming the polygonal jumps seems to be driven , at least in part , by surface tension ( see also supplementary materials @xcite ) . thus , surface tension is included in our model and indeed the instability forming the polygonal jumps seems to be closely related to the rayleigh - plateau instability for the liquid cylinder " formed by the roller . in this paper , we treat both linear and nonlinear properties of our model . for the linear properties , we can with some confidence include the effects of surface tension , but for the fully developed polygonal states , this is very difficult because of the lack of precise data on the height profiles . even without surface tension , our model does have polygonal states , and we compute their shapes assuming that they will be only slightly changed by surface tension effects . the layout of the paper is as follows : we first derive the model and point out the similarities and differences to earlier models . we then solve a nonlinear model with zero surface tension analytically and show that it has many features that are expected from experiments . finally , we compute the temporal stability of the circular state while including surface tension , and show that it has properties close to the rayleigh - plateau instability . in the experiment @xcite , the fluid falls from a nozzle mounted at some height , at constant volumetric flow rate , @xmath3 . we employ cylindrical coordinates , @xmath4 , centered at the point , where the central axis of the cylindrical liquid jet impinges the horizontal plate . the height of the fluid surface above the plate is parametrized by @xmath5 . as the fluid leaves the point of impact , @xmath6 is small and the fluid spreads in a very thin layer ( around 1 mm in our experiments ) with supercritical flow speed ( i.e. a speed larger than the speed of surface waves ) . at a short distance from the point of impact , the boundary layer in the film has become fully developed @xcite and the fluid height is almost constant , i.e. , @xmath7 , as seen in fig . [ fig : heightprofile ] . due to the supercriticality , this inner flow is not affected by the transition from type i to type ii jump , and the flow remains axially symmetric . the surface height of the type ii jump abruptly increases at the jump radius @xmath8 due to the presence of the roller on top of the thin fluid layer . the surface height increases for @xmath9 until a certain radius @xmath10 where it settles at a roughly constant level , @xmath11 . the height difference across the jump is @xmath12 , and since @xmath13 , we use the approximation @xmath14 throughout the paper . we further introduce the _ aspect ratio _ @xmath15 which is typically small , say of the order of 0.2 . in fig . [ fig : rollerflowvisual ] we show a visualization of the flow in a triangular jump created with ethylene glycol . it can be clearly seen that the roller fills the jump region from @xmath8 to @xmath10 , where @xmath8 now has lost its azimuthal symmetry , whereas the outer edge @xmath10 remains circular . we therefore define the normalized local roller width as one of the basic variables of our model @xmath16 which varies with @xmath17 to give the jump its characteristic polygon shapes . the dye dripped into the roller reveals an extremely slow exchange of the fluid in the roller ( see also flow visualization in the supplementary material @xcite ) . in fact the dye can be visible in the roller for several minutes if it is left undisturbed . from this we conclude that the main part of the fluid in the thin layer going under the roller leaves the jump region without ever entering the roller . downstream of the jump region , the flow speed is subcritical ( less than the speed of surface waves ) , and the flow is again purely radial , but it now has an azimuthal dependence : at the corners of the polygon , strong radial jets are observed . within the jump region , an azimuthal transport must therefore exist in the jump region from the sides to the corners of the polygon . indeed a small , spiraling azimuthal flow is observed in the roller from its belly " out to the `` neck '' . from fig . [ fig : rollerflowvisual ] the jets seem to consist primarily of liquid going right through the jump , but they also carry with them the dyed fluid transported to the corners inside the roller . the jets at the corners of an @xmath18-gon are actually so strong that they apparently break up the roller into @xmath18 distinct rollers , touching and interacting in a rather complicated manner in the corners @xcite . this effect is not included in our model , but since it occurs only in narrow regions near the corners , we believe that our model can still give useful results . and the outer height @xmath2 are varied . the height of the nozzle about the surface is set to 2 cm . the number of corners is found to be less sensitive to the nozzle height @xcite . polygonal states with the number of corners @xmath19 are found in this experiment . [ fig : phasediagram ] , title="fig:",scaledwidth=45.0% ] + a phase diagram has been presented in @xcite in terms of three parameters : the flow rate @xmath3 , the outer fluid height @xmath20 and the height of the nozzle . the dependence on the first two parameters is reproduced in fig . [ fig : phasediagram ] . when increasing @xmath2 , the transition to the type ii structure leads to a polygon with many corners ( up to @xmath21 have been observed ) . increasing @xmath2 further , the number of corners decreases and so does the average jump radius , @xmath22 , until at last the jump closes on the vertical jet from the nozzle . throughout this regime the outer radius @xmath10 is surprisingly constant , while the inner border defining the polygon changes . in fact , this polygon shape shows considerable hysteresis , and a whole series of bordering polygon shapes can be stable at the same flow rates . an extra corner can be created by pulling out a polygon side with a needle , and similarly , a corner can be removed by pushing a corner inward . for sufficiently small disturbances the jump simply deforms - just as if a rubber band was pulled - and bounces back to its original shape after being released . in our phenomenological model , we focus on the roller which we consider as a separate object interacting with the fluid beneath and behind it . we look at a control volume defined by the section of the roller inside an infinitesimal wedge @xmath23 $ ] , as illustrated in fig . [ fig : jumpanatomy ] , and label the areas of the bottom and the rear faces of the control volume as @xmath24 and @xmath25 , respectively . the lateral areas are denoted as @xmath26 . let further @xmath27 be the stress tensor , let @xmath28 and @xmath29 be the volume and the boundary , respectively , of the roller slice , and let @xmath30 be the unit normal of @xmath29 . then the momentum equations are @xmath31 initially , we shall consider stationary states and thus the first term on the left hand side is zero . also , the last term on the right vanishes , since the only body force is gravity , which is taken into account by assuming that the pressure @xmath32 is hydrostatic . the inner flow in fig . [ fig : jumpanatomy ] is purely radial and axially symmetric . when it reaches the roller , most of it passes underneath it . when it has passed the roller the flow is still radial , but has acquired an angular dependence . in between , in the region of the roller , the flow thus acquires a small tangential component , which we assume to run inside the roller . our first task is then to write down the continuity equation for the flow . then we shall evaluate the force " terms in ( [ eq : momentumbalance ] ) for the radial and tangential directions , respectively . in the following , we use cylindrical coordinates and denote the respective vector components of dependent variables by indices @xmath33 and @xmath34 . [ fig : massconserv ] ( a ) shows the top view of the entire roller , and fig . [ fig : massconserv ] ( b ) displays an angular section @xmath35 of the roller , together with the mass flux entering and leaving the control volume ( corresponding to the gray shaded volume in fig . [ fig : jumpanatomy ] ) . since the flow before the jump is purely radial and independent of @xmath17 , as evidenced in fig . [ fig : heightprofile ] , the radial flow into the volume must be @xmath36 . the fluid going out of the volume in the radial direction is denoted by @xmath37 . the fluid going into the roller at the cross sectional wall at @xmath17 is @xmath38 , and the corresponding flow out of the roller at @xmath39 is @xmath40 . mass conservation in the region then demands @xmath41 we define the normalized radial and azimuthal flux per azimuthal angle @xmath42 and @xmath43 , respectively . the continuity equation is now written as @xmath44 by integrating from @xmath45 to @xmath46 and using the periodic condition for @xmath47 , we obtain a constraint for the total flux out of the jump region . @xmath48 we now inspect fig . [ fig : vprofile ] , the side view of fig . [ fig : jumpanatomy ] , and construct a force balance equation in the radial direction for the roller which we view as a stationary body of liquid . it experiences a force directed radially inward due to the difference in fluid height at its inner and outer rim . the shear force also acts on the bottom of the roller , caused by the velocity gradient of the radially outward flow along its bottom boundary which we estimate at @xmath49 . we assume that these forces balance , and that surface tension plays a secondary role in the radial direction . under this assumption , we have @xmath50 where @xmath51 and @xmath52 are , respectively , the hydrostatic pressure force and the radial component of the viscous friction force per unit azimuthal angle on the roller slice in fig . [ fig : jumpanatomy ] . . the height inside and outside of the jump are @xmath53 and @xmath2 , respectively . we treat the roller eddy ( shaded region ) as a separate body of fluid from the main stream , and assume the force balance of the hydrostatic force @xmath51 and the shear force @xmath52 in the radial direction ( forces indicated by bold red arrows ) . [ fig : vprofile],title="fig:",scaledwidth=40.0% ] + the pressure force , @xmath51 , is obtained by integrating the pressure over the area of its outer rim assumed to be at @xmath54 and extending from @xmath49 to @xmath2 ( see fig . [ fig : vprofile ] ) . further , assuming the pressure to be hydrostatic , and omitting the ambient pressure which cancels , the force becomes @xmath55 using ( [ eqn : deltah ] ) where @xmath56 is the fluid density and @xmath57 is the gravitational acceleration . if the fluid is newtonian with the dynamic viscosity @xmath58 , the outward force on the roller @xmath59 shown in fig . [ fig : vprofile ] is obtained by integrating the shear @xmath60 over the bottom area of the roller : @xmath61 where @xmath62 is the radial velocity component . to estimate this integral , we assume that the thin radial flow continues forward under the roller in a height of the order @xmath63 , an appropriate characteristic height near the roller . the velocity in this region is related to the radial flux @xmath64 by @xmath65 where @xmath66 and @xmath67 denote some characteristic radius and radial velocity , respectively . we also assume that the shear can be expressed as @xmath68 with a numerical factor @xmath69 of order unity . then , we have @xmath70 for simplicity , we choose @xmath71 and @xmath72 . then the shear force becomes @xmath73 substituting the estimated forces ( [ radialhydrostaticforce ] ) and ( [ radialshearforce ] ) into eq . ( [ radialforcebalance ] ) , we obtain the radial force balance equation @xmath74 with the nondimensional parameter @xmath75 where @xmath76 is the kinematic viscosity @xmath77 . this equation was first derived in @xcite . it shows that the radial flux increases at the corners where @xmath78 decreases . for a very thin corner , the flux would increase to infinity , and although our model remains meaningful in this limit , we do not expect it to represent the physics correctly , since the detailed structure of the corner flow could play an important role . lacking such a more detailed model , we shall thus keep in mind that polygons with very thin or vanishing rollers in the corners are probably somewhat idealized . in the circular state the normalized radial flux is @xmath79 so the dimensionless roller has width @xmath80 where @xmath81 combining ( [ eq : radly_balance ] ) with the mass conservation ( [ eq : masscons ] ) , we obtain @xmath82 whereas prime denotes the derivative with respect to @xmath17 . the constraint ( [ eq : totalmasscons ] ) , along with eq . ( [ eq : radly_balance ] ) , becomes @xmath83 we now discuss the azimuthal transport in the control volume shown in fig . [ fig : massconserv ] . since the flows in the azimuthal direction are slow , we neglect the kinetic term arising from the second term on the left hand side of ( [ eq : momentumbalance ] ) , and consider hydrostatic pressure , viscous , and surface tension terms . the second term on the left hand side of ( [ eq : momentumbalance ] ) is the momentum flux through the boundary of the roller slice . we consider the flux for the vertical sides of the cross sections at @xmath17 and @xmath39 in fig . [ fig : jumpanatomy ] . using the approximation @xmath84 , the areas of these sides are given by @xmath85 we consider the average azimuthal flow velocity over such a surface @xmath86 . then , the kinetic term is estimated as @xmath87_{\theta}^{\theta + d\theta } \approx \rho \overline{u_\theta}^2 r h_o \delta(\theta ) d\theta\\end{aligned}\ ] ] since the mean velocities in the roller always remain small , except , perhaps inside a very thin corner ( where our model breaks down anyway , since the roller breaks ) we shall neglect this term in the following . for the azimuthal component of the first term on the right hand side of eq . ( [ eq : momentumbalance ] ) , we assume the pressure in the roller to be hydrostatic . then , a net azimuthal pressure force on the roller slice arises due to the difference in area of the two vertical sides at @xmath17 and @xmath39 shown in fig . [ fig : jumpanatomy ] . the force on one side is estimated as in eq . ( [ hyd ] ) . for this , we assume a simple surface profile which is linear within the cross - section @xmath88 const . , i.e. , @xmath89 where @xmath90 at @xmath91 , and @xmath92 at @xmath93 . with @xmath13 we get @xmath94\ ] ] with the aspect ratio defined in ( [ alpha ] ) . we then get @xmath95 considering the directions of the forces acting on the sides , the force difference between the two sides is @xmath96 where prime again denotes the derivative with respect to @xmath17 . due to the minus sign , the force acts in the direction from a belly to a corner of the roller . the contributions to the azimuthal component of the second term on the right hand side of eq . ( [ eq : momentumbalance ] ) are the shear forces between the azimuthal flow inside the roller and the radial flow underneath and behind it . just as for the radial force balance we must make some estimate of the velocity gradients involved . the outer rim of the roller has area @xmath97 on which we estimate @xmath98 since @xmath99 for @xmath100 . here , @xmath101 is a positive numerical factors of order unity , and @xmath102 is the average azimuthal flow velocity over @xmath103 . the areas of the sides are given by ( [ at ] ) . thus , @xmath104 likewise , the area on the bottom of the roller is @xmath105 on which @xmath106 since @xmath107 on @xmath108 . here , we have used the height @xmath53 rather than @xmath12 to estimate the shear between the roller and the main flow , and @xmath109 is another positive numerical factor of order unity . note that the radial velocity gradient at the rim is measured from the interior to the exterior of the roller , whereas the velocity gradient on the bottom is in the opposite direction . thus , we reverse the sign of the latter in order to add both contributions for an estimate of the shear force on the control volume : @xmath110d\theta \nonumber \\ & = & - \frac{\mu q_\theta}{r h_o \delta } \left [ \frac{c_2 h_o}{\delta } + \frac{c_3 r \delta ( r+r_j)}{2 h_i}\right]d\theta \nonumber \\ & = & - \mu q \xi_\theta \left[\frac{c_2}{r\delta^2 } + \frac{c_3 r(2-\delta)}{2 h_i h_o}\right]d\theta . \nonumber\end{aligned}\ ] ] here , the relation ( [ averageazimuthalvelocity ] ) is again used . surface tension acts by changing the pressure inside the roller depending on the local curvature . this does not to a first approximation change the radial force balance , so we shall concentrate on the effects in the azimuthal direction . a somewhat similar approach in a different context can be found in @xcite . to model this , we shall include the laplace pressure difference across the surface @xmath111 : @xmath112 where @xmath113 and @xmath114 are the two principal curvatures ; we take their signs to be positive if the center of curvature is on the `` inside '' direction from the surface ( i.e. located under the surface ) . taking the atmospheric pressure to be zero everywhere , the relation ( [ e1 ] ) describes the fluid pressure just beneath the surface . thus , the total pressure inside the fluid becomes @xmath115 to use these expressions we need information about @xmath5 , in the circular state and in the polygonal state and in the family of states between them . of course , we do not have this information , so , in keeping with our simple model , which expresses the geometry of the jump entirely in terms of the local width of the roller , we shall replace the two radii of curvature with averages over the cross - section of the surface . in what follows , we consider small perturbations , i.e. nearly circular jumps . one principal radius of curvature @xmath113 quantifies the curvature roughly in the @xmath116-plane , i.e. , the cross section of the jump . this is the dangerous " one , which may lead to instability . the other one , @xmath114 , is defined in a plane orthogonal to @xmath113 , also including the surface normal . since surface inclinations are small , @xmath114 can be approximated by the radius of curvature of the jump shape seen from above ( cf . fig . [ fig : massconserv ] ) . for slow variations of the plane curve @xmath117 the radius of curvature we get @xmath118 where primes denote differentiation with respect to @xmath17 . to estimate @xmath113 we have to know how the shape distorts when a corner is created . since it is observed that addition of surfactant destroys the polygons ( see also supplemental material @xcite ) , it seems likely that surface tension can make the circular states unstable through a rayleigh - plateau - like instability ( however , it can not explain the observation of circular type - ii states that are sometimes stable . ) . this means that the curvature should increase when @xmath78 is decreased as it does near a corner . this is indeed observed in experiments as seen in fig . [ fig : heightprofile ] , where the height profile ( on a radial path ) of a polygon - state has been measured with a conducting needle @xcite , both through an edge and a corner . it is clear that the curvature is larger ( i.e. the radius of curvature is smaller ) through the corner . thus , we assume @xmath119 where prime denotes differentiation with respect to @xmath78 . when the height @xmath12 and the width @xmath120 are similar , we expect that @xmath113 would have a value close to these , like for a cylindrical surface , but in contrast to the classical cylinder case of plateau and rayleigh , the curvature can not keep increasing when the roller shrinks since the external height is fixed . thus , we expect the curvature to saturate when the width of the roller becomes much smaller than @xmath12 . for future use , we shall use the symbol @xmath121 for the normalized cross - sectional curvature : @xmath122 . the laplace pressure term in ( [ totalpressure ] ) now becomes @xmath123 to get the force , we now have to integrate @xmath124 over the bounding surfaces , where @xmath125 is the local outward normal , and project the result along @xmath17 . this will give the force from the roller on the surroundings . to get the force _ on _ the roller we have to put in a minus - sign . the rear surface does not contribute anything since it is parallel to @xmath126 . treating @xmath127 as constant in the cross - section @xmath128 const . yields @xmath129 where the areas of the sides @xmath103 are given by ( [ at ] ) . to estimate the free surface contribution we need the area of the free surface and the @xmath17-component of the outward unit normal vector . as above , we use the linear height profile ( [ h ] ) , where @xmath130 and @xmath131 where , for brevity , we use a compact notation for partial derivatives throughout this section . the area of the free surface is @xmath132 and the unit normal vector is @xmath133 with a suitable normalization factor @xmath18 . assuming that @xmath134 so that @xmath135 and @xmath136 , we have @xmath137 the @xmath17-component of the unit normal becomes @xmath138 and we obtain ( based on the assumption @xmath139 ) @xmath140 using the average value for @xmath141 , @xmath142 we obtain @xmath143 the force contribution from the free surface is this product times @xmath127 . thus , @xmath144 the total surface tension force in the azimuthal direction can now be computed from ( [ forcesigmavertical ] ) and ( [ forcesigmafree ] ) . note that we need the forces _ on _ the roller , so the direction of the force is _ into _ the roller , i.e. , in the direction @xmath145 at @xmath17 and in the direction @xmath146 at @xmath39 . thus , @xmath147 .\end{aligned}\ ] ] it is seen that if @xmath148 ( like in the case of a cylinder squeezed uniformly ) , the first term gives a contribution similar to the hydrostatic one ( [ g ] ) , but with _ opposite _ sign . this seems to be the case for the polygonal hydraulic jump shown in fig . [ fig : heightprofile ] , where the curvature in the corner is larger than at the edge . for later use , we shall introduce the shape parameter @xmath149 for a cylindrical roller with @xmath150 we would then get @xmath151 . the ratio of the prefactor of the gravitational term ( [ g ] ) to the surface tension term ( without @xmath78 ) is the dimensionless number @xmath152 where @xmath153^{1/2}$ ] is the capillary length and where @xmath154 is the bond number based on @xmath2 . using these estimates , we write down the azimuthal component of eq . ( [ eq : momentumbalance ] ) : @xmath155 which can be written in dimensionless form as @xmath156 \xi_\theta \\ \label{s1 } & + & 3 \alpha { \rm bo^{-2 } } \delta \left({\frac{\rho_1'(\delta)}{\rho_1 ^ 2 } } \delta'(\theta)+ \delta'''(\theta ) \right),\end{aligned}\ ] ] where we have defined the dimensionless numbers @xmath157 the two coupled equations ( [ eq : staticeq1 ] ) and ( [ s2 ] ) close our model for @xmath78 and @xmath47 . an overview of the numerous symbols used in our model is provided in table [ nomenclature ] in sec . [ appendixb ] . we investigate the existence of nearly circular polygons , i.e. for small oscillations around the circular solution @xmath158 and @xmath159 . inserting @xmath160 and @xmath161 into ( [ s1 ] ) and ( [ s2 ] ) , and by using @xmath162 in ( [ b ] ) , we find the first order perturbation equations to be @xmath163 \\ & + & 3 \alpha { \rm bo^{-2 } } \delta_0 \;\delta_1 ' '' , \\\label{eq : linanalysis1b } \xi_1 ' & = & { \frac{1}{2 \pi \delta_0}}\delta_1.\end{aligned}\ ] ] differentiating ( [ eq : linanalysis1a ] ) with respect to @xmath17 and substituting @xmath164 from ( [ eq : linanalysis1b ] ) , we obtain @xmath165\delta_1\\ & + & \delta_0 3 \alpha { \rm bo^{-2 } } \ ; \delta_1''''.\end{aligned}\ ] ] for perturbations with @xmath166 , we obtain the characteristic equation : @xmath167=0.\end{aligned}\ ] ] this equation has the solution @xmath168 where we define @xmath169.\ ] ] when the wave number @xmath170 is an integer , the solution is periodic on @xmath171 $ ] and corresponds to a polygon with @xmath170 corners . in the present case , @xmath172 needs to be positive for the solution to make sense . in the limit of negligible surface tension where @xmath173 ( i.e. @xmath174 ) , we simply get @xmath175 } .\ ] ] to solve the static nonlinear model , eqs . ( [ s1])-([s2 ] ) , we would need to know @xmath176 . at present we do not know the shapes well enough to estimate this , and we shall restrict our attention to the case where surface tension is absent ; despite this simplification , we shall see that we still obtain meaningful results which allow us to explain the structure of the phase diagram . it is instructive to start with a further simplification . by letting @xmath177 , we neglect the term due to the shear against the bottom of the roller . by ignoring this regular perturbation term , the model can be solved analytically , which enables further analysis , as discussed in sec . [ sec : pi3model ] , where we investigate the effect of the @xmath178 term . we obtain @xmath179 where we have introduced the rescaled variables @xmath180 and the non - dimensional parameter @xmath181 we now solve ( [ eq : goveqncase1 ] ) for @xmath182 , differentiate it once with respect to @xmath17 , and eliminate @xmath183 in ( [ eq : goveqncase2 ] ) , i.e. , to obtain the single ( second - order ) equation @xmath184 by substituting @xmath185 , this may be formulated as an equation of motion for a `` mass '' @xmath186 in the conservative force field , @xmath187 where the potential function is given by @xmath188 we introduce the rescaled angle @xmath189 and integrate once in @xmath190 to obtain @xmath191 where @xmath192 is a constant of integration corresponding to the `` energy '' of the mass @xmath186 . solving the equation for @xmath193 and transforming back to @xmath194 , we obtain @xmath195 this equation has the implicit solution @xmath196 with ( [ eq : scaling1 ] ) we recover the expression in terms of the original scaling @xmath197 and similarly , by substitution of ( [ eq : xprime ] ) into ( [ eq : goveqncase1 ] ) , the corresponding azimuthal flux is given by @xmath198 we now consider the parameter space for existence of polygons . such solutions may be considered bound " states of the potential @xmath28 . as shown in fig . [ fig : phdiagcubic ] , such a state exists if and only if @xmath199 . when @xmath192 is within the range , there are two roots @xmath200 and @xmath201 for the equation @xmath202 , satisfying @xmath203 . they serve as turning points for a trajectory , and describe the minimum ( corner ) and the maximum ( belly ) of the jump width , respectively . without loss of generality we may choose @xmath204 when @xmath205 , and then a trajectory can be computed by integrating ( [ eq : xprime ] ) with the plus sign from @xmath206 to @xmath207 , where @xmath208 denotes the period of oscillation in terms of @xmath190 , and extending that part of the solution using symmetry . for a polygon with @xmath18 corners a trajectory must oscillate in the potential @xmath18 times , and this must result in an increase by @xmath46 in terms of @xmath209 . thus , a solution for @xmath18-polygon must satisfy a commensurability condition @xmath210 with the normalized half period @xmath211 as a function of @xmath212 and @xmath213 , respectively . a periodic solution exists only when @xmath199 . the solution corresponds to a polygon with @xmath18 corners if the period is @xmath214 in terms of @xmath17 . [ fig : phdiagcubic],scaledwidth=40.0% ] when @xmath215 the equilibrium solution @xmath216 , corresponding to the circular jump @xmath217 , is obtained . when @xmath218 is only slightly larger than @xmath219 we approximate the potential near the local minimum as @xmath220 . if we express @xmath221 , then @xmath222 and the `` energy '' is approximated by @xmath223 . then , @xmath224 and we find @xmath225 for which the condition ( [ eq : polygoncondition ] ) becomes @xmath226 . naturally , this agrees with the condition ( [ kk ] ) found in the linear analysis when @xmath227 . from @xmath228 to @xmath229 , the oscillation amplitude grows until , at @xmath229 , @xmath230 and @xmath231 . correspondingly , the integral @xmath232 decreases monotonically in @xmath192 until it reaches @xmath233 in this limit the condition ( [ eq : polygoncondition ] ) becomes @xmath234 . to summarize , the condition ( [ eq : polygoncondition ] ) for the existence of a polygon solution becomes @xmath235 where @xmath236 \approx [ 1,1.0472]$ ] . @xmath237 corresponds to nearly circular and @xmath238 to spiky polygons . corners exist in the parameter range shown above ( truncated nonlinear model ) . in the truncated nonlinear model . the jumps bifurcate from the line @xmath239 for the circular jumps ( c ) . when the parameters approach the line @xmath240 , the roller thickness approaches @xmath241 in the corners at some @xmath17 , indicating a rupture ; solutions appear spiky near the line ( s ) . on the border @xmath242 , the roller thickness reaches @xmath243 so the solution becomes unphysical ( up ) . [ fig : bd1a],scaledwidth=40.0% ] we also need to ensure @xmath244 . since @xmath245 , @xmath246 is guaranteed . thus , the remaining condition is @xmath247 for all @xmath17 which is equivalent to @xmath248 on this border the roller thickness is predicted to equal the jump radius . this unphysical behavior is likely to be due to truncation of the terms in this approximation . the border curve is computed in fig . [ fig : bd1a ] together with the lines @xmath226 and @xmath249 . since @xmath201 is determined purely by the choice of @xmath192 , the border ( [ eq : case1bif2 ] ) is identical for all @xmath18 but only scaled and shifted in the @xmath0-direction . apart from this border , an @xmath18-polygon exists in the horizontal strip @xmath250 . it is natural to ask for which @xmath18 we find an overlap of these bands corresponding to multistability of polygons as observed experimentally . consider two neighboring strips with @xmath18 and @xmath251 . the condition for overlap of the strips is @xmath252 , or @xmath253 . this leads to a condition @xmath254 , or @xmath255 . while the model thus in principle predicts overlapping regions and resulting hystereses of polygonal jumps with different @xmath18 , it is an experimental fact that overlap exists for much lower @xmath18 . some typical solutions are displayed in fig . [ fig:1234shapes ] in terms of the variables @xmath256 and @xmath257 for the symmetries @xmath258 , which were obtained by solving eqs . ( [ eq : goveqncase1])-([eq : goveqncase2 ] ) numerically . in fig . [ fig:1234shapes ] , each row shows the transition from nearly circular to spiky solutions this may be achieved by tuning @xmath259 between 1 and @xmath260 . notice that a solution only exists for a special value of the energy @xmath192 , and is determined via eq . ( [ eq : polygoncondition ] ) by the choice of the symmetry @xmath18 and `` mass '' @xmath186 . the strict monotonicity of @xmath261 in @xmath192 ( see above ) means that this mapping is one - to - one . it is meaningful to start integration in a corner where @xmath262 at @xmath200 or @xmath201 , respectively . the initial condition @xmath200 is then given by this @xmath192 , ( see eq . ( [ eq : xprime ] ) which yields @xmath263 for @xmath262 ) . in other words , to obtain the desired symmetry @xmath18 with mass @xmath186 , the initial condition @xmath200 must be chosen accordingly . ) -([eq : goveqncase2 ] ) . the polygon shape ( jump line ) is given by @xmath256 ( solid red ) and the radial flux by @xmath264 ( blue dashed ) , which is computed via the radial balance eq . ( [ eq : radly_balance ] ) . numbers refer to the symmetry @xmath18 and increase down the columns . from left to right , the transition from nearly circular to spiky is shown for each symmetry ( c , m and h refer to center , middle or homoclinic orbit , respectively ) , corresponding to the transition discussed in fig . [ fig : bd1a ] . [ fig:1234shapes],scaledwidth=45.0% ] the spiky polygons go along with a singular behavior of the radial outward flux @xmath257 ( blue ) . this behavior is caused by @xmath265 in eq . ( [ eq : radly_balance ] ) , and it is likely that this behavior disappears by inclusion of some of the neglected terms . however , we note that although this behavior is not physical , it is seen that the polygons exhibit very strong jets in the corners in the outward radial direction , as is seen in the flow visualization in fig . [ fig : rollerflowvisual ] . notice also that the wave number has the right dependence on @xmath2 , i.e. larger @xmath2 leads to fewer corners . thus the increase in the number of corners with increasing @xmath3 as seen in fig . [ fig : phasediagram ] is reproduced . we now represent the shaded region in fig . [ fig : bd1a ] in terms of the physical entities . our aim is to compare the predicted region with the measurement on the @xmath266-plane in fig . [ fig : phasediagram ] . the two parameters in the model can be written as @xmath267 we notice that both @xmath80 and @xmath0 appear to depend on @xmath2 and @xmath3 only through the combination @xmath268 . if so , then we are not able to map the bifurcation curves from fig . [ fig : bd1a ] one - to - one to the phase diagram curves in fig . [ fig : phasediagram ] . however , the jump radius @xmath10 actually depends on other parameters as described before . for an estimate of this dependence , we use the relation @xmath269 ( @xmath270 const . ) proposed in @xcite , since it is based on experimental data and leads to a simple estimate for the mapping . this converts eq . ( [ eq : phirelation ] ) to @xmath271 , and the @xmath3-dependence of @xmath0 is corrected . solving this with ( [ eq : delta0relation ] ) , we obtain @xmath272 thus we are able to map @xmath273 back to the physical parameters @xmath2 and @xmath3 . ) . the regions in which polygonal solutions with each @xmath18 up to @xmath274 are shown . the @xmath2-@xmath3 relationship for the border curves , especially in the enlarged inset , is qualitatively similar to the measured fig . [ fig : phasediagram ] despite quantitative disagreement . the bands are predicted to be thinner and do not overlap in the model , unless @xmath275 , unlike in experimental observations . [ fig : physicalborder],scaledwidth=47.5% ] when only @xmath80 is varied while @xmath0 and other parameters on the right hand sides of ( [ eq : horelation],[eq : qrelation ] ) are fixed , these relations predict @xmath276 and @xmath277 , hence @xmath278 . thus , the top and bottom border lines in fig . [ fig : bd1a ] are mapped to these curves . points on the left border line @xmath279 diverge in the physical plane , and the remaining border @xmath280 is mapped to a curve close to the origin of the physical plane . for each @xmath19 we have mapped the region in fig . [ fig : bd1a ] in this way , and the resulting physical parameter plane is shown in fig . [ fig : physicalborder ] . here , the typical set of parameter values in appendix [ appendixa ] are used for physical parameters other than @xmath2 and @xmath3 . an analysis of eqs . ( [ s2 ] ) and ( [ s1 ] ) where @xmath282 is also possible , by numerical means , and partially also by using more intricate analytic arguments based on a hamiltonian formalism which we only mention here for the sake of brevity . to numerically solve the static equations including @xmath282 , a scheme similar to the one described above must be adopted . for this case , we have no analytic means of determining the period @xmath208 in the same manner as above . however , one may integrate the equation system with an event solver : if we start integration from the corner , a half period is given when @xmath283 ( corresponding to the `` belly '' of the polygon ) , which defines the condition ( event ) to stop integration . while integrating from the corner @xmath284 to the belly @xmath285 , one may also compute the period of the solution , @xmath286 , and so , invoke a secondary solver to solve for a @xmath284 such that the commensurability condition @xmath287 is satisfied . such solutions have no extreme spikes as with @xmath227 and instead yield rounder corners . the periodicity @xmath288 of the solution does not any longer depend monotonically on the corner width of the roller , @xmath284 , which results in a more complex solution space ( some of the solutions look like the clover shapes reported by bush _ et al_. @xcite ) . interestingly , larger overlap of the parameter regions corresponding to polygons with different symmetry number @xmath18 is seen , in particular for lower symmetries as low as @xmath289 . details are discussed in @xcite . we now introduce time dependence to the equations . first , we consider time - dependent version of the continuity equation ( [ eq : masscons ] ) . @xmath290 & = & \left [ \frac{q}{2\pi } d\theta + q_\theta(\theta ) \right ] \nonumber \\ & - & \left [ \,q_r(\theta ) d\theta + q_\theta(\theta+d\theta ) \,\right ] . \label{masscons - dimensional}\end{aligned}\ ] ] the left hand side is an estimated rate of change in the volume of the roller slice . the right hand side describes flux into and out of the volume as described in sec . [ sec : massconservation ] . this equation becomes in dimensionless form : @xmath291 & = & \frac{1}{2\pi } - \xi_r(\theta ) - \frac{\partial \xi_\theta}{\partial \theta}. \label{eq : masscons - t}\end{aligned}\ ] ] now , the roller thickness @xmath292 also depends on @xmath293 , and the characteristic time scale @xmath294 is @xmath295 i.e. , time to fill the whole disk with radius @xmath10 and height @xmath296 . using this together with the radial force balance equation ( [ eq : radly_balance ] ) we get @xmath297 = \frac{1}{2\pi } - \frac{\pi_1}{\delta } - \frac{\partial \xi_\theta}{\partial \theta } .\ ] ] next , we introduce time dependence in the azimuthal momentum balance ( [ s1 ] ) . this becomes @xmath298 \xi_{\theta}\\ \label{t1 } & + & 3 \alpha { \rm bo^{-2 } } \delta \left [ { \frac{\rho_1'(\delta)}{\rho_1 ^ 2 } } { \frac { \partial \delta } { \partial \theta}}+{\frac{\partial^3 \delta}{\partial \theta^3 } } \right ] , \end{aligned}\ ] ] where @xmath299 we now linearize around the circular solution as before : setting @xmath300 and @xmath301 , and taking the first order terms in @xmath302 , we find @xmath303 and @xmath304 \xi_1\\ & - & \left(1 - 3 \alpha b { \rm bo}^{-2 } \right ) \frac{\partial \delta_1}{\partial \theta } + 3 \alpha \delta_0 { \rm bo}^{-2}{\frac{\partial^3 \delta_1}{\partial \theta^3}}\end{aligned}\ ] ] instead of their stationary version ( [ eq : linanalysis1a ] ) and ( [ eq : linanalysis1b ] ) . we now fourier transform by @xmath305 = \left [ \begin{array}{c } z \\ x \end{array } \right ] \exp \left [ \frac{s t}{(1-\delta_0 ) { \cal t } } + ik \theta \right],\ ] ] where , due to the @xmath46-periodicity in @xmath17 , the wavenumber @xmath170 has to be an integer . we find @xmath306 where @xmath307\\ c_3 & = & \delta_0 ( 1-\delta_0 ) 3 \alpha \pi_4 { \rm bo^{-2 } } \\ c_4 & = & \frac{1}{2 \pi \delta_0 } , \label{eq : stabcoefficients}\end{aligned}\ ] ] where @xmath162 is defined in ( [ b ] ) . the characteristic equation is @xmath308 with solution @xmath309.\ ] ] for @xmath310 the solutions are @xmath311 and @xmath312 . since @xmath313 , the spatially homogeneous ( circular ) mode is unstable , leading to an increase or decrease of @xmath78 away from @xmath80 . this is clearly unphysical ( since our circular state is uniquely defined from the outset ) and reflects the fact that flux conservation is not built into our time - dependent model , since ( [ t2 ] ) does not directly respect flux conservation . thus , we must require that the constraint ( [ eq : totalmasscons2 ] ) still holds , whereby the total mass flux in and out of the roller is in balance . to linear order , this condition implies that @xmath314 thus , the mode @xmath310 must vanish and is excluded . in the following we only consider integer values @xmath315 . let us take a look at the dispersion relations in ( [ spec1 ] ) . all the @xmath316 s and the bond number are positive , so all the @xmath192 s are guaranteed to be positive except @xmath317 . when @xmath318 , see eq . ( [ maxwavenumber ] ) . the stability curves @xmath319 ( blue dotted ) and @xmath320 ( red dashed ) from ( [ spec1 ] ) for @xmath321 ( above ) and @xmath322 ( below ) with @xmath323 , @xmath324 and @xmath325 . the dashed lines show the curves for continuous @xmath170 , but only integral values of @xmath170 , corresponding to the crosses , are allowed due to @xmath326-periodicity in @xmath17 . the mode @xmath310 is treated separately due to the global conservation law ( [ globcon ] ) . [ fig : disp1],title="fig : " ] when @xmath318 , see eq . ( [ maxwavenumber ] ) . the stability curves @xmath319 ( blue dotted ) and @xmath320 ( red dashed ) from ( [ spec1 ] ) for @xmath321 ( above ) and @xmath322 ( below ) with @xmath323 , @xmath324 and @xmath325 . the dashed lines show the curves for continuous @xmath170 , but only integral values of @xmath170 , corresponding to the crosses , are allowed due to @xmath326-periodicity in @xmath17 . the mode @xmath310 is treated separately due to the global conservation law ( [ globcon ] ) . [ fig : disp1],title="fig : " ] as discussed in sec . [ surfacetension ] , surface shape parameter @xmath162 is positive when @xmath113 grows with @xmath78 this is the case at least close to the circular jump where @xmath327 . then , the sign of @xmath317 only depends on the relative magnitude of @xmath162 and @xmath328 : when @xmath329 , we have @xmath330 , and surface tension is `` active '' in the sense that it then can cause a rayleigh - plateau - like instability . this instability criterion is equivalent to @xmath331 at present , we do not have much data to check this criterion , but from the case presented in fig . [ fig : heightprofile ] we can estimate the two curvature radii and thereby the magnitude of @xmath332 ( see appendix a ) . note however , that since the values entering @xmath162 should really be averages over the surface , we somewhat overestimate the curvatures and therefore likely underestimate @xmath162 . ethylene glycol has a surface tension of @xmath333 n / m and a density of 1100 kg m@xmath334 ; these numbers yield a capillary length of @xmath335 mm and a bond number bo @xmath336 . in effect , the left hand side of ( [ crit ] ) thus becomes @xmath337 , whereas the right hand side is @xmath338 . this demonstrates that the circular state should indeed be unstable towards the formation of polygons . the flow - rate is @xmath339 ml s@xmath340 and using the parameters listed in appendix a , we have @xmath341 ; this yields a positive prefactor in @xmath317 of @xmath342 . although the two sides of ( [ crit ] ) appear similar , these estimates show that it is plausible that @xmath317 may take on negative values . ( dashed line ) from eq . ( [ spec1 ] ) shown for stability coefficients @xmath343 , @xmath344 , @xmath345 , and @xmath346 , based on experimental parameters and the height profile of a pentagon ( see appendix a ) . note that the constant branch @xmath347 is negative . we then find that the maximally stable wave number ( eq . [ maxwavenumber ] ) is @xmath348 ; thus the most likely symmetry to occur has symmetry @xmath349 or @xmath350 which matches well the expectation . the crosses mark the values of @xmath351 for integer @xmath170 , respecting the @xmath326-periodicity . the mode @xmath310 is treated separately according to the global conservation law ( [ globcon ] ) . [ fig : disp4 ] ] let us look at the dispersion relations in more detail . first we note that @xmath352 has reflective symmetry around @xmath310 . in figs . [ fig : disp1 ] , [ fig : disp4 ] we therefore only display the positive @xmath170-axis . along this positive @xmath170-axis , there are ( at most ) three distinct values that occur for the dispersion relation : ( i ) the maximal value of @xmath352 at @xmath353 , defining the two neighboring modes of strongest growth ( integer values ) , ( ii ) the root of the dispersion relation @xmath354 , and ( iii ) the point @xmath355 where the two branches @xmath356 meet . for @xmath357 , the discriminant @xmath358 becomes negative and the real part assumes simply a constant value , @xmath359 , which defines a further symmetry axis for the dispersion relation . note that , by definition , @xmath360 is identical to eq . ( [ eq : klin ] ) . for the unstable case ( @xmath361 ) we expect the symmetry breaking predominantly to occur at the _ maximally unstable wave number _ @xmath353 . @xmath362 is maximal when the term @xmath363 is maximal , and so @xmath364 clearly , such a maximal value only exists when @xmath317 is negative , i.e. when the condition @xmath365 in ( [ crit ] ) is satisfied , as illustrated in fig . [ fig : disp1 ] . for small bond number , or strong instability , ( [ maxwavenumber ] ) simplifies to @xmath366 and if the roller cross section is close to cylindrical , we find @xmath367 , i.e. a wavelength proportional to the width of the roller as in the rayleigh - plateau instability . this fits nicely with the observed bifurcation sequence , since , experimentally , polygons with many corners ( up to 13 ) are seen close to the transition , where the roller is thin , and lower order polygons only occur later when the width is larger . in fig . ( [ fig : disp4 ] ) , we show the dispersion relation using parameter estimates from appendix a which are based on the height profile of the pentagon shown in fig . [ fig : heightprofile ] . for this case , we obtain @xmath368 and @xmath369 . with this value of @xmath353 , it is most likely that polygons with either @xmath349 or @xmath350 corners occur , in accordance with the experimental observation . we hope that we have been able to convince the reader that the type ii hydraulic jumps and their intriguing polygon offspring can be understood within the phenomenological framework presented here . we have given a detailed presentation of the model and the underlying assumptions about the flow . in particular , we assume that the inner edge of a roller " structure inside the jump describes the shape of the type ii jump . in the circular type ii regime , the internal flow of this roller structure is similar to a toroidal vortex . the local width of this roller is one of three basic variables in our model . the two other variables in our model are the radial and tangential flow rates across the jump and inside the roller , respectively . a polygonal shape is obtained when the width loses circular symmetry and becomes a periodic function of the polar angle , while the outer edge of the roller remains circular , as observed in experiments . also , for the polygonal states , a net flux is carried inside the roller from the sides to the corners . the model is formulated by considering a radial and a tangential force balance across the roller structure and the polygon shapes occur primarily by competition between viscous and gravitational effects . we have shown that one variant of the model is exactly solvable , even in the strongly non - linear limit , and that it has polygonal solutions which are quite similar to those observed experimentally . the resulting bifurcation diagram is qualitatively similar to experimental measurements . the question of stability has been addressed as well . here , surface tension plays an important role . for small deformations around the circular state , we can include surface tension by considering the laplace pressure inside the roller . this pressure can be estimated by introducing a parameter that characterizes how the curvature of the roller changes with deformation . when this parameter is positive , which is expected for the circular type ii hydraulic jump , surface tension effects will tend to destabilize the circular state if the length scales at the jump ( e.g. jump height @xmath2 and radius @xmath10 ) are of the order or smaller than the capillary length . thus a polygon state will emerge with a wave length of the order of the width of the roller , in analogy with the rayleigh - plateau instability . to include surface tension beyond the linear approximation would require more detailed knowledge about the height profile of the polygon states . hopefully such results will soon be available indeed , one of the aims of the present paper is to stimulate renewed research in this direction . in the present study , we have neglected the rotational motion inside the roller which is clearly observed in experiments . the stability of a rotating liquid column with free cylindrical surface has been investigated in other studies @xcite . the relative importance of the rotational kinetic energy to the surface energy is characterized by the weber number @xmath370 , where @xmath371 is the characteristic rotational frequency and @xmath372 is the width of the roller . for an infinitely long cylinder , subject to axisymmetric disturbances ( neglecting disturbances in e.g. azimuthal direction ) , a necessary and sufficient condition for stability is given by @xmath373 @xcite , where @xmath170 is our wave number ( this result is valid for potential vortex flows and solid body rotation ) , and thus rotation could widen the spectrum of unstable modes . compared to the roller in the hydraulic jump , these vortices are of course strongly idealized , and we do not know at present how to include them in a convincing way . we are very grateful to kre stokvad hansen , ditte jrgensen , bjrn tolbod , jesper larsen and clive ellegaard for valuable discussions in the initial phases of this work . we are particularly indebted to johan rnby for valuable discussions and suggestions throughout the entire process . e.m . would also like to thank steven strogatz , paul steen , and richard rand for helpful discussions on issues regarding the stability analysis of strongly non - circular solutions . a ` typical ' set of parameters used in the experiments is : @xmath374 m@xmath375/s , @xmath376 m , @xmath377 m , @xmath378 m@xmath379/s , @xmath380 m / s@xmath379 , @xmath333 n / m , and @xmath381 @xmath375 . clearly , the assumption @xmath382 used throughout the article is then valid . with the above parameters , we have a bond number of @xmath383 ( @xmath384 ) and @xmath385 mm . furthermore , using the values from the height profile in fig . [ fig : heightprofile ] , we read off the following numbers : the outer edge of the roller @xmath386 m ; the width of the roller between two corners ( belly ) @xmath387 mm , and in the corner @xmath388 mm ; and the corresponding curvature radii @xmath389 mm and @xmath390 mm . using the approximations @xmath391 and @xmath392 we estimate that @xmath393 . note that the values entering @xmath162 should really be averages over the surface ; therefore we somewhat overestimate the curvatures and therefore likely underestimate @xmath162 . the scaling factors @xmath394 and @xmath395 are unknown , but using the above experimental parameters and the estimates from the height profile , we can at least estimate some of them . using the scaling @xmath269 proposed in @xcite for the outer edge of the roller , @xmath10 , we estimate that @xmath396 ; moreover , we can match the definition @xmath397 with @xmath398 . the constants @xmath399 originate from the linearization of the frictional force ( [ eq : df_t_mu ] ) , and it is reasonable to assume that they are of order unity , too . with these estimates , the non - dimensional numbers become : @xmath400 , @xmath401 , @xmath402 , @xmath403 , @xmath404 , @xmath405 . the stability coefficients are then @xmath343 , @xmath344 , @xmath345 , and @xmath346 . with these estimated coefficients , we find the spectrum for stable wave numbers as shown in fig . [ fig : disp4 ] , with a maximally unstable wave number @xmath348 and @xmath406 has a root @xmath369 . evaluating the wave number of the linearly perturbed circular jump in eq . ( [ eq : klin ] ) , we have @xmath407 which is at least quite close to a pentagon . however , we note again that these numbers can only serve as rough estimates due to the lack of experimental data ( i.e. to determine @xmath408 and @xmath162 ) . typical frequencies for the rotational motion in our experiment are of the order of 5 hz , so @xmath409 rad / s . the roller structure has a width of order @xmath410 mm and a height of @xmath411 mm . using for the average roller radius an estimate of @xmath412 mm we arrive at an estimated weber number of @xmath413 ; notice though that the weber number scales with the cube of the average roller width @xmath414 and is thus very sensitive to errors in the estimate of @xmath414 . it is not clear that the vorticity extends all the way out to where the surface attains its maximal value ( this is how we estimated @xmath415 ) , and therefore we may be overestimating @xmath414 . we define standard dimensionless numbers as follows . defining the velocity before the jump as @xmath416 we get the reynolds number @xmath417 where we take @xmath2 as the characteristic scale , similar to @xmath120 . the froude numbers are @xmath418 so @xmath419 similarly , we have the downstream froude number @xmath420 these dimensionless numbers are evaluated using the above choice of parameter , with @xmath421 mm for the height profile in fig . [ fig : heightprofile ] . with these definitions , @xmath0 from eq . ( [ eq : phirelation ] ) is @xmath422 the weber number is defined as @xmath423 because @xmath2 is similar to @xmath120 , which is similar to radii of curvature . the variables and dimensionless constants used throughout the paper are summarized in table [ nomenclature ] . a. marcussen : det hydrauliske spring ; et eksperimentelt studie af polygoner og hastighedsprofiler . ( the hydraulic jump ; an experimental study of polygons and velocity profiles ) . master s thesis , cats , niels bohr institute , copenhagen ( 1999 ) . in danish . supplemental materials with a visualization of the flow inside the roller structure , a demonstration of the role of surfactant eliminating polygons are provided at http://eam.webhop.net/si.php5 [ http://eam.webhop.net/si.php5 ] ` http://eam.webhop.net/si.php5 ` .

recently , it has been pointed out that two different excesses of events observed at lep could be interpreted as the cp - even higgs bosons of the mssm with masses of approximately 98 and 114 gev . if this is the case , the entire mssm higgs sector is required to be light . in this article , we explore such a scenario in detail . we constrain the higgs and supersymmetric spectrum using physics constraints as well as the magnetic moment of the muon . we then point out the implications for neutralino dark matter next generation direct detection experiments will be sensitive to all mssm models with such a higgs sector . finally , we find that all models outside of a very narrow corridor of the parameter space have a charged higgs boson which will be observed at the lhc . in those exceptional models which do not contain an observable charged higgs , a light top squark will always be seen at the lhc , and likely at the tevatron . 0.5 cm

hep-ph0506061

the four experiments at lep have searched for higgs bosons up to a mass of approximately 115 gev @xcite . although no strong indication of a higgs has been detected by lep , excesses with statistical significances of 2.3@xmath2 and 1.7@xmath2 have been reported for higgs - like events corresponding to masses of 98 gev and 115 gev , respectively . very recently , it has been pointed out by manuel drees that both of these excesses reported by lep could be accommodated within the minimal supersymmetric standard model ( mssm ) as signatures of the two cp - even higgs bosons @xcite . one of the reason that the lep excess at 98 gev has received so little attention might be that the rate observed corresponding to this mass is about a factor of ten below that which would be expected from a standard model higgs boson . in the mssm , however , the @xmath3-@xmath4-@xmath4 coupling is suppressed relative to the value in the standard model by a factor of @xmath5 , where @xmath6 is the mixing angle between the two cp - even higgs bosons and @xmath7 is the ratio of the two vacuum expectation values . to predict the number of events seen at lep for a 98 gev light higgs boson , we constrain @xmath8 ( the precise range of @xmath9 , @xmath10 and @xmath11 we allow is the same as used in ref . @xcite to allow for easy comparison of results ) . meanwhile , the analogous coupling of the heavy cp - even higgs is only slightly suppressed below the standard model value and is consistent with the number of events reported by lep corresponding to a 115 gev higgs boson . these two excesses together yield a combined statistical significance of 3.1@xmath2 @xcite . the mass scale of the mssm higgs sector is usually fixed by the mass of the pseudoscalar higgs boson @xmath12 . in the well known leading @xmath13 approximation @xcite , we can very roughly link the pseudoscalar mass to the two scalar higgs masses in the limit of non - mixing top squarks : @xmath14 in this approximation @xmath15 is the top squark mass , @xmath16 are the neutral higgs masses and @xmath17 is the charged higgs mass . we immediately see that the existence of two light scalar higgs bosons leads to all of the mssm higgs bosons being fairly light . this can lead to interesting phenomenology in the mssm . in this article , we consider the constraints on this scenario from the @xmath18 and @xmath19 branching fractions and the magnetic moment of the muon . we discuss the implications of such a model on the phenomenology of neutralino dark matter , and find that the prospects for the direct detection of neutralinos in elastic scattering experiments to be excellent . we then study the spectra of the superpartners and the higgs sector with two light higgs scalars and discuss in detail the prospects for discovering the charged higgs boson at the lhc . is shown with the charged higgs mass . for all points shown , @xmath20 , @xmath21 and @xmath22 , corresponding to the range matching the observations at lep . the black points are consistent with measurements of @xmath18 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath23 , but are consistent with @xmath18 . green points violate @xmath18 . these constraints will be discussed further in section [ sec : constraints].,title="fig:",width=192 ] is shown with the charged higgs mass . for all points shown , @xmath20 , @xmath21 and @xmath22 , corresponding to the range matching the observations at lep . the black points are consistent with measurements of @xmath18 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath23 , but are consistent with @xmath18 . green points violate @xmath18 . these constraints will be discussed further in section [ sec : constraints].,title="fig:",width=192 ] in order to study the phenomenology of supersymmetric models with light higgs scalars , we first performed a scan over the relevant parameters of the mssm . we have varied all masses up to 6 tev , and @xmath24 between 1 and 60 . we did not assume any specific supersymmetry breaking scenario or unification scheme . to take into account the radiative corrections to the higgs masses , we have used feynhiggs @xcite , which includes all contributions up to the two - loop level . we find that this level of precision is needed to obtain an accurate representation of the mssm phenomenology within this class of models . we give a detailed analysis of the higgs sector in section [ sec : coll ] . in fig . [ fig : achmuc ] , we confirm the conclusion of ref . @xcite that the entire mssm higgs sector is required to be rather light to accommodate both excesses reported by lep . in the overwhelming majority of models found , the mass of the pseudoscalar higgs bosons is between 95 and 130 gev and the mass of the charged higgs lies between 110 and 150 gev . although we do find some points outside of this range , in particular a trail of models extending to the upper right of the figure ( left frame ) , these are quite rare compared to the dense collection of models near @xmath25 gev and @xmath26 gev . for each point shown , we require @xmath27 , @xmath21 and @xmath28 , corresponding to the range needed to match the observations at lep . the differences between this figure and the corresponding one in ref . @xcite come from the fact that we scan continuously over @xmath29 . in the right frame of fig . [ fig : achmuc ] , we check the mass of the charged higgs boson for correlations with the value of @xmath30 . we have checked that indeed the trails of scattered points at large pseudoscalar masses correspond to the larger values of @xmath29 the larger the corresponding charged higgs masses become . without imposing additional constraints we find relatively little correlation , for example , between the higgs masses and the @xmath30 parameter , apart from the chargino mass limit of 104 gev from lep . branching fraction as a function of @xmath24 , the lightest chargino mass , and the charged higgs mass . shown as horizontal dashed lines are the @xmath32 confidence bounds on this quantity as measured by belle , cleo and aleph . again , all points shown have @xmath27 , @xmath33 and @xmath28 . black points do not violate the tevatron constraint on the @xmath19 branching fraction . green points violate this constraint.,title="fig:",width=192 ] branching fraction as a function of @xmath24 , the lightest chargino mass , and the charged higgs mass . shown as horizontal dashed lines are the @xmath32 confidence bounds on this quantity as measured by belle , cleo and aleph . again , all points shown have @xmath27 , @xmath33 and @xmath28 . black points do not violate the tevatron constraint on the @xmath19 branching fraction . green points violate this constraint.,title="fig:",width=192 ] branching fraction as a function of @xmath24 , the lightest chargino mass , and the charged higgs mass . shown as horizontal dashed lines are the @xmath32 confidence bounds on this quantity as measured by belle , cleo and aleph . again , all points shown have @xmath27 , @xmath33 and @xmath28 . black points do not violate the tevatron constraint on the @xmath19 branching fraction . green points violate this constraint.,title="fig:",width=192 ] constraints from rare @xmath0-decays and the magnetic moment of the muon are some of the most useful tools we currently have to guide our studies of supersymmetric phenomenology . the @xmath0 physics constraints gain their power mostly because of a possible @xmath29 enhancement of the bottom yukawa couplings . in this section , we consider some of these constraints within the context of models with 98 and 115 gev higgs bosons . the branching fraction of @xmath0 decays to a strange state plus a photon has been measured by the belle @xcite , cleo @xcite and aleph @xcite experiments . the weighted average of these experiments results indicate br@xmath34 . in comparison , the standard model prediction for this transition rate is @xmath35 @xcite . the supersymmetric processes which are most likely to contribute substantially to this branching ratio involve a charged higgs or a chargino . these contributions are enhanced by powers of the yukawa coupling @xmath36 for large values of @xmath24 . in passing we emphasize that there are additional @xmath37 corrections @xcite which can have huge effects if the mssm spectrum is split between light gluinos or higgsinos and heavier sbottoms , but we will see that this is not the part of parameter space in which we are interested in . if the gluino mass is smaller than 300 gev , the lhc will be swamped by gluino pair production with cross sections as large as @xmath38 pb and the effect of the gluino mass on the charged higgs boson is negligible . in fig . [ fig : bsg ] , we plot br@xmath39 versus @xmath24 ( left frame ) , the lightest chargino mass ( center frame ) and the charged higgs mass ( right frame ) . shown as horizontal dashed lines are the @xmath32 confidence bounds on this branching fraction . again , all points shown have @xmath40 , @xmath41 and @xmath8 . strictly applying a @xmath42 limit would rule out all points with @xmath43 , but on the other hand allowing for a @xmath44 deviation from the standard model brings all values of @xmath29 back into the allowed region . note that the preference of larger values of @xmath29 is due to a slight bias toward a finite mssm contribution : if we force the higgs sector to be light , the 2hdm diagrams will lead to an increase of the observable br@xmath45 by typically tens of percent up to factor of two . this charged higgs contributions consist of a @xmath29 suppressed term and a constant term , but does not exhibit any @xmath29 enhancement ( if we do not consider anomalously large gluino loops ) . because the measured value of br@xmath46 is actually slightly smaller than the standard model prediction , the chargino has to compensate for the 2hdm contribution . because the chargino contribution to br@xmath39 is enhanced by one power of @xmath29 on the amplitude level , we can achieve this by choosing large values of @xmath29 , as we see in the left panel of fig . [ fig : bsg ] . we can , of course , try to increase the chargino diagram by making the chargino light , but this is much less efficient , because the loop involved is a chargino stop loop , so that just making the chargino light has comparably little effect . again , we see in the center frame of fig . [ fig : bsg ] that very light chargino masses serve this purpose . moreover , the large chargino contribution has to come with the right sign and therefore prefers positive @xmath30 , as is shown in fig . [ fig : achmuc ] . as we can see , the charged higgs mass and therefore the contribution to br@xmath47 is basically fixed by the two light higgs masses . thus merely shifting the charged higgs mass between 120 and 170 gev has even less of an effect . branching fraction versus @xmath24 . the dashed horizontal line is the 90% upper confidence bound placed on this branching fraction by the tevatron experiments . black points are consistent with measurements of @xmath18 at the 3@xmath2 level . green points violate this constraint . right two panels : the contribution to the magnetic moment of the muon . the dashed horizontal lines are the @xmath42 bounds from @xmath48 data . the black points are consistent with measurements of @xmath49 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points do violate @xmath19 , but are consistent with @xmath49 . green points violate @xmath18 . all points shown have @xmath27 , @xmath50 and @xmath51.,title="fig:",width=192 ] branching fraction versus @xmath24 . the dashed horizontal line is the 90% upper confidence bound placed on this branching fraction by the tevatron experiments . black points are consistent with measurements of @xmath18 at the 3@xmath2 level . green points violate this constraint . right two panels : the contribution to the magnetic moment of the muon . the dashed horizontal lines are the @xmath42 bounds from @xmath48 data . the black points are consistent with measurements of @xmath49 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points do violate @xmath19 , but are consistent with @xmath49 . green points violate @xmath18 . all points shown have @xmath27 , @xmath50 and @xmath51.,title="fig:",width=192 ] branching fraction versus @xmath24 . the dashed horizontal line is the 90% upper confidence bound placed on this branching fraction by the tevatron experiments . black points are consistent with measurements of @xmath18 at the 3@xmath2 level . green points violate this constraint . right two panels : the contribution to the magnetic moment of the muon . the dashed horizontal lines are the @xmath42 bounds from @xmath48 data . the black points are consistent with measurements of @xmath49 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points do violate @xmath19 , but are consistent with @xmath49 . green points violate @xmath18 . all points shown have @xmath27 , @xmath50 and @xmath51.,title="fig:",width=192 ] [ fig : g2 ] limits on the branching fraction of the rare decay @xmath23 can also be exploited to limit the allowed mssm parameter space . the standard model prediction for this branching ratio is br=@xmath52 . in particular , the higgs induced mssm contributions to this branching fraction scales with @xmath53 and tends to be especially large in our scenario with a light higgs sector @xcite . if we require the higgs sector to be light , models with very large values of @xmath24 are likely to violate the constraints on this quantity placed at the tevatron @xcite . in fig . [ fig : bmumu ] , we plot br@xmath54 versus @xmath7 . shown as a dashed horizontal line is the 90% upper confidence bound placed by the tevatron experiments . the standard model value is clearly visible as the low @xmath29 nose in the distribution of the mssm parameter points small @xmath29 means small contributions from the mssm higgs sector . the upper limit from the tevatron becomes an issue for @xmath55 but it really only develops seriously destructive power for @xmath56 . combining the upper limit on the @xmath57 decay and the measurement of @xmath58 makes it considerably harder for us to find viable mssm scenarios with two light higgs scalars . on the one hand , we need an mssm contribution to @xmath18 from the chargino sector with the right sign and magnitude to compensate for the charged higgs diagrams . this is possible by exploiting the enhancement by one power of @xmath29 in the chargino amplitude . on the other hand , we do not want too large a contribution to @xmath19 from the same sector which scales as @xmath59 . putting both of these constraints together , we favor fairly large , but not too large , values of @xmath29 and at the same time accept some light superpartners to accommodate @xmath60 . we should stress , however , that this argument is not strict in the sense that these two constraints guarantee a light mssm mass spectrum . by choosing properly tuned parameters , we can still get by with a spectrum where the lightest of the charginos and stops weighs more than 1 tev . looking into the near future , we see that of course the lhc prospects of seeing these light - higgs mssm scenarios in @xmath0 physics are excellent . simulations of cms , atlas and lhcb events probing br@xmath54 predict tens of events for 10 fb @xmath61 and the standard model decay rate . moreover , it has been shown that high luminosity triggering on this search channel is possible , so that the lhc reach should even cover smaller branching fractions than predicted in the standard model . basically all parameter points shown in fig . [ fig : bmumu ] will be clearly visible @xcite . the only question is whether the theoretical and experimental errors will allow us to distinguish between the standard model and the mssm predictions . finally , the magnetic moment of the muon has been measured to be anomalously high in comparison to the standard model prediction . using @xmath48 data , the measured value exceeds the theoretical prediction by @xmath62 , where the error bars correspond to theoretical uncertainties in the leading order hadronic and the hadronic light - by - light contributions as well as from experimental contributions @xcite . this measured value is 2.4@xmath2 above the standard model prediction . experiments using @xmath63 data , on the other hand , find @xmath64 , which is only 0.9@xmath2 above the standard model prediction and likely not in agreement with most recent kloe data @xcite . given this conflict and the marginal statistical significance of these measurements , we do not require all of our scenarios to produce the measured value of @xmath65 . contributions to @xmath31 occur at the one and two loop levels from diagrams involving both higgs bosons @xcite and superpartners @xcite . at the one - loop level the contribution from superpartner exchange is proportional to @xmath66 @xcite . the exchange of a light pseudoscalar higgs contributes like @xmath67 in the leading order of @xmath29 . because we are only considering models with a light higgs sector , the effects from the @xmath68 exchange will dominate . this is what we see in the @xmath29 dependence of the permitted parameters points in fig . [ fig : g2 ] . in a way , the situation is the same as for @xmath18 : both measurements sit very slightly away from their respective standard model predictions ( and from a statistical point of view are not very convincing ) , and the central values can be accommodated by choosing large @xmath29 . in the two right panels of fig . [ fig : g2 ] we see that the @xmath69 constraint goes a long way to also accommodate the @xmath65 measurement in our mssm parameter space . this is particularly striking when we look at the behavior of the points around @xmath70 . the @xmath71 constraint disfavors the ` wrong ' sign of @xmath72 already , so that the impact of the @xmath65 measurement on our light - higgs models is very limited once we allow slightly more than a @xmath42 window . again , we see how very large values of @xmath29 are disfavored by @xmath23 , which drives the mssm parameter points toward lighter superpartners , lighter stops and charginos / neutralinos . this constraint is expected to become more stringent over the coming years , even before the lhc will start operation . to calculate br@xmath39 , br@xmath73 and @xmath74 , we have used the micromegas program @xcite . , which for models with a bino - like lightest neutralino is a good approximation for @xmath75 , and therefore scales with the annihilation cross section to @xmath76 via pseudoscalar higgs exchange . the two solid lines represent the analytic result of this calculation using @xmath77 gev and @xmath7850 ( bottom ) and 500 ( top ) gev . all models shown have @xmath79 , @xmath21 and @xmath28 . the black points are consistent with measurements of @xmath18 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath18 . green points violate @xmath49.,title="fig:",width=192 ] , which for models with a bino - like lightest neutralino is a good approximation for @xmath75 , and therefore scales with the annihilation cross section to @xmath76 via pseudoscalar higgs exchange . the two solid lines represent the analytic result of this calculation using @xmath77 gev and @xmath7850 ( bottom ) and 500 ( top ) gev . all models shown have @xmath79 , @xmath21 and @xmath28 . the black points are consistent with measurements of @xmath18 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath18 . green points violate @xmath49.,title="fig:",width=192 ] one very attractive feature of r - parity conserving supersymmetry is that it naturally provides a stable particle which , in many models , can be a viable candidate for dark matter . the lightest neutralino is particularly appealing in this respect @xcite . neutralinos can annihilate through a variety of channels , including through the exchange of cp - even or odd higgs bosons , charginos , neutralinos , sfermions and gauge bosons . which annihilation channel(s ) dominates varies from model to model . in the scenario we are studying here , however , the presence of light higgs bosons suggest that s - channel higgs exchange to fermion pairs is likely to be a particularly efficient annihilation channel , especially in those models with moderate to large values of @xmath24 . the formulae for these annihilation channels are collected in the appendix . the annihilation cross section can be used to calculate the thermal relic abundance present today : @xmath80 where @xmath81 is the number of relativistic degrees of freedom available at freeze out , @xmath82 and @xmath83 are the amplitudes given in the appendix , and @xmath84 is the value at freeze out : @xmath85 over the range of cross sections and masses expected for the lightest neutralino , @xmath86 . to yield the density of cold dark matter measured by wmap @xcite , we thus require @xmath87 @xmath88/s . the main contribution to @xmath82 and @xmath83 are s - channel cp - odd and cp - even higgs exchange , respectively . from the amplitudes given in the appendix , we find @xmath89 , so we conclude that the @xmath83-term induced through cp - even higgs plays a subdominant role in the relic density calculation to the cp - odd higgs induced @xmath82-term . in the left frame of fig . [ fig : relic ] , we plot the relic density of the lightest neutralino versus its mass . we have calculated this quantity using the micromegas program @xcite . over the vast majority of supersymmetric parameter space , the composition of the lightest neutralino is dominated by its bino component , although a small admixture of wino or higgsino is common . for a bino - like neutralino with large or moderate @xmath7 , the annihilation cross section through pseudoscalar higgs exchange scales as @xmath75 . this up - type higgsino component can be approximated in most models by @xmath90 . with this in mind we plot the neutralino relic density versus the quantity @xmath91 in the right frame of fig . [ fig : relic ] . the two solid lines shown are the analytic results for @xmath77 gev and @xmath7850 ( bottom ) and 500 ( top ) gev in this approximation . the majority of models we find fall within this range , suggesting that their annihilation is in fact dominated by and s - channel pseudoscalar higgs exchange to @xmath76 . as expected , no models fall above this range . in the models which lie below this range , another annihilation mode or modes ( such as t - channel sfermion exchange , s - channel @xmath4 exchange or s - channel pseudoscalar higgs exchange to top quark pairs ) must contribute substantially , thus lowering the relic abundance accordingly . coannihilations between the lightest neutralino and another superpartner may also reduce the relic density . from the right frame of fig . [ fig : relic ] , we can infer that for those models which generate the measured relic abundance ( those which fall along the horizontal dashed line ) , most appear to annihilate substantially through pseudoscalar higgs exchange to @xmath92 . models with a bino - like neutralino with a small higgsino admixture and moderate to high value of @xmath24 are often capable of generating the measured density of dark matter . , @xmath50 and @xmath51 . the left frame includes models with any relic density , while those models shown in the right frame generate an abundance of dark matter within a factor of ten of the observed quantity . the dashed curve is the current limit placed by the cdms experiment . the dotted yellow , solid red , dotted blue and solid magenta curves ( from top to bottom ) are the approximate projected limits of gerda , cdms , zeplin - max and super - cdms ( phase iii ) , respectively . the black points are consistent with measurements of @xmath18 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath93 branching fraction . blue points violate @xmath23 , but are consistent with @xmath18 . green points violate @xmath18.,title="fig:",width=192 ] , @xmath50 and @xmath51 . the left frame includes models with any relic density , while those models shown in the right frame generate an abundance of dark matter within a factor of ten of the observed quantity . the dashed curve is the current limit placed by the cdms experiment . the dotted yellow , solid red , dotted blue and solid magenta curves ( from top to bottom ) are the approximate projected limits of gerda , cdms , zeplin - max and super - cdms ( phase iii ) , respectively . the black points are consistent with measurements of @xmath18 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath93 branching fraction . blue points violate @xmath23 , but are consistent with @xmath18 . green points violate @xmath18.,title="fig:",width=192 ] the prospects for the direct detection of particle dark matter are quite encouraging for a neutralino in conjunction with a light higgs sector . in many of the models we study here , the spin - independent neutralino - nucleon elastic scattering cross section is dominated by the t - channel exchange of cp - even higgs bosons . the cross section for this process is roughly given by @xmath94 ^ 2,\ ] ] where the sum is over quark types and @xmath95 are the matrix elements over the nucleonic state . the quantity @xmath96 should be replaced with @xmath97 for up - type quarks . @xmath98 , @xmath99 and @xmath100 are the bino , wino and higgsino fractions of the neutralino . in the sum over quark species , the strange quark contribution dominates with @xmath101 gev . for models with @xmath102 gev higgs bosons and a bino - like lsp , we estimate @xmath103 of course other elastic scattering channels ( squark exchange in particular ) can also contribute substantially . in fig . [ fig : direct ] we show the spin - independent neutralino - nucleon elastic scattering cross section for models in this scenario . models shown in the right frame each generate an abundance of neutralino dark matter within one order of magnitude of the observed density , while those models shown in the left frame may or may not . the dashed curve is the current limit placed by the cdms ii experiment @xcite . the dotted yellow , solid red , dotted blue and solid magenta curves ( from top to bottom ) are the approximate projected limits of the gerda @xcite , cdms ii , zeplin - max @xcite and super - cdms ( phase iii ) experiments , respectively . edelweiss @xcite should also reach a sensitivity similar to that of cdms . to calculate these elastic scattering cross sections , we have used the darksusy program @xcite . from fig . [ fig : direct ] , it is obvious that the prospects for direct detection are excellent for this class of supersymmetric models with light higgs sectors which often dominate the cross section . most of the favored models are within the reach of the cdms and edelweiss experiments and those remaining models below this level of sensitivity should be within the reach of next generation experiments such as zeplin - max or super - cdms . it is straight forward to understand why models with very small elastic scattering cross sections do not appear in fig . [ fig : direct ] . if we limit the magnitude of @xmath30 to be less than a few tev for fine tuning reasons , the higgsino fraction of the lightest neutralino is then bounded from below to be @xmath104 . from this , we see that @xmath105 can not be smaller than @xmath106 to @xmath107 pb without @xmath108 being unnaturally large . in addition to these excellent prospects for direct dark matter searches , the characteristics of the lightest neutralino in this class of models are fairly ideal for the purposes of indirect detection . since the annihilation cross section during freeze out is dominated by the first term in the expansion , @xmath109 , the annihilation cross section relevant for neutralinos annihilating near the galactic center , throughout the galactic halo , or elsewhere of interest to indirect detection , is the maximum value consistent with the measured relic abundance , @xmath110 @xmath88/s . this makes it rather likely that such models with be within the reach of future indirect detection experiments searching for neutralino annihilation products in the form of gamma - rays @xcite or anti - matter @xcite . the prospects for detecting such a neutralino in future cosmic positron experiments are especially promising . before we begin discussing the prospects of observing a light mssm higgs sector at the lhc , we should point out that it has been known for a long time that one cp - even higgs scalar in the mssm is guaranteed to be seen in weak - boson - fusion production with a subsequent decay @xmath111 @xcite . the search strategy is identical to the standard model search for low higgs masses and has been extensively studied including detector effects @xcite . in the special case of three light mssm higgs scalars , this discovery channel is known to face a challenge : the @xmath112 mass resolution will not be sufficient to resolve the two cp - even scalars if their mass difference is less than @xmath113 gev they will appear as one sightly wider higgs resonance . moreover , the pseudoscalar higgs only couples to gauge bosons through a dimension-5 operator @xcite , thus it will not be produced in weak boson fusion . the case of three almost mass - degenerate light higgs scalars has been specifically studied , with an emphasis on distinguishing the three mass peaks in the decay to muons @xcite . the conclusion is that for higgs masses below 140 gev , the inclusive higgs search is challenging and that the heavy higgs bosons will probably not be separately observable in this channel . the more promising strategy is to look for bottom - higgs associated production with a subsequent higgs decay to muons and at least one @xmath83 tag . the yukawa coupling for the heavy scalar and the pseudoscalar higgs can then be @xmath29 enhanced and the mass peaks might be resolved for @xmath114 and @xmath115 gev . the general lesson we learn from detailed studies by atlas and cms is that it might well be easier to discover a charged higgs boson than the additional neutral scalars . a charged higgs can be searched for either in anomalous top decays , or it can be directly produced as @xmath116 . in both cases the decay to a tau lepton is most promising @xcite . the coverage of models in the ( @xmath12 @xmath29 ) plane from this channel has historically included a hole for charged higgs masses just above the top mass . we emphasize that this hole does not mean that the charged higgs will be missed at the lhc in this parameter range . it just means that we will have to combine the search strategies for ( off - shell ) anomalous top decays with the usual associated production @xcite . for the purpose of this paper we will use a preliminary parton level analysis @xcite . we include this result in fig . [ fig : lhc1 ] , 300 fb@xmath117 ) discovery potential for a charged higgs boson at the lhc . below the nearly horizontal ( magenta ) line , charged higgs bosons can be identified through the @xmath118 , @xmath119 channel . models falling to the upper right of the red line can be discovered through the channel @xmath120 , @xmath119 . models with very light @xmath68 , @xmath121 or very small @xmath24 are excluded by lep . right : the same points shown in the @xmath122 vs @xmath24 plane . all models shown have @xmath27 , @xmath123 and @xmath51 . the black points are consistent with measurements of @xmath124 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath60 . green points violate @xmath18.,title="fig:",width=192 ] , 300 fb@xmath117 ) discovery potential for a charged higgs boson at the lhc . below the nearly horizontal ( magenta ) line , charged higgs bosons can be identified through the @xmath118 , @xmath119 channel . models falling to the upper right of the red line can be discovered through the channel @xmath120 , @xmath119 . models with very light @xmath68 , @xmath121 or very small @xmath24 are excluded by lep . right : the same points shown in the @xmath122 vs @xmath24 plane . all models shown have @xmath27 , @xmath123 and @xmath51 . the black points are consistent with measurements of @xmath124 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath60 . green points violate @xmath18.,title="fig:",width=192 ] phenomenologically speaking , the easiest searches for charged higgs bosons are in the mass range @xmath125 . a light charged higgs will almost exclusively decay to a tau lepton , because the decays to @xmath126 and @xmath127 are kinematically closed in our scenario . this means we can look for top decays to @xmath128 , where there are too many final state taus as compared to muons and electrons . in top pair production we let the second top decay into @xmath129 , where the @xmath130 can decay leptonically and hadronically ( the latter channel dominates the reach in the charged higgs mass ) @xcite . the tau from the charged higgs typically forms a jet . the partial width of a top decay to a charged higgs contains two terms : @xmath131 . hence , there will be large branching fractions both for small and large values of @xmath29 with the weakest point around @xmath132 . because at the lhc these searches are sensitive to anomalous branching fractions of the order of @xmath133 @xcite , we can expect them to probe charged higgs masses up to @xmath134 gev . including off - shell effects , this reach might even be extended by a few gev . at the tevatron the same searches are currently in progress , but are strongly statistics limited @xcite . they only cover anomalous branching fractions of the order of @xmath135 , corresponding to an enhanced yukawa coupling @xmath136 and masses @xmath137 gev . in fig . [ fig : lhc1 ] we overlay our mssm parameter points with the lhc ( atlas ) @xmath138 reach . the magenta line corresponds to the low - luminosity results from anomalous top decays . the slight bending comes from the mass relation between @xmath12 and @xmath17 as well as from the coupling suppression for intermediate @xmath29 . we see that most of the parameter points with two light higgs scalars will be clearly visible in anomalous top decays at the lhc , possibly even at the tevatron . the best - studied regime for charged higgs searches at the lhc is the region of large @xmath139 and charged higgs masses well above the top threshold . following the approximation in eq.([eq : mt_approx ] ) , we see that the required two light scalar higgs bosons imply @xmath140 gev . requiring charged higgs masses above the top threshold translates into @xmath141 gev , or in other words @xmath142 . this could be possible if one stop were much lighter than the top . on the other hand , the @xmath143 approximation is based on the assumption of non - mixing stops , so we postpone this light - stop scenario to the next subsection . the other way of enhancing the mass difference between the average cp - even higgs mass and the charged higgs mass ( beyond the simple @xmath13 approximation ) and to escape the anomalous top decays is to create large bottom yukawa corrections through large values of @xmath29 ( while keeping the @xmath30 parameter small ) . this leads to additional terms contributing to the charged higgs mass squared , the leading terms of which are proportional to @xmath144 . the two kinds of contributions which appear with this mixed yukawa coupling are either proportional to @xmath145 , or they come without this logarithm and are directly proportional to the average stop mass instead @xcite . we see their effect in fig . [ fig : lhc1 ] when we compare the @xmath29 dependence of @xmath12 and @xmath17 . for all value of @xmath29 the requirement of two light higgs scalar limits the pseudoscalar mass to a narrow corridor . we find hardly any susy scenarios with @xmath146 gev . by considering large values of @xmath29 , we can increase the charged higgs mass by 50 gev and avoid the top decay threshold . because large values of @xmath29 are required to at the same time have two light higgs scalars and a heavier charged higgs , we are automatically driven into the region where the charged higgs can be found in @xmath147 production at the lhc . the only issue is the hole which usually appears in the lhc coverage region for charged higgs masses between 160 and 200 gev . as mentioned above , this hole is an artifact of the two search strategies meeting in this mass range . the associated production process and the anomalous ( off - shell ) top decay have to be combined to cover this hole . this kind of study is ongoing in cms and in atlas @xcite and we do not expect this to be a problem once these analyses are actually performed on data . in fig . [ fig : lhc1 ] the densities of points suggests that it is very unlikely to find such a light - higgs mssm scenario where the lhc sees only one neutral higgs scalar and no charged higgs ( we find only one point somewhat close to this region in our scan ) . however , we will study these kinds of parameter points and their discovery prospects at the lhc in more detail next , to ensure there is no hole in the lhc discovery range . , but with parameters varied as described in the text of section [ sec : weird ] . right two panels : the light stop masses for the same set of parameter points . all models shown have @xmath27 , @xmath50 and @xmath51 . the black points are consistent with measurements of @xmath124 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath60 . green points violate @xmath18.,title="fig:",width=192 ] , but with parameters varied as described in the text of section [ sec : weird ] . right two panels : the light stop masses for the same set of parameter points . all models shown have @xmath27 , @xmath50 and @xmath51 . the black points are consistent with measurements of @xmath124 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath60 . green points violate @xmath18.,title="fig:",width=192 ] , but with parameters varied as described in the text of section [ sec : weird ] . right two panels : the light stop masses for the same set of parameter points . all models shown have @xmath27 , @xmath50 and @xmath51 . the black points are consistent with measurements of @xmath124 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath60 . green points violate @xmath18.,title="fig:",width=192 ] as described above , there is this remaining part of the light - scalars parameter space where charged higgs searches in anomalous top decays and in the associated production with a top are challenging . it requires charged higgs masses just above the top threshold and small @xmath29 . note that we do not claim that lhc will not see a charged higgs in this region of parameter space . instead , we emphasize that covering this hole is a crucial task for the near future . we perform a dedicated scan around the one point which appears in fig . [ fig : lhc1 ] close to the hole in the lhc reach for intermediate @xmath148 and @xmath149 gev . we vary each mass parameter up to 50% above or below the value of the single point at @xmath150 , @xmath151 shown in fig . [ fig : lhc1 ] . only @xmath24 we vary over the entire range 1 - 60 . at this specific point , @xmath152 and @xmath153 are both @xmath154 tev , @xmath155 tev , @xmath156 gev , and the bino mass is somewhat light , @xmath157 gev . to see what kind of mssm parameter choice can enhance the charged higgs mass compared to the cp - even scalar masses , we extend the analytical approximation of eq.([eq : mt_approx ] ) to mixing stops @xcite . to avoid the lhc limits , we are now interested in parameter points with as small as possible @xmath29 , so we can neglect the bottom yukawa coupling . instead , we take into account stop mixing . in two limits the expressions are particularly simple : @xmath158 these results point at another way of increasing the charged higgs mass while keeping two light higgs scalars : we choose fairly light stops and increase the trilinear coupling , @xmath159 , or the @xmath30 parameter . we have checked explicitly that for large values of @xmath29 the same can be achieved through increasing @xmath160 . however , keeping both stops light and at the same time increasing one way or another the stop mixing parameters , @xmath161 , means that at least the lighter stop and the lighter sbottom will be very light . this is precisely what we observe in fig . [ fig : lhc2 ] , where we see that the lightest stop mass is always lighter than approximately 340 gev . depending on the neutralino and chargino masses there are two possible stop decays we have to consider when we search for light stops at the tevatron and at the lhc . if the stop is heavier than the lightest chargino , it will decay mostly into @xmath162 . if this decay channel is kinematically forbidden , as the lep2 limits suggest for the run i stop searches at the tevatron , the stop has to decay to @xmath163 . the production cross sections for light stops are large : at the tevatron , a 200 gev stop will be produced with a total cross section of approximately 300 fb , and at the lhc a 300 gev stop will have a production rate around 10 pb . this means that at the lhc , although this kind of mssm parameter point might be hard to find a charged higgs boson , there will be huge numbers of top squarks establishing supersymmetry . ongoing tevatron searches with a projected run ii reach of well above 200 gev ( dependent on the neutralino and chargino masses ) will already severely constrain models with light stops . of the text . for those models in which charged higgs searches at the lhc will be difficult at the lhc ( @xmath164 gev ) , this cross section is rather small , @xmath165 pb . these models each contain a rather light neutralino , however , @xmath166 gev , making most of them accessible to zeplin - max and all of them accessible to super - cdms ( see fig . [ fig : direct ] ) . all models shown have @xmath27 , @xmath123 and @xmath51 . the black points are consistent with measurements of @xmath124 at the 3@xmath2 level and do not violate the tevatron constraint on the @xmath19 branching fraction . blue points violate @xmath19 , but are consistent with @xmath60 . green points violate @xmath18.,width=192 ] one might worry that in these models with lower @xmath24 direct dark matter detection might be more difficult . in fig . [ directweird ] we show that this is in fact the case , with spin - independent elastic scattering cross sections as low as @xmath107 pb . in each of these models , however , there is a rather light neutralino , @xmath167 gev , which makes all of these models observable by super - cdms and many of them observable by zeplin - max , even for @xmath168 pb ( see fig . [ fig : direct ] ) . in this article , we describe the phenomenology of models within the mssm containing cp - even higgs bosons with masses of approximately 98 and 114 gev , motivated by the excesses observed by lep @xcite . we study in detail indirect constraints on such light - higgs models , coming from @xmath71 , @xmath65 and @xmath169 . the first two constraints require fairly large @xmath29 when compared with data , the latter tends to disfavor this regime , and the upcoming tevatron run ii results are going to close in on our light - higgs parameter space . however , it is possible to avoid all indirect constraints by tuning the different weak - scale susy breaking parameters . we find that the kind models we are interested in typically include a light charged higgs boson along with the other light higgs bosons . such a state is observable in anomalous top decays at the tevatron or at the lhc . scenarios with charged higgs above the top threshold are automatically driven in the larger @xmath29 regime , where the charged higgs can be detected in bottom gluon fusion at the lhc . we find exceptionally models which fall between these two mass regions , so that neither of the two searches is optimized and a combination of the two search tactics would be needed . however , these challenging scenarios require one very light top squark ( @xmath170 gev ) , so a large fraction of them will already be ruled out at the tevatron . we also explore the phenomenology for neutralino dark matter in this class of models . the prospects for direct detection are excellent , with the majority of models within this scenario being testable in currently operating experiments , such as cdms - ii . we find that _ all _ models within this class will be testable in next generation direct dark matter detection experiments , such as zeplin - max or super - cdms . the prospects for indirect are also favorable in this scenario . _ acknowledgements _ : we would like to thank prof . gudrun hiller for discussing in detail the indirect constraints from @xmath0 physics and for being of great help . we would like to thank thomas hahn and sven heinemeyer for their help with feynhiggs . dh is supported by the leverhulme trust . the squared amplitudes for the processes , @xmath171 and @xmath172 , averaged over the final state angle are given by @xcite @xmath173 here the label @xmath174 denotes either cp - even higgs boson . @xmath175 , @xmath176 , @xmath177 and @xmath178 are the fermion - fermion - higgs and neutralino - neutralino - higgs couplings . @xmath179 are the widths of the respective higgs bosons . these squared amplitudes can be used to attain the thermally averaged annihilation cross section @xcite @xmath180 t + \mathcal{o}(t^2 ) \nonumber \\ & = & \frac{1}{m^2_{\chi^0 } } \left[1- \frac{3t}{m_{\chi^0 } } \right ] \omega(s)\bigg|_{s\to 4 m^2_{\chi^0}+6 m_{\chi^0 } t } + \mathcal{o}(t^2),\end{aligned}\ ] ] where @xmath181 is the temperature . keeping terms to zeroth and first order in @xmath181 is sufficient for the relic abundance calculation . writing this as an expansion in @xmath182 , @xmath183 , we arrive at @xmath184 ^ 2 \nonumber \\ b_{\chi\chi \to a \to f\bar{f } } & = & 0 \nonumber \\ \to h \to f\bar{f } } & = & 0 \nonumber \\ b_{\chi\chi \to h \to f\bar{f } } & = & \frac{3 g^4_2 c_f m^2_f \sin^2 \alpha } { 16 \pi m^2_w \sin^2 \beta } \frac{(m^2_{\chi^0}-m^2_f ) \sqrt{1-m^2_f / m^2_{\chi^0 } } } { ( 4 m^2_{\chi^0}-m^2_h + i m_h \gamma_h ) ^2 } \nonumber \\ & \times & \left[\epsilon_u \left ( \epsilon_w-\epsilon_b \tan \theta_w \right ) \cos \alpha - \epsilon_d \left ( \epsilon_w- \epsilon_b\tan \theta_w \right ) \sin \alpha \right]^2 \nonumber \to h \to f\bar{f } } & = & 0 \nonumber \\ b_{\chi\chi \to h \to f\bar{f } } & = & \frac{3 g^4_2 c_f m^2_f \cos^2 \alpha } { 16 \pi m^2_w \sin^2 \beta } \frac{(m^2_{\chi^0}-m^2_f ) \sqrt{1-m^2_f / m^2_{\chi^0 } } } { ( 4 m^2_{\chi^0}-m^2_h + i m_h \gamma_h ) ^2 } \nonumber \\ & \times & \left[\epsilon_u \left(\epsilon_w-\epsilon_b \tan \theta_w \right ) \sin \alpha - \epsilon_d \left(\epsilon_w-\epsilon_b \tan \theta_w \right ) \cos \alpha \right]^2 , \label{eq : app}\end{aligned}\ ] ] where @xmath185 is a color factor , equal to 3 for quarks and 1 for leptons . the quantities @xmath186 are the components of the lightest neutralino which are bino , wino , up - type higgsino and down - type higgsino , respectively . these expressions assume that the final state fermions are down - type . for up - type fermions the factor of @xmath187 should be replaced by @xmath188 . r. barate _ [ aleph collaboration ] , phys . b * 565 * , 61 ( 2003 ) ; a. heister _ et al . _ [ aleph collaboration ] , phys . b * 526 * , 191 ( 2002 ) . m. drees , arxiv : hep - ph/0502075 ; for a similar cp violating explanation see m. carena , j. r. ellis , a. pilaftsis and c. e. m. wagner , phys . b * 495 * , 155 ( 2000 ) . see : j. f. gunion and a. turski , phys . d * 39 * , 2701 ( 1989 ) and phys . d * 40 * , 2333 ( 1989 ) ; m. s. berger , phys . d * 41 * , 225 ( 1990 ) ; r. hempfling and a. h. hoang , phys . lett . b * 331 * , 99 ( 1994 ) ; we are using the formulas as in : a. krause , t. plehn , m. spira and p. m. zerwas , nucl . b * 519 * , 85 ( 1998 ) . m. carena , j. r. espinosa , m. quiros and c. e. m. wagner , phys . b * 355 * , 209 ( 1995 ) . s. heinemeyer , w. hollik and g. weiglein , comput . commun . * 124 * , 76 ( 2000 ) . k. abe _ et al . _ [ belle collaboration ] , phys . b * 511 * , 151 ( 2001 ) . d. cronin - hennessy _ et al . _ [ cleo collaboration ] , phys . rev . lett . * 87 * , 251808 ( 2001 ) . r. barate _ [ aleph collaboration ] , phys . b * 429 * , 169 ( 1998 ) . p. l. cho , m. misiak and d. wyler , phys . d * 54 * , 3329 ( 1996 ) ; j. l. hewett and j. d. wells , phys . d * 55 * , 5549 ( 1997 ) ; f. m. borzumati and c. greub , phys . d * 58 * , 074004 ( 1998 ) and phys . d * 59 * , 057501 ( 1999 ) ; g. hiller and f. kruger , phys . d * 69 * , 074020 ( 2004 ) . l. j. hall , r. rattazzi and u. sarid , phys . d * 50 * , 7048 ( 1994 ) ; m. carena , m. olechowski , s. pokorski and c. e. wagner , nucl . b * 426 * , 269 ( 1994 ) ; m. carena , d. garcia , u. nierste and c. e. wagner , nucl . b * 577 * , 88 ( 2000 ) ; j. guasch , p. hafliger and m. spira , phys . rev . d * 68 * , 115001 ( 2003 ) . j. l. hewett , s. nandi and t. g. rizzo , phys . d * 39 * , 250 ( 1989 ) ; m. j. savage , phys . b * 266 * , 135 ( 1991 ) ; c. bobeth , t. ewerth , f. kruger and j. urban , phys . rev . d * 64 * , 074014 ( 2001 ) ; a. dedes , h. k. dreiner , u. nierste and p. richardson , arxiv : hep - ph/0207026 . d. acosta _ et al . _ [ cdf collaboration ] , phys . rev . lett . * 93 * , 032001 ( 2004 ) ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . rev . lett . * 94 * , 071802 ( 2005 ) see : m. calvi , talk at rencontres de moriond , 2005 . g. w. bennett _ et al . _ [ muon g-2 collaboration ] , phys . lett . * 92 * , 161802 ( 2004 ) ; r. r. akhmetshin _ et al . _ [ cmd-2 collaboration ] , phys . b * 578 * , 285 ( 2004 ) . a. aloisio _ et al . _ [ kloe collaboration ] , phys . b * 606 * , 12 ( 2005 ) ; g. rodrigo , h. czyz , j. h. kuhn and m. szopa , eur . j. c * 24 * , 71 ( 2002 ) . m. davier , s. eidelman , a. hocker and z. zhang , eur . j. c * 31 * , 503 ( 2003 ) ; f. jegerlehner , nucl . suppl . * 126 * , 325 ( 2004 ) ; k. hagiwara , a. d. martin , d. nomura and t. teubner , phys . d * 69 * , 093003 ( 2004 ) . m. krawczyk , arxiv : hep - ph/0103223 . a. czarnecki and w. j. marciano , phys . d * 64 * , 013014 ( 2001 ) ; s. p. martin and j. d. wells , phys . d * 64 * , 035003 ( 2001 ) ; s. p. martin and j. d. wells , phys . d * 67 * , 015002 ( 2003 ) . g. belanger , f. boudjema , a. pukhov and a. semenov , arxiv : hep - ph/0405253 . for a review of neutralinos and other particle dark matter candidates , see : g. bertone , d. hooper and j. silk , phys . rept . * 405 * , 279 ( 2005 ) . exact analytic cross sections for neutralino - neutralino annihilations and coannihilations into two and three body states are given in : a. birkedal - hansen and e. h. jeong , jhep * 0302 * , 047 ( 2003 ) . j. r. ellis , t. falk , k. a. olive and m. srednicki , astropart . * 13 * , 181 ( 2000 ) [ erratum - ibid . * 15 * , 413 ( 2001 ) ] . c. l. bennett _ _ , astrophys . j. suppl . * 148 * , 1 ( 2003 ) . d. s. akerib _ et al . _ [ cdms collaboration ] , phys . lett . * 93 * , 211301 ( 2004 ) . gerda proposal , http://wwwgerda.mppmu.mpg.de/ p. gondolo , j. edsjo , p. ullio , l. bergstrom , m. schelke and e. a. baltz , jcap * 0407 * , 008 ( 2004 ) . l. bergstrom , j. edsjo and p. ullio , phys . lett . * 87 * , 251301 ( 2001 ) ; d. hooper and b. l. dingus , phys . d * 70 * , 113007 ( 2004 ) ; l. bergstrom , p. ullio and j. h. buckley , astropart . * 9 * , 137 ( 1998 ) . e. a. baltz and j. edsjo , phys . d * 59 * , 023511 ( 1999 ) s. profumo and p. ullio , jcap * 0407 * , 006 ( 2004 ) ; d. hooper and j. silk , phys . rev . d * 71 * , 083503 ( 2005 ) . t. plehn , d. l. rainwater and d. zeppenfeld , phys . b * 454 * , 297 ( 1999 ) and phys . d * 61 * , 093005 ( 2000 ) . s. abdullin _ et al . . phys . j. c * 39s2 * ( 2005 ) 41 ; s. asai _ et al . . phys . j. c * 32s2 * , 19 ( 2004 ) ; for a recent update see k. cranmer , talk at north american atlas physics workshop t. plehn , d. l. rainwater and d. zeppenfeld , phys . rev . lett . * 88 * , 051801 ( 2002 ) . e. boos , a. djouadi and a. nikitenko , phys . b * 578 * , 384 ( 2004 ) k. a. assamagan and y. coadou , acta phys . polon . b * 33 * , 707 ( 2002 ) ; k. a. assamagan , y. coadou and a. deandrea , eur . j. directc * 4 * , 9 ( 2002 ) ; d. denegri _ et al . _ , arxiv : hep - ph/0112045 ; s. raychaudhuri and d. p. roy , phys . d * 53 * , 4902 ( 1996 ) ; r. m. barnett , h. e. haber and d. e. soper , nucl . phys . b * 306 * , 697 ( 1988 ) ; s. h. zhu , phys . d * 67 * , 075006 ( 2003 ) ; t. plehn , phys . d * 67 * , 014018 ( 2003 ) . f. borzumati , j. l. kneur and n. polonsky , phys . d * 60 * , 115011 ( 1999 ) ; e. l. berger , t. han , j. jiang and t. plehn , arxiv : hep - ph/0312286 . k. a. assamagan , m. guchait and s. moretti , arxiv : hep - ph/0402057 . j. a. coarasa , j. guasch , j. sola and w. hollik , phys . b * 442 * , 326 ( 1998 ) . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . * 88 * , 151803 ( 2002 ) . c. biscarat , atl - slide-2003 - 002 ; s. banerjee and m. maity , j. phys . g * 28 * , 2443 ( 2002 ) ;

it is shown that signal energy is the only available degree - of - freedom ( dof ) for fiber - optic transmission as the input power tends to infinity . with signal dofsat the input , dofs are asymptotically lost to signal - noise interactions . the main observation is that , nonlinearity introduces a multiplicative noise in the channel , similar to fading in wireless channels . the channel is viewed in the spherical coordinate system , where signal vector is represented in terms of its norm and direction . the multiplicative noise causes signal direction to vary randomly on the surface of the unit-sphere in , in such a way that the effective area of the support of does not vanish as . on the other hand , the surface area of the sphere is finite , so that carries finite information . this observation is used to show several results . firstly , let be the capacity of a discrete - time periodic model of the optical fiber with distributed noise and frequency - dependent loss , as a function of the average input power . it is shown that asymptotically as , , where is the dimension of the input signal space and is a bounded number . in particular , in finite - dimensional periodic models . secondly , it is shown that capacity saturates to a constant in infinite - dimensional models where . an expression is provided for the constant , by showing that , as the input , the action of the discrete periodic stochastic nonlinear schrdinger equation tends to multiplication by a random matrix ( with fixed distribution , independent of input ) . thus , perhaps counter - intuitively , noise simplifies the nonlinear channel at high powers to a _ linear _ multiple - input multiple - output fading channel . as signal - noise interactions gradually reduce the slope of the , to a point where increasing the input power returns diminishing gains . nonlinear frequency - division multiplexing can be applied to approach capacity in optical networks , where linear multiplexing achieves low rates at high powers .

1610.06458

several decades since the introduction of the optical fiber , channel capacity at high powers remains a vexing conundrum . existing achievable rates saturate at high powers because of linear multiplexing and treating the resulting interference as noise in network environments @xcite . furthermore , it is difficult to estimate the capacity via numerical simulations , because channel has memory . multi - user communication problem for ( an ideal model of ) optical fiber can be reduced to single - user problem using the nonlinear frequency - division multiplexing ( nfdm ) @xcite . this addresses deterministic distortions , such as inter - channel and inter - symbol interference ( signal - signal interactions ) . the problem is then reduced to finding the capacity of the point - to - point optical fiber set by noise . there are two effects in fiber that impact shannon capacity in point - to - point channels . ( 1 ) phase noise . nonlinearity transforms additive noise to phase noise in the channel . as the amplitude of the input signal tends to infinity , the phase of the output signal tends to a uniform random variable in the zero - dispersion channel ( * ? ? ? * section iv ) . as a result , phase carries finite information in the non - dispersive fiber . ( 2 ) multiplicative noise . dispersion converts phase noise to amplitude noise , introducing an effect which at high powers is similar to fading in wireless channels . importantly , the conditional entropy grows strongly with input signal . in this paper , we study the asymptotic capacity of a discrete - time periodic model of the optical fiber as the input power tends to infinity . the role of the nonlinearity in point - to - point discrete channels pertains to signal - noise interactions , captured by the conditional entropy . the main result is the following theorem , describing capacity - cost function in models with constant and non - constant loss ; see definition [ def : loss ] . consider the discrete - time periodic model of the nls channel described in section [ sec : mssfm ] , with non - zero dispersion . capacity is asymptotically rcl c(p)= ( p)+c , & non - constant loss , + p+c , & constant loss , where @xmath0 is dimension of the input signal space , @xmath11 is the average input signal power and @xmath16 . in particular , @xmath17 in finite - dimensional models . intensity modulation and direct detection ( photon counting ) is nearly capacity - achieving in the limit @xmath11 , where capacity is dominated by the first terms in @xmath9 expressions . [ thm : main ] from the theorem [ thm : main ] and ( * ? ? ? * theorem 1 ) , the asymptotic capacity of the dispersive fiber is much smaller than the asymptotic capacity of ( the discrete - time model of ) the zero - dispersion fiber , which is @xmath18 , @xmath19 . dispersion reduces the capacity , by increasing the conditional entropy . with @xmath0 dofs at the input , @xmath1 dofs are asymptotically lost to signal - noise interactions , leaving signal energy as the only useful dof for transmission . there are a finite number of dofs in all computer simulations and physical systems . however , as a mathematical problem , the following corollary holds true . capacity saturates to a constant @xmath19 in infinite - dimensional models , including the continuous - time model . [ cor : inf ] the power level where signal - noise interactions begin to appreciably impact the slope of the @xmath9 is not determined in this paper . numerical simulations indicate that the conditional entropy does not increase with input in the nonlinear fourier domain , for a range of power larger than the optimal power in wavelength - division multiplexing ( * ? ? ? * fig . 9 ( a ) ) . in this regime , signal - noise interactions are weak and the capacity is dominated by the ( large ) number @xmath13 in the theorem [ thm : main ] . a numerical estimation of the capacity of the point - to - point fiber at input powers higher than those in fig . [ fig : nfdm ] should reveal the impact of the signal - dependent noise on the asymptotic capacity . the contributions of the paper are presented as follows . the continuous - time model is discretized in section [ sec : mssfm ] . the main ingredient is a modification of the split - step fourier method ( ssfm ) that shows noise influence more directly compared with the standard ssfm . a _ unit _ is defined in the modified ssfm ( mssfm ) model that plays an important role throughout the paper . the mssfm and units simplify the information - theoretic analysis . theorem [ thm : main ] and corollary [ cor : inf ] are proved in section [ sec : proof1 ] . the main ingredient here is an appropriate partitioning of the dofs in a suitable coordinate system , and the proof that the achievable rate of one group of dofs is bounded in input . no assumption is made on input power in this first proof . theorem [ thm : main ] is proved again in section [ sec : proof2 ] by considering the limit @xmath11 , which adds further intuition . firstly , it is shown that , as the input @xmath8 , the action of the discrete periodic stochastic nonlinear schrdinger ( nls ) equation tends to multiplication by a random matrix ( with fixed probability distribution function ( pdf ) , independent of the input ) . as a result , perhaps counter - intuitively , as @xmath8 noise simplifies the nonlinear channel to a _ linear _ multiple - input multiple - output ( non - coherent ) fading channel . secondly , the asymptotic capacity is computed , without calculating the conditional pdf of the channel , entropies , or solving the capacity optimization problem . because of the multiplicative noise , the asymptotic rate depends only on the knowledge that whether channel random operator has any deterministic component . the conditional pdf merely modifies the bounded number @xmath13 in the theorem [ thm : main ] . note that we do not apply local analysis based on perturbation theories ( valid in the low power regime ) . the proof of the theorem [ thm : main ] , _ e.g. _ , the asymptotic loss of dofs , is based on a global analysis valid for any signal and noise ; see section [ sec : proof1 ] . the notation in this paper is motivated by @xcite . upper- and lower - case letters represent scalar random variables and their realizations , _ e.g. _ , @xmath20 and @xmath21 . the same rule is applied to vectors , which are distinguished using underline , _ e.g. _ , @xmath22 for a random vector and @xmath23 for a deterministic vector . deterministic matrices are shown by upper - case letter with a special font , _ e.g. _ , @xmath24 . random matrices are denoted by upper - case letters with another special font , _ e.g. _ , @xmath25 . important scalars are distinguished with calligraphic font , _ e.g. _ , @xmath10 for power and @xmath26 for capacity . the field of real and complex numbers is respectively @xmath27 and @xmath28 . a sequence of numbers @xmath29 is sometimes abbreviated as @xmath30 , @xmath31 . a zero - mean circularly - symmetric complex gaussian random vector with covariance matrix @xmath32 is indicated by @xmath33 . uniform distribution on interval @xmath34 is designated as @xmath35 . throughout the paper , the asymptotic equivalence @xmath36 , often abbreviated by saying `` asymptotically , '' means that @xmath37 . letter @xmath38 is reserved to denote a real number bounded in @xmath0 and @xmath10 . a sequence of independent and identically distributed ( ) random variables @xmath39 drawn from the pdf @xmath40 is presented as @xmath41 . the identity matrix with size @xmath0 is @xmath42 . the euclidean norm of a vector @xmath43 is rcl |x|=(|x_1|^2++|x_n|^2)^. this gives rise to an induced norm @xmath44 for matrix @xmath45 . we use the spherical coordinate system in the paper . here , a vector @xmath43 is represented by its norm @xmath46 and direction @xmath47 ( with convention @xmath48 if @xmath49 ) . the direction can be described by @xmath50 angles . when direction is random , its entropy can be measured with respect to the spherical measure @xmath51 , @xmath52 , where @xmath53 is the @xmath54sphere rcl s^m=\{^m+1 : ||=1 } . it is shown in the appendix [ app : one ] that the differential entropy with respect to the lebesgue and spherical measures , denoted respectively by @xmath55 and @xmath56 , are related as rcl h(x)=h(||)+h_(| ||)+m|x| . [ eq : sph - leb ] the entropy power of a random direction @xmath57 is rcl ( ) = ( h _ ( ) ) . it represents the effective area of the support of @xmath4 on @xmath53 . signal propagation in optical fiber is described by the stochastic nonlinear schrdinger ( nls ) equation ( * ? ? ? 2 ) rcl = l_l(q)+l_n(q)+n(t , z ) , [ eq : nls ] where @xmath58 is the complex envelope of the signal as a function of time @xmath59 and space @xmath60 and @xmath61 is zero - mean circularly - symmetric complex gaussian noise with rcl ( n(t , z)n^*(t,z))=^2_w(t - t)(z - z ) , where @xmath62 , @xmath63 , and @xmath64 is noise bandwidth . the operator @xmath65 represents linear effects rcl l_l(q)= _ k=0^j^k+1 -_r(t , z)q(t , z ) , [ eq : l - l ] where @xmath66 are dispersion coefficients , @xmath67 is convolution and @xmath68 is the residual fiber loss . the operator @xmath69 represents kerr nonlinearity , where @xmath70 is the nonlinearity parameter . the average power of the transmit signal is rcl p= _ t_-t/2^t/2|q(t,0)|^2 t . [ eq : power - cont ] the residual loss in accounts for uncompensated loss and non - flat gain of the raman amplification in distance and is generally frequency dependent . the constant loss model refers to the case where @xmath71 is constant in the frequency @xmath72 , _ i.e. _ , @xmath73 , where @xmath74 is the fourier transform with respect to @xmath75 . in realistic systems , however , loss varies over frequency , polarization or spatial models . this is the non - constant loss model . channel filters act similar to a non - constant loss function . [ def : loss ] we discretize in space and time . divide a fiber of length @xmath76 into a cascade of a large number @xmath77 of pieces of discrete fiber segments of length @xmath78 ( * ? ? ? * section iii . a small segment can be discretized in time and modeled in several ways . an appropriate approach is given by the split - step fourier method ( ssfm ) . the standard ssfm splits the _ deterministic _ nls equation into linear and nonlinear parts . in applying ssfm to the _ stochastic _ nls equation , typically noise is added to the signal . we introduce a modified split - step fourier method where , instead of noise addition , the nonlinear part of is solved in the presence of noise analytically . in the linear step , is solved with @xmath79 . in the discrete - time model , linear step in a segment of length @xmath80 consists of multiplying a vector @xmath2 by the dispersion - loss matrix @xmath81 . in the constant loss model , @xmath82 , where @xmath83 is a unitary matrix . in the absence of loss , @xmath84 is unitary . the values of @xmath85 depend on the dispersion coefficients , @xmath80 and @xmath0 . in general , all entries of @xmath84 are non - zero , although in a small segment , the off - diagonal elements can be very small . matrix @xmath84 is fully dispersive , _ i.e. _ , @xmath86 , for all @xmath87 . [ ass : u ] in the nonlinear step , is solved with @xmath88 resulting in ( * ? ? ? 12 ) , ( * ? ? ? 30 ) : rcl q(t , z)=(q(t,0)+w(t , z))e^j(t , z ) , [ eq : zd ] in which rcl ( t , z)= _ 0^z|q(t,0)+w(t , l)|^2l , where @xmath89 is wiener process . the modified nonlinear step in the mssfm is obtained by discretizing . divide a small segment @xmath90 into @xmath91 sub - segments of length @xmath92 . define @xmath93 as rcl ( x , n)&=&_signal - noise interactions , unknown + + & & + _ conditionally known , [ eq : phase ] where @xmath94 , @xmath95 . the nonlinear step in a segment of length @xmath80 maps vector @xmath2 to vector @xmath96 , according to rcl y_k=(x_k+n_k1++n_kl)e^j(x_k , n_k ) , [ eq : discrete - zd ] where @xmath97 , @xmath98 . the nonlinear step is a deterministic phase change in the ssfm . in this form , nonlinearity is entropy - preserving and does not interact with noise immediately ( * ? ? ? * lemma 23 ) unless several steps in the ssfm are considered , which complicates the analysis . in the mssfm , noise is introduced in a distributed manner within each nonlinear step . this shows noise influence more directly . note that , conditioned on @xmath99 , the last term in is known . other terms in represent signal - noise interactions . they are conditionally unknown and are responsible for capacity limitation . the mssfm model for a fiber of length @xmath76 consists of the cascade of linear and modified nonlinear steps ( without noise addition between them ) . a _ unit _ in the mssfm model is defined as the cascade of three segments of length @xmath80 : a modified nonlinear step @xmath100 , followed by a linear step @xmath101 , followed by another modified nonlinear step @xmath102 ; see fig . [ fig : mssfm ] . a unit of length @xmath103 is the smallest piece of fiber whose capacity behaves qualitatively similar to the capacity of the full model with length @xmath76 . in the appendix [ app : in - out - mssfm ] it is shown that the input output relation @xmath104 in one unit is given by rcl y = mx+z,[eq : one - seg ] where @xmath105 is a random matrix with entries rcl m_kl= r_kle^j_k+j_l , [ eq : mkl ] in which rcl _ l=(x_l , n_l^1),_k=(v_k , n_k^2 ) . here @xmath106 and @xmath107 are @xmath108 gaussian ensembles with entries drawn from @xmath109 , independent of any other random variable . the additive noise @xmath110 is in general non - gaussian but bounded in @xmath3 ; see . finally , vector @xmath111 is the output of the linear step in fig . [ fig : mssfm ] . the input output relation @xmath104 in a fiber of length @xmath76 is obtained by composing @xmath112 blocks @xmath113 : rcl y(k)=m(k)x(k)+z(k ) , [ eq : m - seg ] where @xmath114 is the transmission index , @xmath115 is an stochastic process , and rcl m(k)= _ k=1^|m rm_k , z(k)=rz_|m+_k=1^|m-1 ( _ l = k+1^|m rm_l)rz_k . [ eq : m - m - seg ] the power constraint is discretized to @xmath116 in the discrete - time model . bandwidth , spectral broadening and spectral efficiency in the continuous - time model are discussed in section [ sec : cor ] . note that @xmath117 is a nonlinear random operator . particularly , it depends on input . dimension of the input space is @xmath0 . to approximate the continuous - time model , @xmath118 however , we let @xmath0 be arbitrary , _ e.g. _ , @xmath119 . dimension should not be confused with codeword length that tends to infinity . we first illustrate the main ideas of the proof via elementary examples . consider the additive white gaussian noise ( awgn ) channel @xmath120 , where @xmath121 is input , @xmath122 is output and @xmath123 is noise . applying chain rule to the mutual information rcl i(x;y)=i(x;y)+i(x;y|y ) , where @xmath124 denotes phase . the amplitude channel @xmath125 is @xmath126 where @xmath127 and @xmath128 . it asymptotically contributes rcl i(x ; |y|)p+c to the capacity . phase , on the other hand , is supported on the finite interval @xmath129 . the only way that the contribution of the phase to the capacity could tend to infinity is that , phase noise tends to zero on the circle as @xmath130 . indeed , rcl y&=&x+^-1 ( ) + & & x+ , where @xmath131 . the output entropy is clearly bounded , @xmath132 . however , rcl h(y|x , y)&=&h(z_i)-|x| + & & -p+c , asp . [ eq : cond - ent - awgn ] note that the differential entropy can be negative . the contribution of the phase to the mutual information is rcl i(x;y|y)p+c. condition implies @xmath133 , _ i.e. _ , the effective phase noise on the unit circle asymptotically vanishes . now consider the fading channel @xmath134 , where @xmath121 is input , @xmath122 is output and @xmath135 . to prepare for generalization to optical channel , we represent a complex scalar @xmath20 as @xmath136 . thus @xmath137 , where rcl = m_r & -m_i + m_i & m_r , z= z_r + z_i , in which @xmath138 . as @xmath139 , @xmath140 , @xmath141 , and randomness in @xmath142 does not vanish with @xmath143 . formally , rcl h_(| x , ||)&= & h_(| m^-1y,|| ) + & = & h_(| m^-1 , || ) + & > & - , [ eq : cond - entr - fad ] where follows because @xmath144 does not determine @xmath142 for random @xmath145 : there are four random variables @xmath146 and @xmath147 for three equations @xmath148 and @xmath149 . as a result , @xmath150 , and @xmath151 is the only useful dof at high powers , in the sense that its contribution @xmath152 to the mutual information @xmath153 tends to infinity with @xmath3 . the zero - dispersion optical fiber channel is similar to the fading channel at high powers . the trivial condition rcl h((.,z ) | q(.,0 ) , |q(.,z)|)>- , q(.,0 ) , is sufficient to prove that the capacity of is asymptotically the capacity of the amplitude channel , namely @xmath18 . the intuition from the awgn , fading and zero - dispersion channels suggests to look at the dispersive optical channel in the spherical coordinate system . the mutual information can be decomposed using the chain rule rcl i(q(0 ) ; q(z))&=&i(|q(0)| ; q(z))+i((0 ) ; q(z)||q(0)| ) + & = & i(|q(0)| ; |q(z)|)+i(|q(0)| ; ( z)||q(z)| ) + & & + i((0 ) ; q(z)||q(0)| ) , [ eq : i3 ] where we dropped time index in @xmath58 . the first term in is the rate of a single - input single - output channel which can be computed in the asymptotic limit as follows . let @xmath22 and @xmath154 represent discretizations of the input @xmath155 and output @xmath156 . consider first the lossless model . in this case , @xmath145 is unitary and from , and rcl ||^2&=&|+|^2 + & = & |+^|^2 + & = & |+|^2 , [ eq : chi - squared ] where @xmath157 is the adjoint ( nonlinear ) operator and follows because @xmath158 and @xmath159 are identically distributed when @xmath160 ; see appendix [ app : in - out - mssfm ] . thus @xmath161 is a non - central chi - square random variable with @xmath162 degrees - of - freedom and parameter @xmath163 . the non - central chi - square conditional pdf @xmath164 can be approximated at large @xmath165 using the gaussian pdf , giving the asymptotic rate rcl i(|| ; ||)p+c . [ eq : i(|x|;|y| ) ] the bounded number @xmath13 can be computed using the exact pdf . the case @xmath166 is similar to the lossless case . here @xmath167 , where @xmath168 is a random unitary operator . thus , @xmath169 ; furthermore @xmath170 is deterministic . the loss simply influences the signal power , modifying constant @xmath13 in . in the non - constant loss model , loss interacts with nonlinearity , dispersion and noise . here , @xmath171 is a random variable , and rcl ||=|||||+| , [ eq : multi - chan ] where @xmath172 . taking logarithm rcl ||=||+||+|+| . [ eq : log - fading ] applying lemma [ lemm : decomposition ] , we can assume @xmath173 for a suitable @xmath174 without changing the asymptotic capacity . the last term in is a bounded real random variable because rcl _ ||=1 , ||=1 |+|^2<. thus , the logarithm transforms the channel with multiplicative noise @xmath171 to the channel with additive bounded noise . the asymptotic capacity , independent of the pdf of @xmath171 , is rcl i(|| ; || ) & & ( ( |x|))^2+c + & = & p+c. the last two terms in are upper bounded in one unit of the mssfm using the data processing inequality rcl i(q(0 ) ; q(z)||q(0)| ) & & i(q(0 ) ; q(3)||q(0)| ) , [ eq : dp1 ] + i(|q(0)| ; ( z)| |q(z)| ) & & i(|q(0)| ; q(3)||q(3)| ) . [ eq : dp2 ] we prove that the upper bounds in do not scale with input @xmath175 . let @xmath176 denote discretization of @xmath177 and @xmath178 . in one unit of the mssfm rcl _ ||i ( ; y||| ) & < & , [ eq : hatx - y ] + _ ||i(|| ; |||)&<&. [ eq : absx - haty ] , does not vanish as @xmath179.,scaledwidth=20.0% ] consider first the lossless model , where @xmath145 is a unitary operator . from lemma [ lemm : decomposition ] , as @xmath180 , the additive noise in can be ignored . thus @xmath181 . to prove , rcl i ( ; y|||)&=&i(;||||| ) + & & i(;||| ) + & = & h_(|||)-h_(| , || ) . step @xmath182 follows from the identity rcl i(x ; zy|z)=i(x ; y|z),z0 . [ eq : i - indentity ] we measure the entropy of @xmath183 with respect to the spherical probability measure @xmath184 , @xmath50 , on the surface of the unit sphere @xmath185 . from the maximum entropy theorem ( met ) for distributions with compact support , rcl h_(|||)a_n , where @xmath186 is the surface area of @xmath187 , in which @xmath188 is the gamma function . we next show that the conditional entropy @xmath189 does not tend to @xmath190 with @xmath46 . the volume of the spherical sector in fig . [ fig : spherical - sector ] vanishes if and only if the corresponding area on the surface of the sphere vanishes . this can be formalized using identity . let @xmath191 , where @xmath192 independent of @xmath22 and @xmath154 . from rcl h_(| , ||)=h ( w| , ||)-h(u)-mu . [ eq : interm ] applying chain rule to the differential entropy rcl h(w|.)&=&_k=1^n h(w_k|w^k-1 , . ) + & = & _ k=1^n h(w_k , | w^k-1 , . ) [ eq : chain - rule1 ] + & & + _ k=1^n h(|w_k| |w^k-1 , w_k , . ) , [ eq : chain - rule2 ] where entropy is conditioned on @xmath46 and @xmath4 . for the phase entropies in , note that , from , @xmath193 contains random variable @xmath194 with finite entropy , which does not appear in @xmath195 . formally , @xmath196 for some function @xmath197 , which can be determined from . thus rcl h(w_k| w^k-1 , .)&= & h(_k+f(^n,)|w^k-1 , . ) + & & h(_k+f(^n,))|w^k-1,^k-1,^n , u , . ) + & & h(_k+f(^n,))|^k-1,^n , u , . ) + & & h(_k|^k-1,^n , . ) + & > & - . [ eq : phase - entropies ] step @xmath182 follows from the rule that conditioning reduces the entropy . step @xmath198 holds because @xmath195 is a function of @xmath199 . step @xmath200 follows because @xmath201 determines @xmath202 . for the amplitude entropies in , we explain the argument for @xmath203 : rcl w_k = ue^j_k&&(r_k1x_1e^j_1+r_k2x_2e^j_2+r_k3x_3e^j_3 ) , [ eq : ys ] where @xmath204 . noise addition in implies @xmath205 , @xmath206 ; we thus assume @xmath207 for all @xmath208 . it is clear that @xmath209 . there are 5 random variables @xmath210 , @xmath211 , @xmath212 for two amplitude and phase relations in the @xmath213 equation in . given @xmath213 and @xmath214 , there are 6 random variables and three equations . one could , for instance , express @xmath212 in terms of @xmath210 and @xmath215 . this leaves free at least @xmath210 in @xmath216 , giving rcl h(|w_2|| w_1 , w_2,.)&&h(u)+c + & > & - . the last equation for @xmath217 adds one random variable @xmath218 and one equation for @xmath219 . together with the equation for @xmath220 , the number of free random variables , defined as the number of all random variables minus the number of equations , is 2 ; thus rcl h(|w_3|| w_1 , w_2 , w_3,.)>- . in a similar way , in general , there are @xmath221 random variables in @xmath222 and @xmath223 equations in @xmath224 , resulting in @xmath225 free random variables . thus rcl h(|w_k|| w^k-1 , w_k)>-,1kn . [ eq : amp - entropies ] substituting and into , we obtain @xmath226 . finally , from rcl h_(| , ||)>- . the proof for lossy models , and , is similar . loss changes matrix @xmath84 , which has no influence on our approach to proving the boundedness of terms in . the essence of the above proof is that , as @xmath180 , the additive noise in gets smaller relative to the signal , but phase noise ( and thus randomness in @xmath145 ) does not decrease with @xmath46 . furthermore , @xmath145 has enough randomness , owing to the mixing effect of the dispersion , so that all @xmath227 angles representing signal direction in the spherical coordinate system are random variables that do not vanish with @xmath46 . for some special cases of the dispersion - loss matrix @xmath84 , it is possible to obtain deterministic components in @xmath142 as @xmath180 . these are cases where mixing does not fully occur , _ e.g. _ , @xmath228 . in the mssfm , however , @xmath84 is arbitrary , due to , _ e.g. _ , step size @xmath80 . we fix the power constraint and let @xmath118 in the definition of the capacity . the logarithmic terms depending on @xmath10 in the theorem [ thm : main ] approach zero , so that @xmath229 . consider now the continuous - time model . we discretize the channel in the frequency domain , according to the approach in @xcite . as the time duration @xmath230 in ( * ? ? ? * section ii ) , we obtain a discrete - time model with infinite number of dofs ( fourier modes ) in any frequency interval at @xmath231 . therefore , @xmath229 in the corresponding discrete - time periodic model . it is shown in ( * ? ? ? * section viii ) that , because of the spectral broadening , the capacity of the continuous - time model @xmath232 can be strictly lower than the capacity of the discrete - time model @xmath233 . since @xmath234 , and @xmath235 , we obtain @xmath236 . we do not quantify constant @xmath237 in the continuous - time model , which can be much lower than the constant @xmath13 in the discrete - time model , due to spectral broadening ( potentially , @xmath238 ) . a crude estimate , based on the carson bandwidth rule , is given in ( * ? ? ? * section viii ) for the zero - dispersion channel . to summarize , se is bounded in input power in the continuous - time model with @xmath15 ( with or without filtering ) . the extent of the data rate loss due to the spectral broadening ( @xmath237 versus @xmath13 ) remains an open problem . in this section it is shown that , as @xmath8 , the action of the discrete - time periodic stochastic nls equation tends to multiplication by a random matrix ( with fixed pdf , independent of the input ) . noise simplifies the nls channel to a _ linear _ multiple - input multiple - output non - coherent fading channel . this section also proves theorems [ thm : main ] in an alternative intuitive way . the approach is based on the following steps . _ step 1 ) _ in section [ sec : decomposition ] , the input signal space is partitioned into a bounded region @xmath239 and its complement @xmath240 . it is shown that the overall rate is the interpolation of rates achievable using signals in @xmath241 . lemma [ lemm : i < infty ] is proved , showing that the contribution of @xmath239 to the mutual information is bounded . suitable regions @xmath241 are chosen for the subsequent use . _ step 2 ) _ in section [ sec : fading - model ] , it is shown that for all @xmath242 , the nonlinear operator @xmath243 is multiplication by a uniform phase random variable , _ i.e. _ , rcl l_n(q)= j(t , z ) q , t , z , where@xmath244 . in other words , for input signals in @xmath240 the stochastic nls equation is a simple linear channel with additive and multiplicative noise rcl = l_l(q)+j ( t , z)q+n(t , z ) . [ eq : multiplicative ] discretizing , we obtain that optical fiber is a fading channel when input is in @xmath240 : rcl y&=&x+z,^2p , [ eq : y = hx+n ] in which @xmath145 is a random matrix of the form rcl m = _ k=1^m rd_k , d_k=(e^j_ki ) , [ eq : m - expr ] where @xmath245 and @xmath158 is noise rcl z=_k=1^m(_l=1^kr d_l)z_k , z_k~ _ . in general , @xmath145 and @xmath158 are non - gaussian . however , in the constant loss model , @xmath246 where @xmath247 , @xmath248 . note that @xmath145 and @xmath158 have fixed pdfs , independent of @xmath22 . summarizing , the channel law is rcl p(y|x)= given by the nls equation , & xr^- , + p(mx+z|x ) , & xr^+ . [ eq : law ] _ step 3 ) _ in section [ sec : asymptotic - capacity ] , the capacity of the multiplicative - noise channel is studied . lemma [ lem : cap - y = hx+n ] and [ lem : h(mx ) ] are proved showing that , for any @xmath145 that does not have a deterministic component and is finite ( see ) , the asymptotic capacity is given by the theorem [ thm : main ] . importantly , the asymptotic rate is nearly independent of the pdf of @xmath145 , which impacts only the bounded number @xmath13 . finally , lemma [ lem : h(m - fiber)>-infty ] is proved showing that the random matrix underlying the optical fiber at high powers meets the assumptions of the lemma [ lem : cap - y = hx+n ] . an expression is provided for @xmath13 , which can be evaluated , depending on the pdf of @xmath145 . we begin by proving the following lemma , which is similar to the proof approach in @xcite , where the notion of satellite constellation is introduced . let @xmath249 , @xmath250 , be a conditional pdf . define rcl x= x_1 , & with probability , + x_2 , & with probability 1- , where @xmath251 and @xmath252 are random variables in @xmath253 and @xmath254 . then rcl r_1+(1-)r_2rr_1+(1-)r_2+h ( ) , [ eq : r1-r2-h ] where @xmath255 , @xmath256 and @xmath257 are , respectively , mutual information of @xmath258 , @xmath259 and @xmath20 , and @xmath260 is the binary entropy function , @xmath261 . [ lemm : decomposition ] the pdf of the time sharing random variable @xmath22 and its output @xmath154 are rcl p_x(x)&=&p_x_1(x)+(1-)p_x_2(x ) , + p_y(y)&=&p_y_1(y)+(1-)p_y_2(y ) , [ eq : py = py1+py2 ] where rcl p_y_1,y_2(y)=p(y|x)p_x_1,x_2(x)x . by elementary algebra rcl i(x ; y ) = i(x_1;y_1)+(1- ) i(x_2,y_2)+i , where rcl i & = & d ( p_y_1(y_1)||p_y(y ) ) + ( 1- ) d ( p_y_2(y_2)||p_y(y ) ) . from , @xmath262 which gives @xmath263 . from @xmath264 , @xmath265 , giving @xmath266 . thus rcl i ( x;y ) & & i ( x_1;y_1)+(1- ) i(x_2;y_2)+h ( ) , + i(x;y ) & & i ( x_1;y_1)+(1- ) i(x_2;y_2 ) . define rcl |r=_ni(x;y ) . with definitions in the lemma [ lemm : decomposition ] , we have @xmath267 . [ cor : rate - interpolation ] for the rest of the paper , we choose @xmath239 to be an @xmath0-hypercube in @xmath268 rcl r^-_=\{x^n | |x_k| < , 1kn } , and @xmath269 . we drop the subscript @xmath270 when we do not need it . the following lemma shows that , if @xmath271 , the contribution of the signals in @xmath272 to the mutual information in the nls channel is bounded . let @xmath2 be a random variable supported on @xmath273 and @xmath271 . for the nls channel rcl i(x;y)<. [ lemm : i < infty ] from the met , @xmath274 . let @xmath275 be a markov chain , where @xmath276 is independent of @xmath158 and @xmath277 . then @xmath278 . applying this to the nls channel with an independent noise addition in the last stage , we obtain @xmath279 . alternatively , from @xcite , @xmath280 [ lem : r- ] we begin with the zero - dispersion channel . let @xmath281 and @xmath282 be , respectively , channel input and output in . for a fixed @xmath75 , @xmath20 and @xmath283 are complex numbers . we have rcl _ |x|p(_y|x)&=&_|x|p(_y|x , r_y ) + & = & . thus , the law of the zero - dispersion channel tends to the law of the following channel @xmath284 where @xmath285 , @xmath286 , and @xmath287 are independent . [ lemm : uniform ] the condition pdf is ( * ? ? ? 18 ) rcl p(r_y,_y|r_x,_x)&=&p_0(r_y|r_x ) + & & + _ m=1^(p_m(r_y|r_x)e^jm(_y-_x - r^2_xz ) ) . here rcl p_m(r_y|r_x)&=&2r_xb_m(-a_m(r_x^2+r_y^2))i_m(2b_mr_xr_y ) , where rcl a_m = x_m(x_m),b_m= , in which @xmath288 , @xmath289 . note that @xmath290 . the conditional pdf of the phase is rcl p(_y|r_x,_x , r_y)&= & + & & + & = & _ m=1^ ( d_m(r_x ) e^jm(_y-_x- r_x^2z ) ) + & & + , where step @xmath182 follows from @xmath291 ( see ( * ? ? ? * fig . 6 ( b ) ) ) and rcl d_m(r_x)&= & + & = & + & & \{-b_0(x_mx_m-1)(r_x^2+r_y^2)}. [ eq : dm ] the following three inequalities can be verified : rcl ||^2&= & + & & 1,t>0 . [ eq : inq1 ] rcl || & & || + & & 1 . [ eq : inq2 ] rcl f(t)&=&(x_mx_m-1 ) + & = & t-1 + & > & 0 , [ eq : inq3 ] where @xmath292 . using in , we obtain @xmath293 , where rcl e_m(r_x)=\{-f(t_m)(r_x^2+r_y^2)}. [ eq : d<1 ] we have rcl _ r_x|_m=1^d_m(r_x ) e^jm(_y-_x - r_x^2z)| & & _ r_x _ m=1^|d_m(r_x)| + & & _ r_x _ m=1^e_m(r_x ) + & & _ m=1^_r_x e_m(r_x ) + & & 0 . step @xmath182 follows because @xmath294 and @xmath295 is convergent ; thus , by the dominated convergence theorem , @xmath296 is uniformly convergent . step @xmath198 follows from . it follows that rcl _ r_xp(_y|r_x , _ x , r_y)=. furthermore rcl _ r_xp(_y|r_x,_x)&= & _ r_x p(_y|r_x,_x , r_y)p(r_y|r_x,)r_y + & = & . lemma [ lemm : uniform ] generalizes to the vectorial zero - dispersion channel . since noise is independent and identically distributed in space and time , so are the corresponding uniform phases . this is true even if @xmath297 in fig . [ fig : mssfm ] are dependent , _ e.g. _ , @xmath298 . we now consider the dispersive model . to generalize lemma [ lemm : uniform ] to the full model , we use the following notion ( * ? ? ? * section 2.6 ) . a family of pdfs @xmath299 , @xmath300 , is said to _ escape to infinity _ with @xmath301 if @xmath302 for any finite @xmath13 . let @xmath303 and @xmath154 be , respectively , the channel input and output in the dispersive model . the pdf of @xmath304 escapes to infinity as @xmath305 for all @xmath208 . [ lemm : scape ] the proof is based on induction in the mssfm units . we make precise the intuition that , as @xmath306 , the pdf of @xmath307 spreads out , so that an ever decreasing probability is assigned to any finite interval . consider vector @xmath111 in fig . [ fig : mssfm ] , at the end of the linear step in the first unit . setting @xmath308 , we have rcl pr(w < c)&=&_0^c p_w(w)w + & = & _ 0^c p_w()t + & & c _ t_(t ) _ , [ eq : p(w < c ) ] where @xmath309 , @xmath310 and @xmath311 . below , we prove that @xmath312 . fix @xmath313 and define the ( non - empty ) index set rcl i=\{i : |x_i|^1-}. the scaled random variable @xmath314 is rcl t_&=&|v_k| + & = & |_l=1^n e^j_l(x_l,_l^1)r_kl_l+| + & = & | _ li + _ li+ | , where @xmath315 is an additive noise and @xmath316 . as @xmath317 , the second sum vanishes because , if @xmath318 , @xmath319 in the first sum , @xmath320 , thus @xmath321 , where @xmath322 . therefore @xmath323 , in which rcl t_0=|_l ie^ju_lr_klx_l| , [ eq : t - eps ] where @xmath324 . since the pdf of @xmath325 is in @xmath326 on the circle @xmath327 , so is the conditional pdf @xmath328 , _ i.e. _ , rcl _ t_0(t)_<. [ eq : p(t)<infty ] substituting into rcl _ 0pr(w < c)=0 . [ eq : p(w < c)-2 ] in a similar way , can be proved for @xmath111 at the output of the linear step in the second unit , by replacing @xmath329 in with @xmath330 , and noting that , as @xmath317 , @xmath331 tend to random variables independent of input , with a smooth pdf ( without delta functions ) . from the lemma [ lemm : scape ] , as @xmath305 the probability distribution at the input of every zero - dispersion segment in the link escapes to infinity , turning the operation of the nonlinearity in that segment into multiplication by a uniform phase and independent noise addition . we thus obtain an input region @xmath332 for which , if @xmath333 , the channel is multiplication by a random matrix , as described in . the channel converts any small noise into worst - case noise in evolution . in this section , we obtain the asymptotic capacity of the channel . applying lemma [ lemm : decomposition ] to rcl |r(p)=|r_-(p)+(1-)| r_+(p ) , where @xmath334 , @xmath335 and @xmath336 is a parameter to be optimized . to shorten the analysis , we ignore the term @xmath337 in , as it does not depend on @xmath10 . we choose @xmath270 sufficiently large , _ independent of the average input power _ @xmath10 . from the lemma [ lemm : i < infty ] , @xmath338 . the following lemma shows that @xmath339 is given by the logarithmic terms in theorem [ thm : main ] with rcl c|r_-+(1-)_i(;m ) , where @xmath340 is the achievable rate at low powers . if @xmath145 is haar distributed , @xmath341 . define @xmath342 . assume that rcl h(m)>- , |m_ij|^2<,1i , jn . [ eq : h(m)>-infty ] then , the asymptotic capacity of is given by the expressions stated in the theorem [ thm : main ] . [ lem : cap - y = hx+n ] the capacity of the multiple - input multiple - output non - coherent memoryless fading channel is studied in @xcite . here , we present a short proof with a bit of approximation . using chain rule for the mutual information rcl i(x;y)&=&i(|| ; y)+i ( ; y||| ) + & = & i(|| ; ||)+i ( ; y|||)+i(|| ; ||| ) . [ eq : i(x;y ) ] the first term in gives the logarithmic terms in theorem [ thm : main ] , as calculated in section [ sec : proof1 ] . we prove that the other terms are bounded in @xmath3 . from the lemma [ lemm : decomposition ] , the additive noise in can be ignored when @xmath343 , so that @xmath344 . the second term in is rcl i ( ; y||| ) & = & i ( ; |x| m||| ) + & = & i ( ; m| || ) , where we used identity . note that we can not assume that @xmath3 and @xmath4 are independent . for the output entropy rcl h(m||| ) & & h(m ) + & & _ k=1^n h(_l=1^n m_kl_l ) + & & _ k=1^n ( e |_l=1^n m_kl_l|^2 ) + & & _ k=1^n ( _ l=1^n |m_kl|^2)+ne + & & n(||^2_f)+ne + & < & , where @xmath345 is the frobenius norm . step @xmath182 is obtained using the inequality @xmath346 . step @xmath198 is due to the met . cauchy - schwarz and jensen s inequalities are , respectively , applied in steps @xmath200 and @xmath347 . for the conditional entropy rcl h(m||| , ) & = & _ h(m||| , ) + & & _ h(m ) + & & - . [ eq : inter-6-aa ] step @xmath182 holds because , from the lemma [ lem : h(mx ) ] , @xmath348 for any @xmath349 . the third term in can be upper bounded using the second term by setting @xmath350 . we prove it alternatively . since @xmath142 is compactly supported , @xmath351 . the conditional entropy is rcl h_(|||,||)&= & h_(|||,|||m| ) + & = & h_(|||,|m| ) . [ eq : cond - entropy - y = hx+n ] applying identity to @xmath352 and conditioning on @xmath3 rcl h_(||| , |m| ) & = & h(m| ||)-h(|m|| || ) + & & -(2n-1)((|m|)||| ) . [ eq : inter-6 ] for the first term in rcl h(m|||)&&h(m||| , ) + & > & - , [ eq : inter-6-a ] where we used . since @xmath353 , from the met rcl h(|m|||| ) & & h(|m| ) + & & ( 2e|mx|^2 ) + & & ( 2e||_f^2 ) + & < & . [ eq : inter-6-b ] furthermore , rcl ( ( |m|^2)||| ) & & ( |m|^2||| ) + & & ||_f^2 + & < & . [ eq : inter-6-c ] substituting into and , we obtain rcl h_(|||,||)>- . the main ingredient in the proof of the lemma [ lem : cap - y = hx+n ] , as well as theorem [ thm : main ] , is the following lemma . let @xmath145 be a random matrix and @xmath43 a non - zero deterministic vector . if @xmath145 satisfies the assumptions , then rcl h(m x)>- . [ lem : h(mx ) ] since @xmath354 , at least one element of @xmath355 is nonzero , say @xmath356 . we switch the order of @xmath145 and @xmath23 in the product @xmath357 as follows . let @xmath358 denote the vectorized version of @xmath145 , where rows are concatenated as a column vector . define @xmath359 as follows : rcl v_k= m_in , & r=0 , + y_i+r=_l=1^n m_(i+r)lx_l , & r=1 , + m_(i+1)r , & r2 , [ eq : v - vec ] where @xmath360 , @xmath361 , @xmath362 . then @xmath363 is transformed to , which in matrix notation is rcl v = am , [ eq : v = am ] in which @xmath364 , where the deterministic matrix @xmath365 is rcl x= ( x_1 & x_2^n + 0 & i_n-1 ) , x_2^n= ( x_2 , , x_n ) , in which @xmath366 is the @xmath367 all - zero matrix . from rcl h(v|)&=&h(m|x)+|a| + & = & h(m)+n|x_1| + & = & h(m)+n|x_1| . on the other hand , from rcl h(v|)&&h(y , \ { m_ij}_j2| ) + & = & h(y | ) + h(\ { m_ij}_j2|x , y ) + & = & h(mx ) + h(\ { m_ij}_j2|y ) . step @xmath182 holds because conditions @xmath368 and @xmath369 in include , respectively , @xmath154 and @xmath370 . combining the last two relations rcl h(mx)= h(m)+ n| x_1|-h(\{m_ij}_j2|y ) . if @xmath371 , from the met , the last term is bounded from below . since @xmath372 and @xmath373 , @xmath374 . the random matrix @xmath145 , underlying optical fiber at high powers , satisfies the assumptions of the lemma [ lem : cap - y = hx+n ] . [ lem : h(m - fiber)>-infty ] applying the triangle inequality to , @xmath375 , where @xmath376 is the matrix with entries @xmath377 and @xmath378 is the number of stages . we check the entropy condition in . in what follows , let @xmath379 . for one linear and nonlinear steps @xmath380 : @xmath381 in this case , there are four amplitude dependencies @xmath382 , @xmath383 , and two phase dependencies : rcl m_11=m_21+k , m_12= m_22+k , k=0,1 . a dependency means that @xmath145 contains a deterministic component , _ i.e. _ , @xmath384 . for @xmath385 : @xmath386 in this case too , there is a dependency @xmath387 . for @xmath388 : rcl m_11 & = & e^j(_1+_3+_5)r_11 ^ 3 + e^j(_1+_4+_5)r_11r_12r_21 + & & + e^j(_1+_3+_6)r_11r_12r_21 + e^j(_1+_4+_6)r_12r_21r_22 , + m_12&=&e^j_2r_12 ( e^j(_3+_5)r_11 ^ 2 + + e^j(_4+_6)r_22 ^ 2 + & & + ) , + m_21 & = & e^j_1r_21(e^j(_3+_5)r_11 ^ 2 + e^j(_4+_6)r_22 ^ 2 + & & + ) , + m_22 & = & e^j(_2+_3+_5)r_11r_12r_21 + e^j(_2+_4+_5)r_12r_21r_22 + & & + e^j(_2+_3+_6)r_12r_21r_22 + e^j(_2+_4+_6)r_22 ^ 3 . comparing the boxed terms , @xmath389 . there are still 8 equations for 6 variables . in general , the number of entries of @xmath145 is @xmath390 . as @xmath391 steps are taken in distance , sufficient number of random variables @xmath392 are introduced in a matrix with fixed dimension . since @xmath0 is fixed and @xmath378 is free , we obtain an under - determined system of polynomial equations for @xmath393 whose solution space has positive dimension . thus an entry of @xmath145 can not be determined from all other entries . the rate interpolation lemma [ lemm : decomposition ] implies that , replacing @xmath268 by @xmath240 changes the asymptotic capacity by a finite number @xmath13 . from the upper bound @xmath394 in @xcite and theorem 2.5 in @xcite , we think that the asymptotic capacity can be achieved by an input distribution that escapes to infinity . this implies that @xmath395 , so that @xmath13 is indeed zero . we do not investigate this rigorously . multivariate gaussian input distribution is a poor choice for channels with multiplicative noise . indeed , it achieves a rate bounded in power in . log - normal input pdf for the signal norm achieves the asymptotic capacity of the non - constant loss model . an information - theoretic analysis of the full model of the optical fiber does not exist . even in the special case of the zero - dispersion , spectral efficiency is unknown . in the full model , we do not know anything about the capacity in the high power regime , let alone the spectral efficiency . the state - of - the - art is still lower bounds that are good in the nearly - linear regime . this situation calls for basic research , in order to make progress on these open problems . the present paper builds on earlier work . we acknowledge ( * ? ? ? 12 ) for the equation , @xcite for the pdf of the zero - dispersion channel , @xcite for the analysis of the zero - dispersion model , @xcite for noting that shannon entropy is invariant under the flow of a broad class of deterministic partial differential equations and for highlighting the usefulness of the operator splitting ( in numerical analysis ) in the analysis of the nls equation . furthermore , we acknowledge @xcite for helpful insight leading to the rate interpolation lemma [ lemm : decomposition ] , @xcite for the study of the fading channels and section ii of @xcite for unfolding the origin of the capacity limitations in fiber particularly the finding that signal - signal interactions are not fundamental limitations in the deterministic model if communication takes place in the right basis ( _ i.e. _ , the nonlinear fourier basis ) , which led us to the study of the remaining factor in this paper , namely the signal - noise interactions . we do not intend to survey the literature in this paper . there is a good review in ( * ? ? ? * section i - a ) . the achievable rates of 1- and multi - solitons is studied , respectively , in @xcite and @xcite . there is also a myriad of lower bounds that hold good in the low power regime ; see , _ e.g. _ , @xcite . the achievable rates of the nonlinear frequency - division multiplexing for multi - user communication are presented in @xcite for the hermitian channel . [ fig : nfdm ] compares the nfdm and wdm rates ( * ? ? ? * fig . 6 ) . the gap between the wdm and nfdm curve reflects signal - signal interactions . the gap between the nfdm and awgn curve reflects signal - noise interactions . we conjecture that the nfdm rate is close to the capacity . at the power levels shown in fig . [ fig : nfdm ] , @xmath396 and @xmath397 , @xmath398 . although more gains are expected at @xmath399 db , the slope of the blue curve will gradually decrease , converging , in the limit @xmath11 , to the asymptotic form in theorem [ thm : main ] . it is interesting to compare the extent of the signal - noise interactions in the time domain @xcite and in the nonlinear fourier domain ( * ? ? ? * section iv . the asymptotic capacity of the discrete - time periodic model of the optical fiber is characterized as a function of the input power in theorem [ thm : main ] . with @xmath0 signal dofsat the input , @xmath1 dofs are asymptotically lost , leaving signal energy as the only available dof for transmission . the appropriate input distribution is a log - normal pdf for the signal norm . signal - noise interactions limit the operation of the optical communication systems to low - to - medium powers . the research was partially conducted when the author was at the technische universitt mnchen ( tum ) . the support of the tum institute for advanced study , funded by the german excellence initiative , and the support of the alexander von humboldt foundation , funded by the german federal ministry of education and research , are gratefully acknowledged . the author thanks luca barletta for comments . let @xmath400 and @xmath401 be the volume and surface element at point @xmath402 in the spherical coordinate system . then rcl v(x)&=&|x|^n-1v ( ) + & = & |x|^n-1s()|x| . thus the jacobian of the transformation from the cartesian system with coordinates @xmath23 to the spherical system with coordinates @xmath403 is @xmath404 . as a consequence rcl h(x)&=&h_(||,)+||^n-1 + & = & h(||)+h_(|||)+(n-1)|| . define clearly @xmath407 . however @xmath408 and @xmath409 are generally non - gaussian due to the signal and noise terms in @xmath194 and @xmath410 . but , 1 ) in the constant loss model , if 2 ) @xmath206 @xmath411 , then m. i. yousefi and f. r. kschischang , `` information transmission using the nonlinear fourier transform , part i : mathematical tools , '' _ ieee trans . theory _ , vol . 60 , no . 7 , pp . 43124328 , jul . 2014 , also published at arxiv , feb . 2012 . [ online ] . available : http://arxiv.org/abs/1202.3653 , `` information transmission using the nonlinear fourier transform , part ii : numerical methods , '' _ ieee trans . theory _ , vol . 60 , no . 7 , pp . 43294345 , jul . 2014 , also published at arxiv , apr . 2012 . [ online ] . available : http://arxiv.org/abs/1204.0830 , `` information transmission using the nonlinear fourier transform , part iii : spectrum modulation , '' _ ieee trans . theory _ , vol . 60 , no . 7 , pp . 43464369 , jul . 2014 , also published at arxiv , feb . 2013 . [ online ] . available : http://arxiv.org/abs/1302.2875 m. i. yousefi , g. kramer , and f. r. kschischang , `` upper bound on the capacity of the nonlinear schrdinger channel , '' in _ ieee 14th canadian workshop on inf . theory _ , st . john s , newfoundland , canada , jul . 2015 , pp . 15 . i. tavakkolnia and m. safari , `` signalling over nonlinear fibre - optic channels by utilizing both solitonic and radiative spectra , '' in _ european conf . networks and commun . _ , paris , france , jul . 2015 , pp . 103107 . k. s. turitsyn , s. a. derevyanko , i. v. yurkevich , and s. k. turitsyn , `` information capacity of optical fiber channels with zero average dispersion , '' _ phys . _ , vol . 91 , no . 20 , p. 203901 , nov . 2003 . h. ghozlan and g. kramer , `` models and information rates for multiuser optical fiber channels with nonlinearity and dispersion , '' _ arxiv:1503.03124 _ , pp . 118 , mar . 2015 . [ online ] . available : https://arxiv.org/abs/1503.03124 e. meron , m. feder , and m. shtaif , `` on the achievable communication rates of generalized soliton transmission systems , '' _ arxiv:1207.0297 _ , pp . 113 , jul . [ online ] . available : https://arxiv.org/abs/1207.0297 p. kazakopoulos and a. l.moustakas , `` transmission of information via the non - linear schrdinger equation : the random gaussian input case , '' _ arxiv:1210.7940 _ , pp . 19 , oct . [ online ] . available : https://arxiv.org/abs/1210.7940 m. secondini , e. forestieri , and g. prati , `` achievable information rate in nonlinear wdm fiber - optic systems with arbitrary modulation formats and dispersion maps , '' _ ieee j. lightw . _ , vol . 31 , no . 114 , dec . 2013 . s. a. derevyanko , j. e. prilepsky , and s. k. turitsyn , `` capacity estimates for optical transmission based on the nonlinear fourier transform , '' _ nature commun . _ , vol . 12710 , pp . 19 , sep .

we study time evolution of critical fluctuations of conserved charges near the qcd critical point in the context of relativistic heavy ion collisions . a stochastic diffusion equation is employed in order to describe the diffusion property of the critical fluctuation arising from the coupling of the order parameter field to conserved charges . we show that the diffusion property gives rise to a possibility of probing the early time fluctuations through the rapidity window dependence of the second - order cumulant and correlation function of conserved charges . it is pointed out that their non - monotonic behaviors as functions of the rapidity interval are robust experimental signals for the existence of the critical enhancement around the qcd critical point .

1703.08008

the search for the qcd critical point is one of the most intriguing topics in the physics of qcd in medium . the existence of the critical point(s ) in the qcd phase diagram in the temperature ( @xmath0 ) and baryon chemical potential ( @xmath1 ) plane is suggested by effective models @xcite and lattice qcd monte carlo simulations @xcite . however , the number @xcite , locations , and even existence itself are still controversial . the experimental search for the critical point is one of the fundamental purposes of the relativistic heavy ion collision experiments @xcite . because the medium created by these experiments pursues different trajectories in the @xmath0@xmath1 plane depending on collision energies @xmath2 , this search would be achieved by comparing the characteristics of collision events at various @xmath2 . for this purpose , active experimental studies in the beam - energy scan ( bes ) program at the relativistic heavy - ion collider ( rhic ) are ongoing @xcite . future heavy - ion program at j - parc @xcite , fair @xcite and nica @xcite will also contribute to this subject . in the experimental search for the qcd critical point , fluctuation observables , such as the cumulants of the net - baryon number , are believed to be natural and promising observables @xcite . at the critical point in the thermodynamic limit , equilibrated fluctuations of various observables diverge reflecting the softening of the effective potential . such singular behavior is expected to be found in event - by - event analyses in heavy ion collisions @xcite . the experimental analyses of the fluctuation observables , especially focusing on non - gaussianity , have been performed actively recently @xcite . in interpreting the experimentally - observed fluctuations , however , it should be remembered that the fluctuations observed in these experiments are not those in an equilibrated medium near the critical point . first , the more closely the system approaches the critical point , the longer becomes the relaxation time toward equilibrium because of the critical slowing down @xcite . owing to this effect , the enhancement of fluctuations is limited even if the medium passes right through the qcd critical point . second , the critical fluctuations are to be distorted during the subsequent evolution until the detection @xcite . remember that particles are measured only in the final state . to extract the information on the critical point from the experimental data , proper understanding and description of these effects are indispensable , in addition to various restrictions from the real experimental settings @xcite . the time evolution of critical fluctuations has been discussed in the literature @xcite . in refs . @xcite , the effect of the critical slowing down is discussed for fluctuation of a uniform order parameter field @xmath3 . it is known , however , that the critical mode of the qcd critical point is not the pure @xmath4 mode , but rather is given by a linear mixing of the @xmath4 mode and conserved charges @xcite . the soft mode of the critical point thus is a diffusion mode . as we will see below , the time evolution of a diffusion mode depends on the length scale , and this property is crucial for the description of its dynamics . one of the main goals of the present study is to reveal the dynamical evolution of the critical fluctuation with the diffusion property . another aim of this paper is to make a connection between the critical fluctuation of thermodynamics and experimental observation . this connection will be clearest for the fluctuations of conserved charges , especially that of the net - baryon number ( see ref . @xcite for detailed discussions ) . in order to understand the evolution of the conserved - charge fluctuations , we first need to know that of the charge density itself in a collision event . this problem has been discussed for a hadronic medium in a simple setup @xcite , but to the best of our knowledge it has not yet been studied for the case where a critical enhancement is encountered during the evolution . carrying out this subject is the second goal of this study . in this study , in order to address these issues we employ a simple stochastic diffusion equation ( sde ) . in this approach , the singularity associated with the critical point is encoded in the time - dependent susceptibility and diffusion coefficient . the sde is a counterpart of the stochastic hydrodynamics @xcite , and is a good model to describe fluctuations of conserved charges . moreover , as shown in refs . @xcite , the critical mode at the qcd critical point has the diffusion character and should be described by the sde . in this study , we first write down a formal solution for the density fluctuation within rapidity window @xmath5 . next , with a phenomenological parametrization for the susceptibility and diffusion coefficient , we analyze numerically the time evolution of the fluctuation . we find that the time evolution of the fluctuations strongly depends on the size of the rapidity window @xmath5 . we also show that the second - order cumulant can have a non - monotonic @xmath5 dependence only when the medium fluctuation undergoes a critical enhancement during the time evolution . it is argued that this non - monotonic behavior serves as a robust experimental signal of the critical enhancement . in the present study we restrict our attention only to the second - order cumulant , and leave the analysis of third and still higher order cumulants for future study , because even at the second order non - trivial outcomes are obtained from the analysis . we also deal with the correlation functions of conserved charges . we show that the same argument on the non - monotonic behavior holds for this function with rapidity window replaced by rapidity separation . the relation between the cumulant and the correlation function is also studied in detail . this paper is organized as follows . in the next section we introduce the sde , and argue that this model is suitable for describing the dynamics of conserved - charge fluctuations including the critical ones . we then solve this equation analytically , and discuss general properties in sec . [ sec : analytic ] . in sec . [ sec : time ] , we introduce a phenomenological model , which is numerically solved in sec . [ sec : real ] . the last section is devoted to discussions and a summary . in appendix [ sec : sdec ] , we give a brief review of refs . @xcite , which clarified the diffusion property of the critical fluctuation . in appendix [ sec : discussion ] , we compare the cumulant and correlation function and discuss the condition for the appearance of the non - monotonic behaviors in these functions in detail . in this study , we investigate the time evolution of the second - order cumulant and the correlation function of conserved charges in the hot medium produced in heavy ion collisions . we assume a boost - invariant bjorken model for the event evolution throughout this paper in order to simplify the analysis , and adopt the milne coordinates , the spacetime rapidity @xmath6 and proper time @xmath7 . let us consider a conserved - charge density per unit rapidity @xmath8 , where transverse coordinates have been integrated out . as the conserved charge , we can take specifically the net - baryon number @xcite or net - electric charge in heavy ion collisions . the amount of the charge in a finite rapidity interval @xmath5 at mid - rapidity at proper time @xmath9 is given by @xmath10 the second - order cumulant , or variance , of @xmath11 is written as @xmath12 where @xmath13 is the correlation function with @xmath14 and @xmath15 stands for event average . in a boost invariant system , the correlation function depends only on the rapidity difference @xmath16 , and eq . ( [ eq:<q^2 > ] ) is rewritten as @xmath17 this reveals a close relation between @xmath18 and @xmath19 . we investigate both of these functions in this paper . from eq . ( [ eq:<q^2>cor ] ) , one finds that @xmath20 we consider the time evolution of the conserved charge density @xmath8 at long time and length scales . this is well described by the stochastic diffusion equation @xcite , which is written in the @xmath9@xmath21 coordinates as @xmath22 where the diffusion coefficient @xmath23 is related to the cartesian one @xmath24 as @xmath25 . the noise term @xmath26 represents the coupling with `` short - time '' fluctuations , whose average should vanish @xmath27 . the noise term appears with the @xmath21-derivative in eq . ( [ eq : sde ] ) so as to satisfy the conservation constraint . when the noise correlation is local in time and space , the fluctuation - dissipation relation specifies its value to be @xcite @xmath28 here , @xmath29 denotes the susceptibility of the conserved charge per unit rapidity and is related to the one in the cartesian coordinates @xmath30 as @xmath31 . the susceptibility @xmath29 is related to the second - order cumulant in equilibrium as @xmath32 . furthermore , the noise is independent of the value of the density at earlier times : @xmath33 for @xmath34 . in the rest of this paper , we suppress the subscripts of the diffusion coefficient and susceptibility , and denote @xmath35 and @xmath36 , respectively . the sde ( [ eq : sde ] ) is not only a suitable equation for describing the time evolution of conserved charges in a non - critical system , but also a good phenomenological equation to deal with the slow dynamics near the qcd critical point . this is because there the critical mode is identified as a linear combination of the @xmath4 mode and conserved charges , and therefore its evolution must be consistent with the conservation law @xcite . as a consequence , the equation for the critical mode is given by the same form as in eq . ( [ eq : sde ] ) . see appendix [ sec : sdec ] for more detailed discussion . hence , in the present study we use eq . ( [ eq : sde ] ) solely to describe the time evolution throughout the event trajectories passing near and away from the critical point . in our study , the critical enhancement and slowing down of fluctuations are represented by the @xmath9-dependent susceptibility @xmath37 and diffusion coefficient @xmath38 : near the critical point , @xmath37 grows sharply and @xmath38 becomes vanishingly small reflecting , respectively , the large fluctuation and critical slowing down . in the next section , we derive the formal solution of eq . ( [ eq : sde ] ) with @xmath9 dependent @xmath37 and @xmath38 , and discuss its general property . we then model the trajectories of collision events in terms of possible @xmath9-dependence of the susceptibility @xmath37 and diffusion coefficient @xmath38 including the effect of the critical point , and analyze the time evolution numerically in secs . [ sec : time ] and [ sec : real ] . before closing this section , we comment on the difference of the treatment of the critical fluctuation in the present study from previous ones @xcite . in refs . @xcite , the time evolution of the uniform @xmath4 field was analyzed as the slowest mode of the system mode is studied , which is equivalent to treating the spatially uniform @xmath4 mode , as shown in ref . @xcite . ] . since the @xmath4 mode is non - conserved , it can relax locally in rapidity space and follows a relaxation equation without @xmath39 . this assumption is in contrast to the conserved charge fluctuation discussed in the present study , which can relax only through diffusion . as discussed already , the evolution equation of the critical fluctuation has to be consistent with the conservation law @xcite . in order to respect this property , the evolution has to be dealt with the equations like the sde consistent with the continuity equation ) is the coupling of the soft mode with the momentum - density modes , which plays an important role in describing the critical dynamics more precisely @xcite . this effect is discussed in ref . @xcite . ] . as we will see in the next section , the time evolution of the critical mode then depends on their length scale @xmath5 . this fact makes the problem complicated because the critical mode can no longer be regarded as a spatially uniform mode . at the same time , however , this diffusion property opens a possibility to study the critical fluctuation through @xmath5 dependence in experiments , as we will see later . we also remark that our model can describe the fluctuations throughout the time evolution of the hot medium including the critical region as well as late stages . this property is advantageous in understanding dynamics behind experimental observables . in this section , we formally solve the sde ( [ eq : sde ] ) and study general properties of the second - order cumulant and correlation function analytically . defining the fourier transform of @xmath8 via @xmath40 , the formal solution of eq . ( [ eq : sde ] ) with the initial condition @xmath41 at @xmath42 is obtained as @xmath43 ^ 2/2}\nonumber\\ & & + \int_{\tau_0}^{\tau}d\tau^{\prime } \ iq\xi(q,\tau^{\prime } ) e^{-q^2[d(\tau^{\prime},\tau)]^2/2},\end{aligned}\ ] ] where @xmath44^{1/2 } \label{eq : d}\end{aligned}\ ] ] denotes the diffusion `` length '' in rapidity space from @xmath45 to @xmath46 with @xmath47 . the diffusion length @xmath48 is a monotonically increasing ( decreasing ) function of @xmath46 ( @xmath45 ) , satisfying the boundary condition @xmath49 . the correlation function at proper time @xmath9 is obtained by taking the average of the product of eq . ( [ eq : sol ] ) as ) . ] @xmath50 ^ 2/2 } \nonumber\\ & + \int_{\tau_0}^{\tau}d\tau_1 d\tau_2 \langle i q_1 \xi(q_1,\tau_1 ) i q_2 \xi(q_2,\tau_2 ) \rangle \nonumber \\ & \times e^{-q_1 ^ 2[d(\tau_1,\tau)]^2/2}\ , e^{-q_2 ^ 2[d(\tau_2,\tau)]^2/2},\end{aligned}\ ] ] where we have used @xmath51 for @xmath52 . to proceed further , we assume that the initial fluctuation satisfies the locality condition , @xmath53 indeed , this condition should hold in a thermal medium at the length scale at which the extensive property of thermodynamic functions is satisfied @xcite . we then obtain @xmath54 , and eq . ( [ eq : cor - fourier1 ] ) is calculated to be @xmath55 ^ 2 } \nonumber \\ & + 2 q_1 ^ 2\int_{\tau_0}^{\tau}d\tau^{\prime } \chi(\tau^{\prime } ) d(\tau^{\prime})e^{-q_1 ^ 2[d(\tau^{\prime},\tau)]^2 } \bigg).\end{aligned}\ ] ] the correlation function in @xmath21 space is obtained from eq . ( [ eq : cor - fourier2 ] ) as @xmath56 where @xmath57 and we have defined the normalized gauss distribution @xmath58 we note that the correlation function depends on @xmath38 only through the diffusion length @xmath59 . by substituting eq . ( [ eq : cor ] ) into eq . ( [ eq:<q^2>cor ] ) , the second - order cumulant is calculated to be @xmath60 where @xmath61 with the error function @xmath62 . the behavior of @xmath63 is shown in fig . [ fig : f(x ) ] . as is shown in the figure , @xmath63 is a monotonically increasing function satisfying @xmath64 from eqs . ( [ eq : cum ] ) and ( [ eq : cor ] ) , we find several notable features in the rapidity dependences of the cumulant and the correlation function . first , we consider the behavior of @xmath18 in the small and large @xmath5 limits . using eq . ( [ eq : flimit ] ) , the cumulant in these limits is easily calculated to be @xmath65 { } & \chi(\tau ) , \label{eq : q : dy->0 } \\ \frac{\langle q_{\delta y}(\tau)^2\rangle_{\rm c}}{\delta y } \xrightarrow[\delta y\to\infty ] { } & \chi(\tau ) - \int_{\tau_0}^{\tau}d\tau ' \chi^{\prime}(\tau ' ) \nonumber \\ = & \chi(\tau_0 ) . \label{eq : q : dy->inf}\end{aligned}\ ] ] these results show that @xmath66 takes the local - equilibrium value @xmath37 in the small @xmath5 limit , while it recovers the initial value in the opposite limit . this shows that the relaxation toward the equilibrium state is sufficiently fast as @xmath5 becomes smaller , but it becomes arbitrarily slow with increasing @xmath5 . the latter means that equilibration of the conserved - charge fluctuation over the whole system can not be achieved within a finite time , because it takes infinite time to transport a charge from one end to the other . the analysis here also implies that @xmath66 with smaller ( larger ) @xmath5 bears the information of @xmath37 at later ( earlier ) @xmath9 . this suggests that one can study the @xmath9 dependence of @xmath37 from the @xmath5 dependence of @xmath66 @xcite . second , when @xmath37 increases ( decreases ) monotonically in @xmath9 , @xmath67 ( @xmath68 ) , then @xmath66 for a given @xmath9 is a monotonically decreasing ( increasing ) function of @xmath5 : @xmath69 this can be easily shown from eq . ( [ eq : cum ] ) and the fact that @xmath63 is a monotonically increasing function . taking the contraposition of eq . ( [ eq : dq<0 ] ) , one concludes that @xmath37 must have at least one extremum when @xmath66 is non - monotonic as a function of @xmath5 . in particular , @xmath70 the same argument also applies to the correlation function . from the fact that @xmath71 monotonically decreases as @xmath72 increases , it is again easy to show that @xmath73 for @xmath74 . from the contraposition of eq . ( [ eq : dn<0 ] ) , one obtains @xmath75 the properties ( [ eq : nonmono - cum ] ) and ( [ eq : nonmono - cor ] ) are quite useful in extracting the @xmath9 dependence of @xmath37 in relativistic heavy ion collisions . if the experimental results of @xmath76 and/or @xmath77 show non - monotonic behavior as a function of rapidity , this immediately confirms the existence of non - monotonicity in @xmath37 as a function of @xmath9 . it is known that the susceptibilities of baryon number and electric charge have a peak structure along the phase boundary around the qcd critical point @xcite . the peak in @xmath66 or @xmath77 serves as an experimental signal for this critical enhancement . this is the most important conclusion of this paper . it should be kept in mind that the inverses of eqs . ( [ eq : nonmono - cum ] ) and ( [ eq : nonmono - cor ] ) do not necessarily hold . that is , even if @xmath37 is a non - monotonic function of @xmath9 , there is a possibility that @xmath66 and @xmath77 are monotonic . therefore , from the monotonic behavior of @xmath66 and/or @xmath77 one can not conclude anything about the @xmath9 dependence of @xmath37 . in appendix [ sec : discussion ] , we discuss the condition for the appearance of the non - monotonic behaviors in these functions in more detail . using sde ( [ eq : sde ] ) , it is possible to calculate the time evolution of third and still higher order cumulants and correlation functions . as is easily shown , however , the higher order correlation functions @xmath78 , and accordingly the higher order cumulants @xmath79 , too , vanish in the @xmath80 limit for @xmath81 @xcite ; the fluctuations of @xmath8 in equilibrium described by eq . ( [ eq : sde ] ) obey the gaussian distribution . because of this property , sde ( [ eq : sde ] ) is not capable of describing the relaxation of higher order fluctuations toward nonzero non - gaussian equilibrium values . in relativistic heavy ion collisions , observed higher order cumulants take values close to their nonzero equilibrium values @xcite . if this result is a consequence of the relaxation process in the hadronic medium , the relaxation can not be described by the sde ( [ eq : sde ] ) . this is one of the reasons why we limit our attention to the second - order cumulant and correlation function in the present study , in spite of the useful properties of the higher order cumulants @xcite . to describe the relaxation of higher - order cumulants toward nonzero non - gaussianity , different approaches are needed . in ref . @xcite , for example , the non - interacting brownian particle model is employed to describe this process . in the previous section we showed that non - monotonic behavior of @xmath66 and/or @xmath77 , if observed , is a direct experimental evidence for the existence of a peak structure in susceptibility @xmath37 . in the rest of this paper , we demonstrate the appearance of the non - monotonic behavior by studying the behavior of @xmath66 and @xmath77 with a phenomenological parametrization of @xmath37 and @xmath38 for a collision event evolution passing near and away from the qcd critical point . in this section we first introduce the model for @xmath37 and @xmath38 . then , the time evolution of fluctuation is studied in the next section . in this study , we write the susceptibility and the diffusion coefficient at temperature @xmath0 as a sum of their singular and regular contributions : @xmath82 where @xmath83 and @xmath84 denote the singular and regular parts of susceptibility per unit rapidity , respectively . we also define the singular and regular parts of diffusion coefficients @xmath85 and @xmath86 in cartesian coordinate . we then parametrize the map of the evolution time to the temperature @xmath87 to fix the @xmath9 dependences . first , we discuss the singular parts @xmath83 and @xmath88 . it is known that the qcd critical point belongs to the same static universality class as the 3d ising model . the magnetization @xmath89 of the ising model as a function of the reduced temperature @xmath90 and the dimensionless magnetic field @xmath91 near the critical point is parametrized with the two variables @xmath92 and @xmath93 in the linear parametric model @xcite as @xmath94 where @xmath90 and @xmath91 are expressed as @xmath95 the critical point is located at @xmath96 . the crossover @xmath97 and first - order transition @xmath98 lines correspond to @xmath99 and @xmath100 with @xmath101 , respectively . we adopt approximate values @xmath102 and @xmath103 for the ising critical exponents @xcite . from eq . ( [ eq : m ] ) , one can calculate the magnetic susceptibility as @xmath104 as the susceptibility of a conserved charge @xmath105 near the qcd critical point should share the same critical behavior as @xmath106 , we set @xcite @xmath107 with a dimensionless proportionality constant @xmath108 . the susceptibility in the hadronic medium @xmath109 will be defined in sec . [ subsec : sreg ] . we fix the normalization constants @xmath110 and @xmath111 by imposing @xmath112 and @xmath113 . in reality , the finite system size effect in heavy ion collisions prevents the divergence of @xmath105 @xcite . nevertheless we ignore this effect because the growth of fluctuation would be limited more severely by the finiteness of the evolution time owing to the critical slowing down @xcite . for determining @xmath114 , we employ the dynamic universality argument @xcite . since the qcd critical point belongs to the model h @xcite in the classification of ref . @xcite , the singular part @xmath114 scales with the correlation length @xmath115 as @xmath116 , where the exponents @xmath117 and @xmath118 for model h are obtained by the renormalization group calculation as @xmath119 and @xmath120 @xcite . as the correlation length @xmath115 and the susceptibility @xmath121 are related as @xmath122 , the singular part @xmath114 can be expressed in terms of @xmath121 : @xmath123 ^{(-2+\chi_\eta+\chi_\lambda)/(2-\chi_\eta ) } \label{eq : d^cr}\end{aligned}\ ] ] with a proportionality constant @xmath124 having the dimension of diffusion coefficient . we set @xmath125 fm in what follows . note that @xmath114 vanishes at the critical point , reflecting the critical slowing down . in our model , we leave the strengths of the critical component @xmath108 as a free parameter . the reduced temperature @xmath90 controls the distance of the trajectory from the critical point . we will vary @xmath90 to simulate the change of the collision energy in the next section . to utilize the above universality argument for describing heavy ion collision events , we need a map between the ising variables @xmath126 and the physical variables @xmath127 in qcd , in addition to a map from the proper time @xmath9 to @xmath127 at a given collision energy . to skip these mappings , we follow a simple approach adopted in refs . @xcite : we assume that @xmath127 in qcd are linearly mapped to the ising variables @xmath126 around the critical point , and that only @xmath91 changes while @xmath90 is fixed during the time evolution . we write the linear relation between @xmath91 and @xmath0 as @xmath128 with @xmath129 being the critical temperature . the ratio @xmath130 controls the width of the critical region in the qcd phase diagram . to relate @xmath0 and @xmath9 , we assume the one - dimensional bjorken expansion and conservation of total entropy . the relation is then obtained as @xcite @xmath131 where @xmath132 is the sound velocity and @xmath133 is the initial temperature of the system at the initial proper time @xmath134 . we ignore possible entropy production in the critical region in this work . in our calculation we set the initial temperature @xmath135 mev at proper time @xmath136 fm , the critical temperature of the qcd critical point @xmath137 mev , and the kinetic freeze - out temperature @xmath138 mev @xcite , where we stop the evolution . the parameters for the critical region in eq . ( [ eq : region ] ) are set to @xmath139 mev . for the sound velocity @xmath132 , we adopt @xmath140 , which is indicated in a lattice calculation in the transition region at @xmath141 @xcite . in eq . ( [ eq : t - t ] ) , the effect of transverse expansion is not taken into account . transverse expansion makes the duration of the hadron phase shorter . thus , the following calculation is likely to estimate the effect of diffusion stronger than the actual one . as we will see , the suppression of the diffusion will be advantageous to detecting the critical point in experiments . we assume that the susceptibility per unit rapidity @xmath143 approaches a constant value @xmath144 ( @xmath109 ) at high ( low ) temperature , which we call quark - gluon plasma ( hadronic ) value . we note that the value of @xmath144 depends on the trajectory in the @xmath127 plane as well as the thermal property @xcite ; see also refs . @xcite . we also note that the susceptibility per unit rapidity approaches a constant value in the late stage in heavy ion collisions because the particle abundances are fixed after the chemical freeze - out . in the present study , we use the value estimated in ref . @xcite assuming entropy conservation , @xmath145 for the diffusion coefficient , we assume that the coefficient in the cartesian coordinates approaches constant values , @xmath146 and @xmath147 , at high and low temperatures , respectively . we take @xmath148 fm from an estimate in the lattice qcd calculation @xcite and @xmath149 fm from ref . @xcite . the regular parts @xmath84 and @xmath86 are then constructed by smoothly interpolating these values at high and low temperatures , @xmath150 with @xmath151 here @xmath152 and @xmath153 are determined so that @xmath143 and @xmath154 coincide with the presumed values @xmath155 and @xmath156 at @xmath157 , respectively . we set the width of the crossover region , @xmath158 mev . in fig . [ fig : chid ] , we plot the susceptibility @xmath159 and the diffusion coefficient @xmath154 as a function of @xmath0 for several values of @xmath90 and @xmath108 with @xmath125 . the upper panel of fig . [ fig : chid ] shows that @xmath159 for @xmath160 diverges at @xmath161 . the sharp peak around @xmath129 remains even for @xmath162 with @xmath163 . the regular part eq . ( [ eq : regsus ] ) , labeled `` reg '' , is also shown for comparison . the lower panel of fig . [ fig : chid ] shows that @xmath154 with the singular part vanishes at @xmath161 for @xmath160 , which is a manifestation of the critical slowing down . now , we analyze the time evolution of the fluctuation and correlation function using the parametrization obtained in the previous section , and study the effect of the qcd critical point on observables . one remark concerned with the experimentally - observed fluctuations is that the diffusion described by the sde ( [ eq : sde ] ) proceeds in coordinate space , but the experimental measurements are performed in momentum space . the imperfect correlation between the two rapidities owing to thermal motion gives rise to the `` thermal blurring '' effect @xcite . for nucleons , this effect increases the apparent diffusion length in rapidity space by about @xmath164 at and after the thermal freeze - out @xcite . we take account of this effect in the subsequent analyses . in this section , we show the numerical results of the cumulant and correlation function in the following normalized forms , @xmath165 where @xmath166 is the cumulant in the equilibrated hadronic medium . first , we study the case without the singular parts by setting @xmath167 and @xmath168 . this corresponds to the collision events which pursue a trajectory away from the critical point in the crossover region ( or , in the qcd phase diagram without a first order phase transition ) . as in fig . [ fig : chid ] , @xmath84 behaves monotonically as a function of @xmath0 in this case . in fig . [ fig : reg ] , we show the results of @xmath169 and @xmath170 for several values of @xmath0 from the initial temperature @xmath135 mev to the kinetic freeze - out @xmath138 mev , together with the result after the thermal blurring . note that the result only after thermal blurring can be compared with experimental results . the other results are shown to understand the time evolution of these quantities . at the initial time with @xmath171 mev , @xmath169 is given by a constant , while @xmath170 vanishes , in accord with the locality condition eq . ( [ eq : locality ] ) . as @xmath0 is lowered , nontrivial structures emerge in these functions . as discussed in sec . [ sec : model ] , @xmath169 at @xmath172 is equal to its thermal value , i.e. @xmath159 , which increases monotonically with time in this non - critical case . @xmath169 at nonzero @xmath5 follows this trend but the increase is slower because of the finite diffusion time . as a result , the cumulant of a conserved charge depends on @xmath5 strongly . we also note that @xmath169 decreases monotonically with increasing @xmath5 , which is consistent with the statement , eq . ( [ eq : dq<0 ] ) , which tells us that @xmath169 should be monotonic when @xmath37 is monotonic . @xmath169 approaches its initial value @xmath173 as @xmath5 increases . notice that the behavior of @xmath169 after thermal blurring is qualitatively consistent with the experimental result of the second - order cumulant of net - electric charge observed at the large hadron collider @xcite . the @xmath5 dependence of the second - order cumulant has been studied in refs . the analysis of these studies corresponds to the parameter choice @xmath174 in eq . ( [ eq : regsus ] ) , i.e. @xmath37 jumps discontinuously at @xmath129 . since our result is qualitatively unchanged from the previous one , the nonzero width @xmath175 seems not crucial for the argument here . the lower panel of fig . [ fig : reg ] shows @xmath170 . one finds that this function also behaves monotonically as a function of @xmath72 , which is consistent with eq . ( [ eq : dn<0 ] ) . we also notice that @xmath170 always takes a negative value . this is directly confirmed by substituting @xmath176 into eq . ( [ eq : cor ] ) . let us examine the case where the trajectory in heavy ion collisions passes right through the critical point ( @xmath160 ) . in fig . [ fig : r=0cc=4 ] , we show the evolution of @xmath169 and @xmath170 along the critical trajectory ( @xmath160 ) for several values of @xmath0 and after thermal blurring with @xmath163 . in the upper panel of fig . [ fig : r=0cc=4 ] , @xmath169 at @xmath177 mev shows a remarkable enhancement , which comes from the divergence of @xmath37 at @xmath161 as shown in fig . [ fig : chid ] . this figure shows , however , that the cumulant stays finite for nonzero @xmath5 even at the critical point . this result is a manifestation of the critical slowing down for conserved charges . we remark that the effect of the critical slowing down is dependent on @xmath5 as discussed in sec . [ subsec : property ] . after passing through the critical point , the value of @xmath169 at @xmath172 decreases rapidly in accordance with the suppression of @xmath37 , while the decrease at nonzero @xmath5 is slower because of the slower diffusion for the larger @xmath5 . as a consequence , a non - monotonic structure appears in @xmath169 . in fig . [ fig : r=0cc=4 ] , the non - monotonic behavior continues to exist in @xmath169 until the kinetic freeze - out time and even survives the thermal blurring . therefore , the non - monotonic behavior of @xmath169 can be observed experimentally in this case . as discussed in sec . [ subsec : property ] , this non - monotonic behavior , if observed , is a direct signal for the existence of the critical enhancement of @xmath37 . an important lesson to learn from this result is that the non - monotonic behavior of @xmath169 can survive whereas the magnitude of fluctuation itself is almost smeared to the equilibrated hadronic value @xmath178 ; the maximum value of @xmath169 after thermal blurring is @xmath179 at @xmath180 . this result suggests that the study of the non - monotonicity of @xmath169 is advantageous for the search of the critical enhancement than the value of @xmath169 with fixed @xmath5 . therefore , it is quite an interesting experimental subject to analyze its @xmath5 dependence . from the lower panel of fig . [ fig : r=0cc=4 ] , one can draw the same conclusion on the @xmath72 dependence of @xmath170 : @xmath170 at @xmath181 changes from negative to positive around @xmath129 . triggered by this behavior , the non - monotonic @xmath72 dependence of @xmath170 manifests itself . the non - monotonicity again survives thermal blurring , suggesting that it can be measured experimentally . we note that similar non - monotonic behaviors of correlation functions are also observed in ref . @xcite and that of the mixed correlation function in ref . the appearance of the non - monotonicity in these studies is understood completely the same way as that in sec . [ sec : analytic ] . next , we consider the case of a weaker critical enhancement by setting @xmath182 , but still keeping @xmath160 . we show the results in fig . [ fig : r=0cc=1 ] . although results above @xmath183 mev look almost the same as those for @xmath163 , the non - monotonicity of @xmath169 disappears already at @xmath184 mev . as the growth of the susceptibility with @xmath182 is weaker , the signal is drowned out by the diffusion in the hadronic phase . this exemplifies that , as discussed in sec . [ subsec : property ] , the absence of the non - monotonicity in @xmath169 does not necessarily mean the absence of the peak structure in @xmath37 . the lower panel of fig . [ fig : r=0cc=1 ] shows the result for the correlation function . the figure shows that the non - monotonic behavior of @xmath170 generated at the critical point survives at @xmath184 mev and even the thermal blurring . this result suggests that the non - monotonic signal in @xmath170 is observable even when it disappears in @xmath169 . in fact , as we will discuss in appendix . [ sec : discussion ] , the non - monotonicity in @xmath170 is more sustainable than that in @xmath169 . finally , let us study the time evolution of fluctuations for @xmath185 , which corresponds to the case where the system undergoes a crossover transition . shown in fig . [ fig : r=1cc=4 ] are the results of @xmath169 and @xmath170 for @xmath162 and @xmath163 . the results are qualitatively the same as those in fig . [ fig : r=0cc=4 ] . by closely comparing these results , one finds that the non - monotonic signal with @xmath162 for @xmath186 mev is much clearer than that in fig . [ fig : r=0cc=4 ] , although @xmath143 does not diverge with @xmath162 . there are two reasons behind this result . first , @xmath38 for @xmath162 does not vanish because the trajectory does not pass right through the critical point ( see , fig . [ fig : chid ] ) . therefore , the critical slowing down for @xmath162 is less important than that for @xmath160 , and the fluctuations can grow faster around @xmath129 . second , @xmath159 for @xmath162 and @xmath163 is larger than that for @xmath160 at @xmath187 mev in our parametrization , as seen in fig . [ fig : chid ] . therefore , @xmath143 for @xmath162 approaches @xmath109 more slowly . this behavior makes the non - monotonic peaks in @xmath169 and @xmath170 more prominent . note , however , that the second observation may be dependent on the parametrization of @xmath159 . this argument suggests that the non - monotonic signals can be observed even when the trajectory does not pass right through the critical point . moreover , it is possible that the trajectory off the critical point is more favorable for the emergence of the non - monotonicity . however , the signal of the critical enhancement , of course , weakens and finally disappears as the trajectory departs further off the critical point . in our analysis with @xmath163 , the non - monotonic behavior of @xmath169 and @xmath170 disappears at @xmath188 and @xmath189 , respectively . the most important conclusion of the present study is eqs . ( [ eq : nonmono - cum ] ) and ( [ eq : nonmono - cor ] ) , i.e. the non - monotonic behaviors of @xmath169 and/or @xmath170 , if observed , are direct experimental signals of the critical enhancement in the susceptibility . now , let us consider the application of this conclusion to real heavy ion collisions . throughout this study we assumed the bjorken space - time evolution . this assumption , however , is violated in lower energy collisions . in particular , in the energy range of the bes program at rhic , one has to consider the effect of the violation severely . for lower energy collisions , the effects of global charge conservation @xcite arising from the finite system size would also affect fluctuation observables seriously . it is expected , however , that these effects do not alter our conclusions on non - monotonicity as long as the system size in rapidity space is sufficiently large compared to the diffusion length . first , the violation of the bjorken picture makes the correspondence between the coordinate and momentum space rapidities worse . this makes the thermal blurring effect stronger @xcite , and the experimental measurement of non - monotonicity would become difficult . even in this case , however , the relations eqs . ( [ eq : nonmono - cum ] ) and ( [ eq : nonmono - cor ] ) should hold , because the thermal blurring effect only acts to enhance the diffusion length @xcite . from the analysis in ref . @xcite , it is expected that the fluctuation around mid - rapidity is not affected by the effect of global charge conservation for sufficiently high energy collisions as long as the system size is sufficiently large . however , our conclusion may need to be altered for collisions at bes energies and lower energies such as those at fair , nica , and j - parc . in this study , we described the critical fluctuation by the stochastic diffusion equation ( [ eq : sde ] ) . although this model well describes sufficiently long and slow fluctuations , the fluctuations in heavy ion collisions may not be slow enough compared to the time scale of the medium evolution . to take account of these effects , the sde ( [ eq : sde ] ) has to be modified . one direction is to include higher order derivative terms . another interesting extension is to include the @xmath4 field as a dynamical field , and solve the coupled equation of @xmath190 and @xmath4 . near the critical point , the coupling of @xmath4 with momentum density would also be important @xcite . to deal with these subjects , numerical simulations of medium evolution will be needed with adopting a certain model , for example , the chiral fluid model @xcite . in this study , we investigated the time evolution of the second order cumulant and the correlation function of conserved charges in heavy ion collisions which pass through or near the critical point focusing on the effects of critical slowing down near the critical point and dissipation in the late stages . we adopted the stochastic diffusion equation with critical nature being encoded in the time - dependent susceptibility and diffusion coefficient . this model can describe the dynamics of the critical mode respecting its diffusion property , which was not considered in previous studies on the critical slowing down . we have pointed out that the critical enhancement in susceptibility can be observed as the non - monotonic behaviors in the second - order cumulant and correlation function . our numerical results suggest that these non - monotonic behaviors are a more robust experimental signal than the value of these functions themselves . it is , therefore , quite an interesting experimental subjects to analyze the rapidity dependences of these functions in heavy ion collisions . m. a. and m. k. thank v. koch and m. lisa for inviting them to int workshop `` exploring the qcd phase diagram through energy scans '' , sep . 19 - oct . 14 , 2016 , seattle , usa , and stimulating discussions . the authors thank m. nahrgang and m. bluhm for fruitful discussions . they also thank a. bzdak , k. redlich , and m. stephanov for useful conversations . this work was supported in part by jsps kakenhi grant numbers 16j01314 , 26400272 , and 16k05343 . in this appendix , we explain that the appropriate equation to describe the time evolution of the soft mode near the critical point is the sde ( [ eq : sde ] ) , on the basis of refs . @xcite . at sufficiently long distance and time scales , the dynamics of a finite temperature system near equilibrium is described by hydrodynamic theory , which only contains the modes whose excitation energies vanish in the long wavelength limit `` hydrodynamic modes '' . near the critical point , the hydrodynamic variables are given by the fluctuations of the order parameter and the densities of conserved charges @xcite . the qcd critical point shares the same dynamical universality class with the model h in the classification of ref . @xcite , and the chiral order parameter field @xmath191 has nonzero couplings with baryon number density @xmath190 , and energy and momentum densities . in this appendix , we neglect the energy - momentum density for a simple illustration . we start from the ginzburg - landau free energy functional @xmath192 \nonumber \\ & = \frac12 \int d{\bm x}\left [ a ( \delta\sigma)^2 + 2b(\delta\sigma)(\delta n ) + c ( \delta n)^2 + \cdots \right],\end{aligned}\ ] ] where the coefficients @xmath193 , and @xmath194 are functions of temperature @xmath0 and baryon chemical potential @xmath1 . the neglected terms include higher order terms in @xmath195 and @xmath196 , and derivative terms , which are not important to describe slow modes . here , @xmath197 because the coupling between @xmath4 and @xmath190 is allowed at the qcd critical point because of the finite quark masses and finite baryon density @xcite . deviation of @xmath4 and @xmath190 from the equilibrium values gives rise to relaxation of the system . the evolutions of @xmath4 and @xmath190 are given by the following stochastic hydrodynamic equations : @xmath198 where the noise correlators are local , @xmath199 with @xmath200 . from onsager s principle , we have @xmath201 . in the small momentum limit , @xmath202 is given by a nonzero constant , while the coefficients for @xmath203 , @xmath204 and @xmath205 , are proportional to space - derivative squared because of the conservation law , parity invariance , and analyticity . we thus have in the fourier space @xmath206 with @xmath207 being the momentum squared . inserting eqs . ( [ eq : gamma ] ) and ( [ eq : gl ] ) into eq . ( [ eq : tdgl ] ) , we obtain the hydrodynamic equation to leading order in @xmath207 as @xmath208 solving eq . ( [ eq : hydro ] ) , we obtain two eigenfrequencies , @xmath209 with @xmath210 and the corresponding eigenmodes , @xmath211 which decay with @xmath212 and @xmath213 , respectively . at the critical point , the energy functional develops a flat direction , @xmath214 , and the susceptibilities of @xmath4 and @xmath190 become divergent . the diffusion coefficient @xmath215 goes to zero at the critical point , which represents the critical slowing down . this result tells us that a small fluctuation of @xmath4 and @xmath190 in the system relaxes with two distinct time scales . the first slow mode @xmath216 is just the conserved charge @xmath190 , whose time evolution is described by the sde @xmath217 note that @xmath218 should be proportional to space - derivative as in eq . ( [ eq : sde ] ) so that this equation is consistent with the continuity equation . on the other hand , @xmath219 is a relaxation mode with nonvanishing relaxation time scale @xmath220 . this time scale is fast compared to that of @xmath221 . in the faster time scale , @xmath4 alone relaxes to the value @xmath222 , which minimize @xmath223 $ ] with @xmath190 being fixed to a given value . after that , with the longer time scale @xmath224 the mode @xmath216 relaxes to @xmath225 . in this stage , @xmath4 simply traces the profile of @xmath190 , and the time evolution of the slow mode is described by the sde ( [ eq : sdec ] ) . the effect of the critical point is encoded in vanishing of @xmath226 in the sde . in the above discussion we neglected the energy - momentum density . in model h , with which the qcd critical point shares the same dynamic universality , the hydrodynamic slow modes in the long wavelength limit are in fact given by the baryon number diffusion and the diffusion of two transverse momentum components @xcite having a nonlinear coupling via the nonzero poisson bracket @xcite . the analysis incorporating the nonlinear coupling is left for future works . in sec . [ sec : analytic ] , we showed that the non - monotonic behaviors of @xmath169 and @xmath170 in eqs . ( [ eq : k ] ) and ( [ eq : c ] ) serve as direct experimental evidence for the existence of a peak structure in @xmath37 . in this appendix , we take a much closer look at the conditions for the appearance of the non - monotonic behaviors in these functions and discuss which function is better in sustaining the non - monotonicity . to simplify the problem , in this appendix we consider the functional form of @xmath37 which has only one maximum as a function of @xmath9 . then , @xmath227 changes its sign only once from positive to negative . in this case , @xmath169 ( @xmath170 ) can have only one local maximum ( minimum ) . thus , the necessary and sufficient conditions for the non - monotonic behaviors of @xmath169 and @xmath170 are given by @xmath228 respectively . from eq . ( [ eq : cum ] ) , the @xmath5 derivative of @xmath169 is given by @xmath229 where , @xmath230 . using @xmath231 one obtains @xmath232 here , @xmath233 is a monotonically decreasing function of @xmath234 with @xmath235 . the integrand in eq . ( [ eq : k->0 ] ) is @xmath236 with a weight @xmath237 , which takes larger value for larger @xmath234 . the sign of @xmath236 with later @xmath234 is more strongly reflected to the sign of eq . ( [ eq : k->0 ] ) . on the other hand , in eq . ( [ eq : k->inf ] ) @xmath236 is integrated with a weight @xmath233 taking larger value for earlier @xmath234 . the sign of @xmath236 with earlier @xmath234 is more responsible for that of eq . ( [ eq : k->inf ] ) . next , the @xmath72 derivative of @xmath170 is calculated to be @xmath238 by taking the small @xmath72 limit , we obtain @xmath239 equation ( [ eq : c->0 ] ) shows that the sign of @xmath240 in the small @xmath72 limit is determined by the integral of @xmath236 with a weight @xmath241 . in the large @xmath72 limit , the weight is given by @xmath242 , which concentrates at the initial time @xmath243 in the large @xmath72 limit . the sign in this limit thus is determined only by @xmath244 , which is positive in the present situation . now , let us compare the conditions eqs . ( [ eq : k > < ] ) and ( [ eq : c > < ] ) . as discussed above , the second condition in eq . ( [ eq : c > < ] ) is always satisfied , while that in eq . ( [ eq : k > < ] ) is not necessarily true but dependent on the functional form of @xmath37 . next , the first conditions in eqs . ( [ eq : k > < ] ) and ( [ eq : c > < ] ) are not always satisfied , but the latter is more favored , because the weight of the latter , @xmath245 , is more concentrated at later @xmath234 than the former , @xmath246 . from these observations , one concludes that the appearance of the non - monotonicity is more likely to appear in @xmath170 than in @xmath169 . although the manifestation of non - monotonicity is more robust in @xmath170 than @xmath169 , in experimental analyses it is meaningful to analyze both of these functions . in the above argument , the position of the local extremum is not determined . the ranges of @xmath5 and @xmath72 which can be measured in experiments are limited owing to the coverage of the detector , and the manifestation of the extremum in this range depends on the functional form . therefore , by analyzing both @xmath169 and @xmath170 , the chance to find the non - monotonic behaviors is enhanced . m. asakawa and k. yazaki , nucl . phys . a * 504 * , 668 ( 1989 ) . a. barducci , r. casalbuoni , s. de curtis , r. gatto and g. pettini , phys . b * 231 * , 463 ( 1989 ) . a. m. halasz , a. d. jackson , r. e. shrock , m. a. stephanov and j. j. m. verbaarschot , phys . d * 58 * , 096007 ( 1998 ) [ hep - ph/9804290 ] . j. berges and k. rajagopal , nucl . b * 538 * , 215 ( 1999 ) [ hep - ph/9804233 ] . o. scavenius , a. mocsy , i. n. mishustin and d. h. rischke , phys . c * 64 * , 045202 ( 2001 ) [ nucl - th/0007030 ] . y. hatta and t. ikeda , phys . d * 67 * , 014028 ( 2003 ) [ hep - ph/0210284 ] . z. fodor and s. d. katz , jhep * 0203 * , 014 ( 2002 ) [ hep - lat/0106002 ] . p. de forcrand and o. philipsen , nucl . b * 642 * , 290 ( 2002 ) [ hep - lat/0205016 ] . g. endrodi , z. fodor , s. d. katz and k. k. szabo , jhep * 1104 * , 001 ( 2011 ) [ arxiv:1102.1356 [ hep - lat ] ] . m. kitazawa , t. koide , t. kunihiro and y. nemoto , prog . phys . * 108 * , 929 ( 2002 ) doi:10.1143/ptp.108.929 [ hep - ph/0207255 ] . t. hatsuda , m. tachibana , n. yamamoto and g. baym , phys . lett . * 97 * , 122001 ( 2006 ) [ hep - ph/0605018 ] . x. luo and n. xu , arxiv:1701.02105 [ nucl - ex ] . star collaboration , `` studying the phase diagram of qcd matter at rhic , '' star notes sn0598 , https://drupal.star.bnl.gov/star/starnotes/public/sn0598 ( 2014 ) . m. asakawa and m. kitazawa , prog . phys . * 90 * , 299 ( 2016 ) [ arxiv:1512.05038 [ nucl - th ] ] . m. a. stephanov , k. rajagopal and e. v. shuryak , phys . d * 60 * , 114028 ( 1999 ) [ hep - ph/9903292 ] . s. jeon and v. koch , [ hep - ph/0304012 ] . m. a. stephanov , phys . lett . * 102 * , 032301 ( 2009 ) [ arxiv:0809.3450 [ hep - ph ] ] . m. asakawa , s. ejiri and m. kitazawa , phys . lett . * 103 * , 262301 ( 2009 ) [ arxiv:0904.2089 [ nucl - th ] ] . l. adamczyk _ et al . _ [ star collaboration ] , phys . lett . * 112 * , 032302 ( 2014 ) [ arxiv:1309.5681 [ nucl - ex ] ] . l. adamczyk _ et al . _ [ star collaboration ] , phys . lett . * 113 * , 092301 ( 2014 ) [ arxiv:1402.1558 [ nucl - ex ] ] . b. berdnikov and k. rajagopal , phys . d * 61 * , 105017 ( 2000 ) [ hep - ph/9912274 ] . m. kitazawa , m. asakawa , and h. ono , phys . lett . b * 728 * , 386 ( 2014 ) . [ arxiv:1307.2978 ] ; m. kitazawa , nucl . a * 942 * , 65 ( 2015 ) [ arxiv:1505.04349 [ nucl - th ] ] . m. sakaida , m. asakawa and m. kitazawa , phys . c * 90 * , no . 6 , 064911 ( 2014 ) [ arxiv:1409.6866 [ nucl - th ] ] . c. nonaka and m. asakawa , phys . c * 71 * , 044904 ( 2005 ) [ nucl - th/0410078 ] . j. i. kapusta and j. m. torres - rincon , phys . c * 86 * , 054911 ( 2012 ) [ arxiv:1209.0675 [ nucl - th ] ] . s. mukherjee , r. venugopalan and y. yin , phys . c * 92 * , no . 3 , 034912 ( 2015 ) [ arxiv:1506.00645 [ hep - ph ] ] . c. herold , m. nahrgang , y. yan and c. kobdaj , phys . c * 93 * , no . 2 , 021902 ( 2016 ) [ arxiv:1601.04839 [ hep - ph ] ] . h. fujii , phys . d * 67 * , 094018 ( 2003 ) [ hep - ph/0302167 ] . d. t. son and m. a. stephanov , phys . d * 70 * , 056001 ( 2004 ) [ hep - ph/0401052 ] . h. fujii and m. ohtani , phys . d * 70 * , 014016 ( 2004 ) [ hep - ph/0402263 ] . y. minami , phys . d * 83 * ( 2011 ) 094019 [ arxiv:1102.5485 [ hep - ph ] ] . j. i. kapusta , b. muller and m. stephanov , phys . c * 85 * , 054906 ( 2012 ) [ arxiv:1112.6405 [ nucl - th ] ] . h. fujii and m. ohtani , prog . suppl . * 153 * , 157 ( 2004 ) [ hep - ph/0401028 ] . m. kitazawa and m. asakawa , phys . c * 85 * , 021901 ( 2012 ) [ arxiv:1107.2755 [ nucl - th ] ] ; phys . c * 86 * , 024904 ( 2012 ) [ erratum - ibid . c * 86 * , 069902 ( 2012 ) ] [ arxiv:1205.3292 [ nucl - th ] ] . c. gardiner , _ stochastic methods _ , springer , 2009 . r. guida and j. zinn - justin , nucl . b * 489 * , 626 ( 1997 ) [ hep - th/9610223 ] . p. schofield , phys . * 22 * , 606 ( 1969 ) . m. a. stephanov , phys . * 107 * ( 2011 ) 052301 . a. bzdak , v. koch and n. strodthoff , arxiv:1607.07375 [ nucl - th ] . p. c. hohenberg and b. i. halperin , rev . phys . * 49 * , 435 ( 1977 ) . l. kumar [ star collaboration ] , nucl . a * 931 * , 1114 ( 2014 ) [ arxiv:1408.4209 [ nucl - ex ] ] . a. bazavov _ et al . _ [ hotqcd collaboration ] , phys . rev . d * 90 * , 094503 ( 2014 ) [ arxiv:1407.6387 [ hep - lat ] ] . m. asakawa , u. w. heinz , and b. mller , phys . * 85 * , 2072 ( 2000 ) [ arxiv : hep - ph/0003169 ] . s. jeon and v. koch , phys . lett . * 85 * , 2076 ( 2000 ) [ arxiv : hep - ph/0003168 ] . v. koch , arxiv:0810.2520 [ nucl - th ] . g. aarts , c. allton , a. amato , p. giudice , s. hands and j. i. skullerud , jhep * 1502 * , 186 ( 2015 ) [ arxiv:1412.6411 [ hep - lat ] ] . y. ohnishi , m. kitazawa and m. asakawa , phys . c * 94 * , 044905 ( 2016 ) [ arxiv:1606.03827 [ nucl - th ] ] . b. abelev _ et al . _ [ alice collaboration ] , phys . lett . * 110 * , 152301 ( 2013 ) [ arxiv:1207.6068 [ nucl - ex ] ] . e. v. shuryak and m. a. stephanov , phys . c * 63 * , 064903 ( 2001 ) [ hep - ph/0010100 ] . s. pratt , phys . lett . * 108 * , 212301 ( 2012 ) [ arxiv:1203.4578 [ nucl - th ] ] .

the definition of relative accelerations and strains among a set of comoving particles is studied in connection with the geometric properties of the frame adapted to a fiducial observer . " we find that a relativistically complete and correct definition of strains must take into account the transport law of the chosen spatial triad along the observer s congruence . we use special congruences of ( accelerated ) test particles in some familiar spacetimes to elucidate such a point . the celebrated idea of szekeres compass of inertia , arising when studying geodesic deviation among a set of free - falling particles , is here generalized to the case of accelerated particles . in doing so we have naturally contributed to the theory of relativistic gravity gradiometer . moreover , our analysis was made in an observer - dependent form , a fact that would be very useful when thinking about general relativistic tests on space stations orbiting compact objects like black holes and also in other interesting gravitational situations .

1408.4283

in a series of papers , long ago , de felice and coworkers @xcite defined and studied the relative strains among a set of comoving particles in black hole spacetimes , confined to a normal neighborhood of the observer s world line and in a well specified state of motion . the particles were considered as test with respect to the background geometry . starting from that analysis , we consider here how the definition of relative accelerations and strains is affected by the geometric properties of the frame adapted to the fiducial observer ( e.g. transport law of the spatial triad along the observer s congruence ) . within this more general context we reconsider previous works and extend that discussion in view of possible quasi - local experiments in space laboratories moving in different gravitational environments . the basis of our observer - dependent analysis of relative strains is the concept of gravitational compass introduced by szekeres @xcite ( see also @xcite ) and the related discussion about the problem of setting up a preferred frame within which to study the gravitational field . according to szekeres , a gravitational compass consists in an arrangement of three test particles joined by springs to a central observer ; their relative deviation is then investigated via the geodesic deviation equation to deduce the physical significance of the weyl tensor components . at the instant of measurement the reference particle drops the apparatus observing the strains on the springs . the relative acceleration between two nearby geodesics is completely determined by the electric part of the riemann tensor , which can be thought of as a symmetric force distribution whose six independent components are the strains on the six springs . when the off diagonal terms ( i.e. the transverse strains ) vanish , the springs connecting the test particles to the observer lie along the principal axes of the tidal force matrix , so that the apparatus maps out the local gravitational field , acting just as a sort of compass . expressing the riemann tensor in terms of the weyl tensor ( with which coincides in the case of vacuum spacetimes ) allowed szekeres to discuss the physical meaning of the petrov classification . the electric part of the weyl tensor represents the only direct curvature contribution to the geodesic deviation equation ( as well as the deviation equation for general non - geodesic motion ) , introducing shearing forces due to its property of being symmetric and trace - free . actually szekeres gravitational compass is only valid to describe an idealized situation . for any practical use , in fact , it should be replaced by a gravity gradiometer , " i.e. a device to perform measurements of the local gradient of the tidal gravitational force . the theory of a relativistic gravity gradiometer has been developed by mashhoon , theiss , paik and will @xcite in view of satellite experiments around the earth in the framework of post - newtonian approximation . it should also be noted that a modern observational trend is to use atomic interferometry to build the future generation of highly precise gravity gradiometers ( see @xcite and references therein ) . recently chicone and mashhoon @xcite obtained a generalized geodesic deviation equation in fermi coordinates as well as in arbitrary coordinates as a taylor expansion in powers of the components of the deviation vector , retaining terms up to first order , but without any restriction on the relative spatial velocities . they then investigated in a number of papers @xcite the motion of a swarm of free particles ( in both non - relativistic and relativistic regime ) relative to a free reference particle which is on a radial escape trajectory away from a collapsed object ( a schwarzschild as well as a kerr black hole ) , discussing the astrophysical implications of the related ( observer - dependent ) tidal acceleration mechanism . the further dependence of the deviation equation on the four acceleration of the observer as well as his three velocity has been accounted very recently by mullari and tammelo @xcite . aim of this work is to set up the necessary assessment to assure stability of extended bodies as they move in a given spacetime , identifying which parts of them should be provided of a more rigid structure to resist tidal or acceleration strains , according to their internal constituency . in fact we are not concerned here with internal stresses but with the external field of strains generated by the geometrical environment . special attention will be devoted to type d vacuum stationary axisymmetric spacetimes . in this case , szekeres analysis suggests that the tidal strains cause the distorsion of a sphere of test free particles about the observer into an ellipsoid , as a typical behaviour of particles falling towards the central attracting body . we are thus interested in studying how szekeres picture as well as the one associated with the relativistic gravity gradiometry modifies when the acceleration strains are also present , and which frame is most convenient to measure either tidal or inertial forces experienced by an extended body . the paper is organized as follows . in section 2 we derive the relative acceleration equation and give the ( observer and frame - dependent ) definition of the strain tensor . in section 3 we analyze the strains which affect a bunch of uniformly rotating particles in the flat minkowski spacetime . in section [ kerrcase ] we discuss instead the deviation acceleration and associated strains in vacuum stationary axisymmetric spacetimes from either geometrically or physically motivated timelike congruences . in particular we explore the cases of the born - rigid congruence of static observers , of the irrotational family of zero angular momentum observers ( zamos ) , and ( limiting to the schwarzschild and kerr black hole spacetimes ) of the geodesic and irrotational congruence of painlev - gullstrand @xcite orbits . in what follows greek indices refer to coordinate components while latin indices refer to tetrad components . the formers run from 0 to 3 , the latters from 1 to 3 . the index ( apex ) 0 is often used with the meaning of time tetrad component too . units are choosen so that @xmath0 and the adopted metric signature is @xmath1 . let us consider a bunch of test particles , i.e. a congruence @xmath2 of timelike world lines , with unit tangent vector @xmath3 ( @xmath4 ) parametrized by the proper time @xmath5 . let @xmath6 be the reference world line of the congruence , which we consider as that of the fiducial observer . " in general , the lines of the congruence @xmath7 as well as that of the observer are accelerated with acceleration @xmath8 . the separation between the line @xmath6 and a general line of the congruence is represented by a connecting vector @xmath9 , i.e. a vector undergoing lie transport along @xmath3 : [ eq:1 ] _ u y=0 _ uy=_yu . the term @xmath10 in eq . ( [ eq:1])@xmath11 involves the covariant derivative of @xmath3 , which can be written in terms of the kinematical fields of the congruence as follows : [ eq:1b ] _ u^= - a(u)^u_-k(u)^ _ , where @xmath12 is the kinematical tensor , which summarizes the vorticity of the congruence @xmath13}$ ] and the expansion @xmath14 . here square and round brakets denote antisymmetrization and symmetrization of tensor indices , respectively . from eq . ( [ eq:1b ] ) we also have @xmath15}\ , \qquad \theta(u)_{\alpha\beta } = p(u)_\alpha^\mu p(u)_\beta^\nu \nabla_{(\mu}u_{\nu)}\ , \end{aligned}\ ] ] where @xmath16 is the operator of the orthogonal projection with respect to @xmath3 . ( [ eq:1 ] ) becomes = _ y u= -(yu)a(u)-k(u)y , where @xmath17 denotes index contraction @xcite , so that @xmath18^\beta = k(u)^\beta{}_\alpha y^\alpha$ ] . the covariant derivative along @xmath3 of both sides of eq . ( [ eq:1])@xmath11 gives rise to the relative acceleration equation `` [ eq:2 ] = -r(u , y)u+_y a(u ) , where @xmath19 represents the tidal force contribution to the relative acceleration , whereas @xmath20 is the inertial '' contribution due to the observer s acceleration . we notice that @xmath21 is the electric part of the riemann tensor ( as measured by the observer @xmath3 ) . equation ( [ eq:2 ] ) can then be conveniently rewritten as follows : [ eq:3 ] = -y , = e(u)-a(u ) . let us now set up an orthonormal frame @xmath22 adapted to the congruence @xmath3 and write both the lie transport equation ( [ eq:1])@xmath11 and the relative acceleration equation ( [ eq:2 ] ) with respect to this frame . the spatial triad is generic in the sense that it rotates with a certain angular velocity @xmath23 with respect to gyro - fixed axes along @xmath3 : [ spat_frame ] p(u)_u e_a _ ( fw)(u)e_a= _ ( fw , u , e)e_ac_(fw , u , e)^b_a e_b . here the subscript @xmath24 means that we are referring to a @xmath25 tetrad and a fermi - walker derivative operation ; the cross product @xmath26 is defined in the local rest space of @xmath3 by @xmath27 and the fermi - walker structure functions are defined by @xmath28 ; @xmath29 is the _ spatial fermi - walker derivative along @xmath3 _ @xcite , such that for a generic spatial vector @xmath30 ( i.e. @xmath31 ) [ spatfwder ] _ ( fw)(u)x^a = x^a + c _ ( fw , u , e)^a_b x^b , the overdot denoting differentiation with respect to the proper time @xmath5 . introduce the frame components of @xmath9 , i.e. the decomposition @xmath32 ( and the notation @xmath33 , @xmath34 ) . the lie transport equation ( [ eq:1])@xmath11 then becomes @xmath35u+\omega_{({\rm fw},u , e)}\times \vec y \nonumber \\ & = & y^0 a(u)-k(u)\rightcontract \vec y\ , \end{aligned}\ ] ] where the relation @xmath36 has been used , whence u+ y^a e_a + _ ( fw , u , e)y = -k(u)y , yielding @xmath37^a + k(u)^a{}_by^b=0\ .\end{aligned}\ ] ] from the definition of @xmath38 the relative velocity equation " ( [ conseq2 ] ) can be written as y^a + [ ( _ ( fw , u , e)-(u))y]^a -(u)^a_by^b=0 , implying that @xmath39 when @xmath40 and @xmath41 . as we will see in sec . 4.4 the latter condition is satisfied by a frenet - serret frame along a born - rigid congruence . let us turn to the relative acceleration equation ( [ eq:2 ] ) . substituting eq . ( [ conseq1 ] ) into the first line of ( [ cond_lie ] ) leads to _ u y= y^0 a(u)+ y^a e_a + _ ( fw , u , e)y . taking the covariant derivative along @xmath3 of both sides of the previous equation gives the left hand side of the relative acceleration equation ( [ eq:2 ] ) : @xmath42u-[\vec y\cdot a(u)]a(u)+\nonumber \\ \fl\quad & & + y^0 \left[\dot a(u)^a e_a+\omega_{({\rm fw},u , e)}\times a(u)\right]+ \ddot y^a\ , e_a + \nonumber \\ \fl\quad & & -2 \omega_{({\rm fw},u , e)}\times [ k(u)\rightcontract \vec y]-\omega_{({\rm fw},u , e)}\times [ \omega_{({\rm fw},u , e)}\times \vec y]+\nonumber \\ \fl\quad & & + \dot \omega_{({\rm fw},u , e)}\times \vec y\ , \end{aligned}\ ] ] where eq . ( [ conseq1 ] ) has been taken into account and @xmath43 stands for @xmath44 . it is also easy to evaluate the term @xmath20 on the right hand side of eq . ( [ eq:2 ] ) ; a direct calculation shows that @xmath45+\nonumber \\ & & + y^b\nabla(u)_b a(u)-a(u)\cdot ( k(u)\rightcontract \vec y ) u\ , \end{aligned}\ ] ] where @xmath46 . as a result , from eqs . ( [ eq : secder ] ) and ( [ nablaya ] ) we obtain @xmath47a(u)+\nonumber \\ \fl\quad & & -\omega_{({\rm fw},u , e)}\times [ \omega_{({\rm fw},u , e)}\times \vec y]\nonumber \\ \fl\quad & & -2 \omega_{({\rm fw},u , e)}\times [ k(u)\rightcontract \vec y]+\dot \omega_{({\rm fw},u , e)}\times \vec y \nonumber \\ & = & -{\mathcal e}(u)\rightcontract \vec y \ .\end{aligned}\ ] ] de felice @xcite introduced the relative strains as components of the following tensor : [ eq : strainsdef ] s(u)=(u ) a(u)+a(u)a(u ) , namely @xmath48 . the tensor @xmath49 will be termed strain tensor ( actually fermi - walker strain tensor , see below ) ; it depends only on the congruence @xmath3 and not on the chosen spatial triad @xmath50 . to make our formulas more compact we also introduce the notation @xmath51^a{}_b-2c_{({\rm fw},u , e)}{}^a{}_c\ , k(u)^c{}_b\nonumber \\ \fl\quad & = & \delta^a_b \omega_{({\rm fw},u , e)}^2-\omega_{({\rm fw},u , e)}^a \omega_{({\rm fw},u , e)}{}_b -\epsilon^a{}_{bf}\dot \omega_{({\rm fw},u , e)}^f\nonumber \\ \fl\quad & & -2\epsilon^a{}_{fc}\omega_{({\rm fw},u , e)}^f k(u)^c{}_b\ , \end{aligned}\ ] ] where @xmath52^a{}_b = c_{({\rm fw},u , e)}{}^a{}_c\ , c_{({\rm fw},u , e)}{}^c{}_b$ ] . the relative acceleration equation ( [ eq:2 ] ) ( or equivalently ( [ eq:3 ] ) ) then becomes [ eq : secder3bis ] y^a + k_(u , e)^a_b y^b=0 , where _ ( u , e)^a_b=[t_(fw , u , e)-s(u)+e(u)]^a_b . ( [ eq : secder3bis ] ) are our master equations " which we shall analyze in the following special cases : * _ flat spacetime _ : @xmath53 so that @xmath54 . in this case we have @xmath55 . * @xmath50 _ spatial triad fermi - walker dragged along _ @xmath3 : @xmath56 . this implies @xmath57 , so that @xmath58 . * @xmath3 _ geodesic _ : @xmath59 . in this case @xmath60 and hence @xmath61 . * @xmath3 _ irrotational _ : @xmath62 , so that @xmath63 . * @xmath3 _ born - rigid _ : @xmath64 , so that @xmath65 . clearly we can also consider combinations of the above special cases . for example a congruence of geodesic and irrotational orbits : * @xmath3 _ geodesic and irrotational _ : @xmath59 and @xmath62 ( @xmath63 ) . in this case @xmath60 , so that @xmath61 . finally , the case @xmath66 corresponds to @xmath67 , i.e. absence of relative accelerations among the particles of the congruence . the derivation of the relative acceleration equation ( [ eq : secder3bis ] ) for the frame components of the deviation vector has been obtained by using a generic spatial triad @xmath50 characterized by its rotation with respect to a fermi - walker transported triad along @xmath3 according to eq . ( [ spat_frame ] ) . actually , one can characterize the generic triad @xmath50 in different ways ; for instance by its rotation with respect to other ( not fermi - walker ) geometrically meaningful frames dragged along @xmath3 . in @xcite two more derivatives of spatial vectors along @xmath3 have been introduced , namely 1 . _ spatial co - rotating fermi - walker transport along @xmath3 , _ defined by the relation @xmath68\times e_a = c _ { ( { \rm cfw},u , e)}{}^b{}_ae_b\ , \end{aligned}\ ] ] whence _ ( cfw)(u)x^a = x^a + c_(cfw , u , e)^a_bx^b ; 2 . _ spatial lie transport along @xmath3 , _ defined by the relation _ ( lie)(u)e_a=_(cfw)(u)e_a-(u)^b_a e_b= c_(lie , u , e)^b_ae_b , whence _ ( lie)(u)x^a = x^a + c_(lie , u , e)^a_bx^b . the different structure functions are then related each other by c_(cfw , u , e)^b_a = c _ ( fw , u , e)^b_a+(u)^b_a = c_(lie , u , e)^b_a+(u)^b_a . due to the fact that these three spatial temporal derivatives " are the only true geometrically motivated operators , a useful notation to handle with them contemporarily has also been introduced in @xcite \{_(tem)(u)}_tem = fw , cfw , lie=\{_(fw)(u),_(cfw)(u),_(lie)(u ) } . here the subscript @xmath69 means that we are referring to a @xmath25 tetrad and a tem = fw , cfw , lie derivative operation . for example , the lie transport equation ( [ conseq2 ] ) can be written in the equivalent forms _ ( fw)(u)y+k(u)y=0 , _ ( cfw)(u)y-(u)y=0 , _ ( lie)(u)y=0 . let us now turn to the relative acceleration equation ( [ eq : secder3bis ] ) ; a straightforward calculation shows that _ ( fw)(u)^2 y^a = y^a + t_(fw , u , e)^a_by^b , hence clarifying the meaning of the spatial tensor @xmath70 . equation ( [ eq : secder3bis ] ) can then be written as [ strain_fw ] _ ( fw)(u)^2 y + [ e(u)-s(u)]y = 0 , so that the de felice s strain definition corresponds in the present treatment to a fermi - walker strain definition . let us explore this circumstance in detail . analogously to eq . ( [ strain_fw ] ) one can write _ ( cfw)(u)^2 y^a = y^a + t _ ( cfw , u , e)^a_by^b , _ ( lie)(u)^2 y^a = y^a + t _ ( lie , u , e)^a_by^b , and derive the relations between @xmath71 ( all defined by the first of eqs . ( [ tdef ] ) using the proper functions @xmath72 ) so that @xmath73 with @xmath74 , @xmath75 . hence besides eq . ( [ strain_fw ] ) one has the companion relations [ strain_cfw ] _ ( cfw)(u)^2 y + t _ ( cfw - fw , u , e)y+ [ e(u)-s(u)]y = 0 , and [ strain_lie ] _ ( lie)(u)^2 y + t _ ( lie - fw , u , e)y+ [ e(u)-s(u)]y = 0 . the latter equation can be further simplified recalling that @xmath76 and hence @xmath77 : _ ( lie - fw , u , e)+e(u)-s(u)=0 . this analysis would allow a more involved definition of strains in terms of the kinematical properties of the frame with respect to which they are measured . for instance , one can summarize eqs . ( [ strain_fw ] ) , ( [ strain_cfw ] ) and ( [ strain_lie ] ) in a single one [ strain_tem ] _ ( tem)(u)^2 y + [ e(u)-s _ ( tem , u , e)]y = 0 , , and more properly identify -dependent " strains @xmath78 clearly , when the spatial frame @xmath50 undergoes a _ tem _ transport along the congruence the quantities @xmath72 and @xmath79 vanish identically and _ ( tem)(u ) y^a = y^a , _ ( tem)(u)^2 y^a = y^a , so that eq . ( [ strain_tem ] ) implies [ strain_tem2 ] y^a + [ e(u)-s_(tem , u , e)]^a_b y^b = 0 , _ ( tem)(u)e_a=0 . the _ tem_-dependent analysis of strains is not just of academic interest . as a matter of fact , it is a consequence of a systematic use of spacetime splitting techniques in general relativity : the latters not only reproduce the observer point of view but also play a key role in the assessment of a nonlocal measurement in the exact theory of general relativity . however , since a fermi - walker frame is operationally easier to set up , we shall confine our attention to the fermi - walker strain definition given by de felice and coworkers @xcite . consider first the simplest case of flat spacetime . let us write the minkowski metric in standard cilyndrical coordinates @xmath80 [ munkmetr ] s^2= -t^2 + r^2 + r^2 ^ 2 + z^2 , and introduce the orthonormal frame e_t=_t , e_r=_r , e_z=_z , e_= _ . consider a family of uniformly rotating particles with angular velocity @xmath81 ; the four velocity @xmath3 of the generic particle of the congruence is then given by [ uflat ] u=(_t + _ ) = ( e_t + e _ ) , = ( 1-^2)^-1/2 , where =( 1-r^2 ^ 2 ) ^-1/2 , = . a frame adapted to @xmath3 can be fixed as [ circmink ] e(u)_1=e_r , e(u)_2=(e_t + e _ ) , e(u)_3=e_z . we note that this frame is of frenet - serret type as will be introduced in sec . the orbits are accelerated : @xmath82 , with vanishing expansion ( @xmath41 ) and with vorticity vector @xmath83 . it is easy to show that the deviation equations ( [ eq : secder3bis ] ) reduce to @xmath67 , since @xmath66 resulting from the balancing between the strain tensor and the fermi - walker tensor , namely @xmath84 with only nonvanishing components s(u)_11=s(u)_22=-^4 ^ 2 . the relative velocity equation ( [ conseq2 ] ) implies that in addition @xmath39 , so that the spatial components of the deviation vector remain all constant along the path with respect to the frame ( [ circmink ] ) . rotating the spatial triad in the 2-plane @xmath85 by an angle @xmath86 ( @xmath5 denoting proper time parametrization along @xmath3 ) one obtains a fermi - walker triad @xmath87 with respect to this new triad @xmath88 , so that @xmath89 and the only nonvanishing components are _ ( u , e)_11=k_(u , e)_22=^4 ^ 2 , implying harmonic oscillations for the deviation vector components @xmath90 and @xmath91 with frequency @xmath92 ( see fig . [ fig:1 ] ) . this also follows from the relative velocity equations y^1=-^2 y^2 , y^2=^2 y^1 , y^3=0 . the corresponding solution is straightforward : @xmath93 where @xmath94 are the components of the deviation vector at the starting point . this implies that an initially circular bunch of particles on the @xmath95-@xmath96 plane remains always circular for increasing values of the proper time . the behaviour of the magnitude of the fermi - walker angular velocity as a function of the linear velocity @xmath97 is shown in fig . [ fig:1 ] for a fixed value of the radial coordinate . is shown as a function of the linear velocity @xmath97 for a given circular orbit at @xmath98 . ] practically , we may think of a rigidly rotating disk , say a merry - go - round . on such a disk one may have bodies which are either fixed with the platform itself and comoving with it ( say a car `` ) or have only a fixed point and an axis parallel to the rotation axis of the disk about which it can rotate freely ( say a montgolfier '' ) . the connecting vector @xmath9 of our previous analysis can be taken in this case as the distance between the bodies ( for the montgolfier `` this distance is meant with respect to its fixed point ) . the components of this vector are constant when they are referred to a frame triad at rest with the rotating disk , but they undergo oscillations when they are referred to a frame triad fixed with respect to the infinity ( e.g. , axes at rest with the mongolfier '' ) . these two cases are exactly what we have considered in the above discussion , namely the car `` example on the merry - go - round corresponds to the first frenet - serret frame , while the mongolfier '' example corresponds to the fermi - walker frame . consider the case of vacuum stationary axisymmetric spacetimes . using a coordinate system @xmath99 adapted to the spacetime symmetries , i.e. with @xmath100 ( timelike ) and @xmath101 ( spacelike ) a pair of commuting killing vectors the metric can be written as [ metr_gen ] s^2= g_tt t^2 + 2 g_t t + g_^2 + g_rrr^2 + g _ ^2 , where all the metric coefficients depend only on @xmath102 and @xmath103 . due to the spacetime symmetries it is quite natural to consider two families of observers which are described by two geometrically motivated congruences of curves : 1 . _ static observers _ , at rest at a given point in the spacetime ; their four velocity is aligned with the killing temporal direction m=_t , m= , with dual m^=-m(t - m _ ) , m_=-g_t / g_tt ; we denote by @xmath104 the proper time parameter along @xmath105 defined by @xmath106 . we notice that @xmath107 and @xmath108 are called the lapse and shift functions for the static observers respectively , according to a terminology due to wheeler @xcite . introducing the lapse and shift notation , the spacetime metric ( [ metr_gen ] ) can also be written as [ metr_thd ] s^2= -m^2 ( t - m_)^2 + _ ^2 + g_rrr^2 + g _ ^2 , where @xmath109 ; in this case an orthonormal frame adapted to the static observers is given by [ thdframe ] e(m)_1= _ r , e(m)_2= _ , e(m)_3= [ _ + m__t ] , and in general is not a fermi - walker frame since @xmath110 . the congruence of static observers is born - rigid : @xmath111 , but has in general a nonzero vorticity . + in order to be at rest in the spacetime ( [ metr_gen ] ) a particle must be accelerated to balance the gravitational dragging which would force it to co - rotate with the source . evidently , if we have a bunch of particles and we want them to be at rest with respect to the infinity forming a rigid body they must be differentially accelerated . in the case of co - rotation we should consider instead the following family of observers : 2 . _ zero angular momentum observers _ or zamo , a family of locally nonrotating observers with four velocity [ n ] n = n^-1(_t - n^ _ ) , n=(-g^tt)^-1/2 , n^=g_t / g _ , with dual n^=-nt ; here @xmath112 and @xmath113 are the lapse and shift functions associated with the zamos , respectively ; we denote by @xmath114 the proper time parameter along @xmath115 defined by @xmath116 . the spacetime metric ( [ metr_gen ] ) can then be written as [ metr_slic ] s^2= -n^2t^2 + g_rrr^2 + g _ ^2 + g_(+n^t)^2 , so that a suitable orthonormal frame adapted to the zamos is now fixed by the triad [ zamoframe ] e(n)_1=_r , e(n)_2= _ , e(n)_3=1 _ . for the zamo spatial triad we use the more standard notation e_r = e(n)_1 , e_=e(n)_2 , e_=e(n)_3 . this frame in general is not a fermi - walker frame since @xmath117 . the congruence of zamos is irrotational : @xmath118 , but is not born - rigid in general . let now the spacetime be given by the kerr solution . in standard boyer - lindquist coordinates the metric writes as @xmath119 with inverse @xmath120 ^ 2 + \frac{1}{\sigma \sin^2 \theta } \left [ \partial_\phi+ a \sin^2 \theta \partial_t\right]^2\nonumber\\ \fl\quad & & + \frac{\delta}{\sigma}(\partial_r)^2+\frac{1}{\sigma}(\partial_\theta)^2\ , \end{aligned}\ ] ] where @xmath121 and @xmath122 . here @xmath123 and @xmath124 are the total mass and specific angular momentum characterizing the spacetime . the ( outer ) event horizon is located at @xmath125 . the explicit expressions for the quantities entering the relative acceleration equation ( [ eq : secder3bis ] ) and corresponding to a congruence of static observers are listed in [ appkthread ] . we find that the matrix @xmath126 relative to the frame ( [ thdframe ] ) is identically zero . as a result , the system of deviation equations ( [ eq : secder3bis ] ) reduces to y^a=0 . the first order system of lie transport equations ( [ conseq2 ] ) implies that in addition y^a=0 , so that all the spatial components of the deviation vector remain constant along the path . this means that the system under consideration is rigid in the sense that the tidal deformation induced by the geometry is balanced by a suitable choice of added strains and triad frame . let us now consider how the set of deviation equations modifies when referred to a fermi - walker transported frame adapted to @xmath105 . preliminarly , define the pair of orthogonal unit vectors @xmath127 , the former being aligned with the fermi - walker angular velocity @xmath128 : @xmath129\ , \nonumber\\ \fl\quad \vec \omega_{({\rm fw},m , e)}^{\perp}&=&||\omega_{({\rm fw},m , e)}||^{-1}[-\omega_{({\rm fw},m , e)}{}_{2}e(m)_1+\omega_{({\rm fw},m , e)}{}_{1}e(m)_2]\ .\end{aligned}\ ] ] a fermi - walker transported spatial triad is then given by @xmath130 with respect to such a frame the matrix components @xmath131 are identically zero ; thus the deviation matrix is given by @xmath132 , where the components of @xmath133 and @xmath134 are now evaluated with respect to the new frame . the matrix @xmath135 results to be diagonal , with nonvanishing components @xmath136 , leading to oscillating behaviours of the components @xmath96 and @xmath137 of the deviation vector with the same frequency @xmath138 . the component @xmath95 , directed along the fermi - walker angular velocity , remains instead constant along the path . this follows from the relative velocity equations ( [ conseq2 ] ) , namely y^1=0 , y^2=- ||_(fw , m , e)|| y^3 , y^3= ||_(fw , m , e)|| y^2 , which can be easily integrated @xmath139 where @xmath94 denote the components of the deviation vector at the starting point . it implies that an initially circular bunch of particles on the @xmath96-@xmath137 plane does not change its shape as the proper time varies . the behaviour of the magnitude of the fermi - walker angular velocity as a function of the radial coordinate is shown in fig . [ fig:2 ] . in general the oscillating variation of @xmath96 and @xmath137 is due to the relative motion with respect to the chosen fermi - walker frame . in our case the @xmath95 component being stably aligned with the @xmath140-leg of the triad locally identifies the fermi - walker rotation axis . this particular behaviour suggests a possible experiment to measure the gravitational dragging . is shown as a function of @xmath141 , for @xmath142 and @xmath143 . note that the static observers do not exist any longer inside the ergosphere , located at @xmath144 , where @xmath138 diverges . ] the explicit expressions for the quantities entering the relative acceleration equation ( [ eq : secder3bis ] ) and corresponding to a congruence of zamos are listed in [ appkzamo ] . we find that the matrix @xmath145 is identically zero . as a result , the system of deviation equations ( [ eq : secder3bis ] ) trivially reduces to [ eq:417 ] y^a=0 . the first order system of the lie transport equations ( [ conseq2 ] ) leads to [ eq1ordzamo ] y^1=0 , y^2=0 , y^3=2[_(fw , n , e)_2y^1_0-_(fw , n , e)_1y^2_0 ] , where the components @xmath146 and @xmath147 of the deviation vector , calculated at the initial time , remain constant along the path . the solution for the component @xmath148 can thus be written as y^3=y^3_0_n+y^3_0 , where the quantity @xmath149 is given by the right hand side of the last equation ( [ eq1ordzamo ] ) evaluated at the starting point , implying linear deviation along the direction @xmath150 as a function of the proper time . this is clearly related to the fact that close - by particles moving along circular orbits at different values of the radial coordinate @xmath102 have different angular velocity , the latter being that of the gravitational dragging . evidently , since @xmath148 changes uniformly the system of particles is not internally accelerated as from eq . ( [ eq:417 ] ) ; however , as stated above , it is not a rigid frame . as in the case of a static observer it is interesting to write the set of deviation equations with respect to a fermi - walker transported frame adapted to @xmath115 . as before , let us introduce the pair of orthogonal unit vectors @xmath151 , the former being aligned with the fermi - walker angular velocity @xmath152 : @xmath153\ , \nonumber\\ \fl\quad \vec \omega_{({\rm fw},n , e)}{}^{\perp}&=&||\omega_{({\rm fw},n , e)}||{}^{-1}[-\omega_{({\rm fw},n , e)}{}_{2}e(n)_1+\omega_{({\rm fw},n , e)}{}_{1}e(n)_2]\ ; \end{aligned}\ ] ] a fermi - walker transported spatial triad is then given by @xmath154 the only nonvanishing components of the deviation matrix @xmath155 with respect to the new triad ( [ fwzamo ] ) turn out to be @xmath156\ , \nonumber\\ \fl\quad { \mathcal k}_{(n , e')}{}_{23}&=&-4||\omega_{({\rm fw},n , e)}||^2\sin(||\omega_{({\rm fw},n , e)}||{}\tau_n)\cos(||\omega_{({\rm fw},n , e)}||{}\tau_n)\ , \nonumber\\ \fl\quad { \mathcal k}_{(n , e')}{}_{33}&=&-||\omega_{({\rm fw},n , e)}||^2[3 - 4\cos^2(||\omega_{({\rm fw},n , e)}||{}\tau_n)]\ .\end{aligned}\ ] ] the relative velocity equations ( [ conseq2 ] ) are given by @xmath157\ , \nonumber\\ \fl\quad \dot y'{}^3&= & -||\omega_{({\rm fw},n , e)}||\ , [ \cos(2||\omega_{({\rm fw},n , e)}||\tau_n)y'{}^2-\sin(2||\omega_{({\rm fw},n , e)}||\tau_n)y'{}^3]\ , \end{aligned}\ ] ] and can be easily integrated @xmath158 where @xmath94 are the components of the deviation vector at the starting point . the behaviour of the components @xmath137 vs. @xmath96 for increasing values of the proper time @xmath114 is shown in fig . [ fig:3 ] for different setting of the initial conditions . the resulting squeezing of an initially circular bunch of particles on the @xmath96-@xmath137 plane is depicted in fig . [ fig:4 ] . figure [ fig:5 ] shows instead the behaviour of the magnitude of the fermi - walker angular velocity as a function of the radial coordinate . @xmath159{fig3a.eps}&\qquad \includegraphics[scale=0.42]{fig3b.eps}&\\[.2 cm ] \mbox{(a ) } & \qquad \mbox{(b ) } \end{array } $ ] + @xmath160{fig4a.eps}&\quad \includegraphics[scale=0.4]{fig4e.eps}&\\[.2 cm ] \includegraphics[scale=0.4]{fig4b.eps}&\quad \includegraphics[scale=0.4]{fig4f.eps}&\\[.2 cm ] \includegraphics[scale=0.4]{fig4c.eps}&\quad \includegraphics[scale=0.4]{fig4g.eps}\\[.2 cm ] \includegraphics[scale=0.4]{fig4d.eps}&\quad \includegraphics[scale=0.4]{fig4h.eps } \end{array } $ ] + is shown as a function of @xmath141 , for @xmath142 and @xmath143 . the horizon is located at @xmath161 , where @xmath162 . ] in the kerr spacetime it is also interesting to study the painlev - gullstrand geodesic and irrotational family of orbits @xcite . the associated four velocity 1-form , denoted by @xmath163 , is given by [ pg4vel ] ^=-t - r . it is easy to see that [ caln ] = _ ( , n)[n+(,n ) e_r ] , so that the painlev - gullstrand geodesics observers move radially with respect to the zamos with a relative speed ( , n)=-=- , _ ( , n)= . we denote by @xmath164 the proper time parameter along @xmath165 defined by @xmath166 . a frame adapted to the painlev - gullstrand observers can be fixed with the triad [ framepg ] e()_1=_(,n)[(,n ) n+e(n)_1 ] , e()_2= e(n)_2 , e()_3= e(n)_3 . the most relevant quantities associated with the painlev - gullstrand observers are listed in [ appkpg ] . the spatial components of the deviation vector are then obtained solving the coupled system of second order differential equations ( [ eq : secder3bis ] ) , where the nonvanishing components of @xmath167 are given by eq . ( [ kabpg ] ) . they are all functions of the radial coordinate @xmath102 which in the case of painlev - gullstrand observers is depending on the proper time @xmath164 . therefore , the integration of the equations can be performed numerically . the corresponding first order system of lie transport equations ( [ conseq2 ] ) writes as @xmath168\ .\end{aligned}\ ] ] as above , one can also write the set of deviation equations with respect to a fermi - walker transported frame adapted to @xmath169 . a straigthforward calculation shows that a fermi - walker transported spatial triad is given by @xmath170-\cos\beta e(\mathcal { n})_2\ , \nonumber\\ \fl\quad e'({\mathcal n})_2&=&-\cos\beta [ -\sin\alpha e(\mathcal { n})_1+\cos\alpha e(\mathcal { n})_3]+\sin\beta e(\mathcal { n})_2\ , \nonumber\\ \fl\quad e'({\mathcal n})_3&=&\cos\alpha e(\mathcal { n})_1+\sin\alpha e(\mathcal { n})_3\ , \end{aligned}\ ] ] where =( ) , = + ( ) . the only nonvanishing components of the deviation matrix @xmath171 coincide with @xmath172 , whose frame components are given by eq . ( [ kabpgfw ] ) . the relative velocity equations ( [ conseq2 ] ) turn out to be @xmath173 where @xmath174 are the components of @xmath175 with respect to the fermi - walker triad . this system of first order deviation equations can be integrated numerically treating as integration variable the radial coordinate @xmath102 . the components @xmath176 in fact can be made explicitly depending on the radial coordinate @xmath102 instead of the proper time using the relation @xmath177 . their behaviours as functions of @xmath102 are shown in fig . [ fig:6 ] . of the deviation vector are shown as functions of @xmath141 for the choice of parameters @xmath178 and @xmath143 ( the horizon is thus located at @xmath179 ) . the initial conditions are set to @xmath180 and @xmath181 . we see that @xmath95 increases for decreasing @xmath102 , whereas the deviations along the angular directions are both decreasing . ] the situation greatly simplifies on the equatorial plane ( @xmath182 ) , where the deviation matrix becomes diagonal _ ( n , e)= , and the only nonvanishing components of the expansion tensor are given by = . the system ( [ eq1ordpgfw ] ) then writes as [ eq1ordpgfweq ] = - , = , = y^3 , and can be easily integrated : [ sol4383 ] y^1=y^1_0 , y^2=y^2_0 , y^3=y^3_0 . note that the solution ( [ sol4383 ] ) for @xmath137 is valid everywhere , not only on the equatorial plane , since eq . ( [ eq1ordpgfweq])@xmath183 is actually general . the behaviours of the components @xmath176 as functions of @xmath102 are practically the same as those shown in fig . [ fig:6 ] . figure [ fig:7 ] shows instead the squeezing of an initially circular bunch of particles on the @xmath95-@xmath96 plane for decreasing values of the radial coordinate . no significative differences appear for different values of the polar angle @xmath103 ; the squeezing on the @xmath95-@xmath137 plane also exhibits the same features although with a slight but not negligible difference from the @xmath96 component at small values of @xmath102 as can be seen from fig . [ fig:6 ] . @xmath160{fig7a.eps}&\quad \includegraphics[scale=0.4]{fig7d.eps}&\\[.2 cm ] \includegraphics[scale=0.4]{fig7b.eps}&\quad \includegraphics[scale=0.4]{fig7e.eps}\\[.2 cm ] \includegraphics[scale=0.4]{fig7c.eps}&\quad \includegraphics[scale=0.4]{fig7f.eps } \end{array } $ ] + we notice that the painlev - gullstrand family of geodesics may be considered suitable to describe free fall type of accretion by a set of rotationally dragged particles . consider a family @xmath3 of circular orbits at different radii around a certain reference value @xmath184 , all rotating with the same angular velocity , which we also assume to be constant . the constant angular velocity ensures that @xmath3 is a born - rigid congruence : @xmath41 , that is @xmath65 ( vorticity tensor ) and @xmath185 ( vorticity vector , with an abuse of notation ) . let us use a frenet - serret triad @xmath186 along each @xmath3 of the family , which rotates with respect to a fermi - walker transported frame along @xmath3 with angular velocity [ omegafs ] _ ( fs)=_1 e_3 + _ 2 e_1 , curvature @xmath187 as well as the first and second torsions @xmath188 and @xmath189 are constant along the orbit in this case @xcite . it results that @xmath190 , that is the frenet - serret angular velocity coincides with the vorticity of the congruence . moreover , the alignment of the orbit s four velocity with a killing vector implies _ ( fw , u , e)=0 , so that eq . ( [ eq : secder2 ] ) simplifies to @xmath191+\nonumber \\ & & + \ddot y^a\ , e_a-[y\cdot a(u)]a(u)\ .\end{aligned}\ ] ] it can be shown that the electric part of the weyl tensor for vacuum stationary spacetimes is given by @xcite ( u)=-(u ) ( u ) + a(u)a(u ) + ( u)a(u ) + ( u)^2 p(u ) , whence -e(u)y= ( u)-[a(u ) y]a(u)-y^b(u)_ba(u ) . equating both sides of the previous equation with the corresponding ones of eq . ( [ eq : secder3 ] ) leads to [ der2y ] y^a0 . the same result can be obtained by direct evaluation of the lie transport equation ( [ eq:1])@xmath11 for the connecting vector @xmath9 , which in the fs frame reads [ fslieeqs ] y^0=0 , y^a+[(_(fs)-(u))y]^a-[(u)y]^a=0 . since @xmath41 and @xmath192 in this case , the previous equations simplify to y^0=0 , y^a=0 , trivially implying eq . ( [ der2y ] ) . this case is the most interesting since it naturally describes the orbital motion of a system which we may require to remain rigid as for example an artificial satellite around the earth or a planet " orbiting a star . the physical constraints which assure rigidity and therefore stability of the orbiting system are deduced directly from eq . ( [ fslieeqs ] ) . finally , one can set up a fermi - walker transported frame adapted to @xmath3 also in this case , following exactly the same procedure as in section [ thread ] . it will result that the component of the deviation vector along the fermi - walker angular velocity remains constant along the path , while the other spatial components oscillate with the same frequency @xmath193 . in the case of vanishing rotation parameter the family of zamos coincides with that of static observers , @xmath194 . the expansion tensor @xmath195 as well as the fermi - walker angular velocity @xmath152 are identically zero , whence the matrix @xmath196 too . furthermore we have that ( n)=s(n)=[-2,1,1 ] , implying that also the deviation matrix @xmath145 is identically zero . hence the spatial components of the deviation vector remain all constant along the path , since both @xmath67 and @xmath39 ( note that they are only functions of @xmath102 ) . in the case of painlev - gullstrand observers , instead , the limit of vanishing rotation parameter leads to the radial geodesic motion . the matrix @xmath197 results to be diagonal and completely determined in terms of the surviving components of the electric part of the weyl tensor _ ( n , e)=e()=[-2,1,1 ] . as a result , the system of deviation equations ( [ eq : secder3bis ] ) decouples as follows : y^1+k_(n , e)_11y^1=0 , y^2+k_(n , e)_22y^2=0 , y^3+k_(n , e)_33y^3=0 , where the coefficients @xmath198 are functions of the proper time @xmath164 . we notice that the frame ( [ framepg ] ) is already fermi - walker transported also in this case , since @xmath199 ; moreover , the fermi - walker triad ( [ fwpg ] ) reduces to the spatial triad of ( [ framepg ] ) in the limit @xmath200 . the corresponding first order system of lie transport equations ( [ eq1ordpg ] ) thus simplifies to [ schwpg1ord ] y^1=()_11y^1 , y^2=()_22y^2 , y^3=()_33y^3 , where = , from eq . ( [ thetapg ] ) , and can be easily integrated . after introducing the explicit dependence on the radial coordinate instead of the proper time parametrization eqs . ( [ schwpg1ord ] ) then become = - , = , = , whose solution is straightforward : y^1=y^1_0 , y^2=y^2_0 , y^3=y^3_0 . in contrast with the kerr case the deviations along the angular directions @xmath91 and @xmath148 have thus exactly the same behaviour as functions of the radial coordinate . as a result , the squeezing of an initially circular bunch of particles on the @xmath90-@xmath91 plane is the same as the one on the @xmath90-@xmath148 plane ; its shape changes for decreasing values of the radial coordinate in a way very similar to the corresponding one for kerr , as shown in fig . [ fig:7 ] . we have defined relative accelerations and strains among a set of comoving particles ( nongeodesic congruence of timelike world lines ) with respect to a fiducial observer " world line ( i.e. a single reference line of the congruence ) in terms of the geometric properties of both the congruence and the fiducial observers . we have provided an operational definition of strains , based on spacetime splitting techniques , evidentiating the role of the observer as well as the choice of the spatial triad associated with it , otherwise arbitrary and arbitrarily dragged along the observer world line . strains and deviation vectors may have strongly different behaviours according to the preferred reference frame set up for their measurements , a point that is somehow missing in the literature . a first attempt due to szekeres as well as further developments ( more recent ) due to mashhoon and coworkers ( concerning the theory of the relativistic gravity gradiometer ) and de felice and coworkers ( concerning the definition of strains ) have been encompassed . to better specify the results of our analysis we have studied certain special congruences in the kerr spacetime ( static observers , zamos , painlev - gullstrand , circular orbits ) . particular attention has been devoted to the fermi - walker frame , since it represents a reference system operationally defined in terms of three gyroscopes dragged along the observer world line . in the kerr case , when using fermi - walker axes , static observers and zamos experience certain relative strains which result in harmonic oscillations for the deviation vectors ; differently , the painlev - gullstrand observers ( free - falling and locally nonrotating , spiraling towards the singularity ) feel decreasing deviation along the angular directions ( approaching the singularity ) , while increasing deviation in the radial direction . this analysis can be repeated in other interesting gravitational situations and it should be very useful when thinking about general relativistic experiments on space stations in black hole gravitational backgrounds . we list below the relevant quantities entering the relative acceleration equation ( [ eq : secder3bis ] ) and corresponding to all families of observers congruences considered in section [ kerrcase ] for the case of a kerr spacetime . the components of the electric part of the weyl tensor as well as the relevant kinematical quantities entering eq . ( [ eq : secder3bis ] ) and corresponding to a congruence of static observers are given by @xmath201 we recall that all the components of the tensor fields in eq . ( [ variousqua_thd ] ) refer to the ( orthonormal ) adapted frame ( [ thdframe ] ) . by using eq . ( [ variousqua_thd ] ) it results @xmath202 , implying that the matrix @xmath126 is identically zero . the components of the electric part of the weyl tensor as well as the relevant kinematical quantities entering eq . ( [ eq : secder3bis ] ) and corresponding to a congruence of zamos are given by @xmath203 we recall that all the components of the tensor fields in eq . ( [ variousqua_zamo ] ) refer to the ( orthonormal ) adapted frame ( [ zamoframe ] ) . by using eq . ( [ variousqua_zamo ] ) it results @xmath204 , implying that the matrix @xmath145 is identically zero . the kinematical properties of the @xmath165 observers are summarized by the expansion @xmath205 . in fact the exterior derivative of ( [ pg4vel ] ) is zero : @xmath206 , implying that @xmath207 and @xmath208 @xcite . in this case @xmath209 , so that @xmath210 and in addition @xmath211 has the form @xmath212 the nonvanishing frame components of the expansion tensor are given by @xmath213 and for the components of the angular velocity vector @xmath214 we find [ omfwpg ] _ ( fw,,e)_1=-()_23 , _ ( fw,,e)_2=()_13 , _ ( fw,,e)_3=()_12 . we notice that the electric part of the riemann tensor associated with the painlev - gullstrand observers can be obtained using the general transformation laws mapping the fiels measured by a family of observers into the corresponding ones measured by another family @xcite . in the special case of vacuum spacetimes , using the decomposition ( [ caln ] ) it is easy to show that @xmath215_\alpha{}^\mu [ p(n,{\mathcal n})^{-1}]_\beta{}^\nu \bigl [ { \mathcal e}(n)_{\mu\nu}-2 { \mathcal h}(n)_{(\mu|\rho } v^\rho{}_{\nu ) } + \nonumber \\ \fl\quad & & + v_\mu{}^\rho { \mathcal e}(n)_{\rho\sigma } v^\sigma{}_\nu \bigr]\ , \end{aligned}\ ] ] with the analogous expression for @xmath216 ( simply obtained with the replacements : @xmath217 and @xmath218 ) . here @xmath219_\alpha{}^\beta = p(n)_\alpha{}^\beta+n_\alpha \nu({\mathcal n},n)^\beta$ ] , with @xmath220 , projects from the zamo local rest space to that of the painlev - gullstrand observers , and we have used the notation v_v(n , n)=(n)_(n , n)^ , ( n)_=n^ _ , together with the standard definitions @xmath221 for the electric and magnetic part of the riemann tensor with respect to the zamo family of observers , given by eq . ( [ variousqua_zamo ] ) and @xmath222 respectively . the long formulas for the components of @xmath223 are listed below : @xmath224\ , \nonumber\\ \fl\quad { \mathcal e}(\mathcal { n})_{23}&=&\nu(\mathcal{n},n)\frac{3a^3\mathcal{m}}{\sigma^4}(4r^2-\sigma)\cos\theta\sin^2\theta\ , \nonumber\\ \fl\quad { \mathcal e}(\mathcal { n})_{33}&=&-\frac{\mathcal{m}r}{\sigma^4}(4r^2 - 3\sigma)\frac{-\sigma^2\delta+2\mathcal{m}r[3(r^2+a^2)-4\sigma]}{(r^2+a^2)^2-a^2\delta\sin^2\theta}\ .\end{aligned}\ ] ] using spacetime splitting techniques @xcite , it is also possible to show that @xmath225^{\rm ( tf)}{}_{\alpha\beta}\ , \end{aligned}\ ] ] where @xmath226 is the the projected lie temporal derivative and @xmath227 denotes the trace - free part of a tensor . the nonvanishing components of @xmath228 are then given by @xmath229\ , \nonumber\\ \fl\quad { \mathcal k}_{({\mathcal n},e)}{}_{22}&=&-\dot\theta(\mathcal { n})_{22}-\theta(\mathcal { n})_{22}^2\ , \nonumber\\ \fl\quad { \mathcal k}_{({\mathcal n},e)}{}_{31}&=&-2[\dot\omega_{({\rm fw},\mathcal{n},e)}{}_2+\omega_{({\rm fw},\mathcal{n},e)}{}_2(\theta(\mathcal { n})_{11}+\theta(\mathcal { n})_{33})]\ , \nonumber\\ \fl\quad { \mathcal k}_{({\mathcal n},e)}{}_{32}&=&2[\dot\omega_{({\rm fw},\mathcal{n},e)}{}_1+\omega_{({\rm fw},\mathcal{n},e)}{}_1(\theta(\mathcal { n})_{22}+\theta(\mathcal { n})_{33})\nonumber \\ \fl\quad & & -2\omega_{({\rm fw},\mathcal{n},e)}{}_2\omega_{({\rm fw},\mathcal{n},e)}{}_3]\ , \nonumber\\ \fl\quad { \mathcal k}_{(\mathcal n , e)}{}_{33}&=&-\dot\theta(\mathcal { n})_{33}-\theta(\mathcal { n})_{33}^2\ , \end{aligned}\ ] ] whose explicit expressions can be easily obtained through eqs . ( [ thetapg ] ) and ( [ omfwpg ] ) . finally , we list the nonzero components of the deviation matrix @xmath230 with respect to the fermi - walker transported frame ( [ fwpg ] ) @xmath231\}\sin\theta\cos\theta\ , \nonumber\\ \fl\quad { \mathcal k}_{({\mathcal n},e')}{}_{22}&=&\frac{\mathcal{m}r}{a^2\sigma^4}\ { 15\sigma^2(\sigma - a^2)-4[4r^2(12r^2 + 5a^2)+3a^4]\sigma\nonumber\\ \fl\quad & & + 24r^2(8r^2-a^2)(r^2+a^2)\}\nonumber\\ \fl\quad & & + \frac{3\mathcal{m}\lambda}{a^2\sigma^5}\ { 2a^2\sin^2\theta[\sqrt{r^2+a^2}\cos^2\theta+r\sin^2\theta]-r(4\sigma+r^2+a^2)\}\ , \nonumber\\ \fl\quad { \mathcal k}_{({\mathcal n},e')}{}_{33}&=&\frac{\mathcal{m}r}{\sigma^3}(4r^2 - 3\sigma)\ , \end{aligned}\ ] ] where @xmath232

the kochen - specker theorem proves the inability to assign , simultaneously , noncontextual definite values to all ( of a finite set of ) quantum mechanical observables in a consistent manner . if one assumes that any definite values behave noncontextually , one can nonetheless only conclude that _ some _ observables ( in this set ) are value indefinite . in this paper we prove a variant of the kochen - specker theorem showing that , under the same assumption of noncontextuality , if a single one - dimensional projection observable is assigned the definite value 1 , then no one - dimensional projection observable that is incompatible ( i.e. , non - commuting ) with this one can be assigned consistently a definite value . unlike standard proofs of the kochen - specker theorem , in order to localise and show the extent of value indefiniteness this result requires a constructive method of reduction between kochen - specker sets . if a system is prepared in a pure state , then it is reasonable to assume that any value assignment ( i.e. , hidden variable model ) for this system assigns the value 1 to the observable projecting onto the one - dimensional linear subspace spanned by , and the value 0 to those projecting onto linear subspaces orthogonal to it . our result can be interpreted , under this assumption , as showing that the outcome of a measurement of _ any _ other incompatible one - dimensional projection observable can not be determined in advance , thus formalising a notion of quantum randomness .

1503.01985

bell s theorem @xcite and the kochen - specker theorem @xcite are perhaps two of the results which have been most influential in developing the modern understanding of quantum mechanics as an irreducibly non - classical theory @xcite . moreover , these two no - go theorems are seen as the strongest argument for quantum mechanics being a fundamentally indeterministic theory , rather than one ruled by a deeper determinism below the level of the quantum mechanical description of reality . bell s theorem , which shows that quantum mechanics predicts statistical correlations between separated particles greater than what would be possible in any local , realistic , classical theory , was the focus of attention for several decades due to its relatively clear ability to be tested experimentally @xcite . the kochen - specker theorem was proved very shortly afterwards , but was largely ignored due to a perceived lack of testability , and perhaps also its formalisation in terms of partial algebras , until it attracted renewed attention with the more recent advances in quantum information theory and foundations . in contrast to the bounds on probability distributions given by bell s theorem , the kochen - specker theorem shows that the hilbert - space structure of quantum mechanics makes it impossible to assign ` classical ' definite values to all quantum observables in a consistent manner . since such a definite value is precisely a ( deterministic ) hidden variable specifying , in advance , the result of a measurement of an observable , this means that the outcomes of all quantum measurements on a system can not be simultaneously predetermined . more recent developments have significantly reduced the size and difficulty of proofs of the kochen - specker theorem @xcite and converted such proofs into testable inequalities @xcite . however , in showing the impossibility of a classical deterministic ` two - valued ' measure ( i.e. , value assignment ) the kochen - specker theorem leaves open several possible conclusions . the kochen - specker theorem , more specifically , finds a contradiction between the following three assumptions , which will be formalised more rigorously a little later : * all observables are assigned a definite value ( i.e. , are ` value definite ' ) ; * this definite value should be noncontextual that is , assigned as a function of the observable alone , and not depending on other compatible observables ; * the definite values for a set of compatible observables must be consistent with the theoretical quantum predictions for the relations between them . condition ( iii ) is largely uncontroversial and hence one must generally conclude that either ( or even both ) ( i ) and ( ii ) must be given up . several alternative interpretations of quantum mechanics are contextual ( e.g. , @xcite ) , and hence discard ( ii ) . perhaps the more popular interpretation , however , is that the inability to simultaneously assign noncontextual definite values , representing predetermined measurement outcomes , to all observables means that measurement outcomes are not determined in advance at all : that quantum mechanics represents a value indefinite reality . this interpretation is often referred to simply as ` contextuality ' in the literature ; however we reserve this term strictly for the contextual behaviour of definite values . if we choose to require ( ii ) to hold , at least for any observables that are assigned definite values , then there remains an oft - overlooked gap between the formal result of the kochen - specker theorem and the general interpretation of it . indeed , the negation of ( i ) is that _ not all _ observables are assigned definite values : it does not prove that no observable can be assigned a definite value and hence , given that definite values represent predetermined measurement outcomes , does not show that all measurements must result in the _ ex nihilo _ creation of an outcome , nor does it allow one to know _ which _ observables in any set are value indefinite . we can , of course , postulate that if some observables are value indefinite , then this should , by symmetry or uniformity considerations , be the case for all observables ( or at least those for which the born rule assigns a probability strictly between 0 and 1 to some outcomes ) . however , it is key to realise that this is not in any sense a formal consequence of the kochen - specker theorem , and constitutes an additional , undesired , assumption . in this paper we address precisely this issue . as is common in modern treatments of the kochen - specker theorem @xcite we focus on one - dimensional ( rank-1 ) projection observables , and we denote the observable projecting onto the linear subspace spanned by a vector @xmath0 as @xmath1 . by using a modified , weakened set of assumptions , we prove that if one such projection observable @xmath2 is assigned the value 1 , then no other such projection observable @xmath3 can be consistently assigned a definite value unless @xmath2 and @xmath3 commute . in interpreting this result physically , we note that if a system is prepared in the state @xmath0 then the outcome of a measurement of the observable @xmath2 is known to be 1 with certainty , and thus any value assignment representing the outcomes of possible measurements on the system should assign the value 1 to @xmath2 . if @xmath3 does not commute with @xmath2 it is therefore value indefinite under such a value assignment and hence can not have a consistently predetermined measurement outcome . this self - contained , analytic proof extends and generalises the results of @xcite . throughout the paper we will assume ( ii ) to hold , as is common in interpretations of the kochen - specker theorem , and our strengthened results and interpretation of the kochen - specker theorem thus rely on this condition . we do not attempt to justify this assumption here , as this is an interpretational choice and the subject of much debate ( see ( * ? ? ? 4 ) for an overview ) , which is beyond the scope of this paper . as usual we denote by @xmath4 the set of complex numbers and use the standard quantum mechanical bra - ket notion ; that is , we denote ( unit ) vectors in the hilbert space @xmath5 by @xmath6 . as mentioned above , we will focus on one - dimensional projection observables and denote by @xmath2 the operator projecting onto the linear subspace spanned by @xmath0 ; that is , @xmath7 . in the following we formalise hidden variables and the notion of value definiteness in a clear and unambiguous fashion . this framework is based on that we developed in @xcite , and similar to standard approaches to the kochen - specker theorem @xcite ; we have made several simplifications since we do not wish to explore contextual definite values or hidden variable theories in any detail here . we fix a positive integer @xmath8 . let @xmath9 be a nonempty set of one - dimensional projection observables on the hilbert space @xmath10 @xcite . a set @xmath11 is a _ context of @xmath12 _ if @xmath13 has @xmath14 elements ( i.e. , @xmath15 ) and for all @xmath16 with @xmath17 , @xmath18 . since distinct one - dimensional projection observables commute if and only if they project onto mutually orthogonal linear subspaces , a context @xmath13 of @xmath12 is thus a maximal set of compatible one - dimensional projection observables on @xmath5 . because there is a direct correspondence ( up to a phase - shift ) between unit vectors and one - dimensional projection observables , a context is uniquely defined by an orthonormal basis of @xmath5 . recall that a partial function is one which may be undefined for some values . if it is defined everywhere , then it is total . [ def : v ] a _ value assignment function ( on @xmath12 ) _ is a partial two - valued function @xmath19 , assigning values to some ( possibly all ) observables in @xmath12 . we note that we could , as in @xcite , allow @xmath20 to be a function of both the observable @xmath21 and the context @xmath13 containing @xmath21 , allowing values to be assigned contextually . it would perhaps be more correct to call @xmath20 , as defined above in definition [ def : v ] , a _ noncontextual _ value assignment function ; however , since we are interested only in the noncontextual case , we avoid this for compactness . an observable @xmath22 is _ value definite ( under @xmath20 ) _ if @xmath23 is defined ; otherwise it is _ value indefinite ( under @xmath20)_. similarly , we call @xmath12 value definite ( under @xmath20 ) if every observable @xmath22 is value definite . with this terminology , we can state the kochen - specker theorem formally . we present it in the following form deliberately in order to draw the comparison to our earlier informal description , and to clarify the following discussion , even though the second condition is redundant because we require , by definition , that a value assignment function be noncontextual . let @xmath24 . then there exists a ( finite ) set of one - dimensional projection observables @xmath12 on the hilbert space @xmath5 such that there is no value assignment function @xmath20 satisfying the following three conditions : * @xmath12 is value definite under @xmath20 ; that is , @xmath20 is a total function . * the value @xmath23 of an observable @xmath22 depends only on @xmath21 and not the context containing @xmath21 . * for every context @xmath13 of @xmath12 the following condition holds : is a _ boolean frame function _ with weight 1 @xcite . ] @xmath25 . the third condition expresses the fact that only one projection observable in a context can be assigned the value 1 . as we mentioned earlier , this is largely uncontroversial : one can simultaneously measure the observables in a context and quantum mechanics predicts precisely that exactly one of these measurements should give the result ` 1 ' , thus any corresponding definite values assigned to these observables should obey this same condition . hence , if we assume ( ii ) to be true at least for observables that are value definite for which the statement makes clear sense then the kochen - specker theorem requires us to conclude the negation of ( i ) : that @xmath12 can not be value definite , and hence _ at least one _ observable must be value indefinite . note that the third condition is not independent of the first : it is not clear how the sum @xmath26 should be evaluated if @xmath23 is undefined . this is one of the key issues we will clarify in attempting to localise value indefiniteness . while the kochen - specker theorem certainly succeeds , as was the original intention , in showing that quantum mechanics must obey an entirely non - classical event structure , it does not , as we have pointed out , show that all observables must be value indefinite and their outcomes intrinsically indeterministic . as a consequence of the global nature of the hypothesis of the theorem that _ all _ observables are value definite one can only draw a global conclusion : that _ not all _ observables are value definite . that is , the theorem , even under the assumption of noncontextuality , can not ` locate ' value indefiniteness to any particular observable . this is an important point , not only for the foundational understanding of quantum mechanics , but also in practical applications : quantum random number generators and cryptographic schemes rely on the indeterminism of quantum mechanics providing ` irreducible randomness ' @xcite . to certify such claims , it is important to be able to localise value indefiniteness to ensure it applies to the observables measured in such applications . we proceed by providing more nuanced and less demanding , localised versions of the kochen - specker assumptions , and use these to localise value indefiniteness ( always under the assumption of noncontextuality for value definite observables ) . our approach is a conservative one : rather than assuming complete value definiteness of the entire set of observables considered , we require observables to be value definite only when their indefiniteness would allow the possibility of measurements violating the quantum predictions specified in condition ( iii ) of the kochen - specker theorem ( see the more detailed discussion and example below ) . in order for this approach to work , we need , as a premise , at least one observable to be value definite . we then show that the assumption that _ any other _ observable is value definite leads to a contradiction . fortunately , there is a justification for this premise : if a system is prepared in an arbitrary state @xmath27 , then measurement of the observable @xmath2 should yield the outcome 1 . thus , it seems reasonable to require that , if @xmath28 , then @xmath29 . we call this the _ eigenstate assumption _ @xcite , which is similar to , although weaker than , the ` eigenstate - eigenvalue link ' discussed in @xcite . furthermore , since the critical feature of a set @xmath12 of projection observables is the orthogonality relations between these observables rather than the specific form of these observables , we can hence choose our basis at will . it is thus not unreasonable to consider that some one - dimensional projection observable in @xmath12 has the value 1 , and to fix the basis used to express @xmath12 to that of the state @xmath0 to make this observable coincide with @xmath2 . let us finally discuss how assumption ( iii ) can be generalised for partial value assignment functions @xmath20 , that is , the case where some observables in @xmath12 may be value indefinite . [ admis - rules ] let @xmath12 be a set of one - dimensional projection observables on @xmath5 and let @xmath30 be a value assignment function . then @xmath20 is _ admissible _ if the following two conditions hold for every context @xmath13 of @xmath12 : * if there exists a @xmath31 with @xmath32 , then @xmath33 for all @xmath34 ; * if there exists a @xmath31 with @xmath33 for all @xmath34 , then @xmath32 . admissibility requires that the quantum predictions of ( iii ) are never violated , while allowing the value indefiniteness of an observable @xmath21 if both outcomes ( 0 and 1 ) of a measurement of @xmath21 would be compatible with the definite values of other observables sharing a context with @xmath21 . for example , if @xmath32 , then a measurement of all the observables in a context @xmath13 containing @xmath21 must yield the outcome 1 for @xmath21 , and hence to avoid contradiction the outcome 0 for the other observables in the context . on the other hand , if @xmath35 , even though measurement of @xmath21 must yield the outcome 0 , any of the other observables in @xmath13 could yield the value 1 or 0 ( as long as only one yields 1 ) , hence we should not conclude the value definiteness of these other observables . let us illustrate the difference between our weakened assumptions , and in particular admissibility , with the hypotheses of the kochen - specker theorem . ccc 01 [ scale=0.6 ] = [ line width=1pt ] = [ circle , minimum size=6 ] = [ rectangle , minimum size=9 ] = [ draw = none , circle , minimum size=35 ] = [ draw = none , circle , minimum size=12 ] ( 0:5 ) coordinate(1 ) ( 4.167,1.44 ) coordinate(2 ) ( 3.33,2.88 ) coordinate(3 ) ( 60:5 ) coordinate(4 ) ( 0.833,4.33 ) coordinate(5 ) ( -0.833,4.33 ) coordinate(6 ) ( 120:5 ) coordinate(7 ) ( -3.33,2.88 ) coordinate(8 ) ( -4.167,1.44 ) coordinate(9 ) ( 180:5 ) coordinate(10 ) ( -4.167,-1.44 ) coordinate(11 ) ( -3.33,-2.88 ) coordinate(12 ) ( 240:5 ) coordinate(13 ) ( -0.833,-4.33 ) coordinate(14 ) ( 0.833,-4.33 ) coordinate(15 ) ( 300:5 ) coordinate(16 ) ( 3.33,-2.88 ) coordinate(17 ) ( 4.167,-1.44 ) coordinate(18 ) ; \(1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 ) ( 7 ) ( 8) ( 9 ) ( 10 ) ( 11 ) ( 13 ) ( 14 ) ( 15 ) ( 16 ) ( 17 ) ( 18 ) ( 1 ) ; \(2 ) ( 9 ) ; ( 11 ) ( 18 ) ; ( 3 ) ( 14 ) ; ( 5 ) ( 12 ) ; ( 6 ) ( 17 ) ; ( 8) ( 15 ) ; \(2 ) arc ( 450:270:2 and 1.44 ) ; ( 3 ) arc ( 90:270:2 and 1.44 ) ; ( 6 ) arc ( 90:270:2 and 1.44 ) ; ( 9 ) arc ( 90:270:2 and 1.44 ) ; ( 12 ) arc ( 90:270:2 and 1.44 ) ; ( 15 ) arc ( 90:270:2 and 1.44 ) ; \(1 ) coordinate[s1,fill , label=0:@xmath36 ; ( 2 ) coordinate[c1,fill ] ; ( 3 ) coordinate[c1,fill ] ; ( 4 ) coordinate[c1,fill ] ; ( 5 ) coordinate[c1,fill ] ; ( 6 ) coordinate[c1,fill , label=90:@xmath37 ; ( 7 ) coordinate[s1,fill , label=90:@xmath38 ; ( 8) coordinate[c1,fill ] ; ( 9 ) coordinate[c1,fill = white , label=150:@xmath39 ; ( 9 ) node[c1,fill = none ] @xmath40 ; ( 9 ) coordinate[c1,draw ] ; ( 10 ) coordinate[c1,fill ] ; ( 11 ) coordinate[c1,fill ] ; ( 12 ) coordinate[s1,fill , label=180:@xmath41 ; ( 13 ) coordinate[c1,fill ] ; ( 14 ) coordinate[c1,fill ] ; ( 15 ) coordinate[s1,fill , label=270:@xmath42 ; ( 16 ) coordinate[c1,fill ] ; ( 17 ) coordinate[c1,fill ] ; ( 18 ) coordinate[c1,fill ] ; ( c1 ) at ( 10 ) ; ( c2 ) at ( 10 ) ; of each observable ( node ) @xmath21 is represented as follows : @xmath32 black square ; @xmath35 filled circle ; @xmath23 undefined ( value indefinite ) hollow circle . ( a ) the contradiction arising when @xmath43 : @xmath20 can not be admissible , since this would require that @xmath44 and @xmath45 simultaneously , as shown by the cross in the diagram . ( b ) a possible admissible value assignment when @xmath46 and @xmath47 . ] & & 01 [ scale=0.6 ] = [ line width=1pt ] = [ circle , minimum size=6 ] = [ rectangle , minimum size=9 ] = [ draw = none , circle , minimum size=35 ] = [ draw = none , circle , minimum size=12 ] ( 0:5 ) coordinate(1 ) ( 4.167,1.44 ) coordinate(2 ) ( 3.33,2.88 ) coordinate(3 ) ( 60:5 ) coordinate(4 ) ( 0.833,4.33 ) coordinate(5 ) ( -0.833,4.33 ) coordinate(6 ) ( 120:5 ) coordinate(7 ) ( -3.33,2.88 ) coordinate(8 ) ( -4.167,1.44 ) coordinate(9 ) ( 180:5 ) coordinate(10 ) ( -4.167,-1.44 ) coordinate(11 ) ( -3.33,-2.88 ) coordinate(12 ) ( 240:5 ) coordinate(13 ) ( -0.833,-4.33 ) coordinate(14 ) ( 0.833,-4.33 ) coordinate(15 ) ( 300:5 ) coordinate(16 ) ( 3.33,-2.88 ) coordinate(17 ) ( 4.167,-1.44 ) coordinate(18 ) ; \(1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 ) ( 7 ) ( 8) ( 9 ) ( 10 ) ( 11 ) ( 12 ) ( 13 ) ( 14 ) ( 15 ) ( 16 ) ( 17 ) ( 18 ) ( 1 ) ; \(2 ) ( 9 ) ; ( 11 ) ( 18 ) ; ( 3 ) ( 14 ) ; ( 5 ) ( 12 ) ; ( 6 ) ( 17 ) ; ( 8) ( 15 ) ; \(2 ) arc ( 450:270:2 and 1.44 ) ; ( 3 ) arc ( 90:270:2 and 1.44 ) ; ( 6 ) arc ( 90:270:2 and 1.44 ) ; ( 9 ) arc ( 90:270:2 and 1.44 ) ; ( 12 ) arc ( 90:270:2 and 1.44 ) ; ( 15 ) arc ( 90:270:2 and 1.44 ) ; \(1 ) coordinate[c1,fill , label=0:@xmath36 ; ( 2 ) coordinate[c1,fill = white ] ; ( 2 ) coordinate[c1,draw ] ; ( 3 ) coordinate[c1,fill ] ; ( 4 ) coordinate[c1,fill = white ] ; ( 4 ) coordinate[c1,draw ] ; ( 5 ) coordinate[c1,fill ] ; ( 6 ) coordinate[c1,fill = white ] ; ( 6 ) coordinate[c1,draw ] ; ( 7 ) coordinate[c1,fill = white ] ; ( 7 ) coordinate[c1,draw ] ; ( 8) coordinate[c1,fill = white ] ; ( 8) coordinate[c1,draw ] ; ( 9 ) coordinate[c1,fill = white , label=150:@xmath39 ; ( 9 ) coordinate[c1,draw ] ; ( 10 ) coordinate[c1,fill ] ; ( 11 ) coordinate[c1,fill ] ; ( 12 ) coordinate[s1,fill , label=180:@xmath41 ; ( 13 ) coordinate[c1,fill ] ; ( 14 ) coordinate[c1,fill ] ; ( 15 ) coordinate[c1,fill = white ] ; ( 15 ) coordinate[c1,draw ] ; ( 16 ) coordinate[c1,fill = white ] ; ( 16 ) coordinate[c1,draw ] ; ( 17 ) coordinate[c1,fill = white ] ; ( 17 ) coordinate[c1,draw ] ; ( 18 ) coordinate[c1,fill = white ] ; ( 18 ) coordinate[c1,draw ] ; of each observable ( node ) @xmath21 is represented as follows : @xmath32 black square ; @xmath35 filled circle ; @xmath23 undefined ( value indefinite ) hollow circle . ( a ) the contradiction arising when @xmath43 : @xmath20 can not be admissible , since this would require that @xmath44 and @xmath45 simultaneously , as shown by the cross in the diagram . ( b ) a possible admissible value assignment when @xmath46 and @xmath47.,title="fig : " ] + ( a ) & & ( b ) + consider the greechie orthogonality digram shown in fig . [ fig : cabello ] , in which vertices depict observables and smooth lines or curves represent contexts . this well known diagram represents the ` orthogonality ' relations between the observables used in a well known proof of the kochen - specker theorem due to cabello et al . @xcite , containing only 18 one - dimensional projection observables on @xmath48 . the kochen - specker theorem implies that there is no way to assign every observable in this diagram a value such that the admissibility requirements hold : exactly one observable in each context should have the value 1 . let us suppose for the sake of example that @xmath43 and that @xmath20 is admissible . then , by working from @xmath49 and @xmath50 and applying the admissibility rule ( a ) one deduces that all observables in a context with @xmath49 or @xmath50 must take the value 0 . one then notices that there are contexts containing 3 observables with the value 0 , so we can deduce from ( b ) that the fourth must have the value 1 . if we follow this line of reasoning , we can continue to assign values to observables with the admissibility requirements , as depicted in fig . [ fig : cabello](a ) , where a black square represents the value 1 , and a black circle the value 0 . as we can see , by considering the contexts @xmath51 and @xmath52 we can infer that @xmath53 must take both the values 1 and 0 respectively : both possibilities contradict the admissibility of @xmath20 , as does the final possibility that @xmath53 is value indefinite . note that , in fig . [ fig : cabello](a ) , the contradiction obtained at @xmath53 marked by the cross is a consequence of a specific succession of applications of the admissibility rules ( a ) and ( b ) in definition [ admis - rules ] . by applying these rules in a different order , one can obtain the contradiction also at @xmath54 , @xmath55 , or @xmath56 . the most important aspect of this reasoning in this context is that it is deterministic : we proceed only by deducing the value definiteness of observables via ( a ) and ( b ) . now let us assume that @xmath46 and @xmath47 , as depicted in fig . [ fig : cabello](b ) . we again apply ( a ) to observables commuting with @xmath49 ; however , we then see that neither ( a ) nor ( b ) can be used again to deduce the value of another observable . normally , in proving that this diagram permits no consistent assignment of definite values , one would then proceed by assuming that one of the unfilled observables , such as @xmath53 , must have either @xmath45 or @xmath44 , and trying both possibilities . one can do this when proving the kochen - specker theorem since one assumes ( i ) : that every observable must have a definite value . however , in order to localise value indefiniteness we do not make this assumption . hence , the value assignment in fig . [ fig : cabello](b ) , with the observables represented by unfilled circles being value indefinite ( e.g. , @xmath57 undefined ) represents an admissible value assignment . thus , under the assumption that @xmath46 , fig . [ fig : cabello ] does not suffice to prove that @xmath58 must be value indefinite , and hence can not be used to localise value indefiniteness . it is not difficult to see that we reach the same conclusion irrespective of our choice of observables as @xmath49 and @xmath50 . in this paper , in proving the main theorem , we give a set of observables for which this is the case . that is , there are observable @xmath49 and @xmath50 such that if @xmath46 then both @xmath47 and @xmath59 lead , via admissibility , to contradictions . let us now state the strengthened theorem which is the focus of this paper . as we mentioned , this generalises the results of @xcite and uses a different proof technique allowing for a more symmetrised analytic approach . the result in @xcite , on the other hand , relies on computational results and the interpretation of graphs . [ thm : main ] let @xmath24 and @xmath60 be unit vectors such that @xmath61 . we can effectively find a finite set of one - dimensional projection observables @xmath12 containing @xmath2 and @xmath3 for which there is no admissible value assignment function on @xmath12 such that @xmath29 and @xmath3 is value definite . before we proceed to prove theorem [ thm : main ] , let us first discuss some important relevant issues . this theorem has a slightly different form from the standard kochen - specker theorem because of the requirement that a particular observable in the set @xmath12 be assigned the value 1 . however , since , as we will see , it is only the orthogonality relations between the observables in @xmath12 which is important , a change of basis can always ensure that the required observable @xmath2 be assigned the value 1 . in order to interpret this result one has to take into account the eigenstate assumption discussed in the previous section : if a quantum system is prepared in a state @xmath0 in @xmath24 dimensional hilbert space , then every one - dimensional projection observable that does not commute with @xmath2 is value indefinite and hence can not have a predetermined measurement outcome . the first question to address is whether existing kochen - specker diagrams ( i.e. , greechie diagrams specifying the orthogonality relations of @xmath12 ) could be used to provide a set @xmath12 of observables proving theorem [ thm : main ] ; it is not _ a priori _ obvious that such diagrams are unable to do so . in section [ sec : cabelloexample ] we showed , as an example , that a particular simple and well - known kochen - specker diagram is not sufficient for this purpose . a careful search through existing diagrams showed that this is the case in general , and we were unable to find an existing kochen - specker diagram in which there are two observables @xmath49 and @xmath50 with the required property that if @xmath46 , both @xmath47 and @xmath59 lead to a contradiction . a second conceptual problem with the use of fixed kochen - specker diagrams as in existing proofs is the following . since , in order to derive a contradiction , we need to assume that an observable @xmath2 in the given observable set has @xmath29 , this limits the observables which can be shown to be value indefinite to , at best , the remaining ones in @xmath62 . however , we wish to prove more : that _ every _ one - dimensional projection observable not commuting with @xmath2 is value indefinite . as a result , we need not only a set of observables with the required properties discussed above , but furthermore an approach to generalise this set of observables to arbitrary other observables . we overcome this apparent lack of generality via a method of reductions , which we present in the next section and will return to discuss later on . we prove theorem [ thm : main ] in three main steps : 1 . we first prove it for the special case that @xmath63 . a similar result ( for @xmath64 ) was shown in ref . @xcite , but this involved two separate diagrams applying to separate cases . here we give a single diagram providing a much more compact , clear proof . 2 . we prove a simple reduction for @xmath65 to the first case . the third and main part of the proof involves finding a reduction in the opposite sense , applying to the final @xmath66 case . it is this final reduction allowing the complete proof that is the most involved technical aspect of this paper . as is standard in kochen - specker proofs @xcite , we will work directly in the three - dimensional case of @xmath67 ( in fact , only @xmath68 is needed ) , since the case for @xmath69 can be simply reduced to this situation . [ lemma : explicitcase ] given any two unit vectors @xmath70 with @xmath71 there exists a finite set of one - dimensional projection observables @xmath12 such that if @xmath46 then @xmath50 is value indefinite under every admissible assignment function @xmath20 on @xmath12 . by choosing an appropriate basis we can assume , without loss of generality , that @xmath72 and @xmath73 . let us consider the set @xmath74 } of one - dimensional projection observables where the vectors @xmath75 for @xmath76 are defined in table [ table : vectorlist ] ( with the normalisation factors emitted for simplicity ) . the orthogonality relations between these vectors gives the 26 contexts shown in table [ table : contextlist ] . note that these observables are ` tightly ' connected : the context - observable ratio is relatively high . the greechie diagram showing the orthogonality relations is shown in fig . [ fig : greechie ] . .the 37 vectors specifying the observables used in the proof of lemma [ lemma : explicitcase ] , with normalisation factors omitted . [ cols= " < , < , < , < , < " , ] [ table : prooftable2 ] hence , we must conclude that @xmath50 can not be value definite if @xmath20 is admissible on @xmath12 . we next show a ` contraction ' lemma that constitutes a simple ` forcing ' of value definiteness : given @xmath49 and @xmath50 with @xmath43 , there is a @xmath77 which is ` closer ' ( i.e. , at a smaller angle of our choosing ; contracted ) to both @xmath78 and @xmath79 , for which @xmath45 also . this result was proved in @xcite , but we reproduce the short proof here for completeness . the form of the vectors @xmath80 specified in the lemma will be used several times in the rest of the paper . [ lemma : reduction1 ] given any two unit vectors @xmath70 with @xmath81 and a @xmath82 such that @xmath83 , we can effectively find a unit vector @xmath77 with @xmath84 , and a finite set of one - dimensional projection observables @xmath12 containing @xmath49 , @xmath50 , @xmath53 such that if @xmath43 , then @xmath45 , for every admissible assignment function @xmath20 on @xmath12 . furthermore , if we choose our basis such that @xmath85 and @xmath86 , where @xmath87 , then @xmath77 can only be one of the following two vectors : @xmath88 , where @xmath89 , @xmath90 and @xmath91 . without loss of generality , we assume the @xmath92 and choose a basis so that @xmath93 and @xmath94 where @xmath87 and @xmath95 . note that , since @xmath96 and thus @xmath97 we have @xmath98 if we let @xmath99 we thus have @xmath100 we can then set @xmath101 , making @xmath102 a unit vector such that @xmath103 . let @xmath104 , @xmath105 and note that @xmath106 also . thus , if we let @xmath107 and @xmath108 , then @xmath109 , @xmath110 and @xmath111 are all orthonormal bases for @xmath68 and thus @xmath112 , @xmath113 and @xmath114 are all contexts in @xmath115 . this construction is illustrated in fig . [ fig : reduction1 ] . if @xmath20 is an admissible assignment function on @xmath12 with @xmath43 then we must have @xmath116 and hence @xmath45 , as required . . once again , the circles and squares represent observables that have the values @xmath117 and @xmath118 respectively.,title="fig : " ] 01 [ scale=0.6 ] = [ line width=1pt ] = [ circle , minimum size=6 ] = [ rectangle , minimum size=9 ] = [ draw = none , circle , minimum size=4 ] ( 5,0 ) coordinate[s1,fill , label=0:@xmath36 ( b ) ( 10,4 ) coordinate[c1,fill , pos=0.5 ] ( c3 ) coordinate[l1,pos=0.7,label=0:@xmath119 coordinate[c1,fill , label=0:@xmath120 ( beta ) ( 0,4 ) coordinate[s1,fill , pos=0.5,label=90:@xmath39 ( c ) coordinate[l1,pos=0.35,label=25:@xmath52 ] coordinate[c1,fill , label=180:@xmath122 ( alpha ) ( 5,8 ) coordinate[c1,fill , pos=0.5 ] ( c2 ) coordinate[l1,pos=0.3,label=180:@xmath51 ] coordinate[s1,fill , label=0:@xmath41 ( a ) ; we now present a proof for the reduction in the opposite direction : finding ( from @xmath124 ) two vectors @xmath125 specifying observables @xmath126 for which @xmath127 , and which are further apart from each other than @xmath78 is from @xmath79 . this is made easier by noting that it is not necessary to find a vector @xmath77 ` further ' from @xmath78 than @xmath79 , but rather just two vectors further from each other than @xmath78 is from @xmath79 . this process is broken into two steps . we first prove an ` expansion lemma ' which , unlike the contraction lemma , does not find two vectors arbitrarily far apart satisfying the required criteria . rather , we then show a further lemma , the ` iteration lemma ' , proving that this expansion can be iterated to meet the required conditions . [ lemma : reduction2 ] given any two unit vectors @xmath70 with @xmath128 , we can effectively find unit vectors @xmath125 with @xmath129 and a finite set of one - dimensional projection observables @xmath12 containing @xmath130 such that if @xmath43 , then @xmath131 , for every admissible assignment function @xmath20 on @xmath12 . let @xmath132 . without loss of generality , we will consider only the positive , real case of @xmath133 . we fix an orthonormal basis such that , written in this basis , @xmath78 and @xmath79 lie in the @xmath134-plane bisected by the @xmath135-axis . in this basis we thus have @xmath136 where @xmath137 it is readily confirmed that @xmath138 as desired . note that we thus have @xmath139 figure [ fig : lemma2sphere ] shows the contour representing all the possible vectors specifying observables which can be forced to take the value 1 from the construction in lemma [ lemma : reduction1 ] . we use two applications of lemma [ lemma : reduction1 ] applied to @xmath124 to give two such vectors @xmath125 lying in the @xmath140-plane . corresponding to the one - dimensional projection observables that lemma [ lemma : reduction1 ] can force to take the value 1 . the bold ( red ; colour online ) curve represents the position on the unit sphere of such vectors for given @xmath124 . note the @xmath77 and @xmath141 are further apart from each other than @xmath78 and @xmath79 . ] we can also see , at least for the chosen values of @xmath124 that are shown in fig . [ fig : lemma2sphere ] , that @xmath142 . indeed it appears that the vectors ` @xmath77 ' , ` @xmath141 ' shown in the @xmath140-plane provide the maximum separation , and the symmetry under exchange of @xmath78 and @xmath79 of lemma [ lemma : reduction1 ] seems to support this . however , it is not necessary to prove this is the case . rather , we will show directly that the vectors @xmath125 provide the required expansion . to do so , we derive a simple explicit form for @xmath125 and thus @xmath143 . we focus first on finding @xmath77 ; the form of @xmath141 follows immediately . rather than use basis - transformations to attempt to apply lemma [ lemma : reduction1 ] to find the form of @xmath125 in this specific case , we will re - derive the result explicitly making use of our symmetrised basis choice . the vectors @xmath144 need to follow the orthogonality relations shown in fig . [ fig : reduction1 ] in order to conclude that @xmath145 . that is , we need vectors @xmath146 such that @xmath147 is an orthonormal set , and further that @xmath148 . since we choose @xmath77 to be in the @xmath140-plane , we can write it in the parameterised form @xmath149 , where @xmath150 remains to be found . since @xmath151 should be orthogonal to both @xmath78 and @xmath77 , we have @xmath152 similarly , we have @xmath153 further , the orthogonality of @xmath151 and @xmath154 gives us @xmath155 and hence @xmath156 thus , @xmath157 further , it is readily verified that @xmath158 for @xmath159 and hence for all @xmath160 ( recall eqn . [ eq : betabound ] ) . similarly , we find @xmath161 using a further two auxiliary vectors @xmath162 forming the orthonormal set @xmath163 where @xmath164 . thus , if we take @xmath165 , as a result of the orthogonality relationships expressed in fig . [ fig : reduction1 ] , @xmath127 for any admissible @xmath20 on @xmath12 with @xmath43 . it remains then just to show that @xmath166 we note that @xmath167 for @xmath168 . we finish the proof by showing proving eqn . [ eqn : gleb ] , that is , that @xmath169 , or , equivalently , @xmath170 . but since we can write @xmath171 we have from eqn . [ eqn : gamma ] @xmath172 concluding the proof . we note for completeness that we can write @xmath143 directly in terms of @xmath173 from eqns . [ eqn : beta ] , [ eqn : gamma ] and [ eqn : gleb ] as @xmath174 we now prove that by iterating this procedure we can find a pair of vectors arbitrarily far apart from each other . [ iteration lemma ] [ lemma : reduction3 ] given any two unit vectors @xmath70 with @xmath175 , we can effectively find unit vectors @xmath125 with @xmath176 and a finite set of one - dimensional projection observables @xmath12 containing @xmath177 such that if @xmath43 , then @xmath131 , for every admissible assignment function @xmath20 on @xmath12 . we prove by iterating lemma [ lemma : reduction2 ] , and use the notation @xmath178 and @xmath179 , indicating the 0th iteration . we start with @xmath180 and for each @xmath181 , as long as @xmath182 satisfy @xmath183 , apply the construction used in the proof of lemma [ lemma : reduction2 ] to generate @xmath184 for the next iteration . in particular , @xmath184 satisfy the equality for @xmath185 ( in particular , @xmath186 ) . by lemma [ lemma : reduction2 ] , we know that @xmath187 for each iteration @xmath188 . we now prove that the process can not produce an infinite sequence @xmath189 , with @xmath190 for all @xmath188 , that is , for some @xmath188 we have @xmath191 . ( the sequence must stop here , since lemma [ lemma : reduction2 ] can not be applied for @xmath192 . ) from eqn . [ eqn : iprodfn ] we define the function @xmath193 such that @xmath194 giving the inner product of the next pair in the iteration . we thus have @xmath195 and , more generally , @xmath196 . we can thus rephrase the problem : _ does there exist a @xmath197 such that @xmath198 ? _ let us , for the sake of contradiction , assume the contrary . then @xmath199 is an infinite strictly decreasing sequence of reals with @xmath200 for all @xmath188 . for any finite @xmath188 we thus have @xmath201 let us define the function @xmath202 such that @xmath203 so that @xmath204 we can show that @xmath205 for @xmath206 : calculating the derivative we have @xmath207 since @xmath208 is thus a strictly decreasing function on @xmath209 and @xmath210 for all @xmath211 , we have @xmath212 for all @xmath211 . hence we set @xmath213 since @xmath214 is a positive constant , it is not possible that @xmath215 , for all @xmath216 , because in this case we would have @xmath217 , for all @xmath216 , a contradiction . in fact , if @xmath197 is the smallest positive integer greater than @xmath218 , then @xmath219 , as required . we note that @xmath220 is not defined . by lemma [ lemma : reduction2 ] , for each @xmath221 there exists a set @xmath222 of one - dimensional projection observables such that @xmath223 under any @xmath20 admissible on @xmath222 satisfying @xmath224 . hence , if we take the set @xmath225 we must have @xmath226 under any admissible @xmath20 on @xmath12 satisfying @xmath43 , and @xmath227 , as required . with these lemmata proved , we are in a position to combine them to prove theorem [ thm : main ] . if we have @xmath63 then , by lemma [ lemma : explicitcase ] , there exists a finite set @xmath12 of one - dimensional projection observables for which there is no admissible @xmath20 on @xmath12 satisfying the requirements , so we are done . otherwise , we proceed directly to prove that if @xmath12 is a set of one - dimensional projection observables containing @xmath228 then no admissible assignment function @xmath20 on @xmath12 with @xmath29 can have @xmath3 value definite . we show this in two cases : first that @xmath229 and then that @xmath230 . let us first show that there is a set @xmath231 for which @xmath229 if @xmath20 is admissible on @xmath231 . there are two cases : either @xmath232 or @xmath233 . if @xmath232 , then by lemma [ lemma : reduction1 ] there exists a vector @xmath234 such that @xmath235 and a set @xmath236 of observables containing @xmath237 such that if @xmath20 is admissible on @xmath236 , @xmath238 also . but , by lemma [ lemma : explicitcase ] , there exists a set @xmath239 of one - dimensional projection observables containing @xmath240 such that if @xmath20 is admissible on @xmath239 and @xmath29 , @xmath241 must be value indefinite . thus , if we take @xmath242 we can not have @xmath243 as required . if @xmath233 , then by lemma [ lemma : reduction3 ] there exist two vectors @xmath244 such that @xmath245 and a set @xmath246 of observables containing @xmath247 such that if @xmath20 is admissible on @xmath246 then @xmath248 also . but , by lemma [ lemma : reduction1 ] , there exists a vector @xmath249 such that @xmath250 and a set @xmath251 of observables containing @xmath252 such that if @xmath20 is admissible , @xmath253 also . finally , once more by lemma [ lemma : explicitcase ] , there exists a set @xmath254 for which @xmath20 there is no admissible @xmath20 on @xmath251 satisfying @xmath255 . hence , there is no admissible @xmath20 on the set @xmath256 such that @xmath243 as required . this shows that there exists a set @xmath231 of one - dimensional projection observables containing @xmath228 such that we can not have @xmath243 if @xmath29 if @xmath20 is admissible @xmath231 . it remains to show that there exists a set @xmath257 such that we can not have @xmath258 if @xmath20 is admissible on @xmath257 . let us assume , without loss of generality , that @xmath259 and @xmath260 where @xmath261 . then let @xmath262 , @xmath263 and @xmath264 . then @xmath265 and @xmath266 are orthonormal bases for @xmath67 and hence @xmath267 and @xmath268 are contexts in @xmath269 . but if @xmath20 is admissible on @xmath270 and @xmath29 , @xmath258 , admissibility implies that @xmath243 . as we have shown just before , there exists a set @xmath271 of one - dimensional projection observables containing @xmath240 such that there is no admissible assignment function @xmath20 on @xmath271 with @xmath272 , and hence there is no admissible @xmath20 on @xmath273 such that @xmath29 and @xmath258 . having covered all cases , we are forced to conclude that there is a set @xmath274 of observables containing @xmath2 and @xmath3 such that if @xmath29 , @xmath3 can not be value definite if @xmath20 is admissible on @xmath12 . the important difference between theorem [ thm : main ] and the kochen - specker theorem lies in what physical conclusions can be drawn from the theorems which , of course , are purely mathematical results . key to interpreting such theorems is the recognition that a value assignment represents a possible hidden variable assignment for a quantum system , and that the value assigned to an observable thus represents the result that would be obtained upon its measurement . under this interpretation the kochen - specker theorem shows that , given a system prepared in the quantum state @xmath0 in dimension 3 or higher hilbert space , the results of all possible measurements on the state @xmath0 can not be predetermined ( noncontextually ) as they would in a classical theory . it says nothing , however , about whether all , or simply a few , outcomes are not predetermined . on the other hand , theorem [ thm : main ] implies that no one - dimensional projection observable @xmath21 can have a predetermined measurement outcome for the system unless @xmath0 is an eigenstate of @xmath21 . this interpretation relies on the eigenstate assumption discussed earlier in the paper , stating that the observable @xmath2 has a predetermined measurement outcome a very weak assumption . conceptually , this means that theorem [ thm : main ] goes further than the kochen - specker theorem in showing the _ extent _ of non - classicality that the quantum logic event - structure implies . it is possible to generalise this result that formally applies only to one - dimensional projection observables to the value - indefiniteness of more general classes of observables . since an observable @xmath275 ( formally a hermitian operator in @xmath14-dimensional hilbert space ) with a non - degenerate spectrum , distinct eigenvalues @xmath276 and eigenstates @xmath277 can be expressed as its spectral decomposition @xmath278 ( where @xmath279 , as usual ) , it physically has a predetermined measurement outcome if and only if all the projectors @xmath280 , @xmath281 , have predetermined measurement outcomes has the predetermined value @xmath118 then one must obtain @xmath282 upon measurement of @xmath275 ; admissibility then requires that all @xmath283 have the definite value 0 for @xmath284 . ] that is , are value definite . thus , for a system prepared in a state @xmath0 in dimension 3 or higher hilbert space , the outcome of a measurement of an observable @xmath275 with non - degenerate spectra can not be predetermined ( noncontextually ) unless @xmath0 is an eigenstate of @xmath275 . since the first appearance of the kochen - specker theorem @xcite , much attention has been given to reducing the number of observables and contexts needed to obtain a contradiction and prove the theorem . the original result used a set of 117 observables , but more recent results have , to quote some notable examples , shown sets containing 31 observables ( in three - dimensional hilbert space ) @xcite and 18 observables ( in four - dimensional hilbert space ) @xcite . while such results do not affect the interpretation of the theorem , they have merit in showing the depth of the contradiction between the classical and quantum logical structures . more recently , smaller proofs have been of particular interest since these have been used to derive noncontextuality inequalities that can be experimentally tested @xcite in the same vein as bell - inequalities @xcite ; smaller sets of observables lead to smaller and more readily testable inequalities . conceptually , however , the key point is probably that the theorem can be proved using a _ finite _ set of observables ; if a contradiction only arose when an infinity of observables were considered , this would potentially raise questions about the constructive and operational character of the theorem and its use of counterfactuals , hence its interpretation would be more questionable @xcite . the localised nature of theorem [ thm : main ] immediately means that a single finite set @xmath12 of one - dimensional projection observables will never suffice to prove the value indefiniteness of all such projection observables @xmath3 not commuting with @xmath2 for a given state @xmath0 . there are infinitely many such observables , and one must , by definition , include @xmath3 in @xmath12 to localise value indefiniteness to @xmath3 . rather , the nature of theorem [ thm : main ] means we must look for constructive methods to obtain a set @xmath285 for a given @xmath3 , which is precisely what we have done in our proof of the result . of course , a given set of orthogonality relations ( i.e. , a greechie diagram ) may be realisable for an infinity of different such sets @xmath12 , as is the case with the diagram depicted in fig . [ fig : reduction1 ] . thus , it would be preferable to find a given set of orthogonality relations for which a set @xmath285 of observables realising these relations and containing both @xmath2 and @xmath3 for any @xmath3 . since we were unable to give such a set of relations , we had to iterate lemma [ lemma : reduction2 ] a number of times times depending on @xmath2 , with no upper bound ( but only ever finitely many times ) . furthermore , it seems that it is difficult , if not impossible , to succeed in giving a fixed set of orthogonality relations that works in all cases . in order to show an observable @xmath49 has @xmath46 using the admissibility requirements , one must give a context @xmath286 for which it is already known that @xmath287 . this implies two observable @xmath54 and @xmath55 such that @xmath288 and @xmath289 . but this is precisely the case described in lemma [ lemma : reduction1 ] . however , in lemma [ lemma : reduction2 ] we showed the limitations of this process in ` widening the angle ' between vectors whose corresponding projectors both take the value 1 hence the necessity of iterating lemma [ lemma : reduction2 ] . as a result it seems that , in contrast to the kochen - specker theorem , arbitrarily large ( but always finite ) sets of observables are needed to show that a given observable @xmath3 is value indefinite nonetheless , the critical point is once again that for _ any given _ @xmath3 , we can show that @xmath3 is value indefinite with a _ finite _ set of observables , and hence that the counterfactual reasoning used is no more problematic than in the kochen - specker theorem . one of the strengths of the kochen - specker theorem that has been repeatedly emphasised is the fact that the contradiction between its hypotheses is derived independently of the state a quantum system is prepared in ; this is commonly referred to as state - independence . this is in contrast to violation of bell - type inequalities ( which occur only for particular entangled states ) and shows that the non - classicality results from the structure of quantum mechanics itself , rather than features of particular states , such as entanglement @xcite . consequently , various experimental inequalities based on the kochen - specker theorem that , although often simpler , are state - dependent have been criticised , and much effort has been expended to find simple , state - independent inequalities to test @xcite . in contrast to the kochen - specker theorem , the form of theorem [ thm : main ] and , in particular , the interpretation ( relying , of course , on the eigenstate assumption ) that _ for a system prepared in a given state @xmath0 _ , any one - dimensional projection observable @xmath3 not commuting with @xmath2 is value indefinite , may suggest that theorem [ thm : main ] does not share this state - independence . as a result , this issue deserves a little discussion . the state - independence of the kochen - specker theorem ensures that no quantum state in @xmath24 dimensional hilbert space admits a classical assignment of definite values to all observables within certain finite sets . this is true also with theorem [ thm : main ] : for any quantum state @xmath0 , all projection observables not contained within the ` star ' of one - dimensional projection observables commuting with @xmath2 ( see fig . [ fig : star ] ) are value indefinite . with @xmath29 and the ( infinite ) set of compatible observables @xmath3 for which @xmath258 . this is the maximal extent of value definiteness for a system in state @xmath0 no other one - dimensional projection observables on @xmath67 can be value definite.,title="fig : " ] 01 [ scale=0.8 ] = [ line width=1pt ] = [ draw , line width=1pt , inner sep=0 ] = [ rectangle , minimum size=6 ] = [ circle , draw = none , fill , minimum size=2 ] = [ draw = none , circle , minimum size=45 ] ( 0:0 ) ( 135:3 ) coordinate[c1,minimum size=14,fill , at start ] ( 0 ) coordinate[c1,circle , midway , fill ] ( 1 ) coordinate[c1,circle , at end , fill , label=35:@xmath51 ] ( 2 ) ; ( 0.center ) ( 90:3 ) coordinate[c1,at start ] ( 0 ) coordinate[c1,circle , midway , fill ] ( 3 ) coordinate[c1,circle , at end , fill , label=350:@xmath52 ] ( 4 ) ; ( 0.center ) ( 45:3 ) coordinate[c1,at start ] ( 0 ) coordinate[c1,circle , midway , fill ] ( 5 ) coordinate[c1,circle , at end , fill , label=305:@xmath119 ( 6 ) ; ( 0.center ) ( 315:3 ) coordinate[c1,at start ] ( 0 ) coordinate[c1,circle , midway , fill ] ( 9 ) coordinate[c1,circle , at end , fill , label=215:@xmath290 ( 10 ) ; ( 0.center ) ( 270:3 ) coordinate[c1,at start ] ( 0 ) coordinate[c1,circle , midway , fill ] ( 11 ) coordinate[c1,circle , at end , fill , label=170:@xmath291 ( 12 ) ; ( 0.center ) ( 225:3 ) coordinate[c1,at start ] ( 0 ) coordinate[c1,circle , midway , fill ] ( 13 ) coordinate[c1,circle , at end , fill , label=125:@xmath292 ( 14 ) ; ( 0.center ) ( 0:3 ) coordinate[c1,fill , at start ] ( 0 ) coordinate[c1,circle , fill , midway , fill ] ( 7 ) coordinate[c1,circle,,fill , at end , label=260:@xmath293 ( 8) ; \(0 ) at ( 0.center ) ; ( . ) at ( 190:2 ) ; ( . ) at ( 180:2 ) ; ( . ) at ( 170:2 ) ; rather , it is not theorem [ thm : main ] that is state - dependent , but the proof we have given : to show that a given observable @xmath3 is value indefinite from the assumption that @xmath29 , we need a set @xmath12 particular to this @xmath294 . however , as we discussed in the preceding section , this is perfectly reasonable given the form of the theorem . one can emphasise further the state - independence of theorem [ thm : main ] by restating the theorem in the following form : _ `` only a single one - dimensional projection observable on the hilbert space @xmath5 for @xmath24 can be assigned the value 1 by an admissible , noncontextual value assignment function''_. in this form the state - independence is clear ; the illusion of state - dependence enters because of the connection , via the eigenstate assumption , between the `` one observable assigned the value 1 '' and the particular state @xmath0 ( and corresponding observable @xmath2 with @xmath29 ) which is necessary for the physical interpretation of the theorem . the importance of the state - independence of the kochen - specker theorem arises , in part , in the use of kochen - specker sets of observables in testable inequalities . it is important to note that , even though these inequalities are sometimes referred to as `` kochen - specker inequalities '' @xcite , they are better seen simply as noncontextuality inequalities . these inequalities are derived under the assumption only of noncontextuality , ignoring the admissibility requirements , and bounds on quantities are calculated over all possible noncontextual value assignments . a key result shows that one can derive such an inequality from any kochen - specker set @xcite . it is clear that these value assignments can not obey the admissibility requirements , since the kochen - specker theorem shows precisely that no classical value assignment can do so . the strength of theorem [ thm : main ] , on the other hand , relies precisely on the use of the admissibility requirements to determine when definite values should be assigned . hence , while one can use the methods of @xcite to derive inequalities from the constructions in the proof of theorem [ thm : main ] , these bounds would be calculated over all noncontextual value assignments ( subject to @xmath29 ) , without paying heed to admissibility , and hence would offer no conceptual advantage over existing inequalities . furthermore , since our construction in lemma [ lemma : explicitcase ] , for example , contains 37 observables , these would pose no experimental benefit to existing , simpler inequalities either @xcite . nonetheless , the state - independence of the result shows that the value indefiniteness of almost all one - dimensional projection observables in quantum mechanics is indeed a deep feature of the theory of the logical structure of hilbert space rather than a property of particular states . in summary , we proved a variant of the kochen - specker theorem showing that the non - classicality implied by the kochen - specker theorem is , in a specific sense , maximal . specifically , under the assumptions that ( 1 ) any value definite observables behave noncontextually , and ( 2 ) contexts obey weak ` admissibility ' rules on any value definite observables they contain , we show that only one one - dimensional projection observable on the hilbert space @xmath5 ( for @xmath24 ) can be assigned consistently the definite value 1 . if a quantum system is prepared in a state @xmath0 , subject to the assumption that any value assignment function for the system must assign the value 1 to @xmath2 since @xmath295 , the theorem can be interpreted as showing that the measurement of any observable @xmath3 projecting onto the linear subspace spanned by a state @xmath294 that is neither orthogonal nor co - aligned with @xmath0 must be value indefinite that is , indeterministic . this interpretation , which shows that almost all one - dimensional projection observables are value indefinite for a given system @xcite , is stronger than what can be drawn from the kochen - specker theorem , which , in contrast , shows only that not all observables can be value definite . this result justifies further the general belief that quantum mechanics is indeterministic that there is no hidden variable or definite value determining the outcome of a measurement in advance . this eliminates the need to assume that the non - classicality shown by the kochen - specker theorem should apply uniformly , instead deriving this global value indefiniteness . as with the kochen - specker theorem , this result relies on the assumption that classical values , should they exist , must behave noncontextually . finally , these results help theoretically certify quantum random number generators @xcite , since the promises of such devices rely on the indeterministic nature of quantum measurements @xcite . by localising value indefinite observables , one can be sure that the measurements producing the output bits do not yield any pre - existing element of physical reality . we emphasise that these results do not hold for two - dimensional systems a class into which many current quantum random number generators unfortunately fall . 22ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) link:\doibase 10.1512/iumj.1968.17.17004 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.65.803 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.49.91 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.101.210401 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physreva.86.062109 [ * * , ( ) ] link:\doibase 10.1103/physreva.89.032109 [ * * , ( ) ] _ _ , @noop ph.d . thesis , ( ) @noop _ _ ( , , ) link:\doibase 10.1512/iumj.1957.6.56050 [ * * , ( ) ] link:\doibase 10.1103/physreva.75.032334 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase physrevlett.109.150401 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( )